Feasibility of ex vivo fluorescence imaging of angiogenesis in (non-) culprit human carotid atherosclerotic plaques using bevacizumab-800CW

Vascular endothelial growth factor-A (VEGF-A) is assumed to play a crucial role in the development and rupture of vulnerable plaques in the atherosclerotic process. We used a VEGF-A targeted fluorescent antibody (bevacizumab-IRDye800CW [bevacizumab-800CW]) to image and visualize the distribution of VEGF-A in (non-)culprit carotid plaques ex vivo. Freshly endarterectomized human plaques (n = 15) were incubated in bevacizumab-800CW ex vivo. Subsequent NIRF imaging showed a more intense fluorescent signal in the culprit plaques (n = 11) than in the non-culprit plaques (n = 3). A plaque received from an asymptomatic patient showed pathologic features similar to the culprit plaques. Cross-correlation with VEGF-A immunohistochemistry showed co-localization of VEGF-A over-expression in 91% of the fluorescent culprit plaques, while no VEGF-A expression was found in the non-culprit plaques (p < 0.0001). VEGF-A expression was co-localized with CD34, a marker for angiogenesis (p < 0.001). Ex vivo near-infrared fluorescence (NIRF) imaging by incubation with bevacizumab-800CW shows promise for visualizing VEGF-A overexpression in culprit atherosclerotic plaques in vivo.

Bevacizumab-800CW showed first of all to be safe, to have a high sensitivity and specificity, and allowed for accurate and reproducible imaging of angiogenesis. Previously, our research group has shown that radiolabeled 89 Zr-bevacizumab is an effective and specific method for the detection of VEGF-A in carotid endarterectomy specimens 31 . The advantage of bevacizumab-800CW over 89 Zr-bevacizumab is that the former has no radiation burden and is relatively less expensive than the latter. In this study, the feasibility of bevacizumab-800CWbased optical imaging for the detection of carotid plaque instability was assessed. The aim of this study was to determine whether the optical tracer bevacizumab-800CW is suitable for ex vivo visualization of intraplaque angiogenesis by means of identifying VEGF-A presence, and thereby able to distinguish between culprit and non-culprit atherosclerotic plaques.

Results
Patient characteristics. Carotid artery plaque specimens from 17 patients were included in the study (n = 17), of which three were used for dose-finding analysis. The optimal dose was used in 15 plaques: 11 plaques in the culprit group and three in the non-culprit group, based on the period of time between symptom onset and carotid endarterectomy (CEA), and one plaque was retrieved from an asymptomatic patient and was analyzed separately. Patient characteristics of the culprit and non-culprit groups are shown in Table 1. Specimens of the common, internal and external carotid artery were used. Of all patients, six suffered from a cerebrovascular accident (CVA), Table 1. Patient characteristics and cardiovascular risk factors in culprit and non-culprit group, the asymptomatic patient not included. CEA, carotid endarterectomy. a Defined as systolic blood pressure ≥ 140 mmHg and/or diastolic blood pressure ≥ 90 mmHg.  Fig. 1a shows a representative example). Based on the H&E staining, fluorescence microscopy revealed the presence of bevacizumab-800CW to be mainly intracellular (Fig. 1b). The corresponding median MFI of the  VEGF-A immunohistochemistry. VEGF-A staining was performed in 30 sections of the total number of CEA specimens, with two sections per specimen. The calculated mean H-score of culprit plaques was 223 ± 25.1; the mean of the non-culprit group was 107 ± 6.34. In 91% of the culprit plaques a strong staining intensity (H-score: 201-300) was found. The H-score of the plaque from the asymptomatic patient was 236, indicating an intense staining for VEGF-A. VEGF-A expression levels were observed to be clearly higher in the culprit plaques than in the non-culprit plaques (Fig. 2B). In the culprit plaques, the pattern of fluorescence overlapped with VEGF-A expression in 82% of the specimens (n = 9/11) (Fig. 2C).

Correlation parameters.
