In fluorescence image-guided surgery (FIGS) research and practice, fluorophores operating in the NIR region are most often used, because of superior imaging characteristics compared to light in the visible spectrum. More specifically, the lower background and improved tissue penetration -a consequence of diminished scattering, endogenous fluorophore absorption and autofluorescence- makes NIR light more suited (Weissleder and Ntziachristos, 2003; Kovar et al., 2007). The NIR domain applicable for in vivo optical imaging can actually be separated in two distinct areas: the NIR-I (650–900 nm) and NIR-II window (1,000–1,700 nm).

While the use of NIR fluorescent dyes has led to increased signal-to-background ratios and better overall depth penetration up to several millimeter in tissue, deeper lying, and hidden lesions would still be missed. To this end, the combination of NIR fluorescence and nuclear techniques in a single tracer may be attractive. The nuclear component can be used for pre-operative imaging, as well as for intra-operative guidance using a gamma-detecting probe (coarse navigation), at which point, fluorescence would provide high resolution visual guidance for precise resection (Van Leeuwen et al., 2016). Integration of both modalities onto a tracer can occur in several ways. Conjugation of the fluorophore and chelator for radioactive labeling at different positions is one option, either through co- or sequential incubation of the targeting ligand with the chosen fluorophore and chelator (Lütje et al., 2014b). Several preclinical hybrid single photon emission tomography (SPECT)/fluorescent tracers combining IRDye800CW and ¹¹¹In have been prepared in this manner, including the antibodies BIWA against CD44v6 (Odenthal et al., 2018) (Figure 3A), MN-14 and Labetuzumab against carcinoembryonic antigen (CEA) (Rijpkema et al., 2014; Hekman et al., 2017), and D2B against prostate specific membrane antigen (PSMA) (Lütje et al., 2014a). This technique has also been applied clinically, with the carbonic anhydrase 9 (CAIX) specific antibody Girentuximab (Hekman et al., 2016, 2018). The same manner of conjugation was used to prepare preclinical positron emission tomography (PET)/NIR tracers by labeling the anti-CD146 antibody YY146 with IRDye800CW and ⁸⁹Zr (Hernandez et al., 2016), the anti-prostate stem cell antigen (PSCA) minibody A11cMb with a Cy5.5 dye and ⁸⁹Zr/¹²⁴I (Tsai et al., 2018), or the anti-prostate stem cell antigen (PSCA) diabody A2cDb with IRDye800CW and ¹²⁴I (Zettlitz et al., 2018).

In another integration strategy, the fluorophore and the radiolabel can be combined into a single structure. This is advantageous since a more homogenous tracer is obtained, whose biodistribution can be accurately determined via radioactivity. Furthermore, this is often the only possibility when working with small peptides or molecules, as multiple attachment points may not be available. However, such structures are more complex to design, as opposed to the more straightforward separate labeling procedure. A urokinase-type plasminogen activator receptor (uPAR) specific antibody, called ATN-658, was conjugated with both ZW800-1 and a chelator for ¹¹¹In labeling through a single structure called a “multifunctional single attachment point” (MSAP) that bore both labels (Boonstra et al., 2015c, 2017). The MSAP-strategy was also used to combine an ¹¹¹In/chelator complex and Cy5.5 fluorophore for conjugation to either an integrin binding RGD peptide or the anti-TZ14011 peptide targeting chemokine receptor 4 (CXCR4) (Kuil et al., 2011; Buckle et al., 2013). Additional dual labeling structures have been explored, for example, the construction of a dual-modality linker combining Cy5 and ¹⁸F on the diabody A2cDb (Zettlitz et al., 2019), or for the preparation of a CH1055-⁶⁸Ga-RGD tracer (Sun et al., 2018), ⁶⁸Ga-IRDye650/IRDye800CW conjugated bombesin antagonists (Zhang et al., 2017a; Li et al., 2018), or a Cy5-labeled anti-PSMA KuE peptide with a chelator for ⁶⁸Ga or ¹⁷⁷Lu labeling for PET or SPECT/targeted radionuclide therapy (TRNT), respectively (Schottelius et al., 2018).

