Preclinical Applications of Multi-Platform Imaging in Animal Models of Cancer

Recent technological developments in scanner design and advances in image reconstruction have secured the rapid application of noninvasive imaging for detection, characterization, and monitoring of cancer etiology in a variety of animal models (1–3). Obvious advantages arise from the ability to study structure, metabolism, and function of cancer cells and cancer supporting microenvironment longitudinally, without the need for necropsy. Indeed, imaging is noninvasive and repetitive studies are performed in the same animals, with each animal serving as its own control. Importantly, most imaging platforms can efficiently survey whole animals, opening new horizons for studying metastatic disease. Furthermore, many imaging technologies are intrinsically translational by applying identical imaging protocols, imaging tracers, and image analysis to various species, thereby providing a bridge from laboratory animals to companion animals and ultimately to humans with the goal of easing the burden of human cancer (4–6). There are various imaging platforms, also referred to as imaging modalities, each based on a specific physical principle (Table 1), allowing unique information/data to be generated. The primary reason for applying a multi-platform imaging approach to cancer research is to obtain comprehensive information from a cancer-bearing animal (Table 2). The in vivo cancer imaging modalities are highly complementary, providing a variety of quantitative biomarkers for cancer cell tracking, and assessing tumor dimensions, pathophysiology, metabolism, and molecular composition (Table 2; Fig. 1), but each has specific advantages and weaknesses (6–8). In this review, we highlight the state-of-the-art applications of preclinical multimodal multiscale imaging and focus on the specific applicability to cancer research.

Magnetic resonance (MR) physics is complicated, but offers extraordinary opportunities to manipulate tissue water signals based on relaxation mechanisms, chemical exchange, flow, and diffusion to reveal diverse anatomic, physiologic, and cellular properties of cancer at high external magnetic fields. The most sensitive nucleus is the proton, notably in H₂O. Anatomic MRI: among all imaging modalities, MRI possesses the best soft-tissue contrast, which may still be enhanced further using exogenous paramagnetic contrast agents. Excellent spatial resolution can reveal ultra-small cancer lesions (as small as 0.2-mm diameter with 9.4 T MRI), particularly in well-structured tissues, such as the brain. MRI is the “gold-standard” for orthotopic brain tumors and brain metastases (Fig. 1A, 1; refs. 9–13), and is also widely applied for the detection of other soft-tissue lesions, including liver (Fig. 1A, 2) and lung metastases (Fig. 1A, 3). Physiologic MRI: beyond high-resolution anatomic MRI, tumor cellular density and edema are easily quantified using diffusion-weighted MRI, which is sensitive to restricted or enhanced diffusion of water molecules, respectively (Fig. 1B, 7; refs. 2, 14, 15). Several recent publications reported increased apparent diffusion coefficients (ADC) associated with treatment-induced necrosis (16–19). Tissue oxygenation may be examined using oxygen-sensitive MRI. Notably, apparent transverse relaxation rate (R₂*) is sensitive to the concentration of deoxyhemoglobin, as exploited in blood oxygen level–dependent (BOLD) contrast and forming the basis of functional MRI to assess neurologic activation (20). Meanwhile, so-called tissue oxygen level–dependent (TOLD) MRI exploits the sensitivity of the spin-lattice relaxation rate, R₁, to the paramagnetic oxygen molecule (O₂) itself (Fig. 1B, 8; refs. 21–27). Noting the importance of hypoxia in cancer development, aggressiveness, and response to therapy, an oxygen gas breathing challenge has been shown to provide a simple effective theranostic: well-oxygenated tissues show response to an oxygen gas breathing challenge, whereas hypoxic tissue does not (28). This approach has been demonstrated to provide a prognostic imaging biomarker in rats with respect to stereotactic ablative radiotherapy (24, 28) and is feasible in man (21, 29). Vascular MRI: the use of exogenous MR contrast agents, namely gadolinium chelates as T₁- and iron oxide nanoparticles as T₂-contrast, enables imaging of tumor angiogenesis and changes in tumor vascularity. Intravenous injection of gadolinium contrast agent allows direct visualization of tumor vasculature by MR angiography (MRA, Fig. 1C, 11; ref. 30) or the generation of tumor perfusion/permeability K^trans maps using dynamic contrast-enhanced (DCE)-MRI (Fig. 1C, 12; refs. 31–33). The use of T₂-contrast blood pool agents (based on ferumoxytol and other iron oxide nanoparticles) allows susceptibility-contrast imaging to assess tumor blood volume (32, 34). Cellular and receptor MRI: the same iron oxide nanoparticles can be used for cell tracking. Breast cancer cells prelabeled with ferumoxytol in vitro, could be detected in the brain by T₂-MRI following intravenous injection (Fig. 1E, 21; ref. 35). Meanwhile, injection of ferumoxytol itself leads to extensive uptake by macrophages, which has been observed as reduced T₂ signal, revealing M1 (antitumor) or M2 (pretumor) activity (Fig. 1E, 22; refs. 36–38). Some reports have explored the possibilities of using iron oxide- or gadolinium-based contrast for detecting cell receptors, including HER2 or C2 imaging in mouse models of breast cancer and precancerous renal inflammation (39–41). In mouse prostate cancer models, prostate specific membrane antigen (PSMA) receptors have been successfully imaged using targeted iron oxide nanoparticles by T₂-MRI or a diamagnetic dextran-based chemical exchange saturation transfer (CEST) MRI agent (see below; refs. 42–44). Receptor imaging with MRI poses unique challenges for signal amplification to deposit sufficient MRI contrast per receptor molecule for its detection.

