Methods, such as colorimetry (78), weight measurements (79), atomic force microscopy (80–82), and immunostaining (83–85), to image collagen structures risk their destruction and are limited by their in vivo translatability. The dorsal skinfold (window) chamber setup allows optical measurements by replacing skin with glass but may lead to collagen structural changes due to inflammation and mechanotransduction by the glass (86). The advances in ultrafast optics significantly improved the ability to image fibrillar collagen (the predominant structural protein in mammalian ECM and mostly type I) by second-harmonic generation (SHG) or third-harmonic generation (61) microscopy in vivo and ex vivo (87–91). The strength of SHG imaging is its specificity to fibrillar collagen (62, 87, 89, 92) and that it can be fairly easily combined with other optical imaging methods, in vivo (Figures 3–5A) and ex vivo (49, 90, 93–95). Ability for clinical translation has been demonstrated in breast cancer patients by combining SHG and bright-field high-resolution microscopy with large field of view to design a semi-automated technique to predict survival based on collagen fiber classifications (93). Recently, confocal microscopy has been used in vivo to detect collagen turnover after introduction of fluorescent fibrillar collagen into the dermis of live mice (96). However, all optical imaging methods suffer from their limited imaging depth, rendering them often an invasive tool and limiting their clinical translation (49, 57). Thus, the diagnosis and treatment of pathologies related to collagen remodeling has benefited greatly from the development of collagen-binding or hybridizing peptides, bearing an imaging contrast agent (CA) for, e.g., magnetic resonance imaging (MRI) or fluorescence imaging, or theranostic agents, to image triple-helical, intact, and/or unfolded, denatured collagen and treatment response (97). Other imaging modalities [e.g., ultrasound (US) (98, 99), optical coherence tomography (OCT) (100, 101), Fourier transform infrared spectroscopic imaging (53), or multispectral photoacoustic imaging (PAI) (102)], and various collagen-targeted agents, e.g., quantum dots (84, 85, 103, 104) or collagen-mimetic peptide-based imaging agents (105, 106) are being developed/applied to improve collagen imaging and to measure collagen turnover during tissue remodeling.

The key role that various MMPs play during cancer initiation and progression, with clear links to tumor invasion and metastasis (107), make them desirable treatment and cancer imaging targets (108). Targeted probes to image MMPs in vivo by optical imaging (fluorescence and bioluminescence), positron emission tomography (PET), single photon emission computer tomography (SPECT), and MRI have been developed (108–110), with a “broad-spectrum” MMP-activatable fluorescence probe available commercially (MMPSense, PerkinElmer, Akron, OH, USA). Each modality and imaging probe displays strengths and weaknesses in effectively imaging MMP activity (108, 110). However, optical imaging has limited penetration depth (108) and, while tomography is possible, anatomical information is lacking. Targeted, inhibitor-based PET and SPECT probes harness the excellent sensitivity of radioactive tracers and may be theranostic, but their synthesis may be difficult and, so far, it has not been possible to quantify proteolytic activity in vivo due to non-specific binding (108, 109, 111). Recently, an ¹⁸F-labeled MMP-activatable PET probe has been developed to overcome the lack of specificity of inhibitor-based probes (112). Photoacoustic tomography (PAT or PAI) (58, 113, 114) combines ultrasonic resolution with electromagnetic-enhanced contrast to obtain quantitative information on tissue structure, blood flow, and perfusion, and, through targeted probes (59), on receptor status or enzyme activity. For example, a photoacoustic probe activated by MMP-2 and MMP-9 demonstrated sensing of MMP-2/-9 activity in a follicular thyroid carcinoma model (115). With an imaging depth of >30 mm, depending on setup and desired spatial resolution (58), PAT expands on the tissue penetration of up to 20 mm typical for optical imaging (51, 108) and is thus suitable to monitor non-invasively tumor characteristics in orthotopic preclinical cancer models. The lack of anatomical/morphological information inherent to optical imaging, PAT/PAI, and PET/SPECT can be overcome by multimodal imaging. Using fluorescence molecular tomography (FMT) (51) coregistered with MRI, Salaun et al. (116) found increasing MMP-13 levels as lung tumors progressed. In skin squamous cell carcinoma xenografts, MMP-2, -3, -7, -9, -12, and -13 activities correlated with degree of angiogenesis and tumor invasion, as imaged by FMT combined with μCT (117). PAT/PAI, combined with US and OCT, can provide non-invasively morphological and functional tumor characteristics (60). Molecular MRI to measure MMP activity using protease-modulated CAs is still emerging and is hampered by its insensitivity, requiring long acquisition times but is potentially quantitative, and anatomical information can be obtained in the same setting (108, 109). Of note is that the interpretation of MMP images obtained with the typically broad-spectrum probes (110) is further complicated by the function of MMPs in biological processes beyond ECM remodeling (111).