Fluorophores (and by extension, hybrid labels) can be linked to the targeting moiety using a range of different conjugation chemistries. As opposed to small molecules and peptides that are generally more tolerant toward harsher conditions such as the use of organic solvents and high temperatures, the conjugation of fluorophores to larger protein-based moieties is restricted to softer conditions. The most commonly applied strategy to this end is the formation of an amide bond between primary amines, and an activated carboxylic acid (often activated via a N-hydroxysuccinimide (NHS)-ester, Figure 1B) (Toseland, 2013). While every polypeptide contains a primary amine at its N-terminus, this method is often applied for random conjugation on lysine residues. In larger protein structures, the abundancy of accessible lysines makes the degree of conjugation and the exact positioning of fluorophores difficult to control, resulting in a heterogenous tracer mixture. Furthermore, having many fluorophores per targeting ligand can lead to self-quenching effects (reduced brightness) and if the fluorophores are conjugated in or close to the antigen-binding region, they may negatively impact the binding capacities of the tracer (Debie et al., 2017). This can especially be a problem for smaller compounds, though it has been demonstrated that even the binding efficiency of full antibodies can be affected (Vira et al., 2010; Szabó et al., 2018).

Site-specific labeling methods resulting in homogenous tracer preparations in which the fluorophore is only attached to a specific, pre-determined location situated away from the antigen-interaction site, may offer a solution. The formation of a thio-ether bond between a sulfhydryl residue and a maleimide functionalized fluorophore is an often-employed strategy in FIGS (Figure 1C). To this end, a free cysteine residue is incorporated in the targeting moiety's structure at the location of predilection, such as the C- or N-terminus of protein scaffolds, peptides and small antibody fragments (Sun et al., 2005; Massa et al., 2014). Antibodies and larger fragments such as minibodies and diabodies can furthermore be partially reduced, freeing the sulfhydryl residues of more exposed disulfide bridges for conjugation (Olafsen et al., 2004; Sonn et al., 2016; Tsai et al., 2018; Zettlitz et al., 2018; Zhang et al., 2018). Many other site-specific methods have gained popularity over the last years, such as the azide-alkyne cycloaddition-click chemistry (Li et al., 2014), enzymatic modification of peptide tags (e.g., transglutaminase-, sortase-, or intein-mediated conjugation) (Massa et al., 2016a), incorporation of unnatural amino acids for biorthogonal conjugation or the use of linking peptides such as the SPY/SPYcatcher method (Massa et al., 2016b; Alam et al., 2018). While these techniques are promising for the future development of FIGS tracers, to date, they have not yet been applied in clinical tracers, and have seen only limited development in preclinical FIGS research.

Conjugation of a fluorescent dye to a targeting ligand can possibly alter the pharmacokinetics of the ligand. The physicochemical properties of the dye (e.g., hydrophilicity and global/local charge distributions), the degree of conjugation, as well as the chosen conjugation method all play a role in the general biodistribution of the FIGS tracer. It has been observed that after labeling with IRDye800CW, tracers most often exhibit more extensive non-specific targeting in vivo, including higher liver uptake. This is likely caused by the interaction of IRDye800CW with serum proteins. Further hepatic metabolization of the tracer can in addition be a source of intestinal background signal which could be contrast-limiting in abdominal applications (Choi et al., 2011, 2013; Debie et al., 2017). This is in accordance with findings for other heptamethine fluorophores, where a high number of negative charges on the fluorophore -as is the case for IRDye800CW- are seen to cause partial hepatobiliary clearance (Sato et al., 2016b) (Figure 2A). Other than merely causing increased background signals, interaction with serum proteins will furthermore cause a change in the imaging window of the tracer. While this effect is already observed for long-circulating antibodies (Sato et al., 2016b; Cilliers et al., 2017), the impact on smaller compounds that usually show fast pharmacokinetics and renal clearance can be considerably more pronounced, in some cases delaying the optimal imaging window from a few hours to up to a day or longer. Indeed, increased liver uptake and delayed optimal tumor-to-background ratios (TBR) were observed for randomly IRDye800CW conjugated nanobodies (Figure 2B), scFv's, Fab fragments and diabodies (Yang et al., 2014; Boonstra et al., 2015b; Debie et al., 2017; El-Sayed et al., 2018; Boogerd et al., 2019). Interestingly, the gravity of this effect has been found to also be related to the used conjugation chemistry. For instance, we found that the pharmacokinetics of site-specifically IRDye800CW or IRDye680RD conjugated nanobodies normalized compared to randomly-conjugated nanobodies (Figure 2B). These nanobody-based tracers could be used for high contrast imaging as early as 1 h post-injection (Kijanka et al., 2016; Debie et al., 2017). Of note, while not directly compared to their randomly-conjugated analog, affibodies expressing a C-terminal cysteine and conjugated at this position with IRDye800CW also enabled imaging at the expected early timepoints (Sexton et al., 2013; de Souza et al., 2017; Ribeiro de Souza et al., 2018).