Beyond proton MRI of tissue water, spectroscopic imaging can detect several endogenous metabolites that occur at sufficiently high concentrations, such as lactate, glutamine, glutamate, creatine, N-acetyl aspartate, γ-aminobutyric acid, citrate, choline, and, most recently, 2-hydroxyglutarate (2HG; ref. 45). The oncometabolite 2HG accumulates in low-grade glioma, secondary glioblastoma, and acute myeloid leukemia, owing to mutations in the metabolic enzymes isocitrate dehydrogenase (IDH) 1 or 2. Mutant IDH1/2 aberrantly produces 2HG (instead of ketoglutarate), which is detectable by ¹H-MRS or ¹³C-magnetic resonance spectroscopic imaging (MRSI) following hyperpolarized [1-¹³C]-glutamine administration (Fig. 1D, 17; ref. 46). For ¹³C-MRSI, the most developed hyperpolarized probe today is [1-¹³C]-pyruvate, which enables the detection of activated lactate dehydrogenase in tumors (47). Isotopically labeled substrates and metabolites are clearly seen against naturally low abundance endogenous signals (e.g., 100% enriched isotopomers vs. 1.1% naturally abundant ¹³C). Furthermore, hyperpolarization of ¹³C substrates can be achieved by various techniques, including dynamic hyperpolarization (48) or parahydrogen-induced polarization (49) and leads to a significant boost in the naturally low ¹³C MR spectroscopy (MRS) signal. However, magnetization decays rapidly within minutes, necessitating fast ¹³C MRI techniques. It has been shown that hyperpolarized ¹³C-pyruvate/lactate MRS(I) is superior to 2[¹⁸F]fluoro-2-deoxy-D-glucose ¹⁸FDG-PET (another metabolic imaging technique, see below) in detecting treatment response to novel targeted therapies and radiation (50, 51). Another approach to amplify MRS signals uses CEST MRI, which detects the exchange of protons from hydroxyl, amine, and amide groups to tissue water through the transfer of signal loss, with repeated proton exchange enhancing the effective signal in endogenous (52) and exogenous compounds (Fig. 1D, 18; ref. 53). Amide proton transfer contrast, which is the CEST signal from endogenous cellular proteins and peptides, differentiates viable glioma from radiation necrosis (54). The use of d-glucose administration as a contrast agent for noninvasive CEST detection of tumors has been termed glucoCEST, and offers cancer detection with glucose as a biodegradable, nontoxic contrast agent (55). CEST measurements of regional pH, on the basis of the clinically approved X-ray contrast agent, iopamidol, have been applied in kidney and lung cancer models (56, 57). Another important nucleus for cancer characterization by MRS is ³¹P for detection of phospholipid precursors, high energy phosphates, and inorganic phosphate, which exhibits a pH-sensitive chemical shift in the physiologic range (58), although it can be difficult to discriminate intra- versus extracellular components. Meanwhile, ¹⁹F-MR agents can offer superior chemical shift response (59). ¹⁹F-MRI with perfluorocarbon agents has been used as an alternative to iron oxide T₂-MRI (see above and Fig. 1E, 22) to detect tumor-associated macrophages with the benefit of no endogenous background signal (60). Perfluorocarbons exhibit very high gas solubility and can serve as molecular amplifiers, as exploited to assess tumor pO₂ providing evidence for hypoxia, heterogeneity, and differential regional response to interventions (28, 59, 61).

