CRYO-IMAGING ASSESSMENT OF IMAGING AGENT TARGETING TO DISPERSING AND METASTATIC TUMOR CELLS
by
Mohammed Q. Qutaish
Submitted in partial fulfillment of the requirements For the degree of Doctor of Philosophy
Dissertation advisor: David L. Wilson, PhD
Department of Biomedical Engineering Case Western Reserve University
August, 2014
CASE WESTERN RESERVE UNIVERSITY SCHOOL OF GRADUATE STUDIES
We hereby approve the thesis/dissertation of
Mohammed Q. Qutaish
candidate for the Ph.D. degree*.
Committee Chair David L. Wilson
Committee Member Susann Brady-Kalnay
Committee Member Zheng-Rong Lu
Committee Member Andrew Rollins
Committee Member Zhenghong Lee
Date of Defense June 23, 2014
*We also certify that written approval has been obtained for any proprietary material contained therein. Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Table of Contents
1 Chapter 1: Introduction ...... 10 1.1 Tumor Metastases ...... 10 1.1.1 Glioblastoma Multiforme...... 12 1.1.2 Breast Cancer ...... 13 1.2 Clinical Detection of Metastases ...... 14 1.3 Targeted Imaging Agents ...... 16 1.3.1 SBK2 peptide ...... 17 1.3.2 CREKA peptide ...... 18 1.4 Preclinical Evaluation of Targeted Imaging Agents ...... 18 1.4.1 Animal Models...... 19 1.4.2 Whole mouse imaging methods ...... 19 1.4.3 Multimodality imaging ...... 23 1.4.4 Literature Review of Whole Mouse Image Registration Algorithms ...... 24 1.5 Cryo imaging ...... 27 1.6 Overview of the Thesis ...... 29 1.6.1 Combining cryo with Other Imaging Modalities ...... 30 1.6.2 Aims ...... 32 1.6.3 Significance...... 33 1.7 Organization of the Thesis ...... 33 1.8 My Specific Contributions ...... 34 2 Chapter 2: Analysis Algorithms for Glioblastoma Multiforme Cell Dispersal . 36 2.1 Introduction ...... 36 2.2 Materials and Methods ...... 37 2.2.1 Experimental method and materials...... 37 2.2.2 Image Processing Algorithms ...... 40 2.3 Results ...... 49 2.4 Discussion ...... 63 2.5 Conclusions ...... 67
Case Western Reserve University Page 2 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
3 Chapter 3: Analysis of Glioblastoma Multiforme Tumor Models ...... 68 3.1 Introduction ...... 68 3.2 Materials and Methods ...... 70 3.2.1 Orthotopic xenograft intracranial tumors ...... 70 3.2.2 Cryo-imaging of tissue samples ...... 71 3.2.3 Image processing algorithms for visualization of tumor cells and vasculature ...... 71 3.2.4 Dispersal on white matter ...... 72 3.3 Results ...... 72 3.4 Discussion ...... 86 4 Chapter 4: Assessment of SBK2 Targeting to GBM Dispersing Tumor Cells .. 89 4.1 Introduction ...... 89 4.2 Material and Methods...... 92 4.2.1 Experimental methods ...... 92 4.2.2 Image analysis algorithms...... 94 4.3 Results ...... 97 4.4 Discussion ...... 108 5 Chapter 5: Validation of Targeting for MRI Molecular Imaging Agents...... 112 5.1 Introduction ...... 112 5.2 Materials and Methods ...... 116 5.2.1 Image Processing and analysis...... 118 5.2.2 Experimental materials ...... 126 5.3 Results ...... 133 5.4 Discussion ...... 151 6 Chapter 6: Discussion and Future work...... 160 7 Appendix...... 167
Case Western Reserve University Page 3 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
List of Tables
Table 1 Comparison of small animal imaging modalities...... 22
Table 2. Pseudo code to find optimal parameters for cost function using lungs ...... 131
Case Western Reserve University Page 4 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
List of Figures
Figure 1. Comparison between BLI and cryo...... 29
Figure 2. Blood vessel visualization using bright field images...... 44
Figure 3. Tumor visualization using fluorescence images...... 45
Figure 4 Steps for blood vessel visualization for tumor 1...... 52
Figure 5. Steps for visualization of tumor and dispersing cells for tumor 2...... 54
Figure 6. Results of the dispersed cell detection algorithm for tumor 1...... 56
Figure 7. LN-229 tumor cell dispersal in 2-D...... 57
Figure 8. Projection of tumor growth along a blood vessel shown for tumor 5...... 58
Figure 9. Tumor cells dispersing along blood vessels shown for tumor 1...... 60
Figure 10. Dispersal distances from the main tumor mass...... 62
Figure 11. Glioblastoma xenograft tumor growth characteristics in 2 dimensions...... 74
Figure 12 Tumor cell dispersal on blood vessels in 2 dimensions...... 76
Figure 13. Three-dimensional reconstruction of Gli36Δ5 and U-87 MG tumors indicates limited cell dispersal...... 78
Figure 14 Three-dimensional reconstruction of LN-229 tumors shows cell dispersal on blood vessels in the tumor microenvironment...... 80
Figure 15 Three-dimensional reconstruction of CNS-1 tumors showing their high dispersal along blood vessels...... 81
Figure 16. Cell dispersal distance measurements...... 83
Case Western Reserve University Page 5 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 17. Three-dimensional reconstruction of U-87 MG, LN-229, and CNS-1 cell dispersal on white matter tracts in the brain...... 84
Figure 18. Analysis of tumor blood vessel density...... 85
Figure 19 CNS-1 glioma tumors and dispersing cells are labeled by the PTPµ probe. .... 98
Figure 20. The PTPµ probe specifically labels CNS-1 glioma cells that have dispersed
from the main tumor...... 102
Figure 21. Three-dimensional views of dispersed cells from CNS-1 intracranial tumors
labeled by the PTPµ probe...... 104
Figure 22. Histogram of the average number of GFP-positive dispersing cells colabeled with the PTPµ probe per unit distance from the main tumor, ± standard error (n = 4 tumors analyzed)...... 105
Figure 23. Dispersing cells from LN-229 intracranial tumors are specifically labeled by
the PTPµ probe...... 107
Figure 24. Pre-clinical platform methodology for metastases targeting analysis...... 115
Figure 25. Experimental design for evaluation of CREKA peptide as MR imaging agent.
...... 117
Figure 26. Registration quality measures for lungs...... 135
Figure 27. Registration quality measures for selected (w1 and w2)...... 136
Figure 28. Volume change due to freezing using CT...... 137
Figure 29. Determinant of Jacobian maps...... 139
Figure 30. Visual evaluation of registration quality for whole mouse...... 141
Case Western Reserve University Page 6 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 31. Metastases distribution in mouse...... 143
Figure 32. Multi-modality view of a slice through mouse after registration...... 144
Figure 33. Multi-modality evaluation of micro-metastases targeting...... 146
Figure 34. Rose Criterion of detection for lungs MRI metastases...... 148
Figure 35. Histology and cryo imaging example...... 150
Figure 36. Metastases targeting by CREKA-Gd over multiple scans...... 156
Figure 37. The mold that was used in cryo-MRI registration to reduce mouse deformation during MRI and freezing...... 168
Figure 38. Whole tumor growth and metastases formation over time using BLI...... 169
Case Western Reserve University Page 7 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
List of Abbreviations
1D: One dimension 2D: Two dimension 3D: Three dimension AJCC: American Joint committee on cancer BLI: Bioluminescence Imaging Case: Case Western Reserve University CNR: Contrast to noise ration CT: Computed tomography CTC: Circulating tumor cells DoOG: Difference of offset Gaussians ECM: Exctracellular matrix EGFP: Enhanced GFP FFD: Free form deformation FFT: Fast fourier transform FMT: Fluorescence molecular tomography FNIII: Fibronectin Type III GB: Gigabyte GBM: Glioblastoma multiforme GFP: Green fluorescence protein GPU: Graphic processing unit H&E: Hematoxylin and Eosin HER2: human epidermal growth factor receptor 2 ICP: Iterative closet point Ig: Immunoglobulin MRI: Magnetic resonance imaging NMI: Normalized mutual information O.C.T: Optimal Cutting Temperature compound PCR: Polymerase chain reaction PET: Positron emission tomography PTP: Protein tyrosine phosphatase RGB: Red green blue ROI: Region of interest RPM: Robust point matching SDF: Stromal cell-derived factor SNR: Signal to noise ratio TNF: Tumor necrosis factor TPS: Thin plate splines VEGFR: Vascular endothelial growth factor receptor
Case Western Reserve University Page 8 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
By
MOHAMMED QUTAISH
Abstract:
The goal of this work is to create a platform methodology consisting of cryo-imaging experimental methods and specialized image analysis software to provide unique 3D, quantitative characterization of tumor models, including spread via dispersal and metastasis, and assessment of imaging agent targeting to dispersed tumor cell and metastatic tumor. Analyses used microscopic, 3D cryo-imaging which has the sensitivity to detect fluorescently labeled single cells over large volumes of tissues. Specifically, tumor cell dispersal in Glioblastoma multioforme (GBM) and micrometastases in breast cancer mouse models were detected, quantified, and visualized. Targeting of the fluorescently labeled SBK2 peptide to GBM dispersing cells, and CREKA peptide multiplexed Gd-MR probe to breast cancer metastases, were analyzed. For GBM tumor cell dispersal analyses in cryo-images, algorithms were developed to detect blood vessels, dispersing tumor cells, white matter tract and main tumor mass, as well as measure cell dispersal distance. Multiple GBM cell lines were characterized to find those that showed high dispersive patterns similar to the human disease. Software was created to assess how far from the main tumor mass, SBK2 efficiently labeled dispersing cells. Results showed that LN-229 and CNS-1 cell lines are highly dispersive, and cells mainly dispersed along blood vessels and white matter tract. Dispersal distance was as far as 562µm in LN-229 and >3mm in CNS-1. Fluorescently labeled SBK2 peptide labeled more than 99% of dispersing cells, and as far as 3.5 mm. For breast cancer metastasis analyses, software was created to quantify number and size of metastases using cryo-imaging volumes. Multimodality 3D deformable image registration was employed to register MRI and cryo-imaging volume. This enabled the validation of CREKA peptide targeting in MRI using the high resolution cryo volumes, and provided information about limitation and efficiency of the developed MRI agent. The 4T1 cell line was used to create a metastatic breast cancer model. Results showed an average of 156 metastases in cryo-volumes ranging in size of 0.1–8mm in diameter. Metastases were mainly found in lungs, liver, bones and adrenal gland. Rose criterion showed >73% of micrometastases in lungs were labeled by the MRI CREKA. Results were shown visually and quantitatively. Analysis methods and software demonstrated in the thesis should be applicable to a wide range of studies of cancer, imaging agents and theranostics.
Case Western Reserve University Page 9 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
1 Chapter 1: Introduction
The need to target tumor cells at the molecular level led to the development of molecular
targeted agents. Validation of these developed agents is required in order to assess their efficiency in targeting all tumor cells and at a whole body level. Current imaging modalities lack the resolution, sensitivity and/or field of view required in order to provide adequate assessment about the targeting of these agents especially to dispersing tumor cells and micrometastases. In this work, I present a unique preclinical platform methodology including imaging protocols and analysis software to evaluate the newly developed targeted agents at molecular level. In this chapter, I provide necessary background information about tumor metastases, clinical effort to detect and image metastases, and advantages and disadvantages of the current preclinical imaging methods.
1.1 Tumor Metastases Malignant tumor cells can detach from the primary tumor mass and intravasate into blood
and/or lymphatic vessel walls to form circulating tumor cells (CTC) [1]. These
disseminated cells will navigate the circulatory system until they find an appropriate
microenvironment where they can extravasate and grow to form secondary tumors or
metastases. Although tens of thousands of tumor cells may enter the circulation, it has
been reported that the mechanical stress and abundance of immune cells in the circulation
affect circulating tumor cell survival. As a result, less 0.02 % of these cells are able to
form metastases [2, 3].
Case Western Reserve University Page 10 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Depending on the tumor type, circulating tumor cells tend to metastasize to
specific organs. For example, breast cancer metastasizes to lungs, bone marrow, liver and
brain; prostate cancer to bones; lung cancer to liver, bones and brain. Some other tumors
do not result in metastases at distant sites. For example, tumor cells in glioma invade and
migrate within the brain in a process called cell dispersal [4-6]. Ovarian tumors are restricted in the peritoneal cavity [7]. Muscles are a rare site for metastasis formation [2].
Multiple hypotheses address the selection of site for metastasis development. One of the
most popular is the ‘seed and soil’ hypotheses by Paget [8], which states that certain
tumor cells (seeds) have affinity to certain organs (soil) and metastases form only if the
seed and soil are compatible [9], suggesting that the “soil” has unique characteristics that
help in metastases recruitment and growth. Another hypotheses explaining tissue
selective formation, at least for tumor cells circulating in blood vessels, is simply
explained by the passive cells’ entrapment in capillaries of organs [3]. This hypothesis
suggests that depending on main mass location, disseminated cells extravasate to organs
which come first in blood circulation after blood passes through the main tumor mass,
carrying disseminated cells that get entrapped in the capillaries of subsequent organs.
Examples that support this are breast cancer which metastasizes to lungs, lymph nodes
and bone as well as colon cancer which metastasizes to the liver [2, 3]. Nevertheless,
answers to the reasons of why some tumors metastasize to specific organs are still not
definitive and are under investigation [10].
Case Western Reserve University Page 11 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Significance of metastases emerges from the fact that, it is the development of
metastases that leads to the majority of cancer deaths [11]. However, not all cells in a
metastatic lesion can contribute to the growth of secondary tumors. In fact, it has been
reported that only a small population of these cell have stem cells-like properties, where
they express markers similar to stem cells, and these cells are the ones that serve as seeds
and contribute to metastatic growth, maintenance and heterogeneity [11, 12]. They can even persist in a dormant non-proliferating state for many years [13, 14], are more resistant to therapy [15-18], and they can lead to tumor recurrence even after aggressive surgical resection, and radio and chemotherapies [19-22]. Therefore, detection and quantification of metastases are important to identify the residual disease, and survival of metastases is considered an indicator of the failure or success of anticancer therapy [14,
23, 24].
Tumor cell dispersal in Glioblastoma multiforme (GBM), and metastases in breast cancer were utilized in this work. A brief description for both follows.
1.1.1 Glioblastoma Multiforme
Glioblastoma multiforme (GBM) is the most common and the most biologically aggressive primary brain tumor, with a median patient survival time of one year after diagnosis [25, 26]. This poor prognosis is due in part to the significant heterogeneity of
GBM, highly dispersive nature the cells and the uncontrolled cellular proliferation inside the cranial space [27-29]. Disseminated cells in GBM do not migrate to other organs to form distant metastases. Rather, they disperse along characteristics pathways within the
Case Western Reserve University Page 12 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
brain. These pathways include myelinated fiber tracts, perivascular regions, ventricular
linings and meninges [6, 28-31]. Multiple theories exist to explain cell dispersal of GBM
along these pathways [27]. One explanation is that GBM cell dispersal occurs along the
pathways of least resistance with larger inter cellular spaces. This could explain cell dispersal along myelinated fiber tracts where the fibers are aligned offering a reduced barrier for cells migration. Moreover, these pathways are rich in cell adhesion molecules
(CAMs) and extracellular matrix molecules (ECM). Since GBM cells have been shown to upregulate integrin receptor expression, this gives them the ability to interact with
ECM molecules which are associated with the brain vasculature [29, 32]. Hence, detection and targeting of GBM cell dispersal is important in order to develop successful targeted therapeutics and increase patient survival rate.
