We use the FRET-based cGMP biosensor cGi500 (“cGMP indicator with an EC₅₀ of 500 nM”) generated by Russwurm et al. (2007). This monomolecular biosensor consists of the two cGMP-binding sites from bovine cGK type I sandwiched between cyan and yellow fluorescent protein (CFP and YFP). Detection of cGMP with cGi500 is based on FRET from CFP to YFP, whose spectra show sufficient overlap for efficient FRET (Figure 1A). In the absence of cGMP, CFP and YFP are in favorable distance and orientation for FRET to occur, whereas cGMP binding induces a reversible structural change of cGi500 that reduces FRET efficiency (Figure 1B). The ratio of CFP and YFP fluorescence (R = CFP/YFP) recorded upon CFP excitation can be taken as a measure for the FRET efficiency (Clegg, 2009) and, therefore, for intracellular cGMP concentrations measured with cGi500. It has been shown in sensor-containing cell extracts and escin-permeabilized cells that cGi500 detects cGMP concentrations as low as 50–100 nM and is saturated at approx. 3 μM cGMP. Similar experiments have shown that cGi500 detects cAMP only at levels higher than 30 μM, which exceeds typical physiological cAMP concentrations (Russwurm et al., 2007; Thunemann et al., 2013a; Milde and Feil, unpublished results). Based on its high selectivity for cGMP over cAMP, its dynamic range for cGMP detection, fast kinetics, and relatively large CFP/YFP ratio changes of up to ~50%, the cGi500 sensor appears to be suitable for in vivo use.

Animal experiments were performed in compliance with the guidelines for human care and use of laboratory animals. They were approved by the Regierungspräsidium Tübingen, the Ministerium für Energiewende, Landwirtschaft, Umwelt und ländliche Räume des Landes Schleswig-Holstein, and local authorities in Boston, MA. The generation of transgenic cGMP sensor mice has been described previously (Thunemann et al., 2013b). Transgene structures are shown in Figure 1C. Briefly, the SM22-cGi500 mouse line was generated by random transgenesis using a 445-bp promoter fragment of the Transgelin/SM22 gene (Li et al., 1996) to drive cGi500 expression. The R26-CAG-cGi500(L2) and R26-CAG-cGi500(L1) lines were generated by targeted mutagenesis of the Rosa26 (R26) locus with a construct derived from the R26-mT/mG targeting vector (Muzumdar et al., 2007). IVM data shown and discussed in this article were obtained with the R26-CAG-cGi500(L1) mouse line that expresses cGi500 ubiquitously (Thunemann et al., 2013b).

Image processing is performed using the open source software ImageJ or ImageJ-derived software (Schneider et al., 2012). In establishing this protocol, the data were evaluated manually, although ImageJ and other software packages (e.g., Metamorph, IDL, Matlab) have built-in functions and routines for automation. Handling of time-lapse image series may require large amounts of free working memory; therefore, data evaluation should be performed on computers equipped with ≥8 GB RAM.

Two designs of genetically encoded biosensors are commonly used. One class includes monomolecular FRET-based biosensors like cGi-type cGMP sensors that contain donor and acceptor fluorophore in a single polypeptide chain. Other biosensors, such as FlincG-type cGMP sensors (Nausch et al., 2008; Bhargava et al., 2013) or Ca²⁺ sensors of the GCaMP family (Tian et al., 2009; Chen et al., 2013), contain circularly permutated green fluorescent protein (cpEGFP). Monomolecular FRET-based biosensors can be measured by simultaneous acquisition of donor and acceptor emission upon donor excitation. In camera-based systems, beam splitters should be used to direct donor and acceptor fluorescence to two separate cameras or onto the halves of a single camera chip. In scanning microscopes, emitted light is detected by two separate photomultiplier tubes. Emission of cpEGFP-based biosensors can be measured at a single wavelength. Potential advantages and disadvantages of FRET- and cpEGFP-based biosensors are discussed in more detail in Newman et al. (2011), Kaestner et al. (2014). Ideally, any change in biosensor fluorescence should originate from ligand binding, but not from other processes that affect fluorescence intensity recordings, such as motion of the target structure. In this regard, ratiometric FRET-based biosensors are advantageous over cpEGFP-based biosensors, because not a single fluorescence intensity (e.g., EGFP) but the ratio between two simultaneously recorded fluorescence intensities (e.g., CFP/YFP) is taken to estimate ligand binding to the sensor. Artificial intensity changes will not affect quantification as long as both donor and acceptor fluorescence are affected to the same extent. This should be the case, for instance, upon movement of the imaged tissue due to vasodilation, heartbeat, blood flow, or breathing.