X-ray CT (microCT) is a high-resolution 3-dimensional (3D) imaging technique; the physical principle of CT is based on scattering and absorption of X-rays by tissues based on their electron density. There are essentially three levels of attenuation yielding color-coded contrast in CT: air (black), soft-tissue (gray shades), and bones (white). Anatomic microCT: compared with MRI, CT is inferior in distinguishing soft tissues/organs, but the major strength of microCT lies in supreme high-resolution (<50 μm) fast imaging of lungs and bones revealing cancer lesions. Because bones are the most common metastatic site for major cancers (including breast and prostate), several studies reported the use of high-resolution (10 μm) microCT for detecting engrafted breast cancer cells in the bone (Fig. 1A, 4; ref. 62). Inhibition of the development of osteolytic bone lesions by zoledronic acid has been reported in MDA-MB-231 breast xenograft mice, also identifying IL1 as one of the key players for metastatic development (63–65). Because of the inherent contrast between air and tissue structures and the resulting attenuation of the X-rays passing through tissue, microCT is particularly well suited for providing high-quality anatomic information in the lung. With the development of precancerous lung conditions, including inflammation (66), fibrosis (67), and emphysema (68), and their progression to lung tumors (69–71), tissue structure becomes dense and can be easily differentiated from both normal lung and airspace. The use of 3D analysis to quantify tumor number, size, and progression is advantageous over traditional histology (69) or macrodissection of the lung to isolate tumors (70). Vascular microCT: gated respiratory-holding techniques, fast acquisition times, and the introduction of novel metal nanoparticles, such as exitron, allow lung microvasculature to be easily visualized, simultaneously with lung tumor detection (Fig. 1C, 13; refs. 72, 73). The low radiation dose of modern instruments makes longitudinal microCT possible without long-term harm to animals (74). Recently, contrast-enhanced microCT has been applied to visualize and map tumor vasculature in brain tumor and neuroblastoma mouse models (75–77).

Ultrasound (US) uses high-frequency sound waves and captures the US energy reflected from interfaces in the body (“echoes”) that separate tissue with different acoustic impedances, where the acoustic impedance is the product of physical density and velocity of sound in the tissue. Typically, a cyst appears sonolucent, because it gives few, if any, echoes (being mostly water), while liver and spleen have solid homogenous echo texture due to medium-level echoes from the fibrous interstitial tissues. High-intensity echoes (increased echogenicity) are caused by calcification, fat, and air interface; however, they do not propagate through bone. Among real-time modalities, US features the highest frame rate up to 20,000 fps, enabling US-guided animal procedures, such as orthotopic cell tumor injections and left ventricular infusion of cancer cells, to generate models of metastasis while avoiding lung engraftment (78, 79). Anatomic US: pancreatic cancer is one of the most challenging mouse models for preclinical imaging. US provides fast, precise quantification of pancreatic tumor burden longitudinally and without contrast administration (Fig. 1A, 6; refs. 80, 81). Vascular US: US is also an excellent technique to assess tumor vasculature, for example, Doppler US measures the speed and direction of flowing blood and has revealed vascular response to antiangiogenic and Notch therapies in an orthotopic renal cell carcinoma mouse model (82), as well as in irradiated rat fibrosarcoma tumors (83). Considerable attention has been given to the development of US-specific nanoparticles and microbubbles, which may be used both for vascular imaging and as theranostic drug carriers. The latest include VEGFR2-targeted microbubbles (84), oxygen microbubbles (85, 86), and US-destructible microbubbles for better delivery of paclitaxel-loaded nanoparticles in pancreatic cancer models (87). Acoustic angiography (AA) is another contrast-enhanced US technique, which uses the super-harmonic signals from microbubbles to produce high-resolution maps of vasculature with exceptional contrast because tissue yields no signal. Furthermore, AA can provide quantitative measurements of vascular density, blood perfusion, and vessel morphology, helpful to evaluate response to antiangiogenic therapy in cancer (82). Quantitative US (88) is obtained from B-mode images and raw radiofrequency data and has been used to examine treatment response. Attenuation coefficients and backscatter coefficients can be derived (89). On the other hand, US elastography can visualize and quantify tissue stiffness noninvasively (90). These data can be used as a potential biomarker to assess changes in the tumor microenvironment, particularly changes affecting the extracellular matrix, which may affect treatment efficacy (91, 92).

