Aluminum fluoride-18 labeled folate enables in vivo detection of atherosclerotic plaque inflammation by positron emission tomography

Inflammation plays an important role in the development of atherosclerosis and its complications. Because the folate receptor β (FR-β) is selectively expressed on macrophages, an FR targeted imaging agent could be useful for assessment of atherosclerotic inflammation. We investigated aluminum fluoride-18-labeled 1,4,7-triazacyclononane-1,4,7-triacetic acid conjugated folate (18F-FOL) for the detection of atherosclerotic plaque inflammation. We studied atherosclerotic plaques in mice, rabbits, and human tissue samples using 18F-FOL positron emission tomography/computed tomography (PET/CT). Compound 2-deoxy-2-[18F]fluoro-D-glucose (18F-FDG) was used as a comparison. Firstly, we found that the in vitro binding of 18F-FOL co-localized with FR-β-positive macrophages in carotid endarterectomy samples from patients with recent ischemic symptoms. We then demonstrated specific accumulation of intravenously administered 18F-FOL in atherosclerotic plaques in mice and rabbits using PET/CT. We noticed that the 18F-FOL uptake correlated with the density of macrophages in plaques and provided a target-to-background ratio as high as 18F-FDG, but with considerably lower myocardial uptake. Thus, 18F-FOL PET/CT targeting of FR-β-positive macrophages presents a promising new tool for the in vivo imaging of atherosclerotic inflammation.

sient ischemic attack and severe (≥80%) stenosis of the carotid artery, incubated cryosections with 18 F-FOL in vitro, and then detected its binding with high-resolution digital autoradiography. Following this, we immunohistochemically stained the same sections and observed co-localization of 18 F-FOL binding, and macrophages positive to FR-β and cluster of differentiation 68 (CD68; Fig. 2A). The quantitative in vitro autoradiography analysis revealed significantly lower 18 F-FOL binding (88% ± 8.9% reduction) in sections incubated with folate glucosamine (blocked binding) than with the total 18 F-FOL binding without blocking (Fig. 2B), thus indicating that tracer binding in lesions was specific to FR.
In addition, we cultured the purified monocytes from healthy human blood under M1-polarising (interferon γ [IFN-γ] and lipopolysaccharides [LPS]) or M2-polarising (macrophage colony-stimulating factor [M-CSF], interleukin 4  and IL-10) conditions and evaluated the binding of 18 F-FOL to macrophages during in vitro differentiation. We found that 18 F-FOL binds to both M1 and M2 macrophages although clearly more to M2-polarized macrophages (Fig. S1).

Intravenously administered 18 F-FOL accumulates in inflamed atherosclerotic lesions in mice.
We used 6-month-old, high-fat fed, low-density lipoprotein receptor deficient, male mice, which expressed only apolipoprotein B100 (LDLR −/− ApoB 100/100 ), as an atherosclerosis model. Regular chow-fed 9-month-old C57BL/6N male mice were used as healthy controls (Table 1). We evaluated the biodistribution and kinetics of i.v. administered 18 F-FOL in atherosclerotic and healthy control mice using PET/CT imaging, gamma counting, and digital autoradiography. A blocking study with co-injection of a 100-fold excess of folate glucosamine was performed to clarify tracer uptake specificity.
We performed in vivo 18 F-FOL PET/CT, which showed that the highest maximum TBR (maximum standardized uptake value (SUV) of aortic arch divided by mean SUV of blood measured from the vena cava; SUV max, aortic arch / SUV mean, blood ) of 1.5 ± 0.34 occurred in atherosclerotic mice 60-90 min after 18 F-FOL injection. For comparison, the maximum TBR in aortic arch of healthy control mice was significantly lower at 0.71 ± 0.18 (P = 0.00019; Figs 3A and S2). A comparison revealed that the myocardial uptake of 18 F-FDG was significantly higher than that of 18 F-FOL (SUV max,60-90 min 12 ± 4.3 vs. 0.36 ± 0.069, P < 0.0001, respectively), which hampered the visualization and quantification of 18 F-FDG uptake in mice aortic arches (Fig. 3A). Blood radioactivity concentrations High-fat diet, months 4 ND ND Female/male animals, no. 0/12 0/9 1/3 Weight, g* 37 ± 5 38 ± 6 3200 ± 800 In vivo blocking study, no.
Ex vivo PET imaging of excised aortas 120 minutes after 18 F-FOL injection followed by en face Oil-Red-O staining revealed the highest tracer uptake in the most prominent atherosclerotic lesions, i.e., in the aortic arch of the LDLR −/− ApoB 100/100 mice (Fig. S3).
In the blocking study, i.v. co-injection of 18 F-FOL and 100-fold excess of folate glucosamine reduced average plaque uptake by 92%, as determined by ex vivo autoradiography of aorta cryosections. When blocking was performed in aorta cryosections in vitro, it reduced total binding of 18 F-FOL by 89% ± 3.5%.