Another major ECM constitutent, the proteoglycan hyaluronan [hyaluronic acid (HA), hyaluronate] is a high molecular weight glycosaminoglycan with a significant role in tumor growth and metastasis (118, 119), acting as tumor suppressor or promoter depending on its molecular weight (120).

It is degraded by hyaluronidases (Hyals), with hyaluronidase-1, -2 (Hyal1, Hyal2) currently being the most studied in cancer (119). Hyal1 overexpression has been associated with more aggressive tumors in a variety of epithelial cancers (e.g., bladder, colorectal, breast, and ovary), while Hyal2 may function as a tumor suppressor or promoter (119). The development of various HA probes to image HA turnover and clearance and of theranostic HA probes where encapsulated drugs are released in response to Hyal activity (119, 121, 122) has expanded greatly in recent years, and a detailed review is beyond the scope of this paper. HA probes often exploit the high specificity of HA for the CD44 receptor, a transmembrane receptor overexpressed in many tumor cell types (120, 121, 123–134). Single moiety, HA-based CAs have been used to image Hyals activity by MRI (135) and NIRF (124). Fluorescence correlation spectroscopy and Forster resonance energy transfer of HA-conjugated probes have shown promise in quantitative bladder cancer staging by detecting shedded Hyals in urine samples (119). Also, fluorescent HA probes may be used in an intraoperative to assess Hyal activity or drug delivery of theranostic probes (119). Often, HA probes contain more than one CA moiety to harness the strength of multimodal imaging, such as MRI/optical imaging (123, 136), MRI/computer tomography (CT) (137), NIRF/CT (138), or NIRF/PA imaging (139–141) to improve diagnostic capability and monitoring of therapeutic efficacy (121, 142).

Only a few studies report the in vivo imaging of other ECM constituents, such as fibronectin or laminins, in cancer. Fibronectin, whose expression increases with epithelial–mesenchymal transition, is typically targeted to image tumor-associated angiogenesis (143, 144). Laminins are a family of glycoproteins which interact with other ECM proteins, assuring the ECM organization, and are involved in cellular signal transduction pathways (145), cell adhesion, migration, and proliferation (146) and thus affect in cancer, tumor invasion, angiogenesis, and metastasis (145, 147). While laminins and their function have been studied extensively in vitro or ex vivo (29, 148–150), in vivo studies directly imaging laminins have been limited. Cuesta et al. developed a fluorescent trimerbody recognizing an angiogenesis-associated laminin epitope, accumulating in tumors (151). Other studies have used imaging agents targeting laminin cell surface receptors directly or indirectly to detect or treat tumors (152–154).

In preclinical cancer models, the preferential homing of ex vivo cultured MSCs to tumor tissue and metastasis has been imaged non-invasively in vivo after intravenous/arterial injection of MSCs, pre-(multi-)labeled with bioluminescence (162–167), fluorescence (168), MRI (169–171), PET (170–173), or SPECT (171, 174) imaging probes (175). The tumor effects and/or localization of MSCs, pre-labeled with an imaging probe and coinjected with tumor cells, have been monitored in vivo with MRI (176) and bioluminescence (177–180) imaging. The preferential accumulation of MSCs at sites of inflammation and tumors makes them an ideal vehicle for treatment delivery (155, 181, 182), and combined treatment/imaging MSC moieties are being developed for cell tracking and treatment monitoring (175, 177, 178, 183). While overall safe clinically (175, 184), ex vivo culture and pre-labeling of MSCs may lead to secondary tumors (185) and/or impact functionality (186). Using fluorescent, transgenic mice (187), potentially avoids ex vivo culture and pre-labeling of MSCs, with the disadvantage that typically all cell types express the imaging marker, limiting the in vivo identification of different stromal cell types, unless pre-labeled.