Photoacoustic imaging (PAI) represents the newest addition to the commercial armamentarium for preclinical imaging studies and progressively experimental investigations in man (93, 94). PAI exploits spectrally selective pulsed laser excitation of chromophores generating local thermoelastic tissue expansion, which is detected on the basis of the resultant US acoustic waves, analogous to lightning generating thunder. Application of multiple wavelengths allows spectral discrimination, which has been applied to endogenous molecules, such as oxy- and deoxyhemoglobin (HbO₂ and Hb) and melanin, and exogenous agents, such as organic dyes, gold nanoparticles, and genetically encoded proteins (95, 96). Indeed, spectral unmixing allows multiple materials to be detected simultaneously. The technology is particularly rapid, typically achieving single-slice images in <100 milliseconds, but usually images are acquired at multiple wavelengths, and signals may be averaged so that a typical acquisition time is 1–2 seconds. Gating may become relevant for assessing rapid changes in tissues subject to motion (97). Selection of an appropriate nonnegative data reconstruction model is vital and choice of filters can enhance signal to noise (98, 99).

Various commercial instruments are optimized for in vivo microscopic, mesoscopic, whole-mouse tomographic, and human applications, and may incorporate additional US excitation to enhance anatomic discrimination with typical spatial resolution approaching 100 μm at depths up to 5 cm.

The most effective application is assessment of tumor vasculature based on the ability to identify and quantify relative Hb and HbO₂ (Fig. 1B, 9) with effective studies of antiangiogenic therapy (100), acute vascular disruption induced by combretastatin (101, 102), and potentially prognostic observations following tumor irradiation (103). It appears that response to an oxygen breathing challenge characterized as ΔsO₂ is more closely related to perfusion and hypoxia than baseline static parameters (102), for example, low CAIX expression correlated with higher ΔsO₂^MSOT. Blood volume and perfusion may be effectively examined using contrast agents, such as indocyanine green (102, 104), or the liposomal formulation, Genhance (105). Small-molecule dyes may be incorporated in targeted liposomal formulations or used to directly label antibodies for detection of tumors or revealing receptor expression (106). Gold nanoparticles (which could also be used in microCT) exhibit exceptionally high photoacoustic activity based on surface plasmon resonance and may be tuned to wavelengths in the range 600–1,000 nm based on size and shape (96, 107). Additional innovations include “smart” activatable probes, for example, sensitive to β-galactosidase activation (108) and genetically encoded proteins, such as BphP1 (109). PAI essentially bridges two modalities to exploit spectrally selective optical excitation and robust spatial detection using US. It is very much an emerging technology.