According to modified Movat's pentachrome and anti-Mac-3 staining, the aortas of the LDLR −/− ApoB 100/100 mice contained large and inflamed atherosclerotic plaques, which were mostly of the fibroatheroma-type, with a well-defined fibrous cap. The intima-to-media ratio was 3.7 ± 2.1, and the lesions were abundantly infiltrated with Mac-3-positive macrophages (13% ± 4.6% of the neointimal area), which were situated mostly in the fibrous cap (Fig. 3D). The control mice showed no signs of atherosclerosis (Fig. 3E). The atherosclerotic plaques showed positivity for anti-inducible nitric oxide synthase (iNOS) and anti-mannose receptor C-type 1 (MRC-1), indicating M1 and M2 polarized macrophages, respectively (Fig. S4).

In vivo [ 18 F]FOL PET/CT detects atherosclerotic lesions in rabbit aorta.
We next used Watanabe heritable hyperlipidemic rabbits, with subdiaphragmatic denudation in the abdominal aorta, as another atherosclerosis model. The rabbits were 45-months-old and fed on regular chow, and three males and one female were used.
We performed a direct comparison of 18 F-FOL and 18 F-FDG in vivo PET/CT imaging of atherosclerotic aorta in Watanabe rabbits. Our results revealed a rapid clearance of 18 F-FOL from circulating blood, and a comparable maximum TBR (SUV max, aorta /SUV mean, blood ) in the atherosclerotic aorta to that of 18 F-FDG (Fig. 5). There was a high association between the maximum TBR of 18 F-FOL PET and the areal percentage of RAM-11-positive staining of macrophages (linear mixed model P = 0.0031) in the corresponding segment of the aorta (Fig. 6A). Furthermore, there was an association between maximum TBR obtained with 18 F-FOL and 18 F-FDG PET (linear mixed model P = 0.0088) in the same segments of the aorta (Fig. 6B). The average uptake of 18 F-FOL (2.6) in the whole aorta was similar to that of 18 F-FDG (1.9; P = 0.074; Table 3).
Ex vivo gamma counts of excised tissue samples were in accordance with the in vivo 18 F-FOL PET imaging results. Compared with atherosclerotic mice, our 45-month-old rabbits showed the highest 18 F-FOL uptake in the kidneys, spleen and liver ( Table 2).
According to hematoxylin-eosin (H&E) and anti-RAM-11 antibody (macrophages) staining, the aortic segments within the denudated part contained large fibroatheroma-type atherosclerotic lesions. The non-denudated aorta showed some small fibroatheroma-type atherosclerotic lesions in some segments, whereas other segments showed only mild intimal thickening. The infiltration of RAM-11-positive macrophages was widespread in the fibroatheroma-type plaques, and was observed within both the core and fibrous cap. Only limited RAM-11-positive staining was observed in the areas of intimal thickening (Fig. 7B).

18
F-FOL is a novel tracer for in vivo PET imaging of atherosclerotic plaques. Although expression of FR is not an exclusive feature of macrophages, the features of easy radiolabeling, rapid clearance from blood circulation and FR-negative tissues, low uptake in myocardium, and association of high uptake with macrophages in advanced atherosclerotic lesions make 18 F-FOL a potential tool for imaging of vascular inflammation in atherosclerosis.
Several FR targeting radiotracers have been developed. These have seen slight alterations to their chemistry, with the aim of improving their selectivity, bioavailability, and pharmacokinetic properties [7][8][9][10][11] . Although these other folate-based tracers have accumulated in FR-positive tissues and cells, some non-specific binding to FR-negative tissues has been noticed. For example, 18 F-fluorophenylfolate has shown high uptake in the intestine (even higher than in kidneys), and in a rodent inflammatory paw model, significant accumulation also occurred in the stomach, bone, liver, and heart 14 . Currently, two folate-based radioconjugates, 111 In-labeled diethylenetriaminepentaacetic acid (DTPA)-conjugated folate and 99m Tc-EC20, have already been tested for in vivo SPECT  imaging in patients with different cancers 15,16 . Both of the folate agents appeared to be safe and well tolerated, and both exhibited rapid target-tissue uptake and non-target-tissue clearance with urinary excretion. Uptake of 111 In/ 99m Tc-folates in the liver, lungs, and myocardium was low, which is advantageous for imaging of atherosclerosis. In general, PET imaging can offer several advantages over SPECT imaging, including higher sensitivity, better quantitation, and sharper spatial resolution in a clinical setting 17 . These are important for imaging of small targets, such as atherosclerotic lesions.