Various markers, such as α-smooth muscle actin (α-SMA), vimentin, and fibroblast-activation protein α (FAP), have been used to identify CAFs (188). In preclinical models, CAFs have been shown to promote breast tumor growth and metastasis by enhancing the recruitment of immune suppressor cells and TAMs (189) and to mediate collagen remodeling (190). The high expression of FAP in CAFs (191, 192) makes it a desirable target for diagnostic and therapeutic imaging, although it may also be expressed in some other tissues and tumor cells (192–194). One difficulty for the development of FAP-targeted in vivo imaging probes is that FAP shares peptide substrates with other post-prolyl peptidases, resulting in non-specific binding in vivo (192). Thus to track CAFs in vivo by MRI or NIRF imaging, Granot et al. took advantage of caveolae-mediated endocytosis in fibroblasts by pre-labeling CAFs in vitro with the CAs biotin-bovine serum albumin-gadolinium diethylenetriaminepentaacetic acid, Feridex, or 1,1′-dioctadecyl-3,3,3-tetramethylindotricarbocyanine iodide (195, 196). Recently developed FAP-specific, activatable NIR fluorescence probes (193, 194) show promise in in vivo imaging of FAP-expressing tumors. And novel cancer treatments based on the depletion of FAP-expressing stromal cells would greatly benefit from monitoring response in vivo with improved multimodal imaging probes (197).

Multiple recent reviews summarize the imaging techniques applied to track various immune cell types in vivo and to monitor immunotherapy response, including the ability to optimize administration route of therapeutic immune cells (45, 47, 61, 63, 198–202).

To visualize immune cells in vivo, cells may be labeled ex vivo with paramagnetic, fluorescent, or radiochemical probes for MRI, FLI, or PET/SPECT respectively or transfected with reporter genes for PET, SPECT, bioluminescence imaging (BLI), and/or FLI (or FMT) before injection into the host (45, 198, 201, 203–206). Tumor-infiltrating lymphocytes, dendritic cells (DCs), or TAMs can be imaged by FLI or PET/SPECT, using fluorescently labeled (Figure 6) or radiolabeled antibodies or antibody fragments targeting cell-type specific surface receptors (198, 199, 201). Delivery of full-size antibodies may be affected by vascular dysfunction in tumors, thus, potentially affecting image quality and interpretation [false positives, reduced specificity (50)]. Targeted US microbubbles have been used to track B7-H3 expressing TAMs and tumor cells (198). The high endocytic activity of TAMs facilitates their imaging by MRI, PET, and to a lesser extent by CT, FLI, US, Raman imaging, or PAI, using nanoparticles, either uncoated (e.g., ultra-small superparamagnetic iron-oxide nanoparticles for MRI) or coated to increase macrophage affinity (199, 200). Nanoparticle uptake may vary across different macrophage populations and may be detected also in other cell types (e.g., tumor cells) to a variable degree, confounding potentially macrophage tracking, but has been used successfully in clinical lymph node cancer staging by MRI (200). Cellular MRI of DC migration using CAs (iron oxide-based nanoparticles, perfluorocarbon emulsions) permits repeated monitoring (a limitation in PET/SPECT) and does not require pre-labeling with reporter genes, as in BLI or FLI, but has lower sensitivity (63). Intravital microscopy of stroma–tumor cell dynamics (207) has been an essential tool in assessing lymphocytic interactions in tumors and draining lymph nodes by optical imaging (208). The clinical translatability of these imaging methods has been reviewed previously (202, 209), with optical methods currently limited to intraoperative imaging (50).

Vascular function and distribution has been assessed in vivo non-invasively by MRI (45, 217–226), CT (μCT) (45, 52, 227, 228), H₂¹⁵O or ¹¹C PET (45, 217), US (45, 52, 229) [clinical (230)], PAI (45, 58, 114), and intravital optical imaging methods (231), such as FLI (48), second-generation OCT [optical frequency domain imaging (OFDI)] (Figure 5) (45, 48, 52, 68), and FMT (45).