Two decades after its invention, in vivo optical imaging is now a well-established standard method to noninvasively monitor biological activity in mouse (and rat) research models. Optical imaging includes four molecular imaging modalities: bioluminescence imaging (BLI), fluorescence imaging (FLI), chemiluminescence, and Cherenkov imaging. The relatively low threshold of implementation, as well as the high sensitivity of in vivo BLI, make this whole-body, noninvasive imaging technique a go-to method in preclinical research (Fig. 1A, 5 and E, 19; refs. 62, 110). Beyond tracking tumor growth and regression via constitutive firefly luciferase expression for drug efficacy determination, the toolbox for this molecular imaging technique has vastly expanded. Bioluminescence enzymes can be used to genetically tag cells, viruses, bacteria, gene therapy, and, now also, antibodies and their fragments (111). These enzymes, such as firefly, Renilla, Gaussia, and NanoLuc luciferases, can be constitutively or inducibly expressed, and as such used for ratiometric imaging, gene expression studies, or dual labeling purposes (e.g., tracking T cells infiltrating tumor; Fig. 1E, 19; refs. 12, 112–114). Split luciferases to evaluate protein–protein interaction, as well as split luciferin substrates to monitor apoptosis have been designed and are utilized to evaluate mechanism of action (115). Potential drawbacks of BLI are the need for cell transfection and delivery of reactive substrate. Luciferin effectively crosses barriers, such as blood–brain and placenta, and its very delivery to tissue has been used to assess selective vascular destruction in tumors (101, 116). Bioluminescence resonance energy transfer constructs, such as Antares, which red shifts the shorter wavelength NanoLuc luciferase for better in vivo sensitivity, are also available (117). Chemiluminescence compounds, substrates, and sensors are luminophores that emit red shifted light upon chemiexcitation and have been reported for detection of H₂O₂, H₂S, formaldehyde, β-galactosidase, and nitroreductase activity (118–121). Dr. Cherenkov received the Nobel Prize in 1958 for his discovery of the bluish hue of light emitted by decaying radioisotopes. This same light emission can be detected by screening mice injected with diagnostic radioisotopes, such as ¹⁸FDG, in an in vivo optical imaging system, adopting the epithet of a poor man's PET scanner (122), and may also be relevant for radiation dosimetry (123). FLI, on the other hand, features both genetically encoded fluorescent proteins (FP) and fluorescent dyes. The powerful combination of BLI and FLI is exemplified by Zeng and colleagues (Fig. 1F, 23; ref. 124), illustrating the tracking of fluorescent micelles to bioluminescent brain tumors. In comparison with BLI, however, the contrast to noise is less with fluorescence due to nonspecific autofluorescent noise originating from innate proteins in tissue. This issue is being combatted with the discovery of red-shifted FPs for better in vivo sensitivity, an initiative led by Nobel laureate Dr. Roger Tsien (125). A second window of opportunity for in vivo FLI is currently being explored in the short wavelength infrared using ultra-bright near-infrared-IIb rare-earth nanoparticles. Here, tissue absorption and light scattering are significantly reduced (126), rendering higher resolution, higher depth penetration images. Crafty alternatives have also been invented in which fluorescent sensors are quenched until activated by an enzymatic reaction (e.g., cathepsin, matrix metalloprotease, neutrophil elastase, etc.) or in which fluorophores shift wavelength upon binding their target (127). A great advantage of fluorophores is that they are also readily detectable ex vivo for histopathologic evaluation. This is highly translational, and intrasurgical fluorescence imaging is actively being explored to both highlight tumor burden, and also improve tumor margin of resection (128). Preclinical optical cancer imaging begs for anatomic context, prompting coregistration with anatomic imaging modalities, such as X-ray, microCT, MRI, or, the recently developed, robotic US, which features inexpensive, exogenous contrast-free 3D soft-tissue resolution (78).