With the aim of optimizing the pharmacokinetic profile of tracers, the chemical design of fluorophores can be systematically modified. Variations in the overall charge, total number of charges and hydrophilicity of a Cy5 dye have led to a hybrid cRGD-tracer with improved properties regarding non-specific background signals, renal elimination, and tumor uptake (Bunschoten et al., 2016). A similar strategy was applied for the optimization of the cRGD peptide as well as a KUE anti-PSMA peptide, with heptamethine fluorophores. However, conclusions of such studies on the most optimal design of fluorophores for peptide conjugation are difficult to generalize as the obtained results are also highly dependent on the physicochemical properties of the peptide itself (Choi et al., 2013; Bao et al., 2017). It is therefore becoming evident, that selection of an appropriate fluorophore and labeling strategy is an essential part of the fluorescent tracer design. For protein-based ligands such as antibodies, antibody-fragments, and scaffold proteins general properties and guidelines concerning the conjugation of certain fluorophores can be established, whereas in small peptides and molecules optimization in the design of new conjugates may be required on an individual basis.

In the current preclinical and clinical FIGS studies, a wide variety of biomarkers have been investigated as potential targets. Extracellular molecules enable targeting with non-cell penetrating ligands, and cell-membrane bound biomarkers are often preferred, as diffusion of secreted targets will reduce the sharpness of imaging. Biomarkers overexpressed by the cancer cells themselves are popular as they are already widely used for targeted therapeutic or nuclear imaging applications (Boonstra et al., 2016). Targets having shown high-level and quite general overexpression within specific cancer types include folate receptor α (FRα), which is mainly known for ovarian cancer (Kalli et al., 2008), CEA in colorectal and pancreatic cancer (van Oosten et al., 2011; de Geus et al., 2016), and PSMA in prostate cancer (Perner et al., 2007). One of the most highly overexpressed biomarkers is the breast cancer marker human epidermal growth factor 2 (HER2). While high contrast in HER2 overexpressing cells will likely be achieved, the fact that this marker is only overexpressed in a small subset of patients can be problematic in the sense that pre-operative screening would be required (Harbeck and Gnant, 2017). In order to generate more widely applicable FIGS tracers, the search for more general tumor markers is ongoing. Targets such as EGFR, carbonic anhydrase 9 (CAIX) and the epidermal cell adhesion molecule (EpCAM) have been suggested and explored to this end (Boonstra et al., 2015a, 2016).

Another strategy lies in the targeting of the tumor stroma, instead of only the tumor cells. This has several distinct advantages, namely, stroma is present on the tumor periphery/invasive edge and is part of a general physiological process, implying the presence of cancer-type independent markers, such as markers associated with infiltration of angiogenesis, tumor associated immune cells, or fibroblasts (Boonstra et al., 2015a). Examples of potential stromal targets include the soluble, cell-released vascular endothelial growth factor (VEGF-A) as marker for angiogenesis or the fibroblast activation protein alpha (FAP-α) for fibroblast targeting. Several markers also share stromal and tumor cell expression, and may as such be especially interesting. Examples are uPAR and integrins such as α_vβ₃ (Boonstra et al., 2015a; Hamidi and Ivaska, 2018). In all cases, constitutive expression in normal or fibrotic tissue could be a limiting factor as it will cause on-target, off-tumor uptake. Moreover, one should always take intra-tumoral heterogeneity into consideration (Pogue et al., 2018).

