Identification of a suitable molecular target can be the most challenging part of the process to obtain a specific molecular imaging agent. Today, modern biotechnology has changed the odds somewhat as screening techniques have become more sophisticated and advanced. Technologies such as recombinant techniques, genomics and proteomics, robotics, and high-throughput screening have all contributed to the increased effectiveness of this aim. High-throughput screening is particular valuable as thousands of samples can be evaluated simultaneously and quantitatively (Kelly et al. 2009; Weissleder et al. 2005). We have shown that high-throughput screening can be performed both rapidly and reliably on many thousand samples utilizing a clinical MR scanner (Fig. 2) (Grimm et al. 2004; Högemann et al. 2002). Genomic screening has become widely available to enable discovery of suitable targets within tumors. For example, the gene expression profile of K-ras-activated lung tumors in a mouse model has been characterized, revealing a profile similar to that of human lung adenocarcinomas (Sweet-Cordero et al. 2005). The gene expression profile of K-ras tumors showed that several cathepsin proteases were strongly up-regulated in lung tumors compared with normal lung tissue. These proteases are also overexpressed in some patients with lung cancer, and their expression has been associated with tumor invasion and poor prognosis (Joyce 2004; Schweiger et al. 2000). We demonstrated subsequently that cathepsin activity could be used as a means of detecting these tumors with noninvasive in vivo optical imaging (Fig. 3) in the same mouse model (Grimm et al. 2005).

Depending upon the method of target detection, a suitable ligand might be discovered at the same time, for example when a library of small molecules is used to screen for binding to the designated target. Small molecules can be easily coupled to nanoparticles as carriers and incubated with the target, which can be either cellular or molecular (or by computational modeling, in silico) (Grimm and Scheinberg 2011). The resulting best hit can than be pursued further for in vivo studies. Screenings like this have identified the amino acid glycine as a ligand, specifically binding to activated macrophages, which are often associated with tumors (Weissleder et al. 2005; Leimgruber et al. 2009). Another frequent approach is to identify a ligand peptide with phage display, with can be performed to achieve high specificity binding peptide sequences, in vitro or in vivo (Kelly et al. 2006). Once the most suitable ligand is determined based on the sensitivity, specificity and biodistribution of the ligand a label has to be attached (Weissleder and Mahmood 2001; Massoud and Gambhir 2003). For nuclear imaging, a variety of appropriate radioisotopes with different properties are available and already in clinical use. Especially promising is the novel PET tracer Zirconium-89 (⁸⁹Zr). The unique physical decay properties and the energy of ⁸⁹Zr (t _1/2 = 78.43 h, β⁺ = 22.3 %, E _β+,max = 901 keV, E _γ = 909 keV) are ideally suited for targeted imaging probes. The greater half-life provides copious activity at the required circulation times for optimal targeting to diseases sites (Holland et al. 2010a; Ruggiero et al. 2011). The longer half-life even enables imaging at later time points, including 120-h post injection, an imaging window difficult to achieve with the shorter lived PET nuclides ¹⁸F and ⁶⁴Cu (Holland et al. 2010b, c). In MRI, paramagnetic metal ions such as chelated Lanthanides (Grimm et al. 2004; Aime et al. 2002; Montet et al. 2006; Tannous et al. 2006) or iron oxide nanoparticles (Grimm et al. 2004; Montet et al. 2006; Tannous et al. 2006) can be used for labeling. In optical imaging, organic fluorochromes or anorganic compounds such as quantum dots for fluorescence optical imaging are available but mostly exclusively for preclinical studies (Massoud and Gambhir 2003; Weissleder and Ntziachristos 2003), however, there was a recent study with the fluorochrome FITC targeted to folate receptor for intraoperative imaging (van Dam et al. 2011). Perfluorocarbon emulsion nanoparticles and acoustically reflective microbubbles are currently under investigation as ultrasound contrast agents (Dayton and Ferrara 2002; Schutt et al. 2003). For CT, various attempts have been made to develop targeted agents, however, the low sensitivity of CT is a considerable challenge. To obtain a decent contrast in CT with conventional agents, 300 mg/ml or more iodine needs to be present. Thus, considerable effort has been made to develop agents based on particles composed of elements with a high atomic number. These are mainly bismuth (Fig. 4) (Rabin et al. 2006) and gold nanoparticles (Kattumuri et al. 2007), which are particularly interesting as they can also be used for optical approaches, including Raman imaging (Zavaleta et al. 2009; Mu et al. 2010), which has also been proposed for intraoperative imaging (Kircher et al. 2012), however, none of these particles are so far clinically approved.