Histological examination of the asymptomatic plaque revealed clear pathological features of vulnerability similar to culprit plaques as compared to non-culprit plaques such as intraplaque hemorrhage and intraplaque angiogenesis (Fig. 3). In this particular plaque, intraplaque hemorrhage was present as shown by H&E staining (Fig. 3B), a prominent feature of the vulnerable plaque 15 . The asymptomatic plaque could therefore be deemed vulnerable, i.e. culprit, as these features are characteristic for culprit lesions 32 .
In Table 2, an overview of the various associations between VEGF-A staining and correlation parameters is displayed. CD34 staining for microvessel density confirmed the presence of microvasculature in all culprit plaques. In addition, in all culprit plaques, including the asymptomatic plaque, VEGF-A staining was strongest in areas of high microvessel density. A strong association with VEGF-A staining was demonstrated (ϕ = 0.829; p < 0.001). No intraplaque micro vessels were observed in the non-culprit plaques. www.nature.com/scientificreports/ Von Kossa staining for (micro)calcifications was performed in 22 sections of the culprit plaques (2 per specimen), and revealed small calcium deposits around the VEGF-A and micro vessel areas in 73% of the sections (n = 16/22). Of the non-culprit plaques, 6 sections were stained with Von Kossa and 3 specimens revealed microcalcifications in both the intimal and medial layers. No significant association was found between microcalcifications and VEGF-A expression in both the culprit and non-culprit group (p > 0.05).
Macrophage infiltration (CD68 staining) was most abundant in the VEGF-A ROIs in culprit plaques (n = 15/17). Weak to moderate staining was found in non-culprit plaques. A moderate association was found between VEGF-A expression scores and CD68 staining scores (ϕ = 0.606; p < 0.05). The average CD68 score in the culprit group was higher than in the non-culprit group (2.83 ± 0.389 vs 1.27 ± 0.577).
Strong expression of smooth muscle actin (SMA) staining was observed in the non-culprit plaques as opposed to mostly weak staining in culprit plaques. The asymptomatic plaque also displayed low SMA expression. There was a strong inverse correlation between H-scores and SMA staining (ϕ = 0.870; p = 0.023).

Discussion
Optical imaging is a relatively new modality in the clinical cardiovascular field and can be used for realtime detection of pathological processes 13,31,33 . Currently, no decisive method is available to make a distinction between culprit and non-culprit atherosclerotic plaques. This study investigated the feasibility of using www.nature.com/scientificreports/ bevacizumab-800CW for the identification of vulnerable plaques by determination of their VEGF content. We demonstrated a successful method to detect a measurable fluorescent signal in incubated atherosclerotic plaques. Next, the optimal incubation concentration was determined, 1 µg/ml bevacizumab-800CW. We were able to demonstrate that areas of high VEGF-A expression, based on specific immunohistochemistry, correlated with a high fluorescence signal. Strong VEGF-A expression was observed in culprit plaques, in contrast to the weak staining found in non-culprit plaques. We demonstrated the relationships between VEGF-A expression, and macrophage infiltration and intraplaque microvessel density. Although macrophage infiltration by itself does not appear to be predictive of upcoming acute events in the future 15 , we found that it coincides with VEGF-A expression and intraplaque angiogenesis. No correlation was demonstrated between microcalcifications and VEGF-A expression. Furthermore, microcalcifications were present both in culprit and non-culprit plaques. This corresponds to previous research by Hop et al. 13 , who investigated the use of the PET-tracer 18 F-NaF specific for microcalcifications in culprit and non-culprit carotid plaques ex vivo.
The inverse association found between VEGF-A expression and SMA expression demonstrates that SMA expression is more pronounced in non-culprit plaques, indicating these plaques to contain more mature vascular smooth muscle cells (VSMCs). This complements earlier research in which diminished SMA staining was related to unstable coronary plaques 34 . Additionally, Shinde et al. 35 demonstrated that SMA-expressing myofibroblasts are associated with scar contraction in healing myocardial infarcts. A possible explanation why SMA expression in particular is present in non-culprit plaques is that these myofibroblasts take part in wound healing processes and vessel repair, that presumably stabilize the atherosclerotic plaque 36 . Although VSMCs and myofibroblasts are distinct cell types, SMA may have protective properties in the sense that SMA expression stabilizes the atherosclerotic plaque.