In vascular MRI, five acquisition methods are used to measure the enhancement of exogenous (i, ii) or endogenous (iii–v) contrast dynamically: (i) Dynamic contrast-enhanced (DCE)-MRI, which exploits the shortening of the T₁ relaxation time of water protons near a CA, typically Gd-based and low molecular weight (223, 229), (ii) dynamic susceptibility contrast (DSC)-MRI, which measures the effect of the CA (e.g., Gd-based CA or superparamagnetic iron oxide particles) on the T₂ and T₂* relaxation time of nearby water protons (218, 222), (iii) arterial spin labeling (ASL)-MRI, where the dynamic measurement of the in- and out-flow of magnetically labeled water protons, which serve as endogenous “CA,” characterizes the vasculature in a region of interest (223, 232), (iv) blood oxygen level dependent MRI (BOLD-MRI) where oxyhemoglobin confers diamagnetic and deoxyhemoglobin paramagnetic contrast, respectively (223), and (v) diffusion-weighted MRI (DW-MRI), where intravoxel incoherent motion (IVIM) reflects vascular perfusion (64, 224, 229, 233, 234).

In DCE- and DSC-MRI, the first pass of the exogenous CA uptake is characterized by the venous or arterial input function (219, 220, 222, 235–237), after which the CA distributes throughout the vasculature, extravasates at sites of leaky blood vessels into the interstitium, and is ultimately cleared from the body. Rate and path (e.g., liver and kidney) of tissue and vascular clearance are dependent on the specific CA (e.g., size and type of CA) used (223, 225, 238). Hemodynamic parameters, such as vessel density, vascular permeability, vascular perfusion, extravascular space, vessel size and plasma volume, etc., are either derived from semiquantitative measures of signal enhancement (220, 239–241) or from pharmacokinetic modeling of signal-versus-time curves (222, 242) with the underlying principle based on standard tracer-kinetic theory of linear and stationary systems (221, 243) and the model adjusted to account for variations of CA characteristics, e.g., low versus high molecular weight (impacting the ability of the CA to extravasate), receptor targeted (e.g., α_νβ₃-integrin), extra- or intracellular or both, with exchange (218, 223, 235, 244). More recently vascular feature-based analyses have been developed to assess intratumor vascular heterogeneity (245, 246).

Hemodynamic parameters conferred from ASL- and BOLD-MRI are tissue blood flow and volume (223, 232). Tissue perfusion and diffusivity are obtained from IVIM DW-MRI by fitting data with a bi-exponential model (224).

Similarly to DCE-MRI, DCE-CT has been used clinically to obtain blood flow, blood volume, and permeability with the disadvantage of radioactivity limiting serial monitoring and often worse spatial resolution than DCE-MRI (227, 228). In preclinical models, the relatively high radiation dose of μCT limits its in vivo use (45), and vascular networks have been assessed ex vivo (247). Viscosity of CA for DCE-CT may also result in complications, such as vessel rupture, and DCE-μCT still suffers from artifacts generated from, e.g., bones or large vessels (52). Generally, functional vascular parameters similar to DCE-MRI can be extracted from DCE-μCT data (52, 248).

While H₂¹⁵O PET measures specifically tissue perfusion (217), the short half-life of ¹⁵O limits the applicability of vascular perfusion measurements by PET (45). Vascular permeability can be obtained from dynamic PET using the macromolecular CAs ⁶⁸Gd-DOTA-albumin (45), ¹¹C-methylalbumin (limited by the 20.4 min short half-life of ¹¹C), or ⁶⁸Ga-transferrin (217). However, lack of accompanying anatomical information and ionizing radiation limits (especially serial) vascularity measurements by PET, despite its potential higher sensitivity than DCE-MRI and its ability to directly measure tissue perfusion.