1.1.2 Breast Cancer
Breast cancer is the most common cancer in women, and one of the leading causes of death in women in the western world after lungs cancer [33]. The main cause of death is the metastases that form in distant organs from the breast main tumor mass [34]. The risk of the disease depends greatly on personal characteristics such as race, age and family history [35, 36]. Like other malignant tumors, breast cancer shows specificity in metastases formation in distant organs, where it mainly metastasizes to lungs, lymph nodes, bones and brain [34]. Although most women do not show signs of secondary tumors after the first line of treatment which includes surgery and chemotherapy, late metastases can develop even after more than 10 years [37, 38].
Case Western Reserve University Page 13 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
The challenge with breast cancer metastases is that they are very heterogeneous,
which make them difficult to cure and even to assess the malignancy risk associated with
them [34]. Breast cancer metastasis to auxiliary lymph nodes is considered poor
prognosis as they can lead to the disease relapse [39], and they are considered an indication of an aggressive tumor phenotype [39]. Metastases in bones can result in serious bone related problems such as bone fractures, hypercalcemia and spinal cord compression leading to a median patient survival of 2-3 years after detection [40, 41]. In addition, median survival of patients with detected liver metastases is less than six
months [42]. Hence, early detection and targeting of breast cancer metastases and
micrometastases is very important to prevent their growth in essential organs and increase
patient survival rates.
1.2 Clinical Detection of Metastases Metastases are classified depending on their size, and according to AJCC cancer staging
manual [43], metastases that are > 2 mm in diameter are called macro-metastases, less
than 2 mm and greater than 0.2 mm are called micro metastases and metastases that are
less than 0.2 mm in diameter are called single isolated cells. The clinical significance of
micrometastases is still under investigation [44-46]. However, there is no doubt that any residual tumor cells after surgery and chemotherapy can at some point contribute to the recurrence of the disease, as it has been reported, for example, that tumor relapse occurred even after many years in breast cancer patients who had bone marrow micrometastases at the time of diagnosis [47].
Case Western Reserve University Page 14 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Biopsy and dissection for pathological studies remains one of the most valuable
tools that are used clinically to evaluate the prognosis of the disease and detect micro- metastases. However, it has been shown in a review by Rajendra et al., [48] that the
reliability of detection via the dissection of lymph nodes, for example, depends on the
protocol used to acquire the histological sections and in fact can miss some micro-
metastases leading to a false negative examination. Of course, biopsy of tissue from an
organ would be very likely to miss sparse micro metastases.
In contrast, medical imaging is one of the most powerful tools that is clinically
used in cancer detection, therapy response, surgery guidance, and staging of the disease
[49, 50]. Widely used imaging modalities in tumor imaging include positron emission
tomography (PET), magnetic resonance imaging (MRI), computed tomography (CT), and
Mammography. Each of these modalities has specific advantages and disadvantages. In
fact, the research effort is focused to improve the sensitivity and specificity of the
medical imaging modalities [51, 52]. For example, PET, which is usually combined with
CT for structural information, uses a radioactive tracer to visualize tumors with a typical
clinical resolution of 4 mm. PET detects tumor cells based on tumors metabolism which
measured by the tracer uptake. Although this modality is very sensitive to detect
metastases, due to the image resolution and partial volume effect [53], micro-metastases
might be under-detectable. Moreover, there is no one optimal tracer to detect all
metastases with PET. For example, the widely used 18F-FDG for cancer imaging can only
detect tumors that show high proliferation and low differentiation [49]. More research is
Case Western Reserve University Page 15 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
still needed to optimize PET targeted molecules and develop combinations of tracers to
detect more metastases [54].
MRI provides basic anatomical details with very good spatial resolution and no
concerns about radiation dose for the patient with typical clinical spatial resolution of 1
mm. In fact, for breast cancer detection, MRI has higher sensitivity than mammography
or ultrasound [55]. However, the capability of MRI to detect metastases is still limited
without using contrast agents that specifically amplify the MRI signal to background ratio
in the metastases.
Many clinically used MRI contrast agents benefit from enhanced permeability and
retention (EPR) effect where blood vessels are leaky in the tumor region [56-60],
resulting in the accumulation of the contrast agent. However, this passive accumulation
of contrast agent is not very efficient in targeting micro-metastases where high concentration of the agent is needed to improve the signal to noise and overcome the partial volume effect. In addition, there might not be enhanced vasculature in a small metastasis [61]. Therefore, there is still a need to develop targeted contrast agents that can
target tumor cells at the molecular level, with high specificity and ability to accumulate in
high concentrations in order to delineate the extent of the tumors and detect micro- metastases [62-65].
1.3 Targeted Imaging Agents The process of developing newly targeted agents that can be used in diagnostics
and later for targeted therapy, starts by identifying a biomarker related to the tumor type,
Case Western Reserve University Page 16 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells such as a receptor on tumor cells or stromal components associated with tumor microenvironment [66]. Targeting of tumor microenvironment is now widely popular
[67-76], in fact multiple targeted agents to the tumor microenvironment are under clinical trials such as therapeutics that target vascular endothelial cell [77]. In this work, two newly developed peptides that target tumor microenvironment will be evaluated. A short description for each follows.
1.3.1 SBK2 peptide
The SBK2 peptide (GEGDDFNWEQVNTLTKPTSD) was developed in the laboratory of
Dr. Brady-Kalnay to target Glioblastoma multiforme tumors [78]. Specifically, this peptide targets the extracellular fragment of protein tyrosine phosphatase PTPμ that is found in the tumor microenvironment. Full length PTPμ plays a role in cell-cell adhesion and signaling in the central nervous system [79]. In tumor microenvironment such as in
GBM, PTPµ is down regulated through proteolytic cleavage, resulting in the release of the cytoplasmic domain of PTPµ from the plasma membrane [80, 81]. SBK2 homophilically binds to these extracellular fragments which are found in GBM tumors microenvironment but not in normal tissue. Tumor margins can be detected by this probe as PTPµ fragments presents at tumor edges [78].
In preliminary studies, SBK2 was able to label flank and intracranial xenograft tumors of human glioma cells when injected into the tail vein of nude mice [78, 82].
Hence, in this work we evaluated a fluorescent version of the peptide as a potential imaging agent for GBM cell dispersal.
Case Western Reserve University Page 17 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
1.3.2 CREKA peptide
CREKA (Cys-Arg-Glu-Lys-Ala) is a pentapeptide that was identified by Simberg et al.
[83]. The peptide formed distinct meshwork in the tumor stroma when intravenously injected into mouse models with xenografts of human breast cancer. The peptide also highlighted the blood vessels in the tumors. This led to the finding that this peptide targets clotted plasma proteins (fibrin-fibronectin complexes) expressed in the tumor microenvironment. The structure of CREKA is linear and contains only five amino acids which makes it an attractive peptide to use in nanoparticle targeting compared to other clot-binding peptides [83]. Also, the CREKA structure allows it to be coupled to other molecules such as fluorescent molecules or MR contrast enhancing molecules (e.g.
SPIOs and Gd) without affecting its binding affinity [83, 84]
In this work, we evaluated CREKA peptide multiplexed Gd-MR probe (CREKA-
Tris-Gd(DOTA)3) which was developed in the laboratory of Dr. Lu [84] for MR imaging of breast cancer micrometastases. In preliminary studies, the peptide showed promising results as MRI molecular targeted agent in imaging whole mouse metastases [84].
1.4 Preclinical Evaluation of Targeted Imaging Agents The process of evaluating new targeted peptides passes through three major phases including in vitro studies, preclinical studies, and clinical studies [49]. The focus of this work will be on the preclinical phase where we evaluated the targeting efficiency of these two imaging agents using mouse tumor models. During this evaluation, we optimized tumor animal models, imaged metastases and dispersing tumor cells at microscopic
Case Western Reserve University Page 18 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells resolution, and used image registration to perform multimodal evaluations. Below is background information about these steps.
1.4.1 Animal Models
Development of new cancer treatment and understanding of tumor progression start by creating an animal model that resembles the human disease. For example, as mentioned above, in Glioblastoma multiforme (GBM), cell dispersal occurs when cells leave the primary tumor mass and migrate along characteristic pathways then spread all over the brain [28, 30, 31]. In contrast, breast cancer cells can leave the primary tumor tissue and circulate in blood to form metastases in distant organs such as lung, liver and bones [85-
87]. Thus, it is very important during the evaluation of tumor targeted agents to produce mouse tumor models for the human disease, that are not only replicating the primary tumor, but also showing metastases and dispersal similar to the human disease [88, 89].
Moreover, using animal models where orthotopic injection of tumor cells is the source of forming spontaneous metastases is better than the direct injection of cells into the tail vein to form artificial metastases. Although it takes longer, tumor formed using orthotopic injections more closely resemble the human disease and cells can metastasize following a natural path. In contrast, tail vein injection of tumor cells might result in localized metastases in a single organ due to capillary entrapment such as in lungs [90].
1.4.2 Whole mouse imaging methods
Detection of whole mouse metastases is important to further understand tumor progression at the whole body level. It is also important in evaluating agent efficiency in
Case Western Reserve University Page 19 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells targeting all metastases, especially to give feasibility answers for translation to the clinic.
In fact, for drug discovery, it is necessary to achieve single cell resolution in order to provide comprehensive information about targeted drugs [66]. However, detection of whole body metastases is an extremely difficult task, since current imaging modalities lack the sensitivity, resolution, and/or field of view required to identify these tumor cells.
The same applies for GBM cell dispersal detection. Indeed, there are a few studies that ever attempted to study whole body micrometastases.
The need to evaluate new agents, that can be clinically used for improving image contrast and detectability of tumor cells, led to developments in the small animal in vivo imaging modalities such as small animal MRI, CT and PET, as the small animal is the starting point to understand the human disease and develop treatment. However, although these modalities are non-invasive, they are not well suited to study whole body micrometastases and cell dispersal, as they still need contrast agents to specifically target tumor cells and increase the signal to background. Undeniably, although these small animal modalities have better resolution than their clinical counterparts, they still suffer from sensitivity, resolution and/or field of view problems. Additionally, if a comparison between a human size and mouse size is done, one can find that these small animal modalities did not actually add significant improvements [91] to solve detectability issues regarding micro-metastases and cell dispersal, let alone in molecular assessment of new targeted imaging agents.
Case Western Reserve University Page 20 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
All these modalities, as mentioned above, depend on using a contrast agent to
delineate tumors. Even so, PET images lack the temporal and spatial resolution due to the half-life of used isotopes and partial volume effect. CT lacks soft tissue contrast, and the signal to noise ratio of the images might suffer if high resolution is used. Also, although
MRI provides a good contrast of soft tissue with a relatively excellent spatial resolution, whole mouse imaging of micro-metastases, even with a long imaging time, might still be
an issue without compromising the contrast and/or signal to noise at higher resolution.
In contrast, optical imaging technologies include bioluminescence (BLI IVIS,
PerkinElmer), which depends on luciferase/luciferin interaction as the source of light,
fluorescence molecular tomography (FMT, Visen), which gives 3D information by using
near infra-red excitation lasers, and multispectral fluorescence imaging (such as
Maestro™ EX, PerkinElmer), which uses a multispectral camera to deconvolve multiple fluorescent sources to generate 2D fluorescence images and remove background. These optical imaging modalities provide many advantages as they are very sensitive and provide good contrast for labeled tumor. However, the use of these in whole body imaging or even in brain imaging is limited by light scattering and absorption which limits the resolution and signal depth. In fact, previous attempts to image whole mouse
GFP expressing metastases were only efficient at shallow levels from the surface due to light absorption and scattering. Even if tumor cells express high amount of GFP, regardless of the strength of the fluorescence signal, scattering at this wavelength is high which can limit the depth of detection of whole mouse tumor micrometastases [92-95].
Case Western Reserve University Page 21 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Besides light scattering issues in BLI and multispectral in vivo fluorescence, both of these modalities only provide 2-dimensional (2D) information which makes it difficult to study the 3-dimensional (3D) information comprehensively. FMT tried to optimize this by using near infrared fluorescent labels which has better tissue penetration and less scattering and absorption. However, FMT resolution is limited due to the size of the mouse and is also limited by the variety of fluorophores that can be utilized since it uses near infrared excitation lasers.
Two photon and confocal have been optimized to image deeper sections with less photo bleaching by using near infrared lasers [96], but still limited in the field of view.
Although Multiphoton intravital microscopy provides a unique in vivo high resolution optical images that can be used to monitor tumor progression, invasion and response to therapy [97], it is limited to small optical windows and inaccessible to all tissue in the mouse [66]. Below is a table comparing advantages and disadvantages of the widely used small animal imaging modalities.
Table 1 Comparison of small animal imaging modalities. Imaging Advantages Disadvantages Typical Resolution Modality Field of view Bioluminescence sensitive Limited by Whole 1-10 mm resolution and depth. mouse No 3D information Fluorescence Multispectral Resolution is Whole 1-10 mm molecular fluorescence imaging. limited. No mouse tomography 3D information anatomical details (FMT, Visen).
Case Western Reserve University Page 22 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Multispectral Sensitive light scatter and Whole 0.2-2 mm fluorescence Multispectral absorption affect the mouse or imaging (such as fluorescence imaging spatial resolution, dissected Maestro™ EX, No 3D information organs PerkinElmer) Confocal and high resolution very limited in the Typically Subcellular multi-photon fluorescence imaging field of views and 1/30,000 resolution. Microscopy. depth the size of the brain.
Small animal PET Very sensitive. Short half-life. Whole 1-2 mm High penetration depth Very expensive if a mouse cyclotron is needed. Partial volume effect
Small animal High spatial resolution limited in soft tissue Whole 0.5-0.1 mm microCT images for bones, and contrast which limit mouse by adding a contrast the contrast of agent results in high tumors and limited resolution images for sensitivity vasculature. Small animal MRI High spatial resolution Imaging time is long Whole 0.5 mm. and anatomical details. for high resolution. mouse 0.1 mm can Sensitivity, contrast also be and SNR might be acquired but an issue in high imaging time resolution imaging become hours. Serial sectioning Subcellular resolution Time consuming, Multiple Subcellular and 3D with the` ability to use laborious, slices slices resolution. reconstruction of different stains to might be misaligned through histology images visualize different and tissue might the mouse [98-100] structures suffer shrinkage or organs which limit its use in 3D 1.4.3 Multimodality imaging
There is no doubt that combining the result of two or more imaging modalities will improve the diagnostics process, and provide better understanding of tumor microenvironment and better assessment of targeted imaging agents and therapeutics.