The fluorescent signals found at sites of clear visual macrocalcification without VEGF-A expression at those specific sites could be explained by autofluorescence of calcified areas, as described previously in human atherosclerotic arteries 37 . Conversely, the frail integrity of 3 µm tissue slides results in disrupted tissue sections due to tearing and breaking of the tissue. This could attribute for a weaker bevacizumab-800CW fluorescent signal, imaged by the flatbed scanner, at the corresponding sites of VEGF-A expression. Another reason could be the submersion method of incubation with bevacizumab-800CW as the tracer might not be able to penetrate the full thickness of the specimens.
A limitation of this study is the low number of asymptomatic and non-culprit atherosclerotic plaques, which posed a difficulty in comparing symptomatic to non-symptomatic plaques and assessing statistical significance between culprit and non-culprit plaques. Asymptomatic plaques are normally not resected as standard of care because of the patient risk involved in the procedure. The number-needed-to-treat (NNT) to prevent major CVA or death is 6 for symptomatic patients, whereas asymptomatic patients have an NNT of 20, emphasizing patient risk 38,39 . Additionally, symptomatic plaques are normally removed within 2 weeks after symptom onset as standard of patient care in the Netherlands, which could be an explanation for the low number of non-culprit plaques.
This study complements previous research by our group with 89 Zr-bevacizumab and the use of bevacizumab in CEA specimens 31 , as the monoclonal antibody bevacizumab was able to target VEGF-A ex vivo in human carotid atherosclerotic plaques in this study too. In addition, this study adds knowledge as we demonstrate the ability of the conjugate bevacizumab-800CW to identify intraplaque angiogenesis by targeting VEGF-A, and to be capable of distinguishing between culprit and non-culprit plaques. Furthermore, we demonstrate that angiogenesis correlates with macrophage infiltration using CD68 immunohistochemistry. An advantage is that a systemic dose of bevacizumab-800CW is already proven to be safe and effective in humans, according to various cancer studies [27][28][29][30] , which facilitates further in vivo research in the cardiovascular field. In general, the benefit of NIRF-tracers over PET-tracers is that the production process of NIRF-tracers is generally simpler Table 2. correlation parameters for VEGF-A staining and CD34, CD68, SMA and Von Kossa staining. a The Chi Square tests of independence were calculated based on the VEGF-A ROIs found in the specimens and the intensity of the various stainings in those ROIs. The asymptomatic plaque was not taken into account. b For the association between SMA and VEGF-A staining, an inverse Chi Square test was used.

Characteristic
Culprit specimens, n (%) Non-culprit specimens, n (%) Chi square tests of independence a www.nature.com/scientificreports/ and less expensive, due to the long shelf life and the elimination of radioactive infrastructure in the production process. NIRF-imaging allows for non-invasive imaging techniques, which is favorable in this fragile patient group. Additionally, with NIRF imaging, it is easier to image in real-time, there is no radiation burden, and the imaging time itself is relatively short 31 , however whole body imaging and deep signal penetration is lacking.
In conclusion, the NIRF tracer bevacizumab-800CW is promising in visualizing VEGF-A expression in human carotid atherosclerotic plaques ex vivo. Our results demonstrate the potential of bevacizumab-800CW to define vulnerable (VEGF-positive) and non-vulnerable (VEGF-negative) plaques. The validation of the latter hypothesis should take place in a clinical study. Future studies should focus on the clinical proof of principle in vivo of the use of bevacizumab-800CW in this particular patient group, to allow for its validation in vivo. We believe that this tracer might allow for improved risk-stratification of carotid atherosclerosis and therefore a reliable selection method for CEA. Benefits would be prevention of disease and mortality caused by acute vascular events. Ultimately, clinical implementation of this tracer to select for CEA should rely on the predictive value of angiogenesis in high-risk plaques.