Tumor angiogenesis, i.e., perfusion and vascular density have been successfully measured in small animals by Doppler or contrast-enhanced US with microbubbles (non-targeted and targeted to, e.g., VEGFR2 (249), α_νβ₃-integrin, or endoglin) with a spatial resolution of ~50–100 μm (45, 52, 229). One advantage of the combining contrast-enhanced and non-contrast-enhanced high-frequency volumetric power Doppler US is the ability to distinguish mature and immature vessels (52). While US is cheaper than DCE-MRI, it is not suitable for whole-body imaging, has limited soft tissue contrast, and is to some extent user dependent, limiting its applicability, especially for monitoring of antiangiogenic treatments (52).

In PAI of tumor angiogenesis, the intrinsic contrast from hemoglobin permits visualization of microvasculature and quantification of blood oxygen saturation (45, 52), with its usefulness recently extended to flow imaging (114). Submillimeter resolution is achieved, albeit restricted to depth of a few centimeters (58). Reporter genes or endogenous targeted CAs extend the ability to visualize and quantify tumor angiogenesis in vivo (52). The hemodynamic response to external stimuli or treatment is quantified from contrast changes after image reconstruction (52, 58, 114).

Blood vessel diameter, surface area, and branching pattern have been assessed with intravital optical imaging methods during tumor growth and/or treatment (Figure 5) (45, 48, 52, 68, 231).

Imaging of the lymphatic system is here summarized only briefly, as is has been reviewed in detail previously (50, 250). The lymphatic system (Figure 1) drains lymph fluid from interstitial space to the venous circulation, thus, maintaining tissue fluid homoeostatis, transports immune cells to lymphoid organs, and plays a role in lipid absorption (50, 250). While the lymphatic vasculature near tumors provides a route for metastatic dissemination of cancer cells, its role has only been explored by non-invasive imaging methods over the past decade (50, 250). Lymphangiography traces the drainage of a CA for X-ray, CT, MRI, US, PAI, or optical imaging but is lacking specificity, requiring direct injection into a lymph vessel (difficult to perform in preclinical models) or intradermal injection near sites draining into the dermal capillary plexus (50, 250). Identifying sentinel lymph nodes containing cancer cells has been achieved with intravenously injected CAs that identify blocked drainage or directly target cancer cells (50). In vivo OCT (or OFDI) and laser speckle imaging permit CA-free visualization of lymphatics (Figure 5) (68, 250). The most specific imaging approach to identify and characterize the lymphatic system is to use CAs, targeted to lymphatic vascular-specific molecules, i.e., vascular endothelial growth factor receptor-3 (VEGFR-3), lymphatic vessel hyaluronan receptor-1 (LYVE-1), podoplanin, or prospero-related homeodomain transcription factor PROX1, with LYVE-1 the most widely used lymphatic endothelial cell marker (250). LYVE-1-targeted CAs have been developed for PET (251) and optical imaging (252). Preclinically, fluorescent or bioluminescent gene reporters in transgenic mouse models have been also been used to visualize lymphatics, typically by intravital imaging methods (50).

Changes in choline-phospholipid metabolism have typically been assessed preclinically (and clinically) by non-invasive ³¹P and ¹H magnetic resonance spectroscopy or spectroscopic imaging (MRS or MRSI) (14–16, 18, 253). Choline uptake and metabolic conversion have also been assessed with high sensitivity by ¹⁸F-fluoro-, ³H-, or ¹¹C-choline PET and successfully translated to the clinic (254–257). As with any PET tracer though, radiolabeled choline PET does not discriminate between different metabolites (256), limiting its value for pathway studies.

Tumor hypoxia has been imaged non-invasively and visualized directly by hypoxia markers accumulating in hypoxic cells using PET, electron paramagnetic resonance (EPR), or ¹⁹F MRS (65, 248, 258–266). Delivery of a specific hypoxia marker to less vascularized regions, which are typically associated with hypoxia, may impact the intensity of the accumulating hypoxia marker (246, 267). Since evolution of chronic or acute hypoxia in tissue is linked to the vascular delivery of oxygen, several indirect MRI methods based on vascular features associated with tumor hypoxia, such as BOLD, TOLD, or DCE-MRI, have been developed to identify hypoxic areas or hypoxia changes in tumors (64, 246, 268–271). Tissue oxygen tensions have been mapped by ¹⁹F MRI oxymetry using perfluorocarbons; as with any exogenous tracer, these measurements are vascular delivery dependent, and thus, potentially biased toward well-perfused tumor regions (271). While carbonic anhydrase-IX (CA-IX) has been proposed as an intrinsic hypoxia marker in tumors, CA-IX expression, measured by immunohistochemistry, correlated to hypoxia in some and not in other studies (260). Nevertheless, attempts are underway to image CA-IX expression in vivo by PET (272), NIRF (273), and SPECT (274).