Biosensor-expressing transgenic mice are important tools for IVM. Expression of the biosensor must be appropriate for its detection with sufficient signal-to-noise ratios. Initially, we used random transgenesis to generate cGi500-expressing mice, which is a relatively simple method in terms of transgene assembly and mouse generation. However, it often requires screening of many founder lines to identify lines with stable and sufficient transgene expression. Using an appropriate promoter is critical for successful generation of such a transgenic mouse line. Unfortunately, random integration of the transgene into the genome often leads to unpredictable and unwanted expression patterns. We used the 445-bp promoter fragment of the Transgelin/SM22 gene (Li et al., 1996) to generate SM22-cGi500 mice with SMC-specific cGi500 expression (Figure 1C upper). Among 11 founder lines, one mouse line showed stable SMC-specific cGi500 expression (data not shown). The expression level of cGi500 in SMCs derived from aorta, colon, and bladder is heterogeneous with ~25% of the cells showing relatively strong cGi500 expression (Thunemann et al., 2013b). However, SM22-cGi500 mice do not express detectable cGi500 levels in cremaster arterioles (Thunemann et al., unpublished data), so that these mice cannot be used for cGMP imaging in the cremaster microcirculation.

To generate mice with strong and homogeneous cGi500 expression, we used targeted mutagenesis to integrate a Cre-dependent cGi500 expression cassette into the R26 gene locus (Figure 1C middle). This gene locus is known for transcriptional accessibility in most if not all cell types throughout pre- and postnatal development (Soriano, 1999). Although technically more challenging, generation of R26 knock-in mice became a popular approach to express fluorescent proteins including biosensors (Zariwala et al., 2012; Abe et al., 2013; Batti et al., 2013). To increase biosensor expression above the relatively low levels that are usually achieved with the endogenous R26 promoter in adult mice, the strong chicken actin/β-globin (CAG) promoter (Niwa et al., 1991) can be integrated into the R26 knock-in transgene. Activation of sensor expression is achieved by breeding mice carrying the R26 knock-in transgene to cell type-specific Cre-transgenic mouse lines. This strategy should allow one to direct high and homogeneous levels of biosensor expression to almost any cell type of interest. We have generated R26 knock-in mice carrying the Cre-dependent R26-CAG-cGi500(L2) transgene (Thunemann et al., 2013b). We observed strong and tissue-specific cGi500 expression in mice generated by breeding R26-CAG-cGi500(L2) mice with tissue-specific Cre mice (Figure 2D, and Wen, Thunemann, Feil, unpublished results). This finding is in line with numerous reports on other Cre-inducible R26 knock-in transgenes (Casola, 2010). Furthermore, we transfected embryonic stem cells carrying the R26-CAG-cGi500(L2) transgene with a Cre expression plasmid to generate the permanently activated R26-CAG-cGi500(L1) transgene (Figure 1C lower). Indeed, mice established from R26-CAG-cGi500(L1) embryonic stem cells showed strong cGi500 expression in virtually all tissues and cell types examined (Thunemann et al., 2013b). Importantly, use of the CAG promoter resulted in high cGi500 expression levels and sufficient fluorescence intensities for in vivo imaging experiments (Figures 2, 3, 5). We conclude that the wide use of the Cre/lox system and the availability of hundreds of Cre mouse lines combined with the positive experience using the R26 locus as knock-in target, the generation of biosensor-expressing mice via Cre/lox-activatable R26 knock-in transgenes is an attractive alternative to random transgenesis.

