2.1.

Animals

All experiments were in line with institutional guidelines and approved by the local ethics committee on the use of laboratory animals. Mice (C57BL6/J; $n=7$ ) were anesthetized using subcutaneous administration of a mixture of $75\mathrm{mg}∕\mathrm{kg}$ ketamin (Nimatek, Eurovet, Cuijck, The Netherlands) and $15\mathrm{mg}∕\mathrm{kg}$ xylazin (Xylazin, Ceva Sante Animale BV, Naaldwijk, The Netherlands) in a volume of $3\mathrm{ml}∕\mathrm{kg}$ . Anesthesia was maintained by subcutaneous injection of $0.88\mathrm{ml}∕\mathrm{kg}$ of the same xylazin-ketamin mixture every $30\min$ . Rats (Wistar-Kyoto; $n=6$ ) were anesthetized by a single intraperitoneal injection of urethane $(1.7\mathrm{mg}∕\mathrm{g})$ . All anesthetics were dissolved in saline. Body temperature was monitored using a rectal probe (PT100 sensor, Watlow, United Kingdom) and maintained at approximately $37°\mathrm{C}$ by a heating platform (TH60-SMZ, Linkam Scientific Instruments, United Kingdom).

2.2.

Tissue Preparation and Instrumentation

Animals were placed in the supine position. In mice, the right common carotid artery was surgically exposed up to the bifurcation.⁵ In rats, the abdominal region was surgically opened to expose the left renal artery. To simplify discrimination between artery and surrounding tissues in both preparations, a thin black plastic sheet was placed underneath the artery segments without stretching the vessel. Exposed areas were kept moist with saline at all time during the experiment.

In both species, the left jugular vein was cannulated for administration of fluorescent dyes. Blood pressure was measured through a catheter inserted in the left femoral artery and linked to two blood pressure sensors (Baxter Uniflow, Baxter B.V., Utrecht, The Netherlands); the first sensor was used to monitor blood pressure and, hence, the hemodynamic condition of the animal. The second blood pressure signal was used as input for image triggering (see later in the paper). All catheters and blood pressure probes were filled with heparin ( $5\mathrm{U}∕\mathrm{ml}$ in saline).

To stabilize respiration, the trachea was intubated and ventilated with normal air using a ventilator (mouse: Minivent 845, Hugo Sachs Electronic GmbH, Germany; rat: Harvard rodent respirator, Harvard Apparatus, Massachusetts). As heart rate and also vessel motion may vary slightly from changes in intrathoracic pressure due to breathing, a respiration sensor (Graseby, Wicklow, Ireland) was placed on top of the chest to obtain a (relative) respiration signal for the trigger unit. In a number of mice $(n=3)$ , no (forced) ventilation was used to investigate the effect of spontaneous breathing on image motion and triggering.

2.3.

Fluorescent Labeling

Nuclei of viable cells were labeled by topical and/or intravenous administration of the viable cell membrane permeable DNA/RNA markers SYTO13 or SYTO41 (Molecular Probes, Leiden, The Netherlands). Eosin (Molecular Probes) was used as a specific fluorescent marker for elastin. SYTO13 and SYTO41 (both $2.0\mu \mathrm{M}$ ) and eosin $\left(0.5\mu \mathrm{M}\right)$ were dissolved in saline and administered topically on the artery. In four rats and three mice, SYTO13 was infused intravenously ( $0.1\mathrm{ml}$ of a $4.0\mu \mathrm{M}$ solution in saline) to more specifically stain the endothelial cell nuclei and blood cells. Circulating blood platelets, cytoplasm of leukocytes, and cells in the arterial wall were labeled fluorescently by intravenous administration (i.v.) of acridin red (Chroma-Gesellschaft Schmidt GmbH, Germany; $2.0\mathrm{mg}∕\mathrm{ml}$ in saline with 5% ethanol; a bolus of $0.03\mathrm{ml}$ in mice or $0.1\mathrm{ml}$ in rats). Collagen (types I and III) in the tunica adventitia was visualized by second-harmonic generation (SHG). The subendothelial collagen layer (type IV) is too thin and located too deep to be detected by SHG on our system, especially in vivo.^{1, 2} Therefore, subendothelial collagen, which is luminally exposed on damage to the endothelium, was visualized using collagen-specific CNA35^{12, 27} labeled with green-fluorescent quantum dots [CNA35-QD525; $200\mu \mathrm{l}$ i.v. containing $15\mu \mathrm{M}$ CNA35-biotin and $0.5\mu \mathrm{M}$ streptavidin coated quantum dots (Invitrogen)]. Propidium iodide was used to label nonviable cells (PI; Invitrogen; $200\mu \mathrm{l}$ of a $0.01\mathrm{mg}∕\mathrm{ml}$ solution applied topically).

