Preclinical Magnetic Resonance Imaging and Systems Biology in Cancer Research

Abstract

Biologically accurate mouse models of human cancer have become important tools for the study of human disease. The anatomical location of various target organs, such as brain, pancreas, and prostate, makes determination of disease status difficult. Imaging modalities, such as magnetic resonance imaging, can greatly enhance diagnosis, and longitudinal imaging of tumor progression is an important source of experimental data. Even in models where the tumors arise in areas that permit visual determination of tumorigenesis, longitudinal anatomical and functional imaging can enhance the scope of studies by facilitating the assessment of biological alterations, (such as changes in angiogenesis, metabolism, cellular invasion) as well as tissue perfusion and diffusion. One of the challenges in preclinical imaging is the development of infrastructural platforms required for integrating in vivo imaging and therapeutic response data with ex vivo pathological and molecular data using a more systems-based multiscale modeling approach. Further challenges exist in integrating these data for computational modeling to better understand the pathobiology of cancer and to better affect its cure. We review the current applications of preclinical imaging and discuss the implications of applying functional imaging to visualize cancer progression and treatment. Finally, we provide new data from an ongoing preclinical drug study demonstrating how multiscale modeling can lead to a more comprehensive understanding of cancer biology and therapy.

CME Accreditation Statement: This activity (“ASIP 2013 AJP CME Program in Pathogenesis”) has been planned and implemented in accordance with the Essential Areas and policies of the Accreditation Council for Continuing Medical Education (ACCME) through the joint sponsorship of the American Society for Clinical Pathology (ASCP) and the American Society for Investigative Pathology (ASIP). ASCP is accredited by the ACCME to provide continuing medical education for physicians.

The ASCP designates this journal-based CME activity (“ASIP 2013 AJP CME Program in Pathogenesis”) for a maximum of 48 AMA PRA Category 1 Credit(s)™. Physicians should only claim credit commensurate with the extent of their participation in the activity.

CME Disclosures: The authors of this article and the planning committee members and staff have no relevant financial relationships with commercial interests to disclose.

The extensive pharmacological studies that have been aimed at developing and testing new anti-neoplastic compounds have clearly established the need for both highly accurate preclinical models, as well as the methods to longitudinally assess therapeutic response. These translational studies have driven significant technological advances in both anatomical and functional imaging and have resulted in the engineering of small animal imaging platforms, including luminescence, fluorescence, positron emission tomography, and magnetic resonance imaging (MRI). Currently, significant needs exist related to enhancing our ability to use quantitative anatomical, metabolic, and functional imaging, integrated into a systems approach to clinical medicine, to expand our understanding of cancer biology and to identify early markers of effective therapeutic intervention.

Mouse Models

Genetically engineered (GE) mice, since their introduction more than 30 years ago, have become increasingly important to our understanding of the interrelationship between genes, the organism, its environment, and the efficacy of new and existing treatment régimes. The early GE models were relatively simple, primarily relying on the integration of a promoter and transgene into the mouse genome, although technological advances such as inducible gene expression (eg, tetracycline, tamoxifen, or steroid hormones) and the Cre:LoxP gene ablation/modification systems have lead to more biologically relevant models with increased fidelity to the cellular and molecular basis for many human cancers.¹ In fact, the use of GE mice and xeno- and allo-grafting models have become the hallmark of translational platforms for the study of human cancers and their treatment.

Imaging

There are inherent difficulties associated with many GE models (eg, brain,² prostate,³ and gastric⁴), as in humans, in identifying disease initiation, its progression, and its treatment without invasive surgery or sacrificing the animal. Preclinical imaging modalities have been developed to overcome these limitations and to enable longitudinal studies analogous to clinical imaging and trials. Two of these preclinical imaging modalities (ie, MRI and positron emission tomography) allow for functional and metabolic imaging. Importantly, MRI is unique in its ability to combine high resolution anatomical imaging with functional and metabolic analyses, making it one of the most flexible preclinical/clinical imaging platforms.