Previously, two folate-based SPECT imaging agents, 99m Tc-EC20 and 111 In-EC0800, have shown promising results for in vivo detection of atherosclerotic lesions in mice 8,9 . Additionally, Jager and co-workers reported that in tissue sections of atherosclerotic human carotid arteries, fluorescein isothiocyanate-conjugated folate (folate-FITC) targeted FR-β-positive macrophages, and was associated with hypoxia in human atherosclerotic lesions 10 . Later, Müller and co-workers showed that an 18 F-labeled folate derivative, 3′-aza-2′-18 F-fluorofolic acid, also detects FR-β-positive macrophages in human atherosclerotic plaques in vitro 11 . However, the in vivo imaging of atherosclerotic plaque inflammation with a folate-based PET tracer and a comparison with 18 F-FDG have not been reported, so we therefore performed a direct comparison in both mice and rabbit models. For this part of the study, we used two models of atherosclerosis: (1) hypercholesterolemic LDLR −/− ApoB 100/100 mice that spontaneously develop atherosclerotic lesions in the aorta, which can be accelerated by a high-fat diet 18,19 , and (2) surgical injury accelerated atherosclerosis in Watanabe heritable hyperlipidemic rabbits 20 . In comparison with 18 F-FDG, our new 18 F-FOL tracer has lower myocardial uptake and is cleared more rapidly from circulating blood. Recent studies using 18 F-NaF as a marker of microcalcification in atherosclerosis 21 and a somatostatin receptor-binding PET tracer 68 Ga-DOTATATE as a marker of macrophages 22 , have shown that tracers with low retention in the myocardium can provide superior imaging in the coronary arteries as compared with 18    F-fluorocholine 23,24 and 11 C-PK11195 25 have been tested for imaging atherosclerotic inflammation in the carotid arteries as alternatives for 18 F-FDG with potentially less signal from the myocardium. In atherosclerotic mice, the average plaque-to-healthy vessel wall ratio of 18 F-FOL was 2.6 ± 0.58, which is higher than the ratios of 2.1 ± 0.16 and 2.1 ± 0.5, which we recently reported for 18 F-FDG and 68 Ga-DOTATATE, respectively 19,24,26 . The high TBR of 18 F-FOL was due to both the higher target uptake and the lower blood radioactivity in comparison with 18 F-FDG. As anticipated, we observed the highest radioactivity of 18 F-FOL in kidneys, which demonstrate the prominent expression of FR 27 . However, as the excretion of the tracer occurs mainly through the kidneys, it is not possible to distinguish the proportion of tracer specifically bound to kidney FR from that being eliminated by this organ.
Al 18 F-based labeling is one of the most amenable radiosynthesis techniques for clinical settings 28,29 . In this study, we successfully labeled folate with a [ 18 F]AlF-NOTA chelating system; however, the critical part was correct formulation of 18 F-FOL. We always observed a minor radioactive HPLC peak (retention time approximately 6 min) in addition to the end product 18 F-FOL (Fig. S5A). Under some conditions (e.g., at low pH, high radioactivity concentration), the proportion of the side product increased with time. This may be an indication of radioactivity-induced instability. In order to stabilize 18 F-FOL, we formulated the tracer in PBS (pH 7.4) in the presence of 8% ethanol and 4-7% polypropylene glycol, keeping the radioactivity concentration under 400 MBq/ mL at the end of synthesis. Under these optimized conditions, 18 F-FOL was stable for at least 4 hours at room temperature (a longer time was not tested). We chose to use polypropylene glycol as it is routinely used in radiopharmaceuticals in our hospital, and is very effective for prevention of radioactivity-induced tracer instability 30 .
A limitation of the current study was that we did not test how the signal detected with folate imaging changes when plaques stabilize, or the effects of statin treatment on the expression of FR on macrophages. Furthermore, the expression of FR and the binding of the radiotracer were not compared between unstable vs. stable carotid plaques. Neither did we provide co-or adjacent comparisons of 18 F-FOL uptake and FR immunostaining in healthy human aorta or in atherosclerotic plaques from the rabbit model, because we had limited access to normal human tissue and did not find rabbit recognizing anti-FR antibody with optimal immunohistochemical staining of cryosections. We observed higher aorta/blood ratio of 18 F-FOL in C57BL/6N control mice than in LDLR −/− ApoB 100/100 atherosclerotic mice, and higher 18 F-FOL radioactivity in the aorta of control mice than in LDLR −/− ApoB 100/100 -blocked mice. This may be due to non-specific binding, the presence of FR in non-atherosclerotic aorta or differences in tracer biokinetics between the mouse strains. We hypothesize that our C57BL/6N mice had low aortic expression of FR (FR-positive macrophages) sensitively detectable by the 18 F-FOL whereas pre-injection of high dose of folate glucosamine had almost completely occupied FRs and blocked 18 F-FOL binding in the atherosclerotic aorta. According to Müller and co-workers, the FR level in human normal aorta is very low, which supports clinical translation of 18 F-FOL 11 . It should also be highlighted that the 18 F-FOL PET imaging in rabbits used a 90-min dynamic acquisition, but the 18 F-FDG acquisition started 160 min post-injection. This difference may have generated differences in SUV values between 18 F-FOL and 18 F-FDG PET measurements.
In conclusion, we demonstrated the feasibility of folate-based PET imaging of inflamed atherosclerotic plaques in vivo, using the new imaging agent 18 F-FOL. Molecular imaging of disease activity in atherosclerosis is a potential approach to evaluate new therapies and even assess the risk of cardiovascular disease and its complications in at risk individuals 3 . Although mainly M2-like macrophage subtypes express FR-β in atherosclerotic lesions 12 , it is too early to say what pathophysiological processes 18 F-FOL uptake reflects in atherosclerosis. Given the potential translatability of the novel tracer for imaging patients, future studies are warranted to test whether the tracer is suitable for detecting active atherosclerosis in patients.