2.4.

Damage-Induced Subendothelial Collagen Exposure

In a subset of experiments, the carotid artery of mice was pinched with a fine-tipped forceps for $3\mathrm{s}$ to induce damage across the vessel wall. In our ex vivo setup, this procedure has been shown to be effective both in inducing cell damage across the vessel wall,¹ and in uncovering the subendothelial collagen layer.¹² Then, the CNA35-QD525 solution was infused i.v. to label exposed subendothelial collagen.²⁷ Furthermore, propidium iodide was applied topically to label nonviable cells.

2.5.

Image Acquisition

A Nikon E600FN microscope (Nikon Corporation, Tokyo, Japan) connected to a Biorad 2100 MP multiphoton system (Biorad, Hemel Hempstead, United Kingdom) was used for TPLSM as previously described.^{1, 28}

A $120\pm 20\text{-}\mathrm{fs}$ -pulsed Ti:sapphire laser (Spectra Physics Tsunami, Mountain View, California) was used as the excitation source tuned and mode-locked at $800\text{to}840\mathrm{nm}$ . Excitation powers at the sample varied from $60\text{to}80\mathrm{mW}$ . Either a $40\times$ or a $60\times$ water dipping objective was used [numerical aperture (NA) 0.8 and 1.0, respectively] for TPLSM recordings. A $20\times$ water dipping objective (NA 0.5) was used in combination with a $20\text{-}\mathrm{W}$ halogen cold light source (Schott KL 200, Schott AG, Mainz, Germany) to locate the exposed vessels.

To detect the emitted fluorescent signals, three photomultiplier tubes were used. These were tuned to correspond with parts of the emission spectra of the fluorescent markers applied, aiming to find a balance between minimal bleed-through versus maximal signal reception.¹

Images of $256\times 256\text{pixels}$ were obtained for each color channel and combined into one RGB image. Series of 20 to 100 subsequent $xy$ images were recorded (time series). Acquisition of each separate image was started by a trigger pulse derived from a trigger unit (see later in the paper). Optimal image acquisition time was determined empirically depending on heart rate (see later in the paper).

Table 1 summarizes possible image acquisition times and settings of the system. Image acquisition times applied in this paper varied from $0.11\text{to}0.43\mathrm{s}$ , depending on matrix size, pixel dwell-time, line speed, and artery/species. When imaging at $>500\text{lines}∕\mathrm{s}$ , optical zoom is automatically applied to reduce total scan area, and hence, image acquisition time. As an advantage, spatial resolution is increased. Furthermore, line speeds of $>750\text{lines}∕\mathrm{s}$ are achieved through bidirectional scanning, which increases scanning speed (requiring no or less optical zoom), and also results in (correctable) interlacing errors (see later in the paper). Image acquisition time can also be decreased by using double and quadruple line scanning modes, where two or four lines are scanned at the same time. This results in a decrease of spatial resolution in the vertical direction ( $y$ direction), which can be corrected by applying additional zoom at the cost of FOV. Objects of interest should therefore be aligned in such a way that the maximum spatial resolution is obtained in the required direction.

Table 1

Image acquisition characteristics.

2.6.