Need for Informatics and Systems Biology to Complete the Multiscale Investigational Platform in Cancer

Imaging data can both define the location of deep tissue tumors and provide data related to cancer progression and regression (response to therapy), because many longitudinal preclinical studies seek to combine in vivo imaging data with ex vivo pathology, as well as with genetic and molecular changes within the target tissue to quantify complex phenotypic traits. The complexity of the data generated, however, which by necessity must include those of the underlying model, requires cross-platform multiscale approaches. These platforms, by definition, must integrate the imaging data for a region of interest (eg, a tumor), which occurs at the organismal and macroscopic scales, with the data from the cellular or microscopic scale (eg, pathology, vascularity) and with data derived at the subcellular or molecular scale (eg, mRNA, proteomics, metabolomics). Currently, there are National Institutes of Health initiatives such as the cancer Biomedical Informatics Grid, and National Cancer Institute initiatives such as The Cancer Genome Atlas that seek to support the groundwork collaborations between cancer imaging laboratories and molecular and bioinformatics/bioengineering groups that are needed to develop these capabilities. Additional innovations, such as the Georgetown Database of Cancer (G-DOC), a systems medicine data integration, analysis, and visualization platform,⁵ will eventually allow for real-time data analysis. These bioinformatics platforms will allow for correlative data comparisons, such as those available from The Cancer Genome Atlas, and for target validation, as well as for predictive modeling of tumors and their responses to intervention.

In this review, recent advances in the use of MRI in preclinical cancer research are highlighted, and we discuss and give an example of the central role that MRI combined with systems biology can fulfill in imaging-based preclinical studies.

Tumor Imaging

Ultrasound and Computed Tomography

Many outstanding imaging modalities exist for anatomical imaging (eg, ultrasound was one of the earliest translations of human imaging into preclinical modeling). Ultrasound provides a rapid and sensitive method for the identification of changes in skin architecture, volumetric analyses of surface tumors, such as xenografted human cancer cells, oncogene-induced changes in the mammary gland before palpable tumors arise,⁶ and vascular blood flow by Doppler. Ultrasound is limited to some extent by its depth of imaging, its dynamic range, and its lack of contrast in some tissues, and also by its inability to perform functional and metabolic imaging. Similarly, computed tomography is widely used in both clinical and preclinical settings and provides extremely high resolution anatomical imaging, however, the use of computed tomography is limited in its metabolic and functional imaging as a stand-alone modality. Furthermore, computed tomography involves ionizing radiation exposure necessitating judicious use in both research and diagnostic settings.

MRI

MRI has long been used for anatomical imaging to allow the visualization of changes that occur during disease progression. In addition, magnetic resonance-based chemical imaging using hyperpolarized carbon-14 has shown great promise in visualizing alterations in tissue pH⁷ and cellular redox activity,⁸ and proton magnetic resonance spectroscopy (¹H-MRS) is a rapidly developing field of discovery, enabling the measurement of altered energy and metabolic pathways in situ. In combination with anatomical MRI, for example, the measurement of changes in citrate to choline ratios by ¹H-MRS have identified cancerous areas of the human prostate.⁹ Using ¹H-MRS, we found that the GE mice we developed that presented with spontaneous prostate cancer, also exhibited an altered choline-to-citrate profile, which was seen in humans.³ These data served to further validate mice as a biologically relevant model amenable to preclinical studies with translational impact on the clinical prostate disease. Importantly, identifying metabolic changes that correlate with cancer progression allows for a possibility of monitoring intervention through dynamic monitoring of tumor chemistry in vivo.

The recent application of functional magnetic resonance pulse sequences to cancer is also beginning to enhance the noninvasive preclinical and clinical diagnostic accuracy and monitoring of disease progression and treatment efficacy of a number of neoplasms, including those of the brain, prostate, and breast. Diffusion-weighted imaging takes advantage of altered localized diffusion of water in tissue and is being actively supported by the National Cancer Institute as a noninvasive technique for imaging cancer biomarkers.¹⁰ Diffusion tensor imaging can identify differences in directional diffusion (anisotropy), for example, between white-matter tracts and nearby malignant tumors. Perfusion-weighted imaging estimates the rate of regional blood flow, which is higher in regions with higher metabolic demands, and regional blood volume, which is an indirect marker of angiogenesis.¹¹ Susceptibility-weighted imaging uses velocity compensated, radio frequency-spoiled, high-resolution, three-dimensional gradient echocardiography scans. With susceptibility-weighted imaging, signals from substances with different susceptibilities, such as tissue-fat, tissue-tumor, and tissue-water interfaces will become out of phase, enhancing their boundaries.¹² Hori et al¹³ have noted that the ability to detect these changes in the tumor can lead to better volume recognition and tissue segmentation, helping to identify smaller structures and delineate vascular and microvascular lesions. Dynamic contrast-enhanced (DCE)-MRI provides a noninvasive method for evaluating tumor vasculature patterns based on contrast accumulation and washout.¹⁴ DCE-MRI can potentially identify intratumoral heterogeneity of vascular permeability reflecting tumor angiogenic activity. We predict that both spectroscopic and functional/diffusion imaging hold great promise for visualizing responses to both chemical and radiological therapies, nearly in real time.