Multi-modality imaging provides complementary information whether through a single
Case Western Reserve University Page 23 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
integrated scanner, or by fusing the images through computer registration algorithms.
Here are some examples, PET/CT employs the spatial resolution and structural information of CT, and sensitivity of PET [101-103]; MR/PET provides soft tissue contrast and functionality of MRI with the PET sensitivity [104-107]; and fused information from CT/MRI/PET data offer improved functional and anatomical details
[108-110]. Consequently, the list of multimodality studies keep growing as computer registration algorithms develop and scanners that integrate two or more imaging modalities are built [111-113].
1.4.4 Literature Review of Whole Mouse Image Registration Algorithms
The importance of multi-modality imaging has long been recognized in several fields especially in radiation oncology, and this resulted in numerous computer algorithms to fuse different medical images of the subject for different purposes. Below is a short literature review of whole mouse registration approaches.
Registration of PET/CT was performed to combine functional and structural information. Example of such registration was performed by Chow, P et al., [114] where they employed hardware registration of PET and CT. They designed a mouse sized chamber that can be rigidly and reproducibly mounted on both CT and PET scanner by using adapter for each system. They estimated the spatial transformation matrix needed for registration using a 3D grid phantom with the standard deviation of image ratio as the cost function.
Case Western Reserve University Page 24 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Examples of CT/CT registration algorithms include the registration that was done
in [115, 116]. They performed microCT/microCT whole body image registration for
different purposes, including atlas based segmentation and monitoring the progression of
disease. Since in CT images bones contrast is high, their algorithm depends on sample
points on the bone structure where they register these points using RPM algorithm [117].
Then use this as initialization to the non-rigid algorithm. In the non-rigid registration,
they use an adaptive bases algorithm which depends on linear combination of radial basis
functions and NMI as similarity measure along with stiffness maps to limit bones
deformation. In contrast, Baiker, M., et al [118], used a combination of piecewise rigid registration to register CT/CT images, they identified joints and individual bones and registered each part using iterative closest point from coarse to fine. Other methods for
CT/CT include the work that was done in [119], where they registered CT/CT volumes by using ‘point matching based surface registration using a 3D shape context’ where a set of point on the source surface are mapped to the target surface, then one surface is warped to the other using TPS deformable transformation.
MRI/MRI registration was performed by Wang, H et al., [120] for the purpose of monitoring tumor response to therapy. They used a combination of rigid and B-spline registration to perform deformable registration. Their method incorporated robust matching algorithm (TPS-RPM) to estimate point correspondence and surface deformation.
Case Western Reserve University Page 25 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
In an effort combine optical imaging with other modalities, Xia, Z et al., [117]
tried to add anatomical information to FMT images by combining with CT images. They
employed the 2D flat images that can be acquired along with 3D FMT information, and
projected 3D CT to bridge the gap in registration between CT and 3D FMT. They
segmented boundaries in both 2D images and applied affine registration using modified
simplex optimization to align the points. The final result is a projected 3D CT fused with
a projected 3D FMT.
Examples of the registration of more than two modalities include the work
presented in [121]. The authors created what they called the ‘Digimouse’, which is a 3D
whole body mouse atlas for CT, cryo sectioning data and PET images. The purpose of
this atlas includes exploring mouse anatomy across modalities, phantom simulations and
modeling of light propagation. They created a rigid framework attached to Teflon tubes
that were filled with ink and 18F- to serve as fiducial marks for the registration. Then they
performed CT and PET scan followed by freezing and cryo sectioning. They performed piecewise rigid registration on the body, head and forelimbs using RVIEW software
[122], followed by a nonrigid registration using the LEREG (linear elastic registration) software [123]. For PET/cryo registration, they applied the resulting parameters from
CT/cryo on the PET images. Another example of multiple modality registration is the work of Humm, J. et al. [91]. The authors created a stereotactic method for the three- dimensional registration of animal multi-modality images. The method registers NMR,
PET, histology, and autoradiography image volumes. Their method depends on using
Case Western Reserve University Page 26 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Teflon™ rods as fiduciary markers. Advantages of the rods include that they can be
visualized in MRI if they are air or lard filled. They can also be visualized in PET if they
are filled with an isotope, as well as they are soft and easy to section once frozen to
acquire sections for histology and autoradiography. Image registration was done either
manually or by using image correlation to minimize the Euclidian distance between rod
centers.
1.5 Cryo imaging In order to fill the gap between the high resolution optical imaging techniques such as
confocal microscopy and the wide field of view of lower resolution techniques such as
BLI, our lab developed the Case whole mouse cryo-imaging system. The system is
currently being commercialized by (BioinVision Inc., Cleveland, OH) as CryoVizTM.
Cryo-imaging provides 3D microscopic color brightfield anatomy and multispectral fluorescence images over large volumes such as an entire mouse or organ [124-128].
Components of the system include a whole mouse cryo-microtome, robotic positioner,
multi-functional fluorescent microscope, robotic positioner, and hands-off acquisition
software capable of acquiring thousands images. The system is fully automated for re- peated physical sectioning and tiled microscope imaging of a tissue block face, providing anatomical brightfield and molecular fluorescence, as well as 3D microscopic imaging.
Briefly, a whole mouse or an organ is embedded in tissue embedding compound such as
O.C.T. (optimum cutting temperature, Tissue-Tek®), then the specimen is frozen using liquid nitrogen or dry ice, and the whole frozen block is placed inside the cryo-microtome
Case Western Reserve University Page 27 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
for sectioning and imaging. During the imaging session, the fully-automated system alternately sections and acquires tiled color brightfield and molecular fluorescence images of the block face of the frozen tissue. Image volumes can exceed 200 GB per mouse, hence specialized workstations and multi-resolution software are used to allow one to visualize a whole mouse on the screen and zoom to ever greater resolution until one can see single fluorescent cells.
Cryo-imaging is uniquely suited for characterization of tumor models, tumor cell dispersal, micro-metastases, and targeted imaging agent assessment as compared to other small animal imaging modalities for the following reasons: (1) High resolution, (2) large volume of view, (3) anatomical details, (4) fluorescence multispectral imaging, and (6) single cell sensitivity.
An example of comparing cryo-imaging and BLI in the application of imaging whole mouse metastases is shown below in (Figure 1). One can see that cryo provides 3D information, along with the ability to resolve very small tumor metastases as compared to
BLI where no information about depth, number or size of these metastases can be observed.
Case Western Reserve University Page 28 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 1. Comparison between BLI and cryo. One can see the ability of cryo imaging to resolve more cells at high resolution in 3D
1.6 Overview of the Thesis Cryo-imaging was utilized to study GBM tumor models, and breast cancer tumor models.
We assessed the targeting efficiency of SBK2 peptide to tumor cell dispersal in GBM, as
well as the targeting efficiency of CREKA to whole mouse breast cancer metastases.
The mouse model for GBM was evaluated for resemblance to the human disease.
We showed the results of creating GBM mouse models following orthotopic injection,
Case Western Reserve University Page 29 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
using four cell lines (LN-229, CNS-1, Gli36∆5 and U-87 MG). To enable comparison, tumor cells expressed Green fluorescent protein (GFP) for high resolution cryo imaging.
Analysis software was created to characterize cell dispersal among these cell lines and
determine which models show more cell dispersal and migration. Cell lines that showed
high dispersal capability were used to create GBM mouse model to characterize targeting
of SBK2. SBK2 was labeled by Cy5 (SBK2-Cy5) and injected intravenously into a
highly dispersive orthotopic GBM mouse model.
For the breast cancer metastatic mouse model, syngeneic 4T1 mouse model of
breast cancer [87, 129, 130] was used. This mouse mammary cell line has proven to be
an appropriate model because it can replicate human breast cancer metastasis to known
sites such as lungs, liver and bones within 3-6 weeks [87, 129, 130]. The 4T1 cell line
expressed GFP to enable detection of whole mouse metastases by high resolution cryo
imaging. Targeted CREKA peptide was conjugated to gadolinium complex or Cy5 labels
for MRI and fluorescence imaging, respectively. This permitted detection of imaging
agent in high resolution fluorescence and enabled comparison to in vivo MRI.
1.6.1 Combining cryo with Other Imaging Modalities
Cryo-imaging is an ex vivo technique, hence studying tumor progression and determining
optimum time for imaging agent accumulation and clearance requires the use of a lot of
animals and it will increase imaging time and processing. As such, we employed a
multimodal approach for our experiments where we used other in vivo imaging
modalities to monitor tumor growth and determine the time course of imagining agent’s
Case Western Reserve University Page 30 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
uptake to determine optimum agent accumulation time and tissue clearance. Then we
sacrificed the animals for cryo-imaging based on these times.
In the GBM studies, the time course of uptake and signal intensities relative to
controls for SBK2 probe was determined using Maestro™ FLEX In Vivo Imaging
System (Cambridge Research & Instrumentation (CRi), Woburn, MA). In the preliminary
experiments performed by Burden-Gully et al. [78], images of the fluorescence peptide were captured at different time points to determine the optimal time for SBK2 accumulation and background tissue clearance. These dynamic studies guided when the animals should be sacrificed for cryo-imaging high resolution studies after the peptide injection.
In breast cancer metastases studies, we used a tumor cell line that expresses GFP and luciferase (4T1-GFP-Luc2). This enables fluorescence imaging as well as bioluminescence imaging. BLI was used to examine the progression of metastases formation over time, which led to the decision of when to sacrifice the animals and perform cryo-imaging studies.
Histology added more information to the cryo-imaging results. Performing histology along with cryo-imaging is a valuable tool to confirm the identity of metastases, cell dispersal and assess imaging agent targeting at a molecular level. Throughout the chapters of this dissertation, I will show examples of histology slices acquired separately, or during cryo-imaging experiments using a tape transfer method.
Case Western Reserve University Page 31 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
1.6.2 Aims
i. Develop cryo-imaging methods and analysis software for multispectral fluorescent
imaging, and employ it in assessment of Glioblastoma tumor models and fluorescently
labeled targeted molecular imaging agents
I improved the cryo-imaging system to detect multiple fluorophores to enable the
evaluation of the GFP labeled GBM tumor models, and the fluorescently labeled SBK2
peptide. I also created image analysis/visualization software to characterize main tumor
mass and dispersal along pathways such as blood vessels and white matter tracts, and
developed image analysis/visualization software for characterization of targeting of
molecular imaging agent. ii. Use advanced 3D image registration techniques to register cryo-imaging volumes and
MRI volumes, to enable mapping of the fluorescently labeled tumors to MRI volumes
I applied the transformation system rigid affine deformable to perform 3D
multimodal image registration between cryo and MRI volumes. This enabled mapping of
breast cancer GFP metastases to the corresponding locations in the MRI volume. iii. Determine the ability of targeted MRI imaging agent to detect small metastatic tumors.
I created image analysis software for the visual and quantitative assessment of the
targeted MRI imaging agent (CREKA-Gd) based on the registered Cryo/MRI volumes.
This analysis was used to map 4T1-GFP breast cancer metastases to MRI volume and
assessed MRI signal intensities of the targeted agent.
Case Western Reserve University Page 32 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
1.6.3 Significance
The main goal of this project was to develop imaging protocols and provide software for
the image analysis of the cryo-imaging system. The significance of this analysis is that it
provides a unique cellular level characterization and quantification of micro-metastases and cell dispersal in tumor models along with the targeting efficiency of imaging agents to tumor cells. The data analysis presented in this thesis fulfills several requirements, including 3D spatial mapping of tumor cells and targeted imaging agent within a unique macroscopic 3D anatomy over volumes much too large for confocal or two-photon microscopy, and at a resolution and sensitivity much greater than that available using other in vivo imaging techniques such as MRI and PET. In addition, by implementing
multimodality 3D image registration, this enabled the validation of in vivo targeted MRI
imaging agent by comparing to the high resolution cryo-imaging, and provided
information about limitation and efficiency of the developed MRI agents. By assessing
the MRI targeting agents, this analysis provides further assessment to the feasibility of
developed imaging agents for clinical applications as noninvasive diagnostics tool.
Overall, this unique analysis is of great significance for developing diagnostics and therapeutics for effective treatment of tumor.
1.7 Organization of the Thesis The following chapters start by describing image analysis algorithms to characterize
tumor cell dispersal in multiple GBM mouse tumor models in order to find the best
model that resembles the human disease. In chapter 2, I describe the image analysis
Case Western Reserve University Page 33 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
algorithms which include segmentation of the main tumor mass, blood vessel detection,
cell dispersal detection, and measuring cell dispersal distance. These analyses were build
and validated using LN-229 cell line which express GFP as a label. In chapter 3, this
analysis was extended to include more cell lines; specifically we examined CNS-1,
Gli36∆5 and U-8 7 MG cell lines, along with the analysis of cell dispersal along white
matter tract. In chapter 4, the analysis of SBK2-Cy5 imaging agent was presented to
assess targeting to dispersal cell in mouse models of GBM.
Following the analyses of GBM, we created a whole mouse tumor model of
metastatic breast cancer. In chapter 5, I present image registration algorithms along with
visualization and image analysis software to assess targeting of CREKA-Gd to whole mouse metastases. Results of such analyses are shown visually and quantitatively.
1.8 My Specific Contributions This dissertation is composed of multiple collaborative projects between my laboratory
(Dr. David Wilson’s laboratory) and other laboratories at Case Western Reserve
University. This enabled me to do unique multidisciplinary research, and these
collaborations resulted in multiple peer-reviewed publications. Specifically, we collaborated with Dr. Susann Brady-Kalnay’s Laboratory (Molecular Biology and
Microbiology Department) and Dr. Zheng-Rong Lu (Biomedical Engineering
Department). In the Glioblastoma project, which is presented in chapters 2-4, the animal preparation and tumor models were created by my coauthor Dr. Susan Burden-Gulley
(Brady-Kalnay’s Laboratory). She also helped in the writing and revisions of these
Case Western Reserve University Page 34 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells publications along with the other coauthors. My contribution in these chapters includes cryo-imaging experimental design, analysis algorithms design and implementation, and generating all cryo-related figures, along with the writing and revisions. In chapter 5, my coauthor, Dr. Zhuxian Zhou (Dr. Lu’s Laboratory), created the animal model for the mouse breast cancer and conducted the MRI imaging. My contribution was the design of the cryo/MRI experiment, the design and implementation of image analysis/registration algorithms, along with writing.
Case Western Reserve University Page 35 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
2 Chapter 2: Analysis Algorithms for Glioblastoma Multiforme Cell Dispersal
2.1 Introduction In this study of brain tumors, we need to microscopically characterize cell
dispersal over the volume of an entire mouse brain, a need uniquely satisfied with cryo- imaging. Volumes of view from confocal and multi-photon would be ≈1/30,000 the size of the brain, and in vivo methods (MRI, PET, etc.) would not image dispersing cells.
Potentially a whole mouse brain could be imaged using histological serial sections, a digital slide scanner, and 3D software. As compared to cryo-imaging, advantages would be even greater resolution and the potential to use histological stains and immunohistology. Disadvantages include the large number of slides required, very time- consuming manual labor, inaccuracy of registration due to tissue shrinkage, tears, and warping, potential loss of GFP signal, and added autofluorescence from slide processing.