When working with biosensor-expressing cells, tissues, and mice, it is important to consider potential side effects of sensor expression. A biocompatible sensor should not interfere with normal physiology. The cGi500 sensor contains the cGMP-binding sites of cGK type I, but not its catalytic kinase domain nor its N-terminal region that interacts with various proteins (Figure 1B). Although we cannot formally exclude that cGi500 acts as an intracellular cGMP buffer, we did not observe any adverse effects of its expression on the health of cGi500-transgenic mice nor on the functions of cells isolated from these mice (Thunemann et al., 2013b). In this context, it is interesting to note that mice expressing a FRET-based Ca²⁺ sensor in a ubiquitous manner show only relatively mild phenotypes (Direnberger et al., 2012). However, when mice express a biosensor with enzymatic activity, for instance, myosin light chain kinase activity in MLCK Ca²⁺ indicator mice (Isotani et al., 2004), it might be important to exclude interference of the sensor with normal physiology by reducing its expression level as much as possible.

The mouse cremaster is a relatively transparent skeletal muscle (~200 μm thick) that forms a pouch around the testis and contains resistance-type vessels. The first in vivo study of vascular responses within the cremaster was published in 1964 (Grant, 1964). Most current protocols for cremaster IVM, including those presented herein, follow the procedure described by Baez (1973), which includes the opening of the cremaster muscle, its detachment from the testicle, and the spreading of the cremaster to a flat sheet (Figure 2A). A number of studies used biosensors to visualize Ca²⁺ in endothelium and SMCs of cremaster arterioles. However, the functional role of cGMP in the cremaster microcirculation has so far been studied only indirectly, for instance, by analyzing vasodilation or vascular permeability in mouse lines with mutations of components of the cGMP signaling pathway (Koeppen et al., 2004; Chen et al., 2012).

Using the cGi500 sensor, we were able to visualize changes in cGMP concentrations in cremaster arterioles in vivo (Thunemann et al., 2013b). Experiments were performed with the R26-CAG-cGi500(L1) mouse line showing ubiquitous cGi500 expression. Accordingly, we observed cGi500 expression in the vessel wall and in the striated muscle of the cremaster itself, in which vessels are embedded, as well as in circulating blood cells (Figures 2B,C,E, and data not shown). Application of the NO-releasing drugs sodium nitroprusside and DEA/NO, as well as adenosine and acetylcholine induced dilation of cremaster arterioles (Figure 2B, and data not shown). However, attempts to detect DEA/NO-induced cGMP signals in arterioles of the intact cremaster preparation (as shown in Figure 2B) failed due to strong cGi500 fluorescence in striated muscle (data not shown). After surgical removal of the skeletal muscle from arterioles (Figures 2C,E), robust CFP/YFP ratio changes indicating reversible cGMP elevations could be observed in the vessel wall upon superfusion of DEA/NO, while adenosine did not cause a detectable cGMP elevation (Figure 2F). However, because vessels that were isolated from surrounding tissue apparently lost their basal tone and did not dilate upon drug superfusion, this experimental setup did not allow us to correlate cGMP levels with the extent of vasodilation. Moreover, because of ubiquitous cGi500 expression in R26-CAG-cGi500(L1) mice, it is difficult to allocate cGMP signals to specific vascular cell types. Thus, this model has two limitations related to ubiquitous sensor expression. Firstly, FRET recordings are disturbed by fluorescence from striated muscle and, secondly, it is difficult to pinpoint cGMP signals to specific vascular cell types. Both problems could be avoided by cell type-specific cGi500 expression within distinct cell types of the cremaster microvasculature. Therefore, the Cre-responsive R26-CAG-cGi500(L2) mouse line is currently bred with endothelium- or SMC-specific Cre lines (Figure 2D, and data not shown). Furthermore, by using other Cre-transgenic lines, it should be possible to visualize cGMP signals in various other cell types, for instance, in leukocytes during inflammatory processes or in platelets during adhesion to the vessel wall and thrombus formation.