Nuclear medicine images are produced by giving the animal short-lived radioactive isotopes and detecting their decay using a gamma camera (single-photon emission CT, SPECT) or positron emission (PET) scanner, revealing spatial and temporal distribution of target-specific radiotracers and pharmaceuticals. An extensive array of radiopharmaceuticals, or molecular probes exist (based on ¹¹C, ¹³N, ¹⁵O, ¹⁸F, ¹²⁴I, ⁶⁴Cu, ⁶⁸Ga, and ⁸⁹Zr for PET and ¹²³I, ^99mTc, ²⁰¹Ti, and ¹¹¹In for SPECT), to image diverse aspects of tumor physiology and biology. Data can reveal properties such as glucose metabolism, blood volume and flow, tissue uptake, receptor binding, and oxygen utilization. Because both modalities have relatively low spatial resolution, CT is usually added for an anatomic overlay of the biodistribution of the radiolabeled probe. Metabolic PET: ¹⁸FDG-PET is the most established metabolic cancer imaging approach both preclinically and clinically. Most tumors have a highly glycolytic phenotype (the Warburg effect), providing the basis for increased uptake and accumulation of the radioactive glucose analogue ¹⁸FDG, as shown in various mouse models of leukemia, pancreatic, lung, colorectal, breast, prostate cancers (Fig. 1D, 15; refs. 51, 129–132). Other tracers have recently been introduced to elucidate abnormal metabolic phenotypes, including, either ¹¹C- or ¹⁸F-, acetate (mitochondrial metabolism; ref. 133), choline (membrane phospholipids; refs. 133, 134), and amino acids in brain tumors (glutamine, tyrosine, or methionine; Fig. 1D, 16; refs. 135–138). Physiologic PET: several essential ¹⁸F-labeled tracers should be mentioned here as potential (although not entirely specific) markers for tumor cell proliferation (¹⁸F-fluorothymidine, ¹⁸FLT) and hypoxia (18F-fluoroazomycin arabinoside, ¹⁸F-FAZA, and ¹⁸F-fluoromisonidazole, ¹⁸F-MISO). Radioactive thymidine is readily incorporated into DNA synthesis, making an increased uptake of ¹⁸FLT visible on animal PET and correlating with increased ADC on diffusion-weighted MRI, albeit exhibiting low specificity (139–142). ¹⁸F-MISO is trapped in hypoxic areas as compared with BOLD and TOLD MRI (Fig. 1B, 10; ref. 143). While ¹⁸F-MISO has been tested for many years, its uptake selectivity is suboptimal and many other potential hypoxia-imaging agents are under development and evaluation (e.g., ¹⁸F-FAZA shows more rapid background clearance; refs. 144, 145). Cellular PET: with the development of checkpoint inhibitor and immunotherapies, significant efforts have been dedicated to develop so-called “immunoPET.” Several T lymphocyte–targeting molecules were radiochemically labeled with long-lived radionuclides (such as ⁶⁴Cu, ⁶⁸Ga, and ⁸⁹Zr). Following intravenous injection, intratumoral accumulation of T lymphocytes has been noninvasively detected in response to checkpoint inhibitor treatment (Fig. 1E, 20; refs. 146–148). Molecular PET/SPECT: specific molecular targets have been visualized using PET- or SPECT-based peptides, antibodies, and receptor binding ligands. One of the most explored is hormone imaging, ¹⁸F-fluoroestradiol PET, as used for preclinical and clinical imaging of estrogen receptor–positive (ER⁺) breast and ovarian cancer (Fig. 1F, 24; refs. 149–152). Recent examples of hormone imaging include PET of androgen receptor in rat brain (153). Several ¹¹¹In/²⁰³Pb-labeled peptides for SPECT (154) and ⁶⁸Ga-MSH for PET (155) have been developed to target the melanocortin-1 receptor in melanoma mouse models (Fig. 1F, 25). A ²⁰³Pb/²¹²Pb theranostic pair has been reported for PSMA-based α-particle–targeted radiopharmaceutical therapy in advanced prostate cancer (156).

Other notable imaging technologies include magnetic particle imaging (MPI) and electron spin resonance (ESR). MPI is an emerging imaging modality that involves iron oxide nanoparticles. Unlike MRI, MPI measures electronic moment of particles, which is more sensitive than measuring changes in proton relaxation by MRI. The detection is linear, sensitive (ng of iron/voxel), and has a high signal-to-background ratio. Using MPI of iron oxide particles, kinetics of accumulation of nanoparticles in rat tumors (157) and kinetics of drug release in mouse breast tumors (158) were studied. Further applications of MPI are dependent on improving the acquisition speed and resolution, as well as improving circulation and targeting properties of nanoparticles.

Electron paramagnetic resonance (EPR), also termed as ESR, has been a research tool for many years, but remains somewhat esoteric in cancer research, largely due to lack of available instrumentation. It directly detects free radicals, but the extremely high frequencies tend to limit tissue penetration, although effective studies have been performed in mice and human teeth and tattoos (159). The most popular application has been based on imaging signal line width and relaxation mechanisms, which may be directly responsive to the presence of oxygen and hence, pO₂. Reporter agents may be injected directly into tumors (e.g., India ink or chars; ref. 160), or infused systemically (OX63- oxygen-measuring spin probe, coincidently the same material is used to achieve hyperpolarization of ¹³C substrates for nuclear magnetic resonance; ref. 161). Sensitivity to oxygen can be particularly high at very low, radiobiologically relevant pO₂ values (0–15 Torr) and significant correlations have been observed between pO₂ values and radiation response (50, 160–162). A significant drawback of EPR is the lack of integrated anatomic information, generally requiring that separate MRI be performed and coregistered.