In addition, our long range goal is to image signal from targeted imaging agents and theranostics. Such exogenous signals could be lost in histological processing. In addition, histological sections do not allow one to unequivocally show dispersed cells disconnected from the main tumor mass in 3D. cryo-imaging also has several advantages over in vivo imaging modalities such as MRI and PET. Currently, there is no standard reporter gene technology for MRI. Any cell labeling technique used at implantation would dissipate over the time required to develop a tumor. Additionally, although MRI offers good tumor contrast in the brain with contrast agent, the resolution is insufficient to image single cells
Case Western Reserve University Page 36 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
dispersing away from the main tumor mass. Although there are now standard reporter
gene technologies for PET, it does not have the resolution to detect dispersing cells. We
anticipate that cryo-imaging will enable imaging of tumor, dispersed cells, vasculature,
and other structures in the brain. The combined features of high resolution, large volume
of view, and single cell sensitivity of cryo-imaging make it an ideal imaging modality for
the study of tumor cell dispersal.
This study is part of a larger effort at Case Western Reserve University to develop
representative animal models of GBM, targeted imaging agents, and targeted
therapeutics. In this report, we evaluate the migration and dispersal of the green
fluorescent protein (GFP)-expressing LN-229 human glioma cell line, following
orthotopic injection into mouse brains. The goal is to assess migration (the active process
of growth along specific structures in the brain), and dispersal (cells that are physically
disconnected from the main tumor) of the fluorescently labeled cells. To achieve this, we
find it necessary to create specialized software to analyze cell migration and dispersal in
relation to the vasculature. In the next section, we describe specialized software
developed for this project and experimental methods.
2.2 Materials and Methods 2.2.1 Experimental method and materials
Orthotopic xenograft intracranial tumors
Human LN-229 glioma cells were obtained from American Type Culture Collection,
Manassas, Virginia. LN-229 cells were infected with lentivirus to express green
Case Western Reserve University Page 37 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
fluorescent protein (GFP) 48 hours prior to harvesting [78]. NIH athymic nude female
mice (5–8 weeks and 20–25 g upon arrival, NCI-NIH) were maintained at the Athymic
Animal Core Facility at Case Western Reserve University according to institutional policies. All animal protocols were IACUC approved.
LN-229-GFP cells were harvested for intracranial implantation by trypsinization
and concentrated to 1x105 cells per µL of PBS. For intracranial implants, NIH athymic
nude female mice were anesthetized by intraperitoneal injection of 50mg/kg
ketamine/xylazine and fitted into a stereotaxic rodent frame (David Kopf Instruments,
Tujunga, California). A small incision was made just lateral to midline to expose the
bregma suture. A small (0.7mm) burr hole was drilled at AP= +0.5, ML= -2.0 from
bregma. Glioma cells were slowly deposited at a rate of 1µl /minute in the right striatum
at a depth of -2mm to -3mm from dura with a 10 µL syringe (26G needle; Hamilton
Company; Reno, Nevada); each of the mice was injected with 2x105 cells. The needle
was slowly withdrawn and the incision was closed with sutures.
After an appropriate period of tumor growth (20-38 days), the animals were
sacrificed and the brains were embedded in Tissue-Tek OCT compound (Sakura Finetek
U.S.A., Inc. Torrance, CA), rapidly frozen in a dry ice/ethanol slurry and transferred to
the stage of the cryo-imaging device for temperature equilibration.
In addition, we analyzed tissue sections from a brain sample (Tumor 6, 25 days
post-implantation). The brain was fixed with 4% paraformaldehyde, cryoprotected by
incubation in sucrose solutions (10-25%), and then frozen in OCT as above. Tissue
Case Western Reserve University Page 38 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
sections were cut at 5-7 micron thickness on a Leica cryostat. For immunolabeling of
blood vessels, sections were incubated with the endothelial cell-specific antibody CD-31
(B-D Biosciences, catalog number 550274).
Cryo-imaging of tissue samples
Using techniques in this report, we have cryo-imaged and analyzed over 14 mouse brain
tumors. Here we focus on results from five brains at 20, 28, 29, 36 and 38 days post- implantation. In some instances, we perfused the mouse with India ink as a contrast agent for blood vessels. Frozen brains were sectioned and imaged using the cryo-imaging system at two different in-plane resolutions and section thicknesses, 11x11 µm pixels and
15 µm section thickness, and 15.6x15.6 µm pixels and 40 µm section thickness. When
imaging at resolutions of 15.6 x 15.6 µm and 11 x 11 µm, the fields of view were 21.22 x
16.16 mm and 14.96 x 11.40 mm, respectively. At each of these resolutions, the brain fit
within a single field of view and did not require image tiling. Image acquisition took
approximately 3 hours when sectioning with 40 µm slice thickness, and approximately 5
hours when sectioning with 15 µm slice thickness. Brightfield and fluorescence images
were acquired for each of the brains. Color brightfield images were acquired using a
liquid crystal RGB filter and a monochrome camera (Retiga Exi, QImaging Inc.,
Canada). Fluorescence images were acquired using GFP fluorescence filters (Exciter:
HQ470/40x, Dichroic: Q495LP, Emitter: HQ500LP, Chroma, Rockingham, VT), a
fluorescent light source (XCite 120PC, EXFO, Canada), and the same low light digital
Case Western Reserve University Page 39 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
camera. Volumes were obtained in a hands-off, automated fashion on our cryo-imaging
system [126, 128].
Evaluation of image analysis algorithms
In some instances, we compared manual versus algorithm-based segmentation. Results
were compared using the DICE Similarity measure, which is widely accepted as a measure of agreement between segmented regions segmented by two methods [131].
Briefly, the DICE score is given by equation (1) below.
(1)
2 (segmentation A ∩ segmentation B) Dice Similarity Coefficient = | segmentation A | + | segmentation B |
A DICE score of 0 means there is no overlap between the two segmentations. A score of
1 indicates perfect overlap. Typically, a DICE score > 0.700 is deemed acceptable [131].
2.2.2 Image Processing Algorithms
We used MATLAB to perform 2D post-processing prior to 3D visualization using Amira.
Each raw image was approximately 4 MB. Depending upon slice thickness and brain
orientation in the field of view, we acquired from 200 to 450 slices, giving 1.5 to 3.5 GB
of raw image data, respectively.
Blood vessel visualization
To enable rapid evaluation of perivascular migration and dispersal of glioma cells along
blood vessels, we developed specialized methods for blood vessel visualization using
brightfield cryo-image data. Blood- or ink-filled vessels are darker than surrounding brain
Case Western Reserve University Page 40 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells tissue. To enhance contrast of blood vessel segments, we utilized filters matched to small vessel segments. Since the in-plane pixel spacing (as small as 11 µm) was smaller than the distance between slices (15 or 40 µm), we performed 2D filtering on each slice to enhance vessel contrast relative to background. We used the green channel from color brightfield images because this gave higher contrast than other channels. A functional diagram (Figure 2a) shows the algorithm. Main steps are described below.
Noise reduction: To reduce noise while preserving edges, we employed a modified wiener filter that adaptively removes the noise based on the statistics of a local region around each pixel in the image [132]. Filter equations are shown in equations (2a-c):
(2a) 1 µ = ∑ I(n1,n2 ) MN n1 ,n2∈γ
(2b) 2 1 2 2 σ = ∑ I (n1,n2 ) − µ MN n1 ,n2∈γ
σ 2 + v2 (2c) I (n ,n ) = µ + (I(n ,n ) − µ) out 1 2 σ 2 1 2
where:
(n1, n2): Pixel location in the image
µ : Estimated local mean around a pixel
Case Western Reserve University Page 41 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
σ 2 : Estimated local variance around a pixel
ν 2 : The noise variance estimated as the average of all local estimated variances for each
pixel in the kernel.
γ : The local neighborhood window around each pixel
MN : The size of the window used in filtering
This adaptive filter estimates the local mean and the variance for each pixel in a
neighborhood of size MN and uses this estimate to reduce the noise.
Background Removal: To increase the contrast of the blood vessels against the
surrounding tissue background, we employed unsharp mask processing. We used an
arithmetic mean filter to estimate the background in each of the 2-D images. The size of the kernel was much larger than the blood vessels, and therefore retained them in the output image.
Vessel Edge Detection: To detect vessel edges, we used a set of 4 oriented kernels, known as Difference of Offset Gaussian filters (DoOG filters) to measure the local gradient in the image [133, 134]. For example, in order to measure the local gradient along the x-direction in the image, the appropriate DoOG kernel is constructed by taking the difference of two copies of Gaussian kernels displaced along the x-axis.
Figure 2b depicts the kernel used to enhance edges in the x-direction. For kernel construction, there are three parameters to optimize for vessel enhancement: the standard deviation of the Gaussians, the truncation size and the offset between the centers of the
Case Western Reserve University Page 42 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Gaussian kernels. The first parameter determines the intensity of the kernel response at
the blood vessel edge and the pixels adjacent to the edge. The second parameter
determines the thickness of the edge; larger Gaussians will result in a response for
adjacent pixels further away from the edge pixels. The last parameter determines the
slope of the filter around the zero crossing point of the kernel. To find the best parameters
for vessels enhancement, we kept the offset parameter proportional to the standard
deviation parameter, specifically we used offset = 4σ [134]. Then, we experimented to
find the best σ and truncation size. Figure 1b shows the particular kernel used to enhance
the edges with a maximum response in the x-direction. The other kernels used to enhance the edges in the y-direction and in (45,135) directions are just a rotated version of the kernel shown. We processed the 2D background subtracted images using each of the four kernels. Then we took the absolute value of each of the resultant images to account for both dark/light and light/dark edges. Since the pixels at a particular processed edge will have maximum response coming from processing that edge by the kernel that best matches that direction, we kept the maximum response for each pixel among the four to form the final image.
Case Western Reserve University Page 43 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 2. Blood vessel visualization using bright field images. (a) Functional diagram showing the steps to visualize blood vessels. (b) The particular 2D Gaussian (top) with sigma =0.5 used to construct DoOG kernel (bottom) that enhances blood vessels in the x-direction
Visualization: Previous steps gave a gray scale volume with local maxima at edges of blood vessels. To visualize blood vessels in 3D, we created a colored volume that contains the result of the edge enhancement step as the red channel and zero for the green and blue channels. We then performed volume rendering on this colored volume such that the voxel is fully opaque if its red channel value is above a threshold and transparent otherwise. The threshold value is determined interactively during the visualization process. Thresholding removes artifactual noise voxels that are unlikely to be vessel
Case Western Reserve University Page 44 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
edges. We rendered brain image volumes from 1.5 GB to 3.6 GB in Amira on a
workstation configured with 32 GB of RAM and a video card with 2-4 GB RAM.
Image Segmentation and Visualization of Fluorescent Tumor Cells
We developed a semi-automated method for segmenting the main tumor mass and
dispersed cells in a fluorescent cryo-image volume. A functional diagram is shown in
Figure 3. Principal steps are outlined below.
Figure 3. Tumor visualization using fluorescence images. Functional diagram showing the steps to visualize the main tumor mass and dispersed tumor cells.
Segmentation of main tumor mass
To segment the main tumor mass, we used a 3D seeded region growing algorithm developed in our laboratory which works within the Amira platform. The region grows from the seeded region by including 26-connected voxels which have intensity and
Case Western Reserve University Page 45 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
gradient magnitude values within preset high/low constraints. To calculate the gradient
magnitude, we used a 3D gradient operator where the magnitude of the gradient, g, can
be estimated numerically using the central difference operator at location (x, y, z) as
given below in equation (3), where max identifies the maximum gradient magnitude in
the processed volume.
∇I(x, y, z) (3) g(x, y, z) = 255 ∇I(x, y, z) max where:
2 2 2 ∇I(x, y, z) = I x + I y + I z
I x = I(x −1, y, z) − I(x +1, y, z)
I y = I(x, y −1, z) − I(x, y +1, z)
I z = I(x, y, z −1) − I(x, y, z +1)
Following 3D region growth, each 2D image was reviewed and manually edited if
necessary. The main tumor mass volume is visualized in later steps using a green color.
Dispersed tumor cell detection
To detect dispersing single and clustered cells, we created a filter for enhancing fluorescent single tumor cells and clusters. We employed a high pass filter, which attenuates the low-frequency background signal without disturbing the high-frequency
content. Since Gaussian based filters are common with the advantages of smooth filtering
and the lack of ringing, we employed a zero-phase shift high-pass Gaussian filter with the following transfer function shown in equation (4) [132]: Case Western Reserve University Page 46 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
2 2 (4) H (u,v) =1− e−D (u,v)/ 2σ
2 2 2 M N D (u,ν ) = u − + ν − 2 2
Where:
D2 (u,ν ) : The distance from the origin of the Fourier transform of the image
M x N: size of the image
σ : Standard deviation of the Gaussian.
Each of the fluorescence images was converted to frequency domain using the
FFT algorithm, filtered using the high-pass Gaussian filter, and converted back to the spatial domain. We thresholded the result of the high-pass filtering then masked out the main tumor mass to create a binary volume of dispersing tumor cells and clusters.
Visualization: To easily visualize both dispersed cells and the main tumor mass, we rendered these objects in different pseudo colors. The main tumor mass was surface rendered in green. Dispersed tumor cells were volume rendered in yellow. Blood vessels were rendered in red as described previously. Results were fused to show tumor cell migration and dispersal along blood vessels.
Case Western Reserve University Page 47 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
3D migration distance of dispersed tumor cells
To analyze dispersal, we measured the 3D distance from the main tumor mass using a
morphological distance algorithm. Inputs are binary volumes containing the main tumor
mass and dispersed cells, respectively. We resampled the volumes to account for
disparities in section thickness and resolution to result in isotropic data. For example, we
resampled volumes acquired at 11x11x15 µm to 11x11x11 µm, a size comparable to the
size of a tumor cell. To measure distance, we applied a sequence of dilations on the
volume containing the main tumor mass using a sphere of increasing diameter as a
structuring element. After each dilation, the volume containing the dispersed tumor cells
was checked to see if any of the cell-containing voxels were overlapped by the dilation.
Steps of the algorithm are:
1. Read binary volumes containing the main tumor mass and dispersed cells,
respectively. Sample both volumes to give isotropic voxels and threshold to create
binary volumes.
2. In the volume containing dispersed tumor cells, save all the indices of the voxels with
non-zero value.
3. Perform 3D dilation on the volume containing the main tumor mass with a sphere
structuring element of radius i, starting with i=1
4. If saved indices from step 3 are included in the new dilated volume, record a distance
as i x 11 µm or i x 15.6 µm and eliminate it from further consideration.
Case Western Reserve University Page 48 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
5. Increase the radius of the sphere structuring element i by 1 and go back to step 4. The
loop continues until all indices are eliminated.
Histograms of distances and statistical measures (mean, median, etc.) are determined.
Processing speed is an issue because of our large data sets. Processing times of
Matlab codes can be reduced by converting to C++. Alternatively, many of our
algorithms use image filtering, which is easily accelerated using GPU processing as has
recently been done by us for another application [135].