Methods
Study design. For this ex vivo feasibility study, fresh human carotid specimens were obtained from patients undergoing carotid endarterectomy between September 2017 and April 2018 in the University Medical Center Groningen (UMCG). Previous literature demonstrated that 'positive' plaques corresponded to the patients with recent events (i.e. culprit plaques) and 'negative' plaques to patients with a remote history of ischemic stroke (i.e. non-culprit plaques) 32,40 . Odds ratios of present plaque characteristics as risk indicators for stroke were demonstrated to be significantly greater for plaques retrieved from patients operated within 30 days after the ischemic event than in plaques retrieved after 30 days after symptom onset. For high microvessel density, these odds ratios were 4.52 and 1.47, respectively 32 . In this study, we adopted this dichotomy: 'positive' culprit plaques were defined as plaques in patients with recent events (i.e. cerebrovascular accident, transient ischemic attack, amaurosis fugax), and 'negative' non-culprit plaques were defined as plaques in patients with no events in the last 4 weeks. For all patients, clinical and demographic data were collected from medical records, using the hospital's electronic medical record system, including medication use and cardiovascular risk factors that were present before symptom onset. The study was released by the Medical Ethical Committee of the UMCG and is in accordance with the Helsinki Declaration (2013). It was decided that patient informed consent was not required as the study uses waste tissue-material, that is not restricted by the Dutch WMO (Wet Mensgebonden Onderzoek, which translates to 'law on research involving human subjects). The IRB of the UMCG approved this study. For all patients, it was checked if objection statements about the use of waste material were ever recorded by the hospital. Data analysis was performed anonymously.
Primary endpoints were the accumulation of ex vivo incubated bevacizumab-800CW in carotid plaques by fluorescence macro-and microscopy, and its correlation with VEGF-A expression, macrophage infiltration (CD68), microvessel density (CD34), microcalcifications (Von Kossa) and structural integrity of the vessel wall (smooth muscle actin; SMA), determined by semi-quantitative analysis of (immuno)histochemical stainings. For these parameters, culprit and non-culprit plaques were compared.
Fluorescence imaging techniques. Bevacizumab (Avastin; Genentech, South San Francisco CA, USA) conjugation with the fluorescent dye IRDye800CW was performed at the department of Clinical Pharmacy and Pharmacology, as described earlier 41 . An optimal dose of 1 μg/ml was determined by submersing the specimens in various solutions (4 μg/ml, 2 μg/ml, and 1 μg/ml bevacizumab-800CW in PBS) for one hour in a dark room to prevent photobleaching. The used concentrations were based on theoretical blood concentrations after intravenous administration of 4.5 mg, 10 mg, and 25 mg bevacizumab-800CW, respectively. The imaging procedures with the optimal dose were as follows. Immediately after excision, the specimens were rinsed with PBS. Subsequently, images of the whole tissue were acquired with the PEARL imager (LI-COR Biosciences, Lincoln NE, USA). The specimens were then submersed in a 1 μg/ml bevacizumab-800CW solution for one hour at room temperature. PEARL images were obtained from the whole tissue specimen, after which hot spots (i.e. spots with 95% of the highest fluorescent signal (expressed in Mean Fluorescence Intensity [MFI]) were selected. The MFI was only calculated from the subsequent PEARL images. All images of all specimens were scaled to the same intensity. The spots were formalin-fixed, paraffin-embedded (FFPE) and cut in 3 μm sections.
The FFPE blocks were analyzed with the Odyssey fluorescent flatbed scanner (LI-COR Biosciences, Lincoln NE, USA). In order to cross-correlate a fluorescent signal within the corresponding 3 μm slides with additional histopathological staining, the sections were deparaffinized (10 min xylene) and analyzed with the Odyssey scanner. Afterwards, these 3 μm slides were stained with hematoxylin and eosin (H&E) for cell morphology and to determine what part of the plaque displayed fluorescent signal.
In additional consecutive 3 μm slides cell nuclei were visualized with Hoechst staining (Invitrogen, Carlsbad CA, USA) in the same manner as we described previously 29 . Leica NIR fluorescence microscopy was carried out to colocalize bevacizumab-800CW and VEGF-A. For image processing and analysis, LAS-AF software (Leica Microsystems, Wetzlar, Germany) and Fiji/ImageJ (version 1.8.0., LOCI, University of Wisconsin, USA) were used as described previously 42 .