Triggering

The respiration sensor and one of the blood pressure sensors were connected to an electrocardiogram (ECG)-trigger unit (Rapid Biomedical, Würzburg, Germany), originally designed for application in magnetic resonance imaging. However, an ECG could not be used as input, since positioning of an objective lens on the animal strongly disturbed the recorded electrical activity. Therefore, the blood pressure signal derived from the femoral artery was fed into the ECG input, which resulted in modification of the signal. The output trigger signal was coupled to the trigger/synchronizer input port of the instrumentation control unit of the TPLSM system.

The trigger unit displayed both respiration and modified blood pressure signals, and a threshold could be set for each (Fig. 1 ). The blood pressure signal was used as the primary determinant for the trigger moment, aiming to generate one trigger pulse per cardiac cycle. Furthermore, the influence of respiration on motion artifacts was minimized using the respiratory signal. Whenever the respiratory signal exceeded its threshold (during inspiration), triggering was disabled. If, however, the respiration signal was below threshold and the modified blood pressure signal crossed its threshold while rising, a trigger pulse was generated, initiating the acquisition of a single optical section. A trigger delay for both blood pressure and respiration could also be set. This simplified the search for an optimal trigger moment (during diastole and expiration) without the need for adjustment of the actual blood pressure signal, respiration signal, or thresholds.

Fig. 1

Recording of trigger pulse, modified blood pressure, and respiration signals. A recording of the output signals from the trigger unit (modified blood pressure and respiration) and the trigger pulses derived from them. Data were acquired in a ventilated mouse. The blood pressure signal is modified as it is connected to an ECG-input. Nevertheless, it can be used as input for triggering. (a): Trigger settings and generation of trigger pulses; thresholds for respiration and modified blood pressure are indicated by horizontal dashed lines. When the respiration signal is below threshold (expiration), triggering is enabled (gray bars). During the first cardiac cycle a trigger pulse is generated, as triggering was enabled when modified blood pressure crossed threshold upwards (vertical dotted line). The next two cardiac cycles, however, did not lead to the generation of a trigger pulse, as the modified blood pressure rose above threshold when triggering was disabled. In the fourth cycle, a trigger pulse is again generated. A trigger delay could be set and is marked by asterisks. The settings for thresholds and delay were determined empirically. Motion artifacts were smallest when imaging in the diastolic phase of the cardiac cycle. (b): A recording for the duration of $6\mathrm{s}$ , clearly showing that trigger pulses are generated in an irregular pattern.

Trigger pulse, modified blood pressure, and respiration signals from the trigger unit and absolute blood pressure signal were digitally recorded using the acquisition system M-PAQ in combination with the acquisition software package IDEEQ (IDEE, Maastricht, The Netherlands). The sampling rate was $1000\mathrm{Hz}$ . Figure 1 shows such a recording for the duration of $6\mathrm{s}$ . Total frame time, start time of each subsequent image, and time gap between two subsequent optical sections (frames) were recorded using the TPLSM acquisition software (Lasersharp 6.0; Biorad).

2.7.

Image Processing

Image reconstructions were performed using the Image-Pro Plus 6.0 software (Media Cybernetics Inc., Silver Spring, Maryland). To further improve overall image quality, interlacing errors were corrected if needed (see results). Furthermore, Fourier transformation was performed to discriminate the original image frequencies (that lie around the origin, i.e., in the corners of the periodic spectrum plot) from higher noise and interlacing frequencies. The undesirable higher frequencies were then filtered out by multiplying the Fourier-spectra with a mask based on a 2-D Gaussian weighted function, thus improving SNR. Filtering in the Fourier domain and correction for interlacing inaccuracy were performed using software tools developed with the Mathematica 6.0 software package with “Digital Image Processing” (Wolfram Research Inc., Champaign, Illinois), and the “Front-end Vision” and “MathvisionTools” plugins (Eindhoven University of Technology, The Netherlands).

3.1.