2.3 Results Using our image processing algorithms, we characterized an orthotopic brain tumor
model consisting of human LN-229 glioma cells expressing GFP (LN-229-GFP). Images presented here are from three brains, but quantitative dispersal distances were from five
LN-229-GFP brains.
We performed experiments to optimize algorithms for vessel edge enhancement and tumor mass segmentation. Using representative bright field and fluorescence images, we processed images and reviewed images in 2D and 3D to optimize visual and quantitative analysis. A 5x5 window size for the Weiner filter and a 15x15 mean filter for unsharp mask processing removed background variations due to anatomy of brain, reduced noise, and preserved large and small vessels. For vessel enhancement with
DoOG, optimal results were obtained with 3x3 Gaussian kernels and σ=0.5 which gives an offset value of 2 pixels between the centers of the Gaussians. These optimized
Case Western Reserve University Page 49 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
parameters gave a strong response at vessel edges with acceptable noise enhancement.
Figure 1.b shows the particular kernel used for edge enhancement along the x-direction.
When segmenting the main tumor mass in fluorescence images with the 3D
region growth algorithm (Figure 3), we found that the gradient constraint was more
important than the intensity constraint because of overlapping intensity values; this is
consistent with bright edges of the main tumor mass. In most cases, a gradient magnitude
of 70 stopped region growing at the correct edge. In about 15-20% on an average of 400
total image slices, we manually edited at least a small portion of the segmentation. With
this semi-automated solution, we reduced analysis time from 10 hours for fully manual
segmentation to less than 2 hours on our large data sets.
We compared segmentations using our semi-automated approach to independent manual segmentations from experienced analysts using the DICE similarity measure
(equation 1). Analysis was done on a representative 40 slice slab imaged at high resolution, and the algorithm-based analysis was done months apart from manual segmentation. There was excellent agreement between the manual segmentations. Over the three potential pairings, the average DICE score was 0.88 ±0.08, a value much better than 0.7 typically deemed acceptable [131]. Comparing the algorithm-based results to each of the three manual segmentations gave an excellent average DICE score of
0.90±0.08. We infer that the algorithm-based method was as good as fully manual analysis.
Case Western Reserve University Page 50 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
The frequency domain, “high pass” Gaussian filter (Equation 4) gave visually optimal results with σ = 33 for an image size of 1036 x 1360 pixels. This value used over all brains completely removed background and retained only dispersed tumor cells and clusters, after theresholding and masking out the main tumor mass.
We compared the number of detected clusters using our automated method to independent manual detection performed by three observers. We accepted a cluster if 2 of
3 analysts marked it as a cluster. Manual analysis over a slab of 30 slices resulted in 142 clusters while our algorithm detected 140 clusters. To further compare results, we calculated 3D Euclidian distance between the center of mass of clusters detected by our automated method and clusters detected by manual analysis. We considered clusters identical if the distance between centers of clusters was < 2 voxels. There were 128 common clusters. Tallying results for the algorithm, we have 128 true positives (TPs), 12 false positives (FPs), and 14 false negatives (FNs). The algorithm based method compared very favorably to manual detection with sensitivity and precision equal to
90.1% and 91.4%, respectively. Accuracy is not computed because the number of TNs can be misleading when TNs consist of the large number of pixels not containing a cluster.
Even without contrast injection, cryo-imaging allows one to visualize blood vessels (Figure 4).
Case Western Reserve University Page 51 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 4 Steps for blood vessel visualization for tumor 1. (a) The raw color bright field block face image of a brain is shown. (b) Extracted green channel from the image shown in (a) used for further processing. (c) Application of the DoOG filters resulting in a high contrast, two-dimensional image of blood vessel profiles. (d) Three-dimensional visualization of the image stack from this brain specimen, showing the detected vasculature of the brain. (Tumor 1, 20 days post-implantation)
Case Western Reserve University Page 52 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
The 2D color brightfield images (Figure 4a) and corresponding green channel images (Figure 4b) show dark blood vessels. Detailed brain anatomy such as blood vessels and white matter tract is also evident. After applying the DoOG filter, vessel edges are shown (Figure 4c) and we have determined that the vessels with diameters ≥
30 µm are clearly visualized with 11x11x15 µm imaging resolution. Finally, volume rendering enables 3D visualization of all major blood vessels (Figure 4d).
Application of the image processing steps for visualization of the tumor and dispersing cells is demonstrated in Figure 4. In the 2D images from all brains, we observed a very bright main tumor and disconnected small tumor cell clusters (Figure 5).
Case Western Reserve University Page 53 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 5. Steps for visualization of tumor and dispersing cells for tumor 2. (a) The raw color bright field block face image of a brain is shown. Arrow indicates approximate position of the LN-229-GFP tumor within this section. (b) Corresponding fluorescence image showing the LN-229-GFP tumor (green). (c) Higher magnification view of the main tumor shown in (b) with cells dispersing away from the tumor edge (arrows). (d) Deconvolved fluorescence image shown in (c). (Tumor 2, 36 days post-implantation)
Tumor is shown in Figure 5b at the same size as the brightfield image in Figure
5a. In Figures 5c and 5d, we show magnified versions of the unprocessed and deconvolved fluorescence images, respectively. We utilized a previously described
deconvolution algorithm to confirm that tumor cell clusters were distinct from the main
tumor mass [125]. Following deconvolution, scattered light was reduced and we more
clearly discerned cell clusters not connected to the main tumor mass. Although scatter
deconvolution helped make this determination, it was unnecessary to deconvolve all
Case Western Reserve University Page 54 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells image data. The high pass Gaussian filter gave nearly identical results. It removed background signal and enabled reliable detection of dispersed cell clusters. Moreover, deconvolution of our extremely large volumes would have required very excessive processing times, while the filtering operation was quick. Often, the tumor was also visible in bright field images as a light structure (arrow in Figure 5a).
LN-229 clearly shows migrating dispersed cells (Figure 6 and 7). As described above, the GFP-labeled main tumor mass was segmented in 3D with a semi-automated region growing algorithm. Disconnected cell clusters were also detected and segmented.
In Figure 5, we show the two tissue types fused together.
Higher magnification views of Figure 6a are shown in Figure 6 b-c to illustrate that clusters are clearly disconnected from the main tumor mass. Accuracy was confirmed through visual inspection of processed images compared with the original raw data. Standard histological sections of a LN-229-GFP tumor also show cell dispersal from the tumor (Figure 7 a, b).
Case Western Reserve University Page 55 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 6. Results of the dispersed cell detection algorithm for tumor 1. (a) Three dimensional rendering of both the main tumor (green) and the dispersed cells (yellow). (b, c) Higher magnification views of dispersed cells shown in (a). (Tumor 1, 20 days post-implantation)
Dispersed cells are often in close proximity to blood vessels (Figure 7 c, d). From these 2D images one cannot determine whether cells are truly disconnected from the main tumor mass or are a continuous projection. The 3D images from cryo-imaging provide a more complete view of tumor architecture and cell dispersal.
Case Western Reserve University Page 56 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 7. LN-229 tumor cell dispersal in 2-D. (b, c, d) Histological section of a GFP-expressing tumor xenograft of LN-229 cells (green fluorescence) was immunolabeled with the endothelial cell specific antibody CD-31 and visualized with a secondary antibody conjugated to Texas Red. (a) Brightfield image from the same section stained with hematoxylin and eosin is shown. (b) Dispersing cells were observed at varying distances from the main tumor mass, as indicated by expression of GFP. (c, d) Higher magnification views of the boxed regions in (b) indicate a close association of dispersing cells with blood vessels. Scale bar represents 100 µm. (Tumor 6, 25 days post- implantation)
Growth of the main tumor mass often follows blood vessels (Figure 8). In 2D, we see that the fluorescent tumor (Figure 8b) follows along the vessel identified in brightfield in (Figure 8a). In (Figure 8c), the segmented ≈ 50 µm diameter vessel is superimposed on the fluorescent tumor. Migration is more appropriately visualized in 3D
(Figure 8d); one can clearly see the green tumor projection growing along the blood vessel. This substantial tumor projection accounts for about 5% of the tumor mass.
Case Western Reserve University Page 57 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 8. Projection of tumor growth along a blood vessel shown for tumor 5. (a) Dark blood vessels are shown in the color bright field image following perfusion with India ink. (b)The corresponding 2D fluorescence image clearly shows GFP labeled tumor. (c) The vessel is segmented and a 2D fusion shows the projection of cells growing along the vessel. (d) A 3D visualization clearly shows a substantive (≈5% of total tumor volume) projection of the green tumor growing along the blood vessel (red). (Tumor 5, 38 days post-implantation)
Case Western Reserve University Page 58 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
We further examined tumor cell migration leading to dispersal of unattached tumor clusters along blood vessels. In Figure 9, we fuse the main tumor mass (green), dispersed tumor cells (yellow), and blood vessels (red) at different magnifications and orientations. In this brain only 20 days after implantation, tumor volume was 1.78 mm3, the number of dispersed clusters was 761, representing ≈ 1.17 % of the main tumor volume. Arrows in Figure 9c point to examples of tumor cell clusters dispersed along blood vessels. Assuming a cell volume of ≈ 1 voxel (11x11x15 µm) and running a connected components analysis on the clusters, fluorescent clusters along blood vessels ranged from single cells to small clusters of up to 10 cells.
Case Western Reserve University Page 59 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 9. Tumor cells dispersing along blood vessels shown for tumor 1. In (a), cryo-images from a mouse brain containing a LN-229-GFP tumor were processed using the algorithms for visualization of main tumor mass (green), tumor cell dispersal (yellow) and blood vessel visualization (red). (b) Higher magnification image of tumor and surrounding vasculature viewed from a different angle. (c) Higher magnification image of small tumor cell clusters dispersing along blood vessels (arrows). (Tumor 1, 20 days post-implantation)
We characterized cell dispersal as a function of time post-implantation. With increasing time post-implantation (20-38 days), there was increased tumor volume, dispersed cell volume, number of clusters, and number of voxels per cluster. We found a marked increase of large dispersed clusters for tumors with time post-implantation. For example, comparing 20 and 38 days post-implantation, we found a small percent of the
Case Western Reserve University Page 60 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells number of clusters (2.2 %) having a size ≥ 5 voxels/cluster at 20 days post-implantation, as compared to 49.4% at 38 days post-implantation. We also found sporadic instances exclusive to tumors with longer times post-implantation of clusters > 1000 voxels/cluster.
Over the course of 20- 38 days post-implantation, tumor volume increased from 1.78 mm3 to 2.09 mm3 and the volume of dispersed cells increased from 0.021 to 0.119 mm3 giving dispersed cell volumes of 1.17 % and 5.71% of the main tumor mass for day 20 and 38, respectively. Also, number of dispersed clusters increased from 761 to 3605, the average number of voxels per cluster increased from 1.6 to 42.6 voxels/cluster and the median number increased from 1 to 6 voxels/cluster. If we assume the voxel size is approximately equivalent to a cell, we can estimate the number of cells per cluster from the above results.
We measured dispersal distances (Figure 10). Plotted are histograms of distances, normalized as percentages. We show the results of tumors at 20 and 38 days post- implantation. Other brains showed similar patterns of tumor cell migration and dispersal.
Although most tumor cells were within 70 µm, some had dispersed > 100 µm from the main tumor. The mean and median values of dispersal distance for all analyzed brains are
63 µm and 53 µm, respectively. Also, at later time points (36-38 days post implantation); we noticed dispersal distances increased to > 200 µm from the main tumor, a distance that represents ≈ 2.5 % of the brain diameter. Both mean and median of dispersal distance tended to increase with time post-implantation.
Case Western Reserve University Page 61 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 10. Dispersal distances from the main tumor mass. Histograms are shown with the number of cells normalized as a percentage. Specimens are analyzed at 20 (a) and 38 (b) days post-implantation. Clearly most dispersed cells are within 70 μm. Longer distances are recorded with increasing time post-implantation. Other tumors show a similar pattern of dispersal. Total number of dispersed clusters in these brains is 761 and 3605 for tumors 1 and 5, respectively
Finally, there was no evidence of blood vessel density increase in and around the
LN-229 tumor. We analyzed a ROI around the tumor and compared that to the same ROI placed in a comparable position on the contralateral side of the same brain. Both visual inspection and quantitative analysis indicated no significant difference in density for vessels ≥ 30 μm. Quantitative assessment was done by counting blood vessel voxels. The results showed a slight increase in blood vessels ≈ 5%. However, this small increase may be partly due to error in the measurement method used for this analysis. Therefore, the measured difference in blood vessels is not enough evidence of significant angiogenesis in LN-229 tumors over the course of 38 days post-implantation.
Case Western Reserve University Page 62 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
2.4 Discussion The ability to characterize migration and dispersal of glioma tumor cells in orthotopic
tumor models is highly significant. The efficacy of chemotherapeutic agents lies in their
ability to block proliferation, survival, migration, dispersal, invasion, and/or metastasis of
tumor cells. Since metastases are typically not seen in human glioma, tumor cell
migration and dispersal are the “spreading” mechanisms of interest. To be useful, an
orthotopic xenograft animal model of human glioma cell lines should display patterns of
tumor cell migration and dispersal characteristic of human GBM tumors. An appropriate
animal model will have a large impact on gene and biologic drug therapeutics discovery,
as well as improvements in imaging brain tumors. We have great interest in imaging cell
dispersal from brain tumors in these models to assess the ability to mark dispersing tumor
cells with molecular imaging probes. Ultimately, we want to determine if such imaging
probes can be used to enhance surgical resection and to improve survival of patients.
Cryo-imaging and software allow, for the first time, 3D analysis of migration and
dispersal of cells throughout an entire brain. The Case cryo-imaging system is well suited
for this task because it provides 3D microscopic resolution, color anatomy, and molecular
fluorescence images of large tissue specimens. In addition to fluorescence imaging of
tumor cells, brightfield images can be used to visualize blood vessels, white matter tracts,
etc. There is no good single cell sensitivity, whole organ imaging alternative. For example, the size of a typical confocal image stack (170x170x500 µm) is only 1/30,000 of the volume of a mouse brain. Another alternative is the use of 2D histology sections.
Manual labor is significant, sampling errors can occur, and it is impossible to determine if Case Western Reserve University Page 63 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
cells are disconnected from the main tumor mass in 2D. One can obtain serial histology
sections and create a 3D image volume in software. However, such methods are fraught
with problems. Sections can shrink, tear, and otherwise distort the image, confounding
accurate 3D alignment. Consequently, in vivo modalities (MRI, PET, etc.) can benefit
from cryo-imaging by validating the 3D results with the single cell resolution of cryo-
images rather than with histology or confocal imaging.