As for targeting moieties, a host of possible types of molecules are available. These range from full IgG antibodies, to antibody fragments, scaffold proteins, peptides, and small molecules. A comprehensive overview of the different targeting ligands that have been investigated preclinically and clinically in the context of FIGS is given in Table 1.

Antibodies are large (150 kDa) proteins that bivalently attach to their molecular target with high affinity and specificity. The large size of antibodies, as well as the presence of an Fc-domain bestow a long blood half-life and limit tumor penetration. Indeed, as is known from nuclear molecular imaging, the use of antibodies as targeted fluorescent contrast agents for FIGS necessitates extended waiting periods of several days to obtain sufficient target-specific contrast (Freise and Wu, 2015). Antibody-based FIGS tracers represent the largest group of currently evaluated compounds as monoclonal antibodies against a wide variety of targets are readily available and can be straightforwardly repurposed following fluorophore conjugation. For example, the therapeutic EGFR targeting antibodies Cetuximab and Panitumumab, and the anti-VEGF-A antibody Bevacizumab have been, after labeling with IRDye800CW, broadly explored in the context of FIGS in a range of pathologies such as head and neck cancer, colorectal cancer, breast cancer, pancreatic cancer, and glioblastoma (Heath et al., 2012; Day et al., 2013a,b; Gong et al., 2014; Korb et al., 2014; De Boer et al., 2015; Warram et al., 2015, 2016; Rosenthal et al., 2016, 2017; Moore et al., 2017; Gao et al., 2018a,b; Prince et al., 2018; Tummers et al., 2018b; van Keulen et al., 2019). Early phase clinical trials demonstrated the feasibility of intraoperative lesion detection and this was confirmed by histological evidence. The anti-CEA compound SGM-101, consisting of the novel antibody SGM-ch511 and the fluorophore BM104 (Figure 1A) (the NHS-activated version of BM104 is called BM105) is currently being evaluated in a phase III clinical trial for colorectal cancer. In addition, the potential of many more NIR-labeled antibodies (most often labeled with IRDye800CW) has been investigated in a variety of subcutaneous, orthotopic, and metastatic mouse models, including the anti-(human) EpCAM antibody 323/A3 (van Driel et al., 2016) or anti-uPAR antibody ATN-658 (Boonstra et al., 2017) for colorectal, head and neck, and/or breast cancer.

To achieve more rapid pharmacokinetics, antibody fragments, obtained through protein engineering, can be considered. Minibodies, consisting of two scFv-CH3 chains are, with 80 kDa, about half the size of full antibodies. Furthermore, they lack a functional Fc-region, preventing extended circulation due to FcRn recycling. A recent study demonstrated the use of an anti-(human) PSCA minibody, called A11, conjugated with IRDye800CW in different mouse models. Subcutaneous, orthotopic and intramuscular implanted tumors and tumor positive lymph nodes could be clearly visualized, even in knock-in models expressing the human homolog of PSCA (Zhang et al., 2018). Furthermore, fluorescence-guided resection of intramuscular tumors, revealed superior overall survival compared to conventional white light surgery (Zhang et al., 2018).