Methods
Production of 18 F-FOL and 19 F-FOL. We produced the 18 F-FOL alias [ 18 F]AlF-NOTA-folate according to known procedures 13,31 . Briefly, [ 18 F]-fluoride (3.2-3.4 GBq) in physiological saline (50 µL) was added to a reaction vessel containing sodium acetate buffer (pH 4.0, 1 M, 25 µL) and AlCl 3 (2 mM). The reaction mixture was kept at room temperature for 3 minutes. NOTA-folate (250 µg) in water (50 µL) and acetonitrile (125 µL) was then added, and the reaction vessel was heated at 100 °C for 15 min, followed by dilution of the reaction mixture with water containing 0.2% formic acid (1 mL). In the next step, we performed a high-performance liquid chromatography (HPLC) purification with a semi-preparative C18 column (Jupiter 250 × 10 mm, Phenomenex Inc., Torrance, CA, USA) with both UV (254 nm) and radioactivity detection. Solvent A was water containing 0.1% formic acid and solvent B was acetonitrile containing 0.1% formic acid. The elution was programmed as a gradient from 8% B to 21% B over 20 min, with a flow rate of 4 mL/min. A solution of potassium bicarbonate (KHCO 3 ) in water (1 M) was added to adjust the collected HPLC fraction to pH 5.5, and the solvents were evaporated off. The residue was then formulated in phosphate-buffered saline (PBS, 1-3 mL) containing 8% ethanol and 4-7% polypropylene glycol. The radioactivity concentration in the end product bottle was <400 MBq/mL at the end of synthesis. After sterile filtration (0.22 µm, Millipore) into the end product bottle, the tracer was ready for the PET studies. The total synthesis time was 80-95 min, commencing from the end of bombardment. We performed HPLC analysis of the end product on an analytical C18 column (Jupiter 250 × 4.6 mm, Phenomenex Inc., Torrance, CA, USA) with both UV (254 nm) and radioactivity detection (Fig. S5A,B). Solvent A was water containing 0.1% trifluoroacetic acid (TFA) and solvent B was acetonitrile containing 0.1% TFA. The elution was programmed as a gradient from 10% B to 25% B over 15 min, with a flow rate of 1 mL/min. To confirm the identity of 18 F-FOL, the corresponding "cold" counterpart 19 F-FOL was used as a reference (Fig. S5C). In addition, the identity of 18 F-FOL was further confirmed with the analyses of a sample spiked with 19 F-FOL (Fig. S5D,E). Under the same HPLC conditions, the retention time of precursor NOTA-folate (Fig. S5F) was clearly different from 18 F-FOL.
For preparation of 19 F-FOL, NOTA-folate (4.2 mg, 4.9 µmol), aluminum chloride hexahydrate (AlCl 3 ·6H 2 O; 5.3 mg, 20 µmol) and sodium fluoride (NaF; 3.7 mg, 90 µmol) were dissolved in 0.1 mL of 0.1 M sodium acetate pH 4.0, resulting in a clear solution. The mixture was heated in a heating block at 105 °C for 15 min, leading to a suspension, which was diluted to 0.3 mL with Milli-Q water and adjusted to pH 7. The white solid was removed by centrifugation and then filtration, affording a clear and yellow solution. The above mixture was submitted to HPLC for further purification: Sunfire C18 column (5 μm, 10 × 250 mm; Waters Corporation, Milford, MA, USA), A = 0.1% acetic acid in water, B = ethanol, column equilibration 8% B for 10 min, gradient 12% B to 40% B in 25 min, flow rate 2 mL/min. The fractions were combined and concentrated under vacuum. The resulting residue was dissolved in 1 mL Milli-Q water and dispensed into small vials, which served as the standard reference. For the high-resolution mass spectrometry (MS) analysis of 19 F-FOL, the samples were loaded into an offline nanospray emitter (Thermo Scientific, Waltham, MA, USA) and analyzed in positive electrospray ionization (ESI)-MS mode, using an orbitrap mass analyzer in the LTQ Orbitrap Velos Pro mass spectrometer (Thermo Scientific) equipped with a nano-electrospray ionization source. MS data were acquired using Xcalibur 3.0 software (Thermo Scientific). Before the analysis, the mass spectrometer was calibrated with an in-house mixed   Calmix LTQ ESI Positive Ion Calibration Solution. Data were analyzed using Xcalibur 4.0 software (Thermo Scientific; Fig. S6).
To evaluate 18 F-FOL stability in the injectable formulation, we kept the end product at room temperature and took samples for radio-HPLC analysis at time intervals of up to 4 hours. For the quality control of 18 F-FOL, we used the same HPLC protocols as described above.
To determine the lipophilicity of 18 F-FOL, approximately 8 kBq of 18 F-FOL in 500 μL PBS (pH 7.4) was added to 500 μL of 1-octanol in an Eppendorf tube. Mixtures were then vortexed for 3 min and centrifuged (12,000 × g for 6 min at room temperature). Following this, 100 μL aliquots of both PBS and 1-octanol layers were counted using a γ-counter (1480 Wizard 3"; Perkin Elmer/Wallac, Turku, Finland). We performed measurements in triplicate, and calculated the octanol-PBS distribution coefficient log D as log 10 (counts in octanol/counts in PBS).