We now discuss the first of three important algorithms to aid characterization of
migration and dispersal of glioma cells, particularly along blood vessels. To visualize
blood vessels in bright field cryo-image volumes, we use filters optimized to reduce noise and enhance vessels in volume visualizations. The Wiener filter reduces noise while retaining edges and the DoOG filters effectively enhance vessel edges. DoOG suppresses noise and enhances vessel edges much better than simpler filters (e.g. Sobel) and computes faster on our very large data sets than more complex alternatives (e.g. level set based algorithms). Visual inspection of results and comparison to raw images proves that this approach enhances most of the larger vessels in the brain. Detection of smaller vessels is limited by image resolution since contrast of smaller vessels against brain tissue diminishes at lower resolution due to the partial volume effect. With an in-plane resolution of 11 µm, we are able to detect decidedly more vessels as compared to data sets with 15.6 µm in-plane resolution. Similarly, with thinner sections, blood vessel continuity is improved. With 3D rendering of high resolution brains, one can reliably visualize vessels with diameters as small as 30 µm. Because this is a filtering technique,
Case Western Reserve University Page 64 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
processing is relatively fast (≈ 90 minutes) even on extremely large (2 GB) data volumes.
Results are robust with little sensitivity to processing parameters. Processing allows
visualization of blood vessels from the color of accumulated blood. No contrast agent is
needed. This should be quite advantageous for assessing targeted imaging agents and
nanoparticle theranostics because no potentially disruptive perfusion agent must be
applied.
The second algorithm separately segments the tumor mass and dispersed cells.
The 3D region growing method includes both intensity and low edge strength inclusion
criteria. In practice, the low edge strength criterion is most important for getting accurate
results. Nevertheless, we had to manually adjust at least a small portion of the boundary
in about 20% of cases, or typically 100 out of 500 image slices. By implementing our
region growing algorithm within Amira, we have access to very useful editing tools,
including interactive editing in 3D. Typically, editing takes 2 hours as compared to 10
hours for a full manual segmentation. As determined from DICE scoring, the algorithm- based method compared very favorably to independent manual segmentations. As a result, we infer that an algorithm-based segmentation is as good as a fully manual analysis. Highpass Gaussian filtering allowed us to detect the dispersing cells with a simple interactive threshold. Rendering the two volumes in different colors makes the dispersed cells/clusters easy to distinguish from the main tumor.
Visual inspection of rendered 3D volumes clearly shows spaces between the yellow dispersing tumor cell clusters and the main tumor mass. Fusing results with the
Case Western Reserve University Page 65 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells blood vessel volume illustrates the migration and dispersal of the tumor cells along blood vessels. In some instances, we used 3D stereo display (NVIDIA 3D stereo) to quickly assess these spatial relationships. Our algorithms showed improved results for the higher resolution brain as compared to the lower resolution brains. At high resolution we were able to detect and visualize vessels as small as 30 µm and assess the tumor cell migration and dispersal distance more accurately. Although a subset of tumor cell clusters dispersed along blood vessels, other clusters dispersed without following blood vessels. By analyzing the 2D images we noticed that such clusters were either dispersing along the white matter tract, which is one of the characteristic pathways for dispersal, or these clusters were around very small blood vessels that we were not able to detect due to the very low contrast between such vessels and the brain tissue. There is a marked increase in the number of cells/cluster with time post-implantation, indicating that the dispersed clusters are growing. Many of the larger clusters are found along blood vessels that we detected (≥30 µm diameter).
The third algorithm measures tumor cell dispersal distance away from the main tumor mass in 3D. In our study, we determined that tumors at later times post-implantation showed a slight increase in dispersal distance and a larger increase in dispersed cell volume. Dispersal distances from 2D histology sections were comparable to those seen with cryo-imaging. However, with 2D sections, one has no way of knowing if cells are connected to the tumor mass in 3D, and one painstakingly makes 2D measurements made more appropriately in 3D. Dispersal distance histograms and statistical measures provide
Case Western Reserve University Page 66 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
an understanding of the aggressive nature of the LN-229 cell line. We believe that the algorithm will provide a fast, accurate method for comparing cell dispersal among different cell lines, time post-implantation, the mouse host environment, and treatments.
2.5 Conclusions Cryo-imaging and software allow, for the first time, 3D, whole brain, microscopic
characterization of a glioblastoma multiforme tumor model. Not only did we visualize cell migration and dispersal, we quantitatively assessed dispersal distance of tumor cell clusters from the main tumor mass. Further, dispersal along blood vessels was clearly identified. Cryo-imaging uniquely allows us to easily image the entire brain and detect even single dispersing fluorescent cells. It is impossible to make such characterizations
with traditional in vivo imaging (MRI, PET, etc.) or with histological sections. This study
can be extended to characterize the migration, invasion, dispersal, and metastasis in other
tumor types, and to characterize imaging agents and therapeutic effects.
Case Western Reserve University Page 67 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
3 Chapter 3: Analysis of Glioblastoma Multiforme Tumor Models
3.1 Introduction The current cellular theory of GBM dispersal into the surrounding tissue involves cellular detachment from the primary tumor, attachment to and degradation of the
extracellular matrix, and finally migration [4]. A number of molecules have been
implicated in regulating these processes in in vitro experiments, including receptor
tyrosine kinases and phosphatases, cell adhesion molecules, and proteases [4, 5].
To identify the molecular regulation of GBM cell infiltration in vivo, accurate models of GBM dispersal need to be developed. To date, the best animal models that recapitulate the main features of GBM are spontaneous brain tumors observed in dogs
[32]. Spontaneous, transgenic, xenograft, and syngeneic tumor models in rodents have also been characterized histologically [32, 136, 137]. From these studies, various cell
lines have been identified that mimic elements of human GBM pathology. For instance,
human tumor-derived U-87 MG cells injected intracranially into C57/B6 mice are highly
angiogenic [32], whereas rat tumor-derived CNS-1 cells injected into Lewis rats mimic
the pseudopallisading and hemorrhaging of GBM tumors in addition to being infiltrative
[32, 137].
In this article, we describe the analysis of mouse orthotopic xenograft tumors by
using a novel 3-dimensional (3D) imaging technique developed in the laboratory of Dr.
David Wilson at Case Western Reserve University. The Case cryo-Imaging System
acquires high-resolution anatomic color and cellular and molecular fluorescent images Case Western Reserve University Page 68 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
and then 3-dimensionally reconstructs a tissue or an entire organism, such as a mouse.
Cryo-imaging is unique among all in vivo and microscopic techniques in that it allows
micron-level resolution and information-rich contrast over large 3D fields of view. We
used the system to analyze the 3D extent of cell migration and dispersal from orthotopic
glioma tumors along blood vessels and white matter tracts within the brain.
We recently described the development of algorithms to reconstruct the 3D
architecture of blood vessels and tumor cell dispersal within the mouse brain [29]. We
used these algorithms to characterize how commonly used human (Gli36Δ5, U-87 MG,
LN-229) and rodent (CNS-1) glioma cell lines disperse in the mouse brain. In this article, we provide a complete 3D analysis of the dispersal and migration of these 4 tumor cell lines on blood vessels and white matter tracts. Our studies suggest that LN-229 and CNS-
1 are effective cell lines to use to study the dispersive nature of tumor cells along both blood vessels and white matter tracts whereas Gli36Δ5 cells are not dispersive in vivo and instead stay associated with the primary tumor. U-87 MG cells showed limited dispersal only along blood vessels. Our data suggest that either the human LN-229 cell
line or the rat CNS-1 cell line in mouse xenograft models of glioma are the most
appropriate for future studies investigating the molecular regulation of tumor cell
dispersal along particular anatomic structures within the brain. These xenograft systems
evaluated with the Case cryo-Imaging System will allow for future testing of therapeutics
aimed at blocking GBM tumor cell dispersal.
Case Western Reserve University Page 69 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
3.2 Materials and Methods 3.2.1 Orthotopic xenograft intracranial tumors
Human LN-229 and U-87 MG glioma cells were obtained from American Type Culture
Collection. CNS-1 rodent glioma cells were obtained from Mariano S. Viapiano [138].
Human Gli36Δ5 glioma cells constitutively overexpress the vIII mutant forms of the
EGFR gene [139] and were obtained from E.A. Chiocca. The CNS-1 and Gli36Δ5 cell
lines were authenticated by Research Animal Diagnostic Laboratory at the University of
Missouri (Columbia, MO) for interspecies and mycoplasma contamination by PCR
analysis. Five- to 8- week-old NIH athymic nude mice (20–25 g each) were housed in the
Athymic Animal Core Facility at Case Western Reserve University according to
institutional policies. All animal protocols were approved by the Institutional Animal
Care and Use Committee.
Gli36Δ5, U-87 MG, CNS-1, or LN-229-GFP cell lines were infected with green fluorescent protein (GFP) encoding lentivirus, harvested for intracranial implantation by trypsinization, and concentrated to 1x105 cells/µL PBS. Mice were anesthetized by
intraperitoneal administration of 50 mg/kg ketamine/xylazine and fitted into a stereotaxic
rodent frame (David Kopf Instruments). Cells were implanted at AP = +0.5 and ML = -
2.0 from bregma at a rate of 1 µL/min in the right striatum at a depth of - 3 mm from dura. For examination of dispersal along white matter tracts, cells were implanted at a depth of - 2 mm from dura for tumor formation in close vicinity to the corpus callosum.
A total of 50,000 to 200,000 cells were implanted per mouse.
Case Western Reserve University Page 70 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Mice were sacrificed 7 to 38 days after implantation on the basis of tumor burden.
Dissected brains were embedded in Tissue-Tek optimum cutting temperature (OCT) compound (Sakura Finetek U.S.A., Inc.), frozen in a dry ice/ethanol slurry and transferred to the stage of the cryo-imaging device.
3.2.2 Cryo-imaging of tissue samples
Three brains implanted per cell type with either LN-229, CNS-1, U-87 MG, or Gli36Δ5-
GFP were analyzed. Frozen brains were alternately sectioned and imaged using the cryo- imaging system at a section thickness of 15 to 40 µm and a resolution of 11 x 11 x 15 or
15.6 x 15.6 x 40 µm. The cryo-imaging system consists of a mouse-sized stage on a motorized cryostat with special features for imaging, a modified bright-field/fluorescence microscope, and a robotic xyz imaging system positioner, all of which are fully automated. The system images fluorescent agents or cells at a very high resolution and sensitivity. Bright-field and fluorescence images were acquired for each of the brains with a low-light digital camera (Retiga Exi), GFP fluorescence filters (exciter
HQ470/40x, dichroic Q495LP, emitter HQ500LP; Chroma), and an epi-illumination fluorescent light source (XCite 120PC; EXFO).
3.2.3 Image processing algorithms for visualization of tumor cells and vasculature
We recently developed methods for segmentation and visualization of the vasculature, main tumor mass, and dispersing cells [29]. To segment the main tumor mass, we used a recently developed 3D seeded region growing algorithm [29]. Results were reviewed in individual slice images and manually edited if necessary. To segment dispersed tumor
Case Western Reserve University Page 71 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
cells and clusters, we applied a high-pass filter and thresholded the result, excluding the
binary volume consisting of the main tumor. To create a 3D volume of the brain
vasculature, bright-field images from each brain specimen were processed using
algorithms developed for vessel edge detection and volume rendering [29]. These
algorithms resulted in a 3D reconstruction of the brain vasculature. The lower limit for
the blood vessel detection algorithm was approximately 30 µm diameter vessels. Three-
dimensional pseudo-colored volumes were created that included the main tumor (green),
dispersing cells (yellow), and vasculature (red). The location of dispersed cells was
visually inspected by rotation of the composite 3D volume within Amira (Visage Imaging
Inc.) to confirm a distinct nonfluorescent region separating dispersing cells from the main
tumor and to identify cells in close proximity to blood vessels.
3.2.4 Dispersal on white matter
The corpus callosum white matter was manually segmented using bright-field cryo- images and Amira Software. The segmented region was reconstructed as a 3D volume that was merged with the 3D tumor and dispersed cell volumes. The white matter was pseudo-colored gray, and the dispersed cells in contact with the white matter were
pseudo-colored magenta, whereas all other dispersing cells were pseudo- colored yellow.
3.3 Results Orthotopic xenograft models of glioma cells in rodents are useful for assessing tumor
growth characteristics and response to therapeutics. Gli36Δ5 is a human glioma cell line
that grows rapidly to form a large tumor within 2 weeks [139]. However, the tumors
Case Western Reserve University Page 72 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells become encapsulated and cell dispersal from Gli36Δ5 tumors was never observed, as shown in 2-dimensional (2D) histologic sections (Fig. 11A and B). Xenografts of U-87
MG human glioma cells also grow rapidly as closely associated cells with defined margins (Fig. 11C and D). The LN-229 human glioma cell line exhibits slower growth characteristics, and examination of orthotopic xenografts in histologic sections revealed that LN-229 cells disperse from the main tumor at 4 to 6 weeks post implantation (Fig.
11E and F). The CNS-1 glioma cell line was developed in the inbred Lewis rat, and it exhibits very rapid growth [138].
Case Western Reserve University Page 73 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 11. Glioblastoma xenograft tumor growth characteristics in 2 dimensions. Histologic sections of GFP-expressing tumor xenografts of Gli36Δ5 (A and B), U-87 MG (C and D), LN-229 (E and F), or CNS-1 cells (G and H) were stained with hematoxylin and eosin to demarcate the tumor borders within the Case Western Reserve University Page 74 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
brain parenchyma. Gli36Δ5 (A and B) and U-87 MG (C and D) cells remain tightly associated within encapsulated tumors. Cells disperse along the length of LN-229 tumors (E and F). CNS-1 tumors consist of more loosely associated cells that migrate readily through the brain parenchyma (G and H). Scale bar, 200 µm.
Within 1 week following implantation, these tumors were highly vascularized, the
cells of the tumor were loosely associated, and individual cells had dispersed from the
entire perimeter of the main tumor (Fig. 11G and H). By 10 days of growth in vivo, the
dispersed cells were observed throughout the entire frontal lobe of the hemisphere
surrounding the main tumor (data not shown).
Perivascular growth is the most common form of glioma dispersal [140]. To examine whether these cell lines dispersed along blood vessels, histologic sections of
GFP-expressing tumor xenografts of U-87 MG (Fig. 2A–D), LN-229 (Fig. 12E– H), and
CNS-1 (Fig. 12I–L) cells were immunolabeled with the endothelial cell–specific antibody
CD-31. U-87 MG tumor cells dispersed along blood vessels in close proximity to the main tumor (Fig. 12A–D); however, the extent of dispersal was minimal. In contrast, LN-
229 (Fig. 12E–H) and CNS-1 (Fig. 12I–K) cell dispersal occurred primarily along blood vessels.
Case Western Reserve University Page 75 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 12 Tumor cell dispersal on blood vessels in 2 dimensions. Histologic sections of GFP-expressing tumor xenografts of U-87 MG (A–D), LN-229 (E–H), and CNS-1 (I–L) cells were immunolabeled with the endotheli al cell–specific antibody CD-31 (C, G, and K). Bright-field images from the same tissue sections stained with hematoxylin and eosin are shown (A, E, and I). U-87 MG cells (A–D) dispersed along blood vessels that were in close proximity to the main tumor. In contrast, LN-229 (E–H) and CNS-1 (I–L) cells dispersed as streams of cells along vessels, often great distances from the main tumor mass. Scale bar, 100 μm.