However, the size of minibodies does still put them above the cut-off value for renal clearance. Compounds smaller than 60 kDa are generally cleared via the kidneys, resulting in faster blood clearance, and correspondingly, diminished non-specific uptake. High contrast should thus be attained at an earlier timepoint, though at the expense of total tumor uptake (total fluorescent signal) (Freise and Wu, 2015). It has furthermore been shown that an inverse relation exists between the compound's size and its rate of tissue penetration, meaning that smaller fragments will distribute more quickly and more deeply throughout tumor and healthy tissues (Li et al., 2016; Xenaki et al., 2017). Antibody fragment types that are under investigation for FIGS are scFv's (25 kDa), diabodies (scFv dimer, 50 kDa), and Fab fragments (single scFv-CH3 chain, 55 kDa). The PSCA-specific diabody A2cDb, site-specifically conjugated to Cy5 was evaluated in murine subcutaneous and intramuscular prostate cancer models and optimal imaging contrast was achieved within 6 h post-injection (Sonn et al., 2016). The same diabody, site-specifically conjugated to IRDye800CW and directly labeled with ¹²⁴I, was recently used as bimodal tracer for the visualization of PSCA-positive pancreatic patient-derived orthotopic xenografts with PET and NIR fluorescence imaging. High-contrast imaging within 24 h or less was possible. It was furthermore demonstrated that the clearance pattern of the tracer was not significantly different from the iodinated-only form, indicating that the conjugation strategy used for IRDye800CW labeling did not impact the tracer's biodistribution or clearance (Zettlitz et al., 2018). This is in contrast to other studies with small antibody fragments that, as discussed above, often exhibit TBRs that are sufficiently high only between 24 and 72 h post-injection, as well as a significant amount of uptake in the liver. This has been found to be the case at least for an anti-HER3 diabody, anti-EpCAM-Fab fragment, and scFv's targeting CEA, EGFR and the tumor-associated glycoprotein 72 (TAG-72), all randomly labeled with IRDye800CW (Yang et al., 2014; Boonstra et al., 2015b; Alam et al., 2018; Gong et al., 2018; Boogerd et al., 2019).

In addition to the antibody fragments derived from conventional antibodies, the monomeric antigen-binding domains of heavy-chain only antibodies, so called single-domain antibodies or nanobodies, have been investigated for FIGS [recently reviewed in Debie et al. (2019)]. In preclinical studies, it has been shown that site-specifically IRDye800CW conjugated anti-HER2 and anti-CAIX nanobodies could achieve high contrast and specific imaging of tumor lesions within the first hours (1–4 h) after injection in murine breast and ovarian cancer models (Kijanka et al., 2013, 2016; Debie et al., 2017, 2018).

An advantage of antibodies and antibody-based tracers is that they can be generated in high through-put using established immunization strategies, and produced via fermentation in animal, yeast or bacterial cells (Ma and O'Kennedy, 2017). For tracer development, antibodies and antibody fragments represent a platform of compounds with a predominantly constant structure, except for variations in the antigen-binding loops prompting different specificities. This allows the definition of general properties and more versatile protocols for tracer preparation. Another type of platform technology are the non-antibody protein scaffolds that have a fixed backbone and are derived from different naturally occurring proteins. New antigen-specific protein scaffolds are selected from synthetic libraries and further engineered in vitro. As opposed to antibody-derived ligands, these compounds undergo no in vivo affinity maturation steps. Many different types of protein scaffolds have been described, but only affibodies, centyrins, and cystine knottins have been evaluated for FIGS so far.

Affibodies are small proteins of about 7 kDa that have been derived from the B-domain of Staphylococcus aureus protein A. They consist of a three helical structure that can, by mutation of 13 amino acids in the first two helices, bind different molecular targets. The affibody tracer ABY-029, directed toward EGFR and conjugated site-specifically with IRDye800CW via a C-terminal cysteine, is currently undergoing clinical translation for FIGS in glioma, head and neck cancer and soft tissue sarcoma. This tracer was first evaluated in an orthotopic human EGFR-expressing glioma model in rats (Figure 3B). The tracer was found to preferentially accumulate in the tumor, and this from 1 h post-injection onwards (de Souza et al., 2017). Other studies using fluorescently-labeled affibodies targeting EGFR and HER2 show similar optimal imaging timepoints (Zielinski et al., 2012; Tummers et al., 2018a).

Centyrins are 10 kDa fibronectin type III domains, derived from human tenascin, made up of 7 antiparallel β-strands. Binding affinity against different targets is generated via mutations in the connecting loops. The human/mouse cross-reactive anti-EGFR centyrin 83v2 labeled with the fluorophore S0456, was preclinically tested in subcutaneous lung cancer xenografts. Imaging could be performed within the first hours after injection and peak fluorescence was reached between 6 and 12 h pi. (Mahalingam et al., 2017).