To evaluate the in vivo stability and plasma protein binding of 18 F-FOL, blood samples from six atherosclerotic mice and four rabbits were drawn at 5-60 min post-injection of 18 F-FOL. Plasma was separated by centrifugation (2,100 × g for 4 min at 4 °C) and plasma proteins were separated by adding an equal volume of acetonitrile followed by centrifugation as above. The radioactivity concentration in the protein precipitate and supernatant was then measured (1480 Wizard 3″; Perkin Elmer/Wallac, Turku, Finland). The plasma supernatant was then filtered through a 0.45 µm Minispike filter (Waters Corporation, USA) for further analysis by HPLC. A semi-preparative C18 column (Jupiter 250 × 10 mm, Phenomenex Inc., Torrance, CA, USA) was used for HPLC analysis of mouse plasma samples with both UV (254 nm) and radioactivity detection. Solvent A was water containing 0.1% TFA and solvent B was acetonitrile containing 0.1% TFA. The elution was programmed as a gradient from 7% B to 22% B over 15 min, with a flow rate of 5 mL/min. Microscopic equipment. All photomicroscopic images were taken by scanning slides with a digital slide scanner (Pannoramic 250 Flash, 3DHistec Ltd., Budapest, Hungary), with brightfield illumination by a 6 W strobe light, 45x magnification by default, and resolutions of 26x, 41x, and 52x, in an acceptable slide format of 25 × 75 mm, 0.9-1.2 mm thickness. We used Pannoramic viewer software (version 1.15.2) to capture images.
Human carotid artery samples. We conducted this study according to the declaration of Helsinki, and the study protocol was approved by the ethics committee of the Hospital District of Southwest Finland. All study patients gave their written informed consent. We obtained carotid endarterectomy samples from five patients (four females and one male; aged 70 ± 10 years) with transient ischemic attack and severe (≥80%) stenosis of the Experimental animals. Male low-density lipoprotein receptor deficient mice expressing only apolipoprotein B100 were used (LDLR −/− ApoB 100/100 , strain #003000, The Jackson Laboratory, Bar Harbor, ME, USA). Prior to imaging experiments, a Western-type diet (0.2% total cholesterol, TD 88137, Harlan Teklad, Harlan Laboratories, Madison, WI, USA) was started at the age of 2 months and maintained for 4 months to induce development of inflamed atherosclerotic lesions as previously described 19 . Eight-month-old C57BL/6N mice fed with regular chow were used as healthy controls.
We also used adult Watanabe heritable hyperlipidemic rabbits with a mutated LDLR gene resulting in high circulating cholesterol levels. When the rabbits were at the age of 6 months, a trained professional (J.H.) performed balloon endothelial denudation of the abdominal aorta 20 . Animals were anesthetized with a mixture of ketamine (15 mg/kg, Ketalar, Pfizer, New York, NY, USA) and medetomidine (0.3 mg/kg, Domitor, Orion Pharma, Espoo, Finland). A 3F Fogarty embolectomy balloon catheter (120403F, Edwards Lifesciences, Irvine, CA, USA) was inserted via the right femoral artery and advanced proximally for 20 cm up to the lower thoracic aorta. The catheter was inflated with 0.6 mL of air and pulled down to the level of the aortic bifurcation for endothelial denudation, with the procedure being repeated three times. After removal of the catheter, the femoral artery was ligated and the wound was closed in layers with resorbable suture. Rabbits were returned to their cages after recovery from anesthesia. Rabbits were maintained on a normal diet for the whole study and used for imaging studies at the age of 45 months. More detailed characteristics of the study animals are given in Table 1.
All animals were housed in an animal facility throughout the study, under standard conditions with lights on from 6.00 a.m. to 6.00 p.m., and access to water and food ad libitum. All animal experiments were approved by the national Animal Experiment Board in Finland, and the Regional State Administrative Agency for Southern or Eastern Finland, and were carried out in compliance with the European Union directive.

In vitro 18 F-FOL studies.
To determine in vitro binding of 18 F-FOL in atherosclerotic plaques, we used human (24 × 7 µm carotid cryosections from five patients), mouse (30 × 8 µm aorta cryosections from four atherosclerotic mice), and rabbit (46 × 20 µm aorta plaque cryosections from four Watanabe rabbits) samples. Firstly, the tissue cryosections were pre-incubated in PBS (pH 7.4) at room temperature for 15 min, and then with 0.2 nM 18 F-FOL in PBS with or without a 100-fold molar excess of folate glucosamine (C 25 H 30 N 8 O 10 , molecular weight 602.56). After incubation, slides were washed twice with ice-cold PBS for 1 min, dipped in ice-cold distilled water, briefly dried with a hair dryer, and then apposed to an imaging plate (Fuji Imaging Plate BAS-TR2025, Fuji Photo Film Co., Ltd., Tokyo, Japan). After an exposure time of 3.5 to 4 hours, the imaging plates were scanned using Fuji Analyzer BAS-5000 (Fuji Tokyo, Japan; internal resolution of 25 μm). For quantitative analysis of digital autoradiographs, we used Tina 2.1 software (Raytest Isotopemessgeräte, GmbH, Straubenhardt, Germany).