Cardiac and Respiratory Activity

During in vivo imaging, each image should be acquired well within one cardiac cycle to avoid image distortion (in-frame) and between-frame motion artifacts. In anesthetized mice, the duration of one cardiac cycle varied from $0.16\text{to}0.24\mathrm{s}$ ( $4.2\text{to}6.1\mathrm{Hz}$ ; reference values for nonanesthetized mice: $0.09\mathrm{s}∕11.1\mathrm{Hz}$ ). This observed cardiac depression is a known side effect of the applied ketamin/xylazin anesthesia.²⁹ In rats, anesthetized with urethane (which has less profound effects on heart rate), the cardiac cycle duration varied from $0.25\text{to}0.30\mathrm{s}$ ( $3.3\text{to}4.0\mathrm{Hz}$ ; reference values $0.17\mathrm{s}∕5.9\mathrm{Hz}$ ).³⁰

Anesthetized animals were ventilated; rats at a rate of $120{\min }^{-1}$ $\left(2.0\mathrm{Hz}\right)$ and mice at a rate of $200\text{to}240{\min }^{-1}$ $\left(3.3\text{to}4.0\mathrm{Hz}\right)$ with a tidal volume of $200\text{to}250\mu \mathrm{l}$ . Some mice were allowed to breathe spontaneously, showing an anesthesia-induced respiratory depression over time (mean respiration rate $161\pm 35{\min }^{-1}$ at the start of imaging and $134\pm 32{\min }^{-1}$ at the end of the experiment). Therefore, experimentation time was limited in these animals. On the other hand, arterial movement due to breathing had a lower frequency in these mice and could thus be overcome more easily. Results obtained in artificially damaged carotid arteries were obtained this way. A drawback of unventilated triggering is a less regular respiratory activity, which might impede accurate triggering.

3.2.

Nontriggered In Vivo Imaging

In view of the heart rates already given, image acquisition times should be shorter than $0.16\mathrm{s}$ in mice and $0.25\mathrm{s}$ in rats to limit motion distortion. Such short image acquisition times can be achieved either by increasing the line scan rate or decreasing the number of lines per image (matrix size). The first method, however, decreases the pixel dwell-time at the cost of a lower SNR and, at higher rates, is accompanied by an automatic optical zoom resulting in a smaller FOV (Table 1). The second method results in lower pixel resolution, which can be resolved only by using an equivalent optical zoom.

Comparison of several typical acquisition rates for imaging of an ex vivo mounted artery revealed that imaging at $2\text{to}3\mathrm{Hz}$ yielded optical sections with a sufficient large FOV (no optical zoom is induced) and relatively good overall image quality. However, this frequency was not high enough for in vivo imaging since acquisition time exceeded a complete cardiac cycle and therefore all motions within this cycle were captured in each single image (Fig. 2 and Video 1 ; mouse carotid artery).

Fig. 2

Nontriggered in vivo TPLSM at normal (ex vivo) scan rates. Three subsequent optical sections (a) to (c) of a left carotid artery of a C57Bl6/J mouse obtained in vivo without application of external triggering. Image acquisition time was $0.43\mathrm{s}$ ( $1200\text{lines}∕\mathrm{s}$ ; normal scanning mode; $512\times 512\text{pixels}$ ). The schematic drawing on the left indicates the position of the optical sections in the vessel wall; vSMC, vascular smooth muscle cell; FOV, field of view; bars indicate $20\mu \mathrm{m}$ . Cell nuclei (SYTO13, green) and extracellular matrix (SHG of adventitial collagen, blue; autofluorescence of elastin, red) are visible. All sections are disturbed by motion artifacts, which cause the arterial wall (blue arrow) to appear as a curved structure. Moreover, every optical section contains different parts of the (moving) vessel wall. The typical morphology and orientation of the smooth muscle cell nuclei (green) is hardly recognizable; the position of the lumen is unclear.

Nontriggered in vivo TPLSM at normal (ex vivo) scan rates. Time series recorded in a left carotid artery of a C57Bl6/J mouse in vivo without application of external triggering. Image acquisition time was $0.43\mathrm{s}$ ( $1200\text{lines}∕\mathrm{s}$ ; normal scanning mode; $512\times 512\text{pixels}$ ). The FOV was approximately $200\times 200\mu \mathrm{m}$ . Cell nuclei (SYTO13, green) and extracellular matrix (SHG of adventitial collagen, blue; autofluorescence of elastin, red) are visible. All sections are disturbed by motion artifacts which cause the arterial wall to appear as a curved structure. Moreover, every optical section contains different parts of the moving vessel wall. The typical morphology and orientation of the smooth muscle cell nuclei (green) is hardly recognizable; the position of the lumen is unclear (QuickTime, $817\mathrm{KB}$ ). (Color online only.) .