The 2D histologic sections provided a high-resolution snapshot of a single plane through the tumor of interest. However, 3D reconstruction of histology images is time consuming and prone to tissue shrinkage and errors in image alignment. To overcome these obstacles the Case cryo-Imaging System was used to analyze tumor cell dispersal from orthotopic xenografts of the 4 glioma cell lines in 3D at high resolution. The technique utilized bright-field images of the block face for overall brain anatomy and fluorescent images to detect the GFP-expressing glioma cells [29]. Three-dimensional
Case Western Reserve University Page 76 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells volumes were created for the vasculature (pseudo-colored red), white matter (pseudo- colored gray), and the main tumor mass (pseudo-colored green). Glioma cells that were no longer physically connected in any dimension to the main tumor were pseudo-colored yellow to indicate tumor cell dispersal. Multiple xenograft specimens of each cell line were analyzed using this technique, and representative examples are shown. The comparison with standard histology images highlights the tumor biology that can be visualized using this novel technique.
Gli36Δ5 cells are non-dispersive
The growth of Gli36Δ5 xenografts was rapid, resulting in a large encapsulated mass by 2 weeks after implantation (arrow in Fig. 13A). Three-dimensional cryo-image analysis of
Gli36Δ5 tumors showed that the average tumor volume was 40.57 mm3 (n = 3). Tumor growth resulted in the compression of surrounding brain structures (Fig. 13A). In addition, 3D reconstruction of the brain vasculature indicated that although the tumors were highly vascularized, cell dispersal was never observed (Fig. 13B and C), which supported previous observations in 2D histologic sections (Fig. 11A and B).
Case Western Reserve University Page 77 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 13. Three-dimensional reconstruction of Gli36Δ5 and U-87 MG tumors indicates limited cell dispersal. Mouse brains containing xenografts of GFP-expressing Gli36Δ5 (A–C) or U-87 MG (D–F) cells were cryo-imaged and reconstructed in 3 dimensions to show the main tumor mass (pseudo-colored green), dispersed tumor cells (pseudo-colored yellow), and vasculature (pseudo-colored red). Two-dimensional block face images are shown (A and D), with the tumor indicated by an arrow. Three- dimensional reconstruction of the same brain specimens shows that no cell dispersal was observed from the Gli36Δ5 tumor despite its large size (B and C). In contrast, cells from the U-87 MG tumor dispersed on a nearby blood vessel (E and F). Scale bars, 500 μm.
U-87 MG cells are marginally dispersive along blood vessels
In U-87 MG xenografts, tumor cells dispersed along blood vessels in close proximity to the main tumor. This dispersal pattern was observed in both 2D histologic sections (Fig.
12A– D), and 3D tumor reconstructions from cryo-image analysis (Fig. 13E and F).
Analysis of the 3D tumors indicated that the average dispersed cell volume was 0.0028 mm3 and average tumor volume was 1.86 mm3 (n = 3). Therefore, the dispersed cells represented only 0.15 % of the total tumor cell population. However, the cells dispersed up to 300 µm from the main tumor (Fig. 18). Although U-87 MG tumors exhibit Case Western Reserve University Page 78 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
somewhat limited dispersal, they may be good models to analyze effects of
chemotherapeutics on reduction of tumor load or migration on blood vessels.
LN-229 is a dispersive human glioma cell line
The LN-229 xenografts grew at a slower rate than Gli36Δ5 or U-87 MG to form an
average tumor load of 2.76 mm3 (n = 3) after 4 to 6 weeks. Two-dimensional histologic
analysis revealed cell dispersal along the length of the tumor (Fig. 11F), often in
association with blood vessels (Fig. 12E– H). Analysis of 3D cryo-image volumes
illustrated that LN-229 cells frequently disperse as connected strands for several hundred
microns along blood vessels (Fig. 14C and D). It is unclear whether these vessels were
pre-existing or due to tumor-mediated angiogenesis. Small populations of cells released
all along the main tumor to migrate through the brain parenchyma (Fig. 14B and F) and
[29]. These cells may also be dispersing along blood vessels that are below the limits of
detection with our analysis algorithms (< 30 µm diameter). Overall, the average dispersed
cell volume was 0.035 mm3, which represents 1.26% of the total tumor cell population.
LN-229 cells were observed to reach the lateral ventricle from the main tumor in some cases, resulting in spread to distant regions of the brain via passive movement in the cerebrospinal fluid (data not shown). In those instances, the cells became imbedded in and spread along the meninges covering the brain.
Case Western Reserve University Page 79 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 14 Three-dimensional reconstruction of LN-229 tumors shows cell dispersal on blood vessels in the tumor microenvironment. Mouse brains containing xenografts of GFP-expressing LN-229 glioma cells were cryo-imaged and reconstructed in 3 dimensions to show tumor (pseudo- colored green), dispersed tumor cells (pseudo-colored yellow), and vasculature features (pseudo-colored red). Examples of 4 week (A–D) and 6 week (E–H) LN-229 tumors are shown. Two-dimensional block face images are shown (A and E), with the tumor indicated by an arrow. Three-dimensional reconstruction of the same specimens illustrates cell dispersal from several regions of the tumors (B and F). Cell dispersal primarily occurred along major blood vessels in close proximity to the tumors (C and G). D, a higher-power view of the tumor shown in C. H, a higher-power view of the top of the tumor shown in G. Note the clustering of dispersed cells on multiple blood vessels. Scale bars, 500 μm (B and F) and 100 μm (D and H).
The rat glioma cell line CNS-1 is highly dispersive in vivo
The CNS-1 cell line spread aggressively to infiltrate distant regions of the brain from the tumor within 7 to 10 days post-implantation. The main tumor consisted of loosely associated, pseudopallisading cells (Fig. 11G and H), with features of hemorrhage and necrosis (arrow in Fig. 15A and E). Because of the rapid dispersal of CNS-1 cells, the volume of the main tumor at 7 days was only 1.49 mm3 (n = 3). However, the dispersed
Case Western Reserve University Page 80 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells cell volume was 0.18 mm3, which represents 10.6% of the total tumor cell population detected by 3D cryo-image analysis. CNS-1 cells migrated extensively along blood vessels, as shown in the 3D reconstructions from 7-day tumors (Fig. 15D and H). In addition, CNS-1 cells readily dispersed through the brain parenchyma (Fig. 11H).
Dispersed cells are clustered on blood vessels in close proximity to the tumor edge as well as streaming along vessels at a distance.
Figure 15 Three-dimensional reconstruction of CNS-1 tumors showing their high dispersal along blood vessels. Mouse brains containing 7-day xenografts of GFP-expressing CNS-1 glioma cells were cryo-imaged and reconstructed in 3 dimensions to show the main tumor mass (pseudo-colored green), dispersed tumor cells (pseudo-colored yellow), and vasculature (pseudo-colored red). Two examples of CNS-1 tumors are shown (A–H). Two-dimensional block face images are shown (A and E), with the tumor indicated by an arrow. Vascular density and hemorrhaging within the tumor are visible A and E). Three-dimensional reconstruction of the same specimens illustrates dramatic cell dispersal from all surfaces of the tumors (B and F). CNS-1 cells were typically clustered on blood vessels near the tumor surface and dispersed along vessels (C, G, and H) to great distances away from the main tumor mass (D). Scale bars, 500 μm (B and F) and 100 μm (D and H).
Case Western Reserve University Page 81 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
We analyzed the dispersal distance of CNS-1, LN-229, and U-87 MG cells using a 3D morphologic distance algorithm that detects the presence of voxels containing fluorescent cells in a series of dilations from the tumor edge outward [29]. The voxel size was 11 x 11 x 15 µm, approximately the size of a single cell. The first 2 dilations closest to the tumor were discarded to reduce nonspecific error. The results from this analysis indicate that LN-229 and CNS-1 xenografts generate thousands of dispersive cells (Fig.
16), which represent a 13- to 28-fold increase in total number of dispersed cells when
compared with the U-87 MG xenografts. The maximum distance traveled by LN-229
cells was 562 µm away from the tumor edge, whereas CNS-1 cells dispersed more than 3
mm from the tumor (Fig. 16).
Case Western Reserve University Page 82 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 16. Cell dispersal distance measurements. The average number of dispersed cells is shown for various distance ranges. LN- 229 and CNS-1 cells migrate 0.5-4 mm away from the main tumor mass.
LN-229 and CNS-1 cell dispersal along white matter tracts
One of the pathways used by human glioma cells for dispersal is intrafascicular growth on white matter. We analyzed the ability of the 4 cell lines to use white matter for dispersal. Neither Gli36Δ5 nor U-87 MG cells dispersed on the corpus callosum, a large white matter structure in close proximity to the main tumors. U-87 MG cells within the main tumor were observed to realign along the longitudinal axis of the corpus callosum
(Fig. 17A and B). In all 3 xenografts examined by 3D cryo-image analysis, the U-87 MG tumor bulged out onto the corpus callosum (arrow in Fig. 17G). Thus, these cells seem to
Case Western Reserve University Page 83 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells respond to cues on the surface of the myelinated fibers. However, individual U-87 MG tumor cells did not disperse on the white matter (Fig. 17G).
Figure 17. Three-dimensional reconstruction of U-87 MG, LN-229, and CNS-1 cell dispersal on white matter tracts in the brain. Histologic sections from mouse brains containing xenografts of GFP-expressing U-87 MG (B), LN-229 (D), or CNS-1 (F) glioma cells were examined for cell dispersal. Bright-field images from the same tissue sections stained with hematoxylin and eosin are shown in A, C, and E, respectively. U-87 MG cells within the main tumor realigned along the longitudinal axis of the corpus callosum (A and B), a major white matter tract in the brain. However, individual cells did not disperse onto the white matter. LN-229 cells separated from the main tumor to disperse as single cells on the corpus callosum (D). The cells of a CNS-1 tumor were more loosely associated but were observed to reorient to align with the corpus callosum for dispersal (F). Mouse brains containing xenografts of GFP-expressing U-87 MG (G), LN-229 (H), or CNS-1 (I) glioma cells were cryo-imaged and reconstructed in 3 dimensions to show the main tumor mass (pseudo-colored green), dispersed tumor cells (pseudo-colored yellow), and white matter of the corpus callosum (pseudo-colored gray). Dispersed cells in contact with white matter were pseudo-colored magenta (H and I). Scale bars, 500 μm.
LN-229 and CNS-1 cell lines dispersed along the corpus callosum as single cells or small cell clusters, as shown in histologic sections (Fig. 17C–F). Individual cells were Case Western Reserve University Page 84 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
observed to exit the main tumor to disperse directly onto the white matter, evidenced in
the 3D tumor volumes by the magenta cells at the tumor–white matter transition zone
(Fig. 17H and I). Together, these results provide further support that both LN-229 and
CNS-1 cell lines are dispersive in the brain, showing dispersal characteristics comparable
with humans, and thus may be good model systems for testing therapeutic efficacy.
Angiogenesis is correlated with tumor growth and dispersal rate
Angiogenesis is a hallmark feature of tumor growth. To determine whether differences in
angiogenesis existed between the 4 glioma cell lines, we used the 3D cryo-image blood
vessel reconstruction volumes to quantitate blood vessel density within each tumor. The
most significant increase in blood vessel density occurred in tumors with the fastest
growth rate such as Gli36Δ5 and U-87 MG as well as in the most extensively dispersing
cells, CNS-1 (Figure 18).
Figure 18. Analysis of tumor blood vessel density.
Case Western Reserve University Page 85 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
3.4 Discussion The most useful animal models of human disease exhibit characteristics that are highly
representative of those observed in humans. In this article, we characterized the in vivo
growth and dispersal of 4 different glioma cell lines by using orthotopic xenografts in
athymic nude mice. The cryo-imaging and 3D reconstruction methods described here provide extraordinary insight into tumor growth characteristics within the complex architecture of the brain at single cell resolution that is a significant advancement over standard methods such as histologic sections. Our results also indicate that the best cell lines for studying migration and dispersal in the context of GBM are the LN-229 and
CNS-1 cell lines.
Of note, the 4 cell lines we evaluated in this article showed a gradation of
migration within the brain. Gli36Δ5 cells do not migrate, U-87 MG cells disperse
marginally only along blood vessels, LN-229 cells migrate significantly along blood vessels and white matter tracts, and CNS-1 cells are the most migratory along both these secondary structures. Given that GBM tumor cells show a great range of distances and substrates for dispersal [140, 141], it would be ideal to have a range of cell lines to evaluate migration in vivo. Our findings that U-87 MG cells are not highly dispersive in vivo are supported by other recent data [142] and suggest that despite its popularity as a
GBM tumor model in vitro, U-87 MG cells are not the best in vivo model.
Dispersal of LN-229 and CNS-1 cells along blood vessels suggests that the cells respond to migration-promoting cues present on the surface of blood vessels. A number of migration-promoting molecules have been identified in dispersive GBM cells in vivo. Case Western Reserve University Page 86 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Stromal cell–derived factor 1α (SDF-1α) and its receptor CXCR4 [143], Na+/H+
exchanger regulatory factor 1 (NHERF-1; ref. [144]), ephrin B2, ephrin B3, and the
EphB2 receptor [145-147] mRNA and/or protein are all elevated in dispersive GBM cells
versus core GBM tumor cells. Additional molecules, such as the cleaved extracellular
fragment of the receptor tyrosine phosphatase PTPµ [78], the type I receptor of the TNF
superfamily TNFRSF19/TROY [148], Neuropilin 1 and its ligand Sema3A [149], and the
vitronectin receptor Necl5 [150], are also elevated in GBM tumor tissue, although not
necessarily in dispersive cells. Only SDF-1 expression has specifically been localized to
GBM secondary structures, whereas its receptor CXCR4 is expressed on migrating glioblastoma cells themselves [143].
All of the aforementioned molecules regulate GBM cell migration in vitro, as shown in Matrigel invasion or 2D migration assays [80, 81, 143-149] or 3D matrix spheroid assays [144, 150]. Ex vivo models evaluating human GBM cell migration on rodent brain slices have also been employed to investigate the function of specific molecules in GBM dispersal [80, 145-148]. In addition to the different in vitro and ex vivo assays carried out, a wide variety of GBM cell lines were evaluated in these studies, including, but not limited to, A172 [149, 150], U-87 MG [81, 143, 145-149], T98G [144,
145, 148], and SNB19 cells [145, 148]. The reason for selecting these different cell lines differs, but most often it is based on the expression level of a gene or protein of interest.
Both cellular and molecular differences exist between the tumor cells found within the main GBM tumor or "core" and those cells that have migrated away from the
Case Western Reserve University Page 87 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells core or "edge" [151]. Given this fact, it is important for us to develop means of studying dispersal as a separate event from the primary tumor growth and survival. The methodology that we present here and in [29] allows for the evaluation of migrating and dispersing cells in vivo at single-cell resolution. Tumor models such as the ones described here will be increasingly important to achieve the goal of understanding cell dispersal at a molecular level and evaluation of therapeutics targeting GBM dispersal.