To study the binding of 18 F-FOL on macrophages, peripheral blood mononuclear cells (PBMCs) from buffy coats were collected using Ficoll centrifugation method. Monocytes were enriched using magnetic-activated cell sorting (MACS) positive selection kit (Monocyte isolation kit with CD14 MicroBeads; Miltenyi Biotec, Bergisch Gladbach, Germany). MACS-purified monocytes were cultured in 12 well plate (1 × 10 6 cells/well) in the Iscove's Modified Dulbecco's medium (IMDM containing 2% AB serum and 2 mmol l-glutamine; Gibco, Thermo Fisher Scientific, Waltham, MA, USA) with M-CSF (20 ng/mL; Peprotech, London, UK) for 6 days at 37 °C in a CO 2 incubator to obtain macrophages (at day 3 half of the medium was replaced with fresh medium with M-CSF (20 ng/mL). At day 6, to differentiate macrophages to M1, IFN-γ (50 ng/mL; Peprotech) and LPS (100 ng/mL; Sigma-Aldrich, St. Louis, MO, USA) were added to fresh medium, and to differentiate macrophages to M2, M-CSF (20 ng/mL), IL-4 (10 ng/mL; Peprotech) and IL-10 (10 ng/mL; Peprotech) were added to the fresh medium, and incubated for 2 days at 37 °C in a CO 2 incubator. At day 8 medium from the wells were removed, and 18 F-FOL (1 MBq/mL in fresh medium) was added to each well and incubated for 1 hour at 37 °C in a CO 2 incubator. After incubation cells were rinsed with PBS (2 × 1 mL) to remove any unbound radioactive materials. Then, 1% sodium dodecyl sulfate in PBS (1 mL) was added to each well and solubilized cells were collected into individual test tubes 13 . Bound 18 F-FOL was measured in a γcounter (1480 Wizard 3″; Perkin Elmer/Wallac, Turku, Finland).

PET/CT and image analysis.
In the mice studies, animals were fasted for 4 hours before the PET/CT imaging. Mice were then anesthetized using isoflurane (3-4% induction, 1-2% maintenance), placed onto the bed of a dedicated small animal PET/CT (Inveon Multimodality, Siemens Medical Solutions, Knoxville, TN, USA) and administered with 18  was i.v. injected, and a high-resolution CT was performed for anatomical reference. The PET data were iteratively reconstructed with an ordered-subsets expectation maximization 2D algorithm (OSEM2D with two iterations) into 5 × 60 s, 3 × 300 s, 1 × 600 s, and 2 × 1,800 s time frames (matrix size: 128 × 128 × 159, pixel size: 0.776 × 0.776 × 0.796 mm). Vascular contrast-enhanced CT images were reconstructed with a Feldkamp based algorithm (matrix size: 768 × 768 × 923, pixel size: 0.094 × 0.094 × 0.094 mm). Co-registration of PET and CT images was automatic, and confirmed visually on the basis of anatomical landmarks. To analyze PET/CT images, we used the free Carimas 2.9 software (Turku PET Center, Turku, Finland, www.turkupetcentre.fi/carimas/). We defined regions of interest (ROIs) for the aortic arch, vena cava (representing blood), and myocardium in the coronal PET/CT images, using the contrast-enhanced CT as an anatomical reference. The transaxial and sagittal views were used to ensure the correct ROI placement. We expressed results as standardized uptake values (SUVs), normalized for the injected radioactivity dose and animal body weight, and as maximum target-to-background ratios (TBR) calculated as SUV max, aortic arch /SUV mean, blood . For both 18 F-FDG and 18 F-FOL studies, we defined the blood ROI in the same area of the vena cava. In addition to in vivo imaging, we performed ex vivo 18 F-FOL PET imaging for excised mice aorta. After preparation of aortas, they were placed on a Petri dish that was covered with ultrasound gel to avoid dehydration, and PET data were acquired for 30 minutes, followed by reconstruction using the OSEM3D algorithm (2 iterations; matrix size: 128 × 128 × 159, pixel size: 0.776 × 0.776 × 0.796 mm).
In the rabbit studies, animals were fasted for at least 4 hours prior to the scans. They were sedated and anesthetized with subcutaneous injection of medetomidine hydrochloride (Domitor, Orion Pharma; dose, 0.1 mg/ kg) and ketamine (Ketalar, Pfizer; dose, 30 mg/kg), and a urinary catheter was inserted to minimize accumulation of radioactivity in the bladder. The rabbits were placed in a supine position in a clinical PET/CT scanner (Discovery VCT, General Electric Medical Systems, GEMS, Milwaukee, WI, USA). All of the rabbit PET studies started with a low-dose CT for attenuation correction. The tube voltage was 80 or 120 kV, and the tube current was 10−80 mA with tube current modulation. The slice thickness was 3.75 mm. On the first imaging day, 18 F-FDG (140 ± 29 MBq; 45 ± 1.5 MBq/kg) was i.v. administered via a marginal ear vein cannula and the PET acquisition was begun 160 min post-injection. Two bed positions were imaged (10 minutes for each 15.7 cm bed position). The total imaging time was therefore 20 minutes. On the following day, the same rabbits were re-imaged with 18 F-FOL (130 ± 37 MBq; 41 ± 3.7 MBq/kg) and a dynamic PET study was performed with two bed positions. The acquisition times were 4 × 15 s, 6 × 30 s, 6 × 60 s, and 10 × 120 s. Bed positions were imaged in an interleaved manner and the total scanning time was 90 minutes. After PET imaging, CT was performed with a tube voltage of 100 kV and a current of 450 mA. The CT field-of-view covered the area from the aortic arch to the bifurcation, and the slice thickness was 0.625 mm. After scanning, PET data were reconstructed with a three-dimensional iterative algorithm (VUE point) with 28 subsets, 2 iterations, a standard z axis filter, and 2.00 mm full width at half maximum post-filtering. Attenuation correction was performed using low-dose CT. The PET image matrix size was 256 × 256, the pixel size was 1.37 × 1.37 mm, and the field-of-view was 35 cm. We performed analysis of PET/CT images as described above. We divided the aorta into 2 cm segments and determined the maximum SUV in each of them. The mean SUV in the blood was measured from the inferior vena cava to calculate maximum TBRs (SUV max, aorta /SUV mean, blood ). We carefully identified the anatomical landmarks (e.g., renal arteries) from CT images in order to match up the defined 2 cm aorta segments to the excised samples used for ex vivo gamma counting and autoradiography.