Tumor glycolysis is typically assessed by in vivo PET using the cellular entrapment of ¹⁸F-FDG, after uptake of ¹⁸F-FDG by glucose transporters (GLUT-1, GLUT-3; often overexpressed in cancer) and subsequent phosphorylation by hexokinase II (256, 275). While it is quite insensitive, ¹³C MRS has been applied preclinically to evaluate glycolysis and the ¹³C labeling of downstream metabolites (275–277). Detection sensitivity of ¹³C MRS can be significantly improved by magnetization transfer techniques or indirect inverse detection (276, 277). Hyperpolarization of ¹³C-labeled substrates increases detection sensitivity up to 10,000-fold, with the caveats that the hyperpolarization is short lived and currently limited to few substrates, including glucose (64, 275, 276, 278, 279). Compressed sensing can further improve acquisition speed and spatial resolution in hyperpolarized (HP) ¹³C MRSI (280). Lactate production from precursors, such as ¹³C-labeled pyruvate or glucose, can be rapidly assessed globally and localized by HP ¹³C MRSI (278, 279). Steady state levels of tumor lactate have been assessed by ¹H MRS and MRSI, using spectral editing methods to suppress the high lipid signal overlapping lactate (64, 275, 281–285).

Measuring tumor tissue pH non-invasively is challenging, and various methods have been and are being developed to measure preclinically extracellular and/or intracellular pH (pHe or pHi) (64, 286).

Tumor pHe and pHi distributions can be obtained by ³¹P MRS/MRSI using 3-aminopropylphosphonate (287, 288). However, insensitivity of ³¹P MRSI limits its spatial resolution and restricts broad applicability. Thus, ¹H MRSI pHe markers have been developed to improve detection sensitivity (288–291). To shorten acquisition times and improve sensitivity further, with potential for clinical translation, HP ¹³C MRSI of injected bicarbonate has been proposed for pH imaging, with pH (predominantly pHi, though it does not distinguish between pHi and pHe) calculated from the signal intensity ratio of hyperpolarized H13CO3− to ¹³CO₂ (292, 293). However, the reaction is dependent on CA-IX activity in the tissue, thus, calibration has to be performed for each tissue type separately, restricting its applicability (294). Chemical exchange saturation transfer (CEST) MRI detects the pH-dependent chemical exchange between an amide proton and surrounding water molecules (291, 295). In acidoCEST, the pH dependence of the CEST effect ratio of a CEST agent with two amide protons, generating two CEST effects, is used to measure pHe (295–297). Challenges to measure pH by acidoCEST include its low sensitivity, requiring optimization of experimental parameters (295), and it may not always be a given that the CEST effect is solely visible for the amide protons of the selected CEST agent and only affected by pH (291).

Recently, the pH-dependency of cellular membrane insertion of radiolabeled pH (low) insertion peptides has been used to image tumor pH (at the intra-/extracellular interface) preclinically with PET (298–301).

Other non-invasive imaging modalities assessing pH in vivo include optical imaging with pH-sensitive dyes (302–304) or a pH-sensitive reporter gene (305), ratiometric PAI with pH-sensitive nanoprobes (306, 307), MRI using a CA with pH-sensitive (and concentration-dependent) relaxivity, with the difficulty of measuring in vivo the CA concentration (291), and EPR spectroscopy (308, 309).

Additionally, pH-sensitive probes are being developed as theranostic agents, combining treatment with diagnostic and monitoring ability (310–314). Of note is that all exogenous pH markers are delivery dependent and may not be clinically translatable, adding further challenges to pHe/pHi imaging.