Radiation plays an important role in cancer therapy; radiation-based therapy has been applied to animal models for decades and recently has undergone significant improvement in terms of applying multimodality imaging to guide radiation planning (163, 164). Radiation kills cancer cells by damaging DNA, either directly or indirectly, through the creation of reactive oxygen species. Because radiation kills both cancer cells and healthy cells alike, various methods are used to increase the tumoricidal effects of radiation while minimizing damage to the surrounding normal tissue, including spatial modulation of the dose distribution to conform to a specific target region. While such conformal dose distributions allow for significant reductions in normal tissue toxicity, they also require onboard image guidance systems to ensure the tumor is in the correct location when the radiation beam is turned on. Modern animal irradiators incorporate multimodal imaging detectors to precisely guide the radiation, combining the ability to deliver targeted radiation treatments using a 225 kVp, gantry-mounted X-ray tube with digital radiography, fluoroscopy, cone-beam CT, and BLI (164, 165). Image-guided irradiation has been successfully applied even for small orthotopic head and neck and lung lesions in tumor-bearing mice (166, 167). The software also allows import of existing imaging datasets from other modalities, such as MRI, which often plays a crucial role for irradiating intracranial brain tumor models (9, 163).

There is increasing interest in using imaging to develop noninvasive quantitative imaging biomarkers (surrogate endpoints) for cancer characterization. Indeed, most imaging read-outs are provided in both qualitative and quantitative form (Table 2; ref. 168). This is especially true for MRI, CT, and US, due to their high spatial resolution to provide precise tumor dimensions, as well as number of suspicious lesions/metastases (169, 170). The well-established mathematical modeling algorithms for tracer kinetics allow quantification of tumor vasculature based on gadolinium, nanoparticle, and microbubble uptake for MRI, CT, and US, respectively (32, 34). The biomarkers include the exchange rate constants (K^trans), which reflect the efflux rate of gadolinium contrast from blood plasma into the tissue/tumor extravascular extracellular space, the volume of contrast agent distribution V_e, or simply the area under enhancement curves after the administration of contrast (19, 171–174). Finally, physiologic MRI provides established quantitative endpoints in the form of ADCs from diffusion-weighted MRI: low ADC (0.5–0.8 × 10⁻³ mm²/second) indicates densely cellular aggressive tumors, while treatment-induced necrosis results in increased ADC, up to 1.2 × 10⁻³ mm²/seconds, and radiation-induced edema's ADC as high as 2.2 mm²/second (17, 19). PET and SPECT tracer uptake is usually reported as standard uptake values (SUV), which includes normalization to injected dose and accounts for radionuclide decay (129, 130, 175). Several studies report ratios of signal intensities (SI) of the tumor-to-normal tissue (most often for brain tumors as tumor-to-brain ratios; refs. 138, 174). Optical imaging (BLI and FLI) is rather semiquantitative, but can provide SIs related to tumor volume or tissue perfusion (11, 114), for example, the change in light emission from luciferase-expressing tumors following an acute intervention, such as a vascular disrupting agent (VDA) provides an indication of vascular shutdown (101, 176, 177). Multimodality imaging ideally combines the advantages of each modality, while mitigating their deficiencies. Image registration is necessary when more than one imaging modality is used. Histology can often serve as the ground-truth for the validation of image-based biomarkers or new imaging modalities.

Identifying noninvasive biomarkers to be used clinically as surrogate endpoints is not only valuable, but also promising. The advent of machine learning and artificial intelligence in medical imaging has led to the field of radiomics (170, 178–181). Like genomics and other “-omics,” radiomics allows quantifiable characterization of image features that provide a means to identify image-based biomarker surrogates for response to cancer treatment. Cameron and colleagues report a radiomics method, based on morphology, asymmetry, physiology, and size (MAPS) using multiparametric MRI (182). Most radiomics data have been reported for multicenter human studies, because a large number of subjects needs to be enrolled, the number of experimental animals in a single-imaging study often being a limiting factor. As quantitative imaging and radiomics lead to more image-based biomarkers, standardization and assessment of reproducibility are becoming important and will require a centralized image archive for multicenter preclinical studies.