All PET/CT images were exported from Carimas 2.9 to Adobe Photoshop CS5 with a resolution of 1,648 × 948 pixels for preparation of Figs 3 and 5.
Ex vivo biodistribution of 18 F-FOL. Two hours after injection of 18 F-FOL, mice and rabbits were euthanized. Blood samples were drawn from mice by cardiac puncture under deep isoflurane anesthesia, and the mice were then sacrificed with cervical dislocation. The aorta and other tissues were excised and weighed, and their radioactivity was measured with a γ-counter (Triathler 3′′, Hidex, Turku, Finland). Blood samples were drawn from rabbits by cardiac puncture under deep medetomidine hydrochloride/ketamine anesthesia, and the rabbits were then sacrificed with an overdose of pentobarbital (Mebunat vet 60 mg/mL, Orion Pharma, Espoo, Finland). From each rabbit, the aorta from the aortic arch to the iliac artery bifurcation was prepared and cut into 2 cm segments. These were weighed, and the radioactivities of each aorta segment, blood, and various tissue samples were measured using a γ-counter (1480 Wizard 3′′; Perkin Elmer/Wallac, Turku, Finland). To determine the ex vivo biodistribution, we normalized the radioactivity values for injected radioactivity dose, decay of 18 F, and weight of tissue sample. In mice, we compensated for the radioactivity remaining in the tail. The ex vivo biodistribution results were expressed as a percentage of the injected radioactivity dose per gram of tissue (%ID/g).

Autoradiography.
After ex vivo gamma counting of excised tissues, we studied the aortic distribution of 18 F-FOL in more detail using autoradiography as described earlier 19 . Whole excised aorta from the mice was used, as well as two segments of abdominal aorta and one segment of thoracic aorta from the rabbits. These were chosen on the basis of the 18 F-FDG uptake seen on the PET/CT images the day before sacrifice. The excised mouse aortas were frozen in dry-ice cooled isopentane and cut into sequential longitudinal cryosections of 20 μm. The rabbit aortas were divided into 3-6 transverse 2 cm segments, frozen, and cut into 20 μm sequential transverse cryosections. The sections were then apposed on an imaging plate (Fuji Imaging Plate BAS-TR2025), and the imaging plates were scanned (Fuji Analyzer BAS-5000) after an exposure time of at least 4 hours. After scanning, sections were stored at −20 °C until being stained with H&E and scanned with a digital slide scanner (Pannoramic 250 Flash).
After a careful superimposition of the digital autoradiographs and H&E images, we determined 18 F-FOL accumulation in mice in the following ROIs: (1) atherosclerotic plaques (non-calcification, excluding media), (2) normal vessel wall (no lesion formation), and (3) adventitia (mainly adipose tissue around the aorta). In total, 677 ROIs (282 in plaques, 256 in normal vessel wall, and 139 in adventitia) were analyzed from ten atherosclerotic SCIEnTIFIC RePoRtS | (2018) 8:9720 | DOI:10.1038/s41598-018-27618-4 (including two mice used for the blocking study) and six control mice (Table 1). For rabbits, the atherosclerotic plaques were divided into two categories: (1) advanced atherosclerotic plaques (large fibroatheroma lesions mainly observed in denudated abdominal aorta) and (2) mild atherosclerotic plaques (small lesions/intimal thickening observed in non-denudated thoracic aorta) on the basis of their histopathological characteristics seen in H&E, and as assessed by J.M.U.S. and A.S., together with an experienced pathologist, P.S. In total, 322 ROIs (220 in advanced plaques and 102 in mild plaques) were analyzed from four rabbits for 18 F-FOL accumulation. We were not able to measure radioactivity accumulation in totally healthy vessel wall, as some plaque formation was also seen in non-denudated thoracic aorta due to the genetic background and advanced age of the rabbits.