Data Integration, Bioinformatics, and Systems Biology

High-throughput discovery technologies, such as next-generation sequencing, and the core -omics, such as genomics, transcriptomics, proteomics, and metabolomics, have furthered our understanding of cancer initiation and progression. In addition, these approaches have helped to define diagnostic and mechanistic biomarkers, as well as establishing the signaling network rewiring that occurs during tumorigenesis. Most of these studies, however, are performed on excised tumor tissue making them static studies, which are unable to reflect the dynamic changes that occur during therapeutic intervention. To overcome these limitations, computational models of disease are essential to allow for the prediction of disease progression and/or responses to treatment with time. The integration of the static ex vivo data with imaging and the dynamic information from imaging will help to build these models, however, before model building, careful integration of the various forms of data (ie, the -omics, pathology, and/or imaging) and the subsequent knowledge extraction are necessary.

Computational Platforms for the Integration of Molecular, Clinical, and Imaging Data

As previously discussed, a promising new area of cancer research is the integration of image features (eg, shape, size, metabolism, perfusion) and molecular data (eg, gene expression, gene copy number, microRNA, metabolomics). Furthermore, once fully validated, the in-depth understanding of the molecular characteristics underlying image features may result in their use as predictive markers of prognosis and therapeutic response. In the future, these features may also be incorporated into clinical decision-making, which would enhance patient treatment and care.

Currently, the imaging-based oncological decision-making used in assessing responses to chemotherapy is based primarily on changes in tumor size. This approach, although useful, has limitations for predicting recurrence and assessment of minimal residual disease. Imaging platforms, such as combining positron emission tomography with either computed tomography or MRI, can offer a more accurate view of response to therapy at the cellular level by assessing diverse aspects of the tumor and its environment, including tumor metabolic activity, tissue vascularity, cellular apoptosis, growth factor levels, and blood oxygenation to name a few. This functional information can be more effective than conventional anatomical imaging alone for identifying early responses to therapy and may ultimately provide the necessary clinical information so that less effective therapies can be stopped earlier. Such studies will have multiple data points that need to be integrated in a unified system that can process results from multiple studies.

Three independent examples of platforms being built for the purpose of integrating disparate data include the Information Sciences in Imaging group at Stanford, the Investigation of Serial Studies to Predict Your Therapeutic Response With Imaging and Molecular Analysis 1 and 2 trials, and the G-DOC. The Information Sciences in Imaging group is developing several tools to collect and integrate annotated imaging to clinical and molecular data through novel computational models. One such tool is the Electronic Physician Annotation Device, which is an open source tool enabling researchers and clinicians to identify and quantify imaging biomarkers. The Investigation of Serial Studies to Predict Your Therapeutic Response With Imaging and Molecular Analysis 1 was a collaboration of the American College of Radiology Imaging Network, Cancer and Leukemia Group B, and the National Cancer Institute’s Specialized Programs of Research Excellence. This study was initiated to identify the molecular markers of response to conventional neoadjuvant chemotherapy, as well as the imaging markers associated with response to therapy.¹⁵ The G-DOC, developed at Georgetown University, is a tool that currently contains various types of -omics data integrated with clinical metadata and patient outcome data. This promising new model for the storage, integration, and visualization of multiple disparate data types is described as follows.

It is clear that one of the major challenges in analyzing the large and complex datasets generated by the various integrative platforms is making them available, and more importantly useful, to the end user, whether it is for the clinical caregiver or the translational research field as a whole. For example, the generation of reports capable of positively impacting clinical care will require a significantly filtered and reduced set of data points to be useful to a physician who will be making critical decisions regarding how to best treat their patients based on the newest information available. We designed the G-DOC as both a cutting-edge data integration platform and as an integrative knowledge discovery system for oncology and translational research communities. The goal of the G-DOC is to provide cancer researchers with a broad range of tools for data reduction, visualization, and analysis.⁵ In addition, the G-DOC includes manually curated information on small molecules as potential drug candidates for key biomarkers/target proteins. The G-DOC further supports flexible clinical criteria browsing to enable selection of specific patient cohorts, and it facilitates the generation of detailed reports and informative publication-quality plots. Internal chemical compound libraries can be screened easily using the integrated structure and detailed molecular property search functions, with the goal of identifying new therapeutic candidate molecules.