Cystine knot peptides (3–4 kDa) are built around a naturally occurring motif, where the peptide forms a type of knot stabilized by three cysteine-bridges around a core of antiparallel β-strands. Through mutations, the loops that connect the strands can confer specific binding capabilities to the knottin peptide. The knottin peptide Tozuleristide, which consists of the scorpion venom derived ligand chlorotoxin conjugated to ICG, is currently undergoing clinical translation. This probe was able to distinguish tumor from normal tissue in resected human breast cancer specimens between 1 and 26 h post-injection (Dintzis et al., 2018). Tozuleristide is a naturally occurring knottin, whose exact specificity is not fully known, though MMP-2 and AnnexinA2 have been observed as targets. Alternatively, the knottin peptide R01-MG, with engineered affinity for the human and mouse α_vβ₆ integrin, has been tested pre-clinically. After conjugation with IRDye800CW, it was evaluated in subcutaneous, orthotopic, and spontaneous tumor models of pancreatic cancer. Fluorescent signal in a subcutaneous tumor became significantly different from a control knottin at 4 h post-injection, and maximal TBR was reached at 20 h post-injection. Cancer lesions, including small metastases could be clearly discerned from normal tissue in the orthotopic and spontaneous models (Tummers et al., 2018a).

Other than these scaffold proteins, non-scaffold peptides and small molecules can also be used as tracers in FIGS. The advantage of peptidic and small molecule-based tracers is that they can be produced at reduced cost compared to the other types of tracers. Target specific peptides and small molecules can be generated by rational design, or compounds with pre-existing affinities for the envisaged target, such as natural ligands, can be repurposed as fluorescent tracers. However, their structure and physicochemical properties vary greatly as they do not share a common backbone. General pharmacokinetic characteristics are hard to define, though peptide/small molecule-based tracers are expected to be small enough to attain at least partial renal clearance, depending on their hydrophobicity.

By far the most investigated peptide, is (cyclic) RGD, which targets a group of integrin receptors through a naturally occurring binding motif. cRGD peptides have been conjugated with a variety of different fluorophores, including Cy5.5, ICG, IRDye800CW and ZW800-1. After evaluation in subcutaneous and orthotopic tumor models of gastric, head and neck, pancreatic (Figure 3C) and colorectal cancer, tumor-to-background values were generally optimal between 4 and 24 h post-injection, depending on the fluorophore used (see above) (Choi et al., 2013; Cheng et al., 2016; Handgraaf et al., 2017). Other promising peptides for further development are tracers against for example PSMA (Wang et al., 2014; Bao et al., 2017; Baranski et al., 2017; Kularatne et al., 2018; Schottelius et al., 2018), the gastrin releasing peptide receptor (GRPR) (Cai et al., 2013; Suganya et al., 2016; Zhang et al., 2017a; Li et al., 2018; Xu et al., 2018) and uPAR (Yang et al., 2014; Christenen et al., 2017; Kurbegovic et al., 2018). Folate-receptor specific compounds were the first small molecules to be used in clinical trials for FIGS. The compound EC17 was constructed by conjugating folic acid with fluorescein isothiocyanate (FITC) (wavelength of maximal emission and excitation at 495/515 nm, outside the NIR region) (van Dam et al., 2011). EC17 was initially tested during debulking surgery in the case of late stage ovarian cancer, where FRα is widely expressed. To improve imaging contrast and detection of deeper located lesions, folic acid was labeled instead with the NIR dye S0456, to form OTL38. Additional clinical studies with both tracers have since been performed in lung, ovarian, and renal cancer. The surgical procedure is generally performed between 2 and 6 h post-injection of either tracer (Hoogstins et al., 2016; Keating et al., 2017; Mahalingam et al., 2018b; Predina et al., 2018).

SH holds a patent in camelid single-domain diagnostics and therapeutics. The remaining author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.