As published before, increasing the image acquisition rate in vivo eliminates large motion artifacts from at least some of the optical sections in a time series.¹¹ Nevertheless, most of the images obtained this way are of low quality due to movements of the sample, loss of focus on the sample, or even complete disappearance of the vessel from the FOV.

3.3.

Triggered In Vivo Imaging

Triggered image acquisition of each optical section resulted in subsequent images that (almost) all display the same part of the artery during the same phase of movement. Consequently, the obtained optical sections are mutually comparable. As in untriggered imaging, higher frame rates yielded better results than rates of the order of cardiac rates, as images are acquired in a shorter part of the cardiac cycle. Therefore, a more stable period can be found using shorter image acquisition times (Fig. 3 and Video 2 ; rat renal artery, image acquisition time $0.17\mathrm{s}$ ; cardiac cycle duration $0.25\mathrm{s}$ ).

Fig. 3

Triggered in vivo TPLSM at high scan rates. Five successive optical sections (a) to (e) of the left renal artery of a Wistar-Kyoto rat in vivo obtained by triggered acquisition at an image acquisition time of $0.17\mathrm{s}$ ( $1500\text{lines}∕\mathrm{s}$ ; normal scanning mode; $256\times 256\text{pixels}$ ). Cardiac cycle duration was $0.25\mathrm{s}$ . Schematic drawing displays the position of the optical sections; vSMC, vascular smooth muscle cell; FOV, field of view; bars indicate $20\mu \mathrm{m}$ . Artery was labeled topically for cell nuclei (SYTO13, green) and systemically for cytoplasm of (blood) cells and platelets (acridin red, orange). The successive optical sections contain the same part of the artery. In the tunica adventitia, collagen (SHG, blue) and nuclei (arrowhead) are visible; vSMC nuclei (white arrow) are observable in the tunica media. Moreover, an adhesive blood cell (encircled by white dotted line) is visible against the vessel wall in several $(>10)$ successive images. The lumen appears orange due to circulating unbound acridin red.

Triggered in vivo TPLSM at high scan rates. Stable recording of the left renal artery of a Wistar-Kyoto rat in vivo obtained by triggered acquisition at an image acquisition time of $0.17\mathrm{s}$ . The FOV was approximately $200\times 200\mu \mathrm{m}$ . The artery was labeled topically for cell nuclei (SYTO13, green) and systemically for cytoplasm of (blood) cells and platelets (acridin red, orange). The nuclei of the smooth muscle cells are labeled by both probes (yellow). SHG of collagen (blue) is visible in the tunica adventitia. Images show no distortion and between-frame motion is very small due to triggered acquisition. (QuickTime, $709\mathrm{KB}$ ). (Color online only.) .

Series of subsequent optical sections with the smallest impact of image distortion and between-frame motion disturbances were obtained when images were acquired within the diastolic phase of the cardiac cycle. However, trigger pulses were generated at peaks of the modified blood pressure signal. Therefore, a trigger delay was used to start image acquisition in diastole, which could easily be introduced via the trigger unit. The amount of delay required for stable imaging was determined empirically. The timing of image acquisition in (part of) the systolic phase of the cardiac cycle resulted in images with large motion artifacts and strong deformations of the tissue, even at higher imaging rates (not shown).

In rat renal arteries, triggered imaging often resulted in more stable images and smaller motion disturbances than in mouse carotid arteries. This finding was to be expected as the more distally located renal artery suffers less from thoracic movement or deformation due to respiration. Also, the heart rate in rats was lower.

3.4.