Advantages of the open cremaster model for IVM are its outstanding optical quality and relatively minor motion artifacts. The good accessibility of arterioles within the exteriorized cremaster allows direct drug delivery via superfusion or locally via glass pipettes. Applying drugs only to the tissue of interest and not systemically through intravenous infusion can be an advantage. Continuous superfusion almost instantaneously removes drugs, allowing for studies with many successive drug applications, for example, to generate concentration-response curves or to compare effects of different drugs. On the other hand, cremaster IVM requires supported breathing, continuous delivery of anesthetics, temperature maintenance with external heat sources, and constant superfusion with gassed salt solution. The time required to prepare an animal for imaging is around 1 h, and the time for experiments is limited to about 6 h per animal.

The DSC model (Figure 3A) has been established to study tumors implanted below the chamber window. In addition, this method allows imaging of the tumor environment, for instance, to monitor changes in vascularization during tumor growth (Brown et al., 2001). IVM studies with DSCs were initially performed with bright field, epifluorescence, or confocal microscopy, but nowadays, MP microscopy is primarily used. MP excitation has several advantages over single photon excitation: out-of-focus fluorescence above or below the focal plane is avoided, excitation with pulsed infrared lasers allows for larger penetration depths (up to ~0.5 mm tissue), and the infrared light causes comparatively mild damage to fluorescent molecules and surrounding tissue (Helmchen and Denk, 2005).

We implanted DSCs into R26-CAG-cGi500(L1) mice and used them for MP-FRET/cGMP imaging (Thunemann et al., 2013b). Subcutaneous vessels were readily visible and, in contrast to the cremaster model, we were able to identify regions where surrounding tissue did not interfere with the detection of FRET/cGMP signals from vessel walls (Figures 3B,E). Mice received vasoactive drugs as bolus injections via a tail vein catheter. In our experiments, 3–5 drug applications were performed over 15–30 min. In studies with cultured cells and cremaster arterioles, the initial steps in image evaluation are ROI definition and subtraction of a dim/non-fluorescent background region (see Figure 2E). When we subtracted intensities from a background region, vasodilation strongly confounded the analysis of CFP/YFP ratio signals within the vessel wall (Figure 5B left). Vasodilation led to an increase in the proportion of dim/non-fluorescent vascular lumen within the ROI that resulted in strong and parallel reductions of CFP and YFP fluorescence. These drops were much stronger than any intensity changes caused by cGMP binding to cGi500 (Figure 5B left). To overcome this problem, we performed image segmentation with a dynamic binary mask (Figure 4). Here, instead of subtracting a background region, structures with a brightness below a certain threshold are set to non-numeric values (“NaN”). These structures actually can be fluorescent; however, their brightness is lower compared to the target structure, i.e., the vessel wall. To compare subtraction of a background region and image segmentation, we applied both methods to cGi500 recordings in cultured cells. Subtraction of a background region results in greater changes of the baseline-normalized CFP/YFP ratio (ΔR/R) as compared to ratio changes obtained through image segmentation (Figure 5A, compare left and right traces), most likely as a direct result of different background treatment. However, when the dynamic binary mask is used, varying proportions of dim/non-fluorescent background within the ROI do not affect noise levels (Figure 5A, compare right traces from small and large ROI), which can be an advantage over the background subtraction method (Figure 5A, compare left traces from small and large ROI). Consequently, ROIs can now be defined to contain large and/or variable proportions of dim/non-fluorescent background without affecting signal quantification. Using this approach with DSC recordings markedly reduced the confounding drop of CFP and YFP fluorescence upon vasodilation that was observed when a background region was subtracted (Figure 5B). This alternative evaluation strategy eventually allowed us to detect cGMP transients in the vascular wall during vasodilation in vivo. Control injection of saline caused a transient vasodilation, most likely due to an increase in blood volume (data not shown) that was not associated with CFP/YFP ratio changes [Figure 5B, right, application (1)]. Injection of DEA/NO caused vasodilation (Figure 3E) that was associated with a clear ΔR/R increase indicating an increase of cGMP levels [Figures 3E and 5B, right, applications (2) and (3)]. Relative vessel diameters were extracted from dynamic binary masks via kymographs (Figure 3D). The extent of NO-induced cGMP elevations measured with the cGi500 sensor correlated well with the extent of vasodilation (Figure 3E). Thus, data evaluation with the image segmentation method made it possible to observe cGMP transients in dilating vessels (Figures 3B–E).