The G-DOC also allows researchers to securely share knowledge with others through a powerful suite of collaboration-enabling features operating within its secure environment, and the G-DOC supports computationally intensive, high memory-using tasks, such as class comparison, hierarchical clustering, principal component analysis, and network analysis for transcriptomic, genomic, and metabolomic data. The G-DOC is constantly being updated, including our current modifications that allow for the integration of imaging data. By providing a powerful but easy-to-use interface, the G-DOC is intended specifically to address the activation barrier normally encountered by basic, clinical, and translational researchers when trying to make use of biomedical informatics tools.

Detection of Topological Changes in Biological Network between Phenotypes

Gene regulatory networks are both context-specific and dynamic in nature. With different conditions, diverse regulatory components and mechanisms can be activated or repressed, leading to rewired genetic and signaling networks and topological changes. For example, a deviation from the normal regulatory network topology may reveal the mechanism of tumorigenesis,¹⁶ and the pathways that are involved may serve as biomarkers or drug targets.^17,18 The differential dependency network is a graphical, model-based analytical tool for detecting and visualizing statistically significant topological changes in dependency networks.^19,20 As an integrated component of the G-DOC, the differential dependency network can be applied to different tissue types, including normal versus tumor, different cancer subtypes, drug sensitive versus insensitive tumors, as well as the differences in pathways targeted by different drugs or drug combinations, accurately capturing the topological changes of the network.

Multiclass Biomarker Selection for Studying Heterogeneous Cancers

Multiclass gene selection is an imperative task for identifying phenotype-associated mechanistic genes and achieving accurate diagnostic classifications.^21,22 The Phenotypic Up-regulated Gene Support Vector Machine is a cancer biomedical informatics grid analytical tool for multiclass gene selection and classification.²³ The Phenotypic Up-regulated Gene Support Vector Machine provides a simple yet accurate strategy to identify statistically reproducible mechanistic marker genes for the characterization of heterogeneous cancers. Multiscale preclinical modeling of cancer progression and treatment with vector machines such as the Phenotypic Up-regulated Gene Support Vector Machine will be necessary to begin to understand the mechanisms underlying tumor regression, after apparently successful interventional therapy and the possible roles that are played by tumor heterogeneity.

An Example of MRI and Systems Integration for Assessment of Drug Efficacy

Medulloblastoma

Childhood brain tumors represent approximately 25% of all childhood cancers and medulloblastoma (MB) is the most common childhood brain malignancy, accounting for 25% of these tumors.²⁴ Four consensus subtypes of MB have been molecularly defined based on gene expression patterns and chromosomal abnormalities.^25–27 MBs are highly malignant, poorly differentiated tumors that have a propensity to spread throughout the neuraxis, either early in the course of the illness or with disease recurrence. Immediately after surgical removal of the tumor, radiotherapy to the entire brain and spine (craniospinal irradiation) is administered and is followed by nine months of multi-agent chemotherapy. For standard risk MB, the five-year survival rate approaches 80%. In high-risk MB, which is classified by patient age, histological/molecular features, postoperative residual disease, and/or the presence of metastatic disease, survival rates range from 20% to 60%, with significantly more morbidity from intensified radiation and chemotherapy. Unfortunately, the majority of all MB survivors are left with auditory, neuroendocrine, and neurocognitive deficits that can severely affect employability and social achievement.²⁸ It must also be noted that radiation is avoided in children less than three years of age due to its destructive effects on the developing nervous system, confounding effective treatment in the youngest patients. The long-term prognosis for most of these children is dismal.

The ability to predict which cases will ultimately experience disease recurrence despite therapy is the first step in improving the therapeutic index of existing therapies. Recently, progress has been made toward the molecular classification of medulloblastoma subtypes with prognostic implications. Historically, therapeutic intensification has been the only strategy to address the high-risk disease strata. The unfortunate consequence of this has been significantly more toxicity with only marginally better outcomes.

Rationally designed, molecularly targeted therapies, which are based on the underpinnings of preclinical and clinical biological discovery, hold the promise for exploiting the tumor altered networks to provide effective therapy. Such an approach will require clinical trial testing based on a different framework than that developed for cytotoxic therapies due to the fact that cell death resulting in tumor shrinkage will no longer be the only potential positive outcome of a successful drug. A molecularly targeted agent must be judged on its ability to reach the anatomical location of the tumor, maintain effective dosage levels, effect the molecular change for which it was designed, and bring forth the desired change in tumor cell phenotype, for example growth arrest, apoptosis, anti-angiogenesis, or loss of invasion, while minimizing off-target cytotoxicity. The assessment of these outcomes will require more than conventional anatomical imaging; it will require a systems biology-based approach that combines functional imaging with biomarker quantification in a platform capable of providing a real-time assessment of disease state. Paradoxically, the effort to reduce therapy for patients with a better prognostic class of disease may entail the risk of increasing disease recurrence in comparison with those rates achieved with potentially excessive therapy. The scaling back of treatment, while still maintaining patient safety, will require the ability to monitor therapeutic efficacy serially and in real time to identify members of the group whose disease is not responding as expected, in effect creating a stopping rule. Anatomical and functional imaging combined with biomarker quantification holds the promise of creating such a platform.