For autoradiography analysis, we used Tina 2.1 software and expressed 18 F-FOL uptake as photostimulated luminescence per square millimeter (PSL/mm 2 ). The count densities of background radiation were subtracted from the actual ROI data. Additionally, the results for each mouse and rabbit were decay-corrected for injection and exposure time, and normalized for injected radioactivity dose. Finally, we calculated the average 18 F-FOL uptake ratios between atherosclerotic plaque, normal vessel wall and adventitia from each mouse, and the average PSL/mm 2 values in advanced and mild plaques from each rabbit. The digital autoradiographs were exported from Tina 2.1 with a resolution of 766 × 1,008 pixels (mice) and 398 × 416 pixels (rabbits), and the H&E images were exported from Pannoramic viewer with a resolution of 4,729 × 3,230 pixels (mice) and 12,516 × 29,356 pixels (rabbits), to Adobe Photoshop CS5 for preparation of Figs 2-4 and 7.
Characterization of atherosclerotic plaques. We used frozen human carotid samples to investigate co-localization of 18 F-FOL binding, CD68-positive macrophages, and FR-β. For this purpose, macrophages were identified by incubating 8 µm cryosections with mouse monoclonal anti-human CD68 antibody (mo876; 1:200, PG-M1, Dako, Glostrup, Denmark), as previously described 12 . For FR-β immunohistochemical staining, we used a previously described protocol with minor modifications 12 . Briefly, sections were dried at room temperature and fixed with 4% paraformaldehyde. Sections were first incubated with biotinylated anti-human FR-β antibody (m909, 1:100) 32 , and then with secondary antibody (P0397, 1:200, Dako). A color reaction was developed using 3,3′-diaminobenzidine (Dako), and counterstained using Mayer's hematoxylin. For quality control of the staining procedure, we stained adjacent sections without primary antibody. Images were exported from Pannoramic Viewer into Adobe Photoshop CS5 at a resolution of 4,309 × 20,239 pixels to create Fig. 2.
Mice hearts were collected and preserved in 10% formalin at room temperature overnight. They were then dehydrated using 70% ethanol, embedded in paraffin, and cut transversely into 6 µm sections at the level of the coronary ostia. Adjacent sections were stained with modified Movat's pentachrome 19 to detect histopathologic characteristics, or with anti-mouse Mac-3 antibody (1:1000, BD Biosciences, Franklin Lakes, NJ, USA) 33 to detect macrophages. Additionally, sections were stained with anti-iNOS (1:200, Abcam, Cambridge, UK)) or anti-MRC-1 (1:500, Abcam, Cambridge, UK) antibodies to detect M1 (pro-inflammatory) and M2 (anti-inflammatory) polarized macrophages, respectively 24 . After staining, the sections were scanned with a slide scanner (Pannoramic 250 Flash). From each section (n = 3/mouse/staining), we outlined the areas of intima and media using the free GNU Image Manipulation Program v. 2.9 (www.gimp.org), and determined the total area as mm 2 using Image-J software v. 1.46 (National Institutes of Health, Bethesda, MD, USA) 34 . To determine macrophage positive area in the intima, we used the automatic color deconvolution method of Image-J, and expressed results as positive staining area in aortic ostia (%). We used Adobe Photoshop CS5 to convert the aortic ostia and high-power view images (resolutions of 1,685 × 1,642 pixels and 4,232 × 1,502 pixels, respectively) to create Fig. 3. We used the aortic ostia to determine the area of macrophages in each mouse, as it could be easily identified because of the valves, and was therefore the most reliable place to compare plaque deposition between different mice.
On the basis of the 18 F-FDG uptake seen in the PET/CT, certain rabbit aorta segments were used for autoradiography and the other segments were fixed in 10% formalin, embedded in paraffin, and cut into 6 µm sections. For each aorta segment, adjacent sections were stained with either Movat's pentachrome or H&E for morphological characterization. To detect rabbit macrophages, we deparaffinized adjacent aorta sections by immersion in xylene. A graded ethanol series was then used for rehydration. Sections were then treated in 0.25% Triton X-100 in PBS, 3% H 2 O 2 , and 3% protease-free bovine serum albumin. Sections were incubated with primary antibody against mouse anti-rabbit RAM-11 (M 0633, 1:1000, Envision, Dako, Carpinteria, CA, USA), followed by secondary anti-mouse antibody (K4001, Envision, Dako, Carpinteria, CA, USA). For detection, 3,3′-diaminobenzidine was used as a chromogen (K3468, Envision, Dako, Carpinteria, CA, USA) and sections were counterstained with Mayer's hematoxylin. We used GIMP and Image-J software (as described above) to determine the RAM-11 positive area (%) in rabbit plaques, and compared the results with the corresponding 18  Statistical analyses. We presented all values as mean ± SD with two significant figures. To compare groups, we used independent samples t tests. For multiple group comparisons, we used a one-way ANOVA with Tukey's correction. Paired t tests were applied for comparisons of uptake between different tissues in the same animals.
Associations between continuous variables were tested using general linear mixed models, where animal was used as a random effect to account for within-subject correlation, and classified aorta segments were used as a repeated effect to account for the difference between segments when appropriate. The correlations between continuous variables were assessed using Pearson's correlation coefficient (r). We checked residuals for justification of the analyses and considered P values less than 0.05 to be statistically significant. Statistical analyses were conducted with SPSS Statistics 21 (IBM Corp., Armonk, NY, USA) or with SAS software, version 9.4 (SAS Institute Inc., Cary, NC, USA).