Image Processing

The overall image quality suffered from the fast acquisition rates that affect the SNR, due to short pixel dwell-times [Fig. 4 ]. The SNR was enhanced using Fourier spectrum-based noise filtering [Fig. 4]. Additionally, fast image acquisition using scan rates of $>750\text{lines}∕\mathrm{s}$ (Table 1) resulted in images that appeared to exhibit stretched details in the horizontal direction. To reduce total acquisition time, odd scan lines of the matrix are automatically scanned from left to right and even scan lines from right to left (bidirectional scanning), instead of scanning each horizontal line from left to right (unidirectional scanning; standard for lower frame rates). Image analysis of bidirectionally scanned images revealed that in comparison with the odd scan lines, the even scan lines of the image matrix were shifted in the horizontal direction (varying from $3\text{to}10\text{pixels}$ ), causing blurred images. Correction of this interlacing inaccuracy resulted in remarkably sharper images [Fig. 4]. As a consequence, the image matrix size was also reduced in the horizontal direction, reflecting a slightly smaller FOV.

Fig. 4

Image enhancement. The effects of image processing applied to one of the optical sections of the series of images as shown in Fig. 3; (a) raw image of the data file, as shown in fig. 3, the image appears fuzzy and the smooth muscle cell nuclei (white arrow) are hardly recognizable as separate structures; (b) the same optical section after correction for inaccurate line positioning (interlacing error), the image already appears sharper and the smooth muscle cell nuclei (white arrow) appear separated and an adhesive blood cell (encircled by white dotted line) is more clearly visible; and (c) filtering in the Fourier domain further improves SNR of the image. Also, the nuclei of smooth muscle cells (white arrow) are more apparent.

3.5.

In Vivo Molecular Imaging of Exposed Subendothelial Collagen

In a subset of experiments, the carotid artery of mice was damaged to expose the thin subendothelial collagen sheet, which was labeled using CNA35-conjugated quantum dots.²⁷ Propidium iodide (PI) was used as a cell viability assay. The damaging procedure dramatically increased the number of PI positive (i.e., nonviable) cells across the vessel wall, including the endothelium [Figs. 5 and 5 ]. Furthermore, subendothelial collagen, which is normally covered by endothelial cells and can not be reached by CNA35-QD525, was now exposed and could be labeled with CNA35-QD525 resulting in a bright green labeled layer. This thin layer was visualized at a subcellular level in vivo, and is clearly located between the smooth muscle and endothelial cell nuclei [Fig. 5], which can be distinguished based on morphology and location in the vessel wall.¹ The collagen in the adventitial layer (as shown by SHG) was not labeled with CNA35-QD525, which indicates that the elastic laminae were still intact and prevented CNA35-QD525 labeling over the vessel wall.

Fig. 5

Imaging of subendothelial collagen prior to and after mechanical damage. Optical sections of a left carotid artery of a C57Bl6/J mouse obtained in vivo with application of external triggering. Images were recorded at double scanning mode, resulting decreased spatial resolution in the vertical direction ( $y$ direction). Image acquisition time was $0.11\mathrm{s}$ ( $1200\text{lines}∕\mathrm{s}$ ; double scanning mode; $256\times 256\text{pixels}$ ). Cardiac cycle duration was $0.45\mathrm{s}$ . Schematic drawings on the left indicate the position of the optical sections in the vessel wall; PI, propidium iodide; FOV, field of view; ROI, region of interest; bars indicate $25\mu \mathrm{m}$ . Mouse carotid artery in vivo after injection of PI (red) and CNA35-QD525 (green) (a) prior to and (b) after pinching. (a) Prepinching; some PI-positive cells (red) can be found in the adventitia (blue SHG, arrowhead) due to surgery; (b) postpinching; the image distortion (white arrow) in the top part indicates that part of the image was not recorded during the diastolic phase; by selecting an ROI (blue dotted area) a stable window can be defined; and (c) ROI from (b). Almost all cell nuclei are PI-positive (red; i.e., have a compromised outer cell membrane) and, therefore, a labeled collagen sheet (green) can be observed located between endothelial cell nuclei (round and closest to the lumen; blue arrows) and smooth muscle cell nuclei (elongated; purple arrow). Adventitial collagen is not labeled by quantum dots, but can be identified by SHG (blue, arrowhead).