(Immuno)histochemistry. To allow for examination of the pathological status (i.e. vulnerability) of the plaques, for each spot slides were double stained for VEGF-A and either Von Kossa, CD34, CD68 or SMA. The latter three immunohistochemical stainings were performed by standardized machine protocols. For the staining of microvessels, slides were incubated with anti-CD34 (mouse IgG 1 κ monoclonal, Dako, clone QBEnd/10). Slides were incubated with anti-CD68 (mouse IgG 3 κ monoclonal, Dako, clone PG-M1) for the staining of mac- www.nature.com/scientificreports/ rophages. Smooth muscle actin in the vessel walls was stained with anti-SMA (mouse monoclonal, Dako, clone 1A4). All slides were incubated for one hour at room temperature. Consecutively, the slides were incubated with horseradish peroxide (HRP) secondary antibodies, and afterwards chromogenic staining with 3,3′-diaminobenzidine (DAB) was performed for color development. Hematoxylin was used to counterstain the slides. For VEGF-A, slides were deparaffinized and rehydrated with xylene (10 min) and graded alcohols. For heatinduced antigen retrieval (15 min), 100 mM Tris/EDTA buffer (pH 9.0) was used. Endogenous peroxidase was blocked for 30 min with 30% hydrogen peroxide, followed by avidin-biotin blocking. Subsequently, the slides were incubated overnight at 4 °C with a polyclonal rabbit anti-human VEGF-A antibody (RB9031, Thermo Fisher Scientific, Waltham, MA, USA). Slides were incubated with goat anti-rabbit biotin in phosphate-buffered saline with 1% bovine serum albumin [PBS 1%BSA] with 1% AB serum, and consecutively with alkaline-phosphatase conjugated streptavidin in PBS 1% BSA with 1% AB serum, both for 30 min. For color development, alkalinephosphatase activity was visualized using naphthol AS-MX phosphate and Fast Blue salt (30 min). The slides were washed with demineralized water and cover slipped with Kaiser's glycerin. PBS was used for washing, and all steps were performed at room temperature unless indicated otherwise.
For the Von Kossa staining, the slides were first submersed with 1% silver nitrate solution for 60 min in broad daylight, followed by 3% sodium thiosulfate (5 min). The slides were counterstained with nuclear fast red (3 min), after which dehydration was performed with graded alcohols. The slides were then air-dried (30 min) and cover slipped with mounting medium. Demineralized water was used throughout washing.
The stained slides were digitized with the NanoZoomer Digital Pathology Scanner (Hamamatsu Photonics K.K., Hamamatsu City, Japan). The imaged slides were then analyzed and evaluated with Aperio ePathology (Leica Biosystems, Wetzlar, Germany) software. To score the degree of VEGF staining, the H-score was calculated in the same manner as described previously by our group 30 , determined by semi-quantitative combined assessment of the percentage of stained cells and the staining intensity (0-300; continuous scale; 0-100: negative/weak; 101-200: moderate; 201-300: strong). Von Kossa, CD34, CD68 and SMA stainings were categorized semi-quantitatively (score 1 = weak; 2 = moderate; 3 = strong). The pattern of fluorescent signal was compared to corresponding VEGF-A-stained slides by fitting the VEGF-A region of interest (ROI) on the corresponding fluorescence image and determining the extent of overlap between VEGF-A expression and fluorescent signal.
Data and statistical analysis. Data were analyzed with SPSS Statistics 24 (IBM, Armonk, NY, USA) and Fiji/ImageJ (version 1.8.0., LOCI, University of Wisconsin, USA). Regions of interest (ROIs) were drawn based on the VEGF-A staining (95% of the strongest expression, determined by standardized counting algorithms). The mean fluorescence intensity of this ROI was measured quantitatively. Categorical data was tested with Chi-Square association tests. A p value ≤ 0.05 was considered to be statistically significant. www.nature.com/scientificreports/