MRI as a Diagnostic Platform

The development and testing of such a systems-based approach to drug validation for treating MB is not feasible using human subjects in the early stages; animal models exist that faithfully recapitulate childhood MB. In an end-point study using MRI as a diagnostic platform, we recently established that arsenic trioxide, a Food and Drug Administration-approved, second-line therapeutic used for the treatment of acute promyelocytic leukemia, was effective in increasing survivorship in the ND2-SmoA1 GE model of MB.² Longitudinal anatomical imaging studies have now been performed on control and ND2-SmoA1 mice using our 7T Bruker MRI, and ¹H-MRS was also performed using a localized point resolved spectroscopy sequence as previously described^1,29,30 (Figure 1A). The metabolite ratios of normal and MB tissues were established (Figure 1B), showing that altered ratios of choline to n-acetylaspartate, choline to creatine, choline to taurine, and choline to myoinositol were associated with MB, correlating well with previous studies performed in this model.³¹ A representative comparison of the spectra of tumor-bearing ND2-SmoA1 mice after treatment with arsenic trioxide versus those of control-treated mice is shown (Figure 1C). The ratios of these metabolites were found to be dynamic early indicators of tumor targeting by arsenic trioxide, as some changes were observed before any obvious changes in tumor growth kinetics were seen (not shown). Reverse phase protein microarray^32,33 was then performed on small biopsies collected at necropsy to interrogate the functional protein signaling architecture, and the data were used to inform our differential dependency network. Using the pathway activation data acquired by reverse phase protein microarray from the MB progression versus response-to-drug experiments, we performed pair-wise and one versus rest differential network analysis using our differential dependency network modeling tool.¹⁹ Our analyses indicated that signaling networks encompassing the receptor tyrosine kinases ErbB-2 and insulin-like growth factor receptor feeding into cell cycle regulators, such as CDK2 and GSK3β, were perturbed in the tumors versus normal cerebellum (Figure 1D). After arsenic trioxide treatment, these receptor signaling pathway lesions were repaired, leading to the re-establishment of the linkage to cell cycle regulation and apoptosis (Figure 1D).

Figure 1.

Preclinical imaging drug responses in the ND2-SmoA1 mouse. A: ¹H-MRS spectra of the ND2-SmoA1 medulloblastoma; insert, MRI of the tumor and voxel placement used to acquire the spectra. B: ¹H-MRS metabolite ratios of n-acetylaspartate (NAA), myoinositol (mI), creatine (Cre), lactate (Lac), and taurine (Tau) versus choline (Cho) in normal mouse cerebellum and ND2-SmoA1 medulloblastomas. C: Alterations in MB metabolite ratios after treatment with subclincial doses (0.15 mg/kg three times per week) of pharmacological grade arsenic trioxide (ATO) versus control-treated mice. D: Differential dependent network modeling of reverse phase protein microarray performed on samples of normal cerebellum versus MB (red) and those of ATO-treated versus control-treated MB (green) collected from the mice in panels A–C.

Conclusion and Future Directions

Much has been accomplished with respect to integrating functional and metabolic imaging into preclinical and clinical studies, although much remains to be done. Currently, platforms such as the G-DOC provide the means by which existing and emerging -omics data can be interrogated to improve the diagnostics and outcome, both experimentally and for individual cancer patients. Ultimately, it is important to seamlessly combine in vivo imaging-based data associated with tumor progression and/or regression (regardless of the imaging modality), with both ex vivo pathological examinations performed on tissue samples as well as patient outcomes. In addition, the exciting new methodology that was recently developed allows for the continuous culturing of a patient’s own normal and malignant epithelial cells, termed conditionally reprogrammed cells,^34,35 adds yet another powerful approach to cancer research and discovery. The long-term vision for implementing systems-based integration of multimodal imaging and -omics information, performed on matched normal and malignant tissues, requires the establishment of robust and comprehensive informatics platforms that will support hypothesis generation and hypothesis testing to positively impact both cancer research and its clinical practice. Continued collaborative efforts between cancer researchers and clinicians, imaging experts, drug discovery programs, and computational scientists are essential.