Imaging is highly translational by nature and murine models have contributed enormously to the development of oncologic imaging methodologies (183). However, the complex, heterogeneous tumor microenvironment observed in human cancer is challenging to model in an immunodeficient animal system, particularly in terms of immunotherapeutic strategies. Lack of optimal preclinical models for testing is likely responsible for the dismal success rate (5%–8%) of cancer therapeutics developed in murine models to eventually obtain FDA approval for use in human patients (184). Dogs with naturally occurring cancers provide an alternative, complementary system for preclinical cancer research. The recent completion of the sequencing of the dog genome has shown that most of their 19,000 genes are orthologous or similar to humans (185). Companion animals live in our homes and are exposed to similar environmental and lifestyle influences. Their cancers grow slowly in an immunocompetent milieu, allowing for complex carcinogenesis, genomic instability, and immune avoidance to develop. Their size is such that serial biologic sampling can be performed before, during, and after therapy. These patients are imaged in human equipment, allowing for standardization of imaging protocols, improved spatial resolution for more accurate quantitative analysis, and adequate quality assurance of biodistribution for novel imaging probes. Power Doppler US and contrast-enhanced US were used to demonstrate tumor vascular response to antivascular therapy in canine patients with cancer noninvasively (186). There are several success stories to report today: ¹⁸FDG- and ¹⁸F-NaF PET/CT have been successfully used in canine patients with cancer to detect head and neck cancer and bone involvement of the nasal cavity (Fig. 1F, 23; ref. 187). An iodinated nanoparticle CT tracer, initially developed and validated in a murine lung cancer model (described above; ref. 73), has been successfully used in a CT study of companion dogs with spontaneous tumors (188). An anatomic and functional imaging probe for a novel immunotherapeutic was developed in dogs with spontaneous lymphoma (189). A recombinant oncolytic vesicular stomatitis virus that expresses a surface sodium-iodide symporter (NIS) protein and IFNβ was characterized. On the basis of clinical response to VSV-IFNβ-NIS therapy in dogs with T-cell lymphoma, a phase I clinical trial in people has been started (NCT03017820; ref. 189). In a follow-up study, dogs administered with VSV-IFNβ-NIS were evaluated to determine whether ¹⁸F-tetrafluoroborate radiopharmaceutical that binds to the cell surface NIS can be used to confirm successful viral gene replication (190). Veterinary patients with naturally occurring cancers may assist in the development of new molecular imaging probes, shorten the approval process of oncologic therapies, and create a mutually beneficial bridge between the fields of veterinary and human oncology.

In summary, multimodal oncologic imaging has become a cutting-edge necessity in preclinical (animal) cancer research. Understanding the physical principles of each modality is essential for applying the correct noninvasive imaging protocol to an animal-based study. Development of imaging probes for multimodal imaging technologies is also an important scientific and clinical goal. Each imaging modality brings specific insights into oncological questions and allows researchers to follow the biology, dictating the choice of the optimal reporter and imaging modality to best characterize cancer phenotype (191). The future also holds a big promise for PET/MRI (similarly to existing PET/CT), combining two powerful molecular, physiologic, and structural techniques into one scanner. Finally, we anticipate the introduction of novel predictive models and deep-learning algorithms (192) in the near future for managing sophisticated and complex image datasets in animal models of cancer.

M. Farhoud reports personal fees from Emit Imaging (employment) outside the submitted work. A. De Lille reports personal fees from UC Denver (honorarium) during the conduct of the study, SonoVol, Inc (employment and share-holder), and PerkinElmer (employment and share-holder) outside the submitted work. D. Simberg reports grants from NIH during the conduct of the study. L. Griffin reports nonfinancial support from AMAG Pharmaceuticals [provided medication (Feraheme) for a grant from Morris Animal Foundation; not related to this study] outside the submitted work. R.P. Mason reports grants from NCI P30 CA142543, NIH S10 OD018094-01A1, CPRIT RP120670-P3, and NCI R01 CA140674 during the conduct of the study. N.J. Serkova reports grants from NCI P30 CA046934, NIH UL1 TR002535, NIH S10 023485, NIH S10 027023, and SBF 639631. No disclosures were reported by the other authors.

This review was based on the 2019 Animal Imaging Workshop held at the University of Colorado Anschutz Medical Campus. The workshop was partly supported by the NIH Shared Instrumentation Grant Program (S10 OD023485, S10 OD023491, S10 OD027023, and S10 OD018094), the University of Colorado and the University of Southwestern CCSG Cancer Center grants (P30 CA046934 and CA142543, respectively), the University of Colorado Clinical Translational Institute (UL1 TR002535), and the Nutritional and Obesity Research Center (P30 DK048520).