The highest CFP/YFP ratio changes observed in the DSC model were relatively moderate (~10%) compared to changes recorded in the cremaster microvasculature (~50%). Although we cannot exclude biological reasons, this discrepancy is likely caused by different ways of sensor excitation (single photon vs. MP) and data evaluation (treatment of background fluorescence). While image segmentation of DSC recordings was performed without background subtraction, non-specific fluorescence from a background region was subtracted in cremaster recordings. As Figure 5 clearly shows, the evaluation strategy strongly affects ΔR/R values. Therefore, quantitative comparisons are only feasible between experiments evaluated with the same strategy. Note that omitting background subtraction during image segmentation may introduce variability if the background varies considerably between experiments. We are working on further refinements of our evaluation approach and expect substantial improvements with the use of mice expressing cGi500 in a cell type-specific manner (see Figure 2D). Another difference was that single-photon excitation was used for cremaster imaging and MP excitation in the DSC model. For the latter, we tested excitation wavelengths between 820 and 880 nm and we have chosen 850 nm for best imaging results (data not shown). Due to overlapping excitation spectra of CFP and YFP and the non-linearity of MP excitation, MP lasers likely cause direct excitation of the acceptor fluorophore, YFP (Zipfel et al., 2003; Svoboda and Yasuda, 2006). We assume that direct YFP excitation as well as bleed-through of CFP emission into the YFP channel were, at least in part, responsible for the relatively low ΔR/R values observed during MP imaging in the DSC model. In order to apply correction algorithms for tissue-intrinsic background fluorescence or spectral bleed-through, or to perform spectral unmixing, samples containing either no, or only the donor or the acceptor fluorophore need to be recorded (Van Rheenen et al., 2004; Thaler et al., 2005; Thaler and Vogel, 2006; Zhang et al., 2010; Zeug et al., 2012; Mauban et al., 2014). In addition, alternative methods to estimate FRET efficiencies such as fluorescence lifetime imaging microscopy (FLIM) are available (Yasuda et al., 2006; Levitt et al., 2009; Fruhwirth et al., 2011). However, recording appropriate donor- or acceptor-containing specimens, or using FLIM techniques with non-optimized FRET pairs can be challenging under in vivo conditions (Niesner and Hauser, 2011; Geiger et al., 2012; Kamioka et al., 2012).

In the DSC model, animals can be analyzed repeatedly for up to 5 weeks, which is an advantage over the cremaster model. The DSC model allows longitudinal studies in individual animals, for example, to follow disease progression. Thereby, animal numbers can be reduced, while statistical power is maintained. Another advantage over the cremaster model is that tissues within the DSC are not acutely touched when imaging is performed. Under the conditions described here, test drugs are delivered intravenously. Due to volume restrictions, this limits the number of drug applications during an imaging session to about three to five. Depending on the experimental question, systemic drug delivery might complicate data interpretation. If so, the DSC model can also be used for direct topical application of test compounds. To this end, the coverslip is carefully removed to visualize vessels and other structures while applying solutions that contain test compounds (Fukumura et al., 1997). A number of protocols exist to implant windows at different sites of the body (Jain et al., 2010; Ritsma et al., 2012). This flexibility allows studying cGMP and other signaling molecules in diverse vascular beds, such as cerebral vessels and the vasculature of abdominal or thoracic organs. For these experiments, the setup has to reach sufficient scanning speeds to resolve the kinetics of the biochemical/physiological processes under investigation. Acquisition speed is especially important when moving cells (e.g., in the blood) or structures (e.g., the beating heart) are analyzed. Here, camera-based systems have an advantage as they typically allow faster acquisition rates (Stephens and Allan, 2003). For some IVM applications, spinning disk systems might be a suitable alternative to MP systems (Jenne et al., 2011), particularly when combined with MP excitation, thereby achieving large penetration depths and fast acquisition rates (Shimozawa et al., 2013). Alternatively, resonant galvanometer mirror scanning enables MP microscopy at high speeds to monitor fast biological processes (Kirkpatrick et al., 2012).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.