In order to be useful as an imaging agent the Signal-to-Noise ratio (SNR) must be sufficiently high. To be detectable in living subjects, imaging probes must also have the ability to reach the intended target at a sufficient concentration within a useful time frame while at the same time be adequately cleared from the circulation and other organs (Massoud and Gambhir 2003). This makes radioactive iodine such an ideal agent as it achieves uniquely high tumor to background ratios for well-differentiated thyroid cancers (Frangioni 2009). Molecular imaging agents are subject to pharmacokinetic precepts: distribution, absorption metabolism, and elimination. Depending on its localization the target is more or less accessible (Fig. 5). To leave the vascular system, the vascular wall needs to be traversed. Imaging agents below a certain cutoff of molecular weight can leave the vascular bed easily, especially in disrupted endothelial layers as occurs within tumors. However, extravasation can be challenging, especially in the normal brain when the blood–brain barrier is intact (Begley and Brightman 2003). The interstitial space needs to be navigated to reach an interstitial target or targeted cells (Grimm and Scheinberg 2011). Additional barriers are present if the target is located in the cytoplasm of the target cells or even the nucleus, with the cell membrane and nuclear architecture (including the nuclear membrane proteins such as histones) representing additional barriers (Weissleder and Mahmood 2001; Massoud and Gambhir 2003).

The Enhanced Permeability and Retention (EPR) effect (certain sizes of agents tend to accumulate in tumors) can be a blessing or curse as it can either help to increase the binding or to increase nonspecific signal with increased background. The “fate” of an imaging compound in vivo depends on its properties such as molecular size, charge, solubility, avidity, affinity and (in case of nanoparticles) also loading density (Choi et al. 2010). The main process by which the body eliminates “foreign” substances is excretion. Hydrophilic materials are eliminated by the kidneys, whereas lipophilic compounds usually are metabolized by the liver, the principal organ for metabolism (Ratain et al. 1996).

To visualize biological processes in vivo, it requires a very strong imaging signal per unit level of target and probe interaction due to the low amounts of target. Various signal amplification strategies have, therefore, been implemented. This is especially important if utilizing less sensitive imaging modalities or when imaging on the DNA- or mRNA-level where the number of targets per cell may be limited to a few copies. For example, there are only two copies of an individual gene, which is buried deep within the cell behinds membranes and nuclear proteins. In this respect, the strategy to image the actual product of a gene, the protein or its function (“downstream imaging”) is much more viable because of a much larger number of targets per cell, thus creating a natural amplification (Fig. 6) (Weissleder and Mahmood 2001; Massoud and Gambhir 2003). The concentration and the SNR of the imaging agent can be improved by pretargeting (Barbet et al. 1998), by avidin–biotin amplification (Rosebrough 1996; Goodwin et al. 1984), by improving kinetics (Choi 2010), or by subsequent trapping of converted ligands (Tjuvajev et al. 1996; Gambhir et al. 2000). The binding avidity to a particular target can be increased by coupling more than one ligand to the probe, creating a multivalent agent (Montet et al. 2006; Tassa et al. 2010). Another strategy is to design activatable agents (Thorek and Grimm 2012 ). Here, the imaging probe provides little or no signal in its native state and changes its physical properties after interaction with its target, which is typically an enzymatic activity (Fig. 1). Many variants of these probes have been designed for optical imaging, based on “quenching”, which describes energy transfer between neighboring fluorochromes resulting in a considerable decrease in fluorescence (Fig. 7a) (Elias et al. 2008). A classic example is the imaging of proteases such as matrix metalloproteinases (Bremer et al. 2001) or cathepsins (Fig. 3) (Grimm et al. 2005), which are all associated with tumors. Here, a peptide sequence is cleaved by the endopeptidases (Fig. 7a). Another examples are the molecular beacons, which are single stranded strands of DNA, forming a hairpin loop in themselves and thus bringing the two fluorochromes at the 3’ and 5’ end close enough together for quenching. Upon annealing to the complementary target sequence (e.g. mRNA of interest) the loop linearizes and dequenching occurs. Smart probes have been developed for MRI as well. These are based on changes of magnetic relaxation upon assembly of paramagnetic or superparamagnetic particles, resulting in a higher T1 or lower T2 signal, respectively (Fig. 7b, c) (Bogdanov et al. 2002; Perez et al. 2002). This concept has been realized as the magnetic relaxation switch with which iron oxide nanoparticles have been used to measure enzymatic activity in vitro, including proteases, restriction-endonucleases, and telomerase activity (Fig. 2 and 7d) (Grimm et al. 2004; Perez, 2002). Other activatable agents have been realized using modified Gadolinium chelates, which permitted enzyme-mediated polymerization. The mechanism of polymerization is based on the oxidation of substrates in the presence of the enzyme (such as myeloperoxidase) and hydrogen peroxide to yield free radicals, which give rise to oligomers with an increased r₁ relaxivity due to a decreased tumbling time of the larger molecule (Fig. 7c) (Tannous et al. 2006; Querol et al. 2005). In an alternative scheme, enzyme-specific activation has been accomplished by “unmasking” the free gadolinium water coordination site upon galactosidase-mediated cleavage of a galactose moiety that blocked the access of bulk water to the coordination side, and thus decreased the relaxivity (Fig. 7b) (Louie et al. 2000).