Case Western Reserve University Page 88 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
4 Chapter 4: Assessment of SBK2 Targeting to GBM Dispersing Tumor Cells
4.1 Introduction Glioblastoma multiforme (GBM) is the most common malignant adult astrocytoma with
very low short and long term survival rates [152]. It is characterized as having a necrotic main tumor mass accompanied by high dispersal of individual or clusters of tumor cells along blood vessels and white matter tracts in the brain. Currently, the best treatment for
GBM is surgical resection of the main tumor mass, followed by radiotherapy and temozolomide chemotherapy [153]. The extent of surgical resection of the main tumor mass has a significant effect on overall survival [154-156], especially when combined with radiotherapy and chemotherapy [155]. Magnetic resonance imaging (MRI) is currently used to delineate the tumor border before tumor resection. Unfortunately, this is an imperfect tool, as the contrast enhancement agent, gadolinium, has highly variable labeling results and cannot highlight individual GBM cells that have migrated away from the main tumor mass [157], nor can a static image taken before surgery be used as a real- time intraoperative guide. Imaging of the dispersed tumor cell population in addition to the main tumor mass before surgery would provide a significant advantage to the surgeon.
The development of tools to molecularly label tumor cells to aid in surgical resection is a major focus of current cancer research [158]. To this end, we have studied a cleaved and shed extracellular fragment of the receptor protein tyrosine phosphatase mu
Case Western Reserve University Page 89 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
(PTPµ) that is found in the tumor microenvironment. Full-length PTPµ has both cell
adhesion and signaling capabilities. The extracellular segment of full-length PTPµ has a
MAM (meprin, A5/neuropilin and mu) domain, an immunoglobulin (Ig) domain and four
fibronectin type III (FNIII) repeats. The MAM and Ig domains as well as the first two
FNIII repeats are required for efficient cell–cell adhesion [159-165]. The intracellular
segment of full-length PTPµ has two tyrosine phosphatase domains [166] of which only
the membrane proximal tyrosine phosphatase domain is catalytically active [167].
PTPµ extracellular fragment is detected in human GBM tissue and in glioma cells
[78, 81]. It consists of the MAM, Ig and first two FNIII repeats of full-length PTPµ [78],
and thus should be able to bind homophilically to other PTPµ molecules. PTPµ is cleaved
either by a matrix metalloprotease (MMP) or by A Disintegrin And Metalloprotease
(ADAM) to yield the PTPµ extracellular segment and a membrane-tethered fragment
consisting of the transmembrane and intracellular domains [81]. The membrane-tethered fragment is subsequently cleaved by the gamma secretase complex to give rise to a membrane-free intracellular fragment capable of translocating to the cell nucleus where it remains catalytically active [81].
Fluorescently labeled targeted PTPµ peptide probes capable of binding homophilically to the shed PTPµ extracellular fragment label tumor cells in sections taken from human GBM tissue, whereas scrambled versions of these probes do not bind tumors [78]. The edge samples of human GBM tumors contain the PTPµ extracellular fragment and are labeled by these probes [78]. However, the PTPµ probes do not bind Case Western Reserve University Page 90 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
endogenous full-length PTPµ in the normal (epileptic) brain, likely because of steric
hindrance induced by the engagement of full-length PTPµ–PTPµ homophilic binding
[78]. The most effective of the fluorescently labeled PTPµ probes, SBK2 and SBK4, can
also label flank and intracranial xenograft tumors of human glioma cells when injected
into the tail vein of nude mice [78]. Although the main tumor mass and the tumor edge
could be labeled with the probes in vivo, the probes were not tested in an
invasive/dispersive model. As conventional imaging techniques such as MRI are limited
to visualizing the main tumor mass, we examined the ability of the PTPµ probe to label
migrating and dispersive cells that have moved away from the main tumor mass.
To evaluate tumor cell dispersal in the complex architecture of the adult brain, we
developed a cryo-imaging system and analysis algorithms that create a three-dimensional
(3D) rendering of fluorescently labeled cells in both the main tumor and in dispersed cells
[6, 29]. Using this cryo-imaging system, we identified two highly dispersive cell lines,
CNS-1 rat glioma cells and LN-229 human glioma cells, that form intracranial tumors much like those observed in human GBM, with large populations of dispersive cells that migrate great distances along both blood vessels and white matter tracts [6]. Because of the resolution of the cryo-imaging system, single fluorescently labeled cells can be visualized in three dimensions in the brain [6].
In our study, we evaluated the ability of the fluorescent PTPµ probe to label the adjacent microenvironment of dispersed cell populations of CNS-1-GFP and LN-229-
GFP intracranial tumors as assayed using the 3D cryo-imaging system. Live mice bearing Case Western Reserve University Page 91 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
brain tumors were injected intravenously with the Cy5 PTPµ probe. Brains were imaged
postmortem using whole-brain macroscopic fluorescence imaging and the cryo-imaging
system [6, 29, 78]. Fluorescent signals were analyzed to determine the colocalization of
the Cy5-labeled PTPµ probe with the GFP-labeled dispersed tumor cells in the entire mouse brain. The 3D cryo-imaging system and analysis indicates that the PTPµ probe detected 99% of tumor cells at the main tumor site and in the dispersed cell population up to 3.5 mm from the main tumor. Therefore, the PTPµ imaging probe has potential translational significance for molecular imaging of tumors, guiding a more complete resection of tumors and to serve as a molecular targeting agent to deliver chemotherapeutics to the main tumor mass and distant dispersive tumor cells.
4.2 Material and Methods 4.2.1 Experimental methods
Peptide synthesis and conjugation
The SBK2 peptide (GEGDDFNWEQVNTLTKPTSD) and a scrambled sequence of the
SBK2 peptide (GFTQPETGTDNDLWSVDNEK) were synthesized using a standard
Fmoc-based solid-phase strategy with an additional N-terminal glycine residue in the
laboratory of Dr. Z.-R. Lu as previously described [78]. After synthesis, the N-terminal glycine residues of the SBK2 and scrambled probes were conjugated to Cy5 NHS ester dye (GE Healthcare Life Sciences, Pittsburgh, PA).
Case Western Reserve University Page 92 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Orthotopic xenograft intracranial tumors
Approved protocols from the Institutional Animal Care and Use Committee at Case
Western Reserve University were followed for all animal procedures. CNS-1 (a gift from
Mariano S. Viapiano [138]) or LN-229 (American Type Culture Collection, Manassas,
VA) cell lines were infected with GFP-encoding lentivirus and used for intracranial implantation in 4-6 week old NIH athymic nude male mice (NCr-nu/+, NCr-nu/nu, 20–25 g each) as previously described [6]. GFP fluorescence in 100% of cells was verified before use in intracranial implants. The CNS-1 cell line was authenticated by Research
Animal Diagnostic Laboratory at the University of Missouri (Columbia, MO) for interspecies and mycoplasma contamination by PCR analysis. A total of 5 × 104 (LN-
229) or 2 × 105 (CNS-1) cells were implanted per mouse.
In vivo labeling of intracranial tumors
Imaging experiments were performed at 10–11 days for CNS-1 and 4–8 weeks for LN-
229 to allow for optimal tumor growth and cell dispersal of these two cell lines [6, 78].
Live mice were injected via lateral tail vein with either SBK2-Cy5 or scrambled-Cy5
probes diluted to 15 µM in PBS (3 nmol total probe delivered). After a 90-min interval
for clearance of unbound probe, the animals were sacrificed and the brains were removed
for whole-brain imaging using the Maestro™ FLEX In Vivo Imaging System
(Cambridge Research & Instrumentation (CRi), Woburn, MA) [78]. Immediately after imaging, the brains were embedded in Tissue-Tek OCT compound (Sakura Finetek
U.S.A., Torrance, CA), frozen in a dry ice/ethanol slurry and cryo-imaged. Of note,
Case Western Reserve University Page 93 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells delivery of multiple doses of either the SBK2-Cy5 or scrambled-Cy5 probes did not result in deleterious effects on mouse health (data not shown).
Cryo-imaging of tissue samples
Frozen brains were alternately sectioned and imaged using the previously described Case cryo-imaging System [6, 29] at a section thickness of 15 µm and a resolution of 11 µm ×
11 µm × 15 µm. Brightfield and fluorescence images were acquired for each of the brains using a low-light digital camera (Retiga Exi, QImaging, Surrey, British Columbia,
Canada), an epi-illumination fluorescent light source (Lumen200 PRO, Prior Scientific,
Rockland, MA) and fluorescence filters for GFP (Exciter: HQ470/40×, Dichroic:
Q495LP, Emitter: HQ500LP) or Cy5 (Exciter: FF01-628/40-25, Dichroic: FF660-Di01-
25 × 36, Emitter: FF01-692/40-25; Semrock, Rochester, NY). Single GFP-expressing cells were readily detected with this system. Brightfield and fluorescence exposure settings were identical for SBK2-Cy5 and scrambled-Cy5 probe-treated brains. Seven brains implanted with CNS-1-GFP cells and six brains with LN-229-GFP cells were analyzed.
4.2.2 Image analysis algorithms
Image processing algorithms for visualization of tumor cells and vasculature
Methodologies for segmentation and visualization of the main tumor mass, dispersing cells and vasculature have been described [29]. Briefly, the main tumor mass was segmented using a fast 3D region growth algorithm with intensity and gradient-based inclusion/exclusion criteria. Individual slice images were manually edited if necessary.
Case Western Reserve University Page 94 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Dispersed cells and clusters were detected by thresholding a Gaussian high-pass filter
image, discounting the masked main tumor mass. Autofluorescence of normal tissue is
not a concern for false-positive signals in both red and green channels as it is only less
than 1% of the signal found in the vicinity of the main tumor and dispersed cells. Light
scattering in tissue might appear as a false-positive signal in Cy5 volumes. However, we
processed Cy5 volumes with a next-image processing algorithm and used attenuation and
scattering parameters for brain tissues that were previously described [127]. Similar
parameters were applied for both SBK2-Cy5 and scrambled-Cy5 volumes. To find
colabeled cells/clusters, we applied a logical AND operator between the dispersed cell
and a Cy5 volume made binary with a threshold. The system is not overly sensitive to
variations in the threshold for creating the binary Cy5 volume. The lower limit for the
blood vessel detection algorithm was 30 µm diameter. Three-dimensional volumes of
tumor, dispersed cells, Cy5-labeled cells∼ and vasculature were rendered using Amira
software (Visage Imaging, San Diego, CA), with modifications developed expressly for
cryo-image data. Pseudo-colors were chosen to give the best contrast between the
different data volumes. Colors were green (main tumor), yellow (dispersing cells), pink
(PTPµ probe-labeled dispersed cells) and red (vasculature).
Relative quantification of Cy5 fluorescence
Cy5 fluorescence intensity of tumors and dispersed cells was quantified by Matlab
software using next-image processed volumes for CNS-1 and LN-229 tumors labeled with SBK2-Cy5 or scrambled-Cy5 probes, which normalized the signal to background
Case Western Reserve University Page 95 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
fluorescence. Total fluorescence signal for each brain was then normalized for tumor
volume. SBK2-Cy5 analysis: n = 4 tumors for CNS-1 and n = 5 tumors for LN-229.
Scrambled-Cy5 analysis: n = 3 tumors for CNS-1 and n = 1 tumor for LN-229.
Distance analysis of SBK2-Cy5 probe-labeled dispersed cells
The distance between the main tumor mass and dispersed cells was determined using a
3D morphological distance algorithm [29]. Briefly, the algorithm detects the presence of fluorescent voxels in a series of 3D dilations from the tumor edge outward. Voxel size was 11 µm × 11 µm × 15 µm, approximately the size of single cells. Dilations within 500
µm of the tumor edge were discarded to reduce nonspecific errors and to focus on the population of cells that had dispersed the greatest distance from the main tumor mass.
Comparisons were made between the average dispersed cell population and the average
SBK2-Cy5-labeled dispersed cell population for each 500 µm distance increment from a total of four brains.
Statistical analysis
Results from the morphological distance algorithm described above were segmented into incremental distances of 500 µm from the tumor edge, and averaged from four brains.
The standard deviation was calculated in Excel from the range of all values within each plotted distance increment for the four brains. The standard error was calculated in Excel using the standard deviation for each data point shown.
Case Western Reserve University Page 96 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
4.3 Results A PTPµ-targeted probe was developed for the molecular recognition of glioma cells.
Intracranial tumors of glioma cells expressing GFP were labeled with the PTPµ-targeted probe (SBK2-Cy5) or with a scrambled version of the PTPµ probe (scrambled-Cy5) through intravenous administration. The main tumors were brightly labeled with the
SBK2-Cy5 probe in all cases examined, whereas the scrambled probe resulted in no significant signal above background within the tumor (Fig. 19). Ten-day CNS-1-GFP intracranial tumors were labeled in vivo with the PTPµ probe by intravenous injection for
90 min before sacrifice and postmortem imaging. This time frame for probe clearance was determined empirically to provide optimal signal to noise (background) ratio of Cy5 fluorescence. The results from the two-dimensional (2D) block face images demonstrate bright PTPµ probe fluorescence (Figs. 19g and 19k) within the main GFP-positive tumor
(Figs. 19f and 19j) in all cases. Cy5 fluorescence signal from PTPµ probe labeling also closely corresponded with the pattern of cell dispersal away from the main tumor (Figs.
19h and 19l). In contrast, the scrambled probe did not appreciably label GFP-positive tumor cells above background (Figs. 19a–19c). Quantitation of Cy5 fluorescence intensity from the CNS-1 main tumors and dispersed cell populations labeled with the
PTPµ probe was 84.2 ± 15.6 (n = 4 brains) compared to 7.2 ± 2.3 (n = 3 brains) for the scrambled probe.
Case Western Reserve University Page 97 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 19 CNS-1 glioma tumors and dispersing cells are labeled by the PTPµ probe. Unfixed mouse brains containing xenografts of GFP-expressing CNS-1 cells were cryo-imaged after in vivo labeling with scrambled probe (a–c) or PTPµ Case Western Reserve University Page 98 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
probe (e–l). Two-dimensional block face images are shown for brightfield (a, e and i), with boxed regions that correspond to the views shown for GFP fluorescence (tumor; b, f and j) and Cy5 fluorescence (scrambled or PTPµ probe; c, g and k) in each column. An overlay of GFP and Cy5 fluorescence demonstrates extensive labeling of the dispersed glioma cells with the PTPµ probe (h and l). In tumors with minimal cell dispersal (f), labeling with the PTPµ probe is localized to the main tumor (g and h). PTPµ probe labels diffusely dispersing cells in a tightly focused pattern when cells migrate as a stream along a defined structure, such as a blood vessel (k and l—see arrow). Brightfield image of a tumor from a brain that was perfused with India Ink (d) to illustrate leakiness of the vasculature. Asterisk in (d) indicates lateral ventricle. Scale bar in (i) represents 1 mm for panels (a, e and i) and the scale bar in (l) represents 1 mm for panels (b–d, f–h and j–l).
The blood–brain barrier is thought to be leaky within the main tumor in humans, a feature that is required for agents such as gadolinium to contrast brain tumors against the normal brain background. We examined the blood–brain barrier integrity in our mouse model system by perfusing with dilute India Ink before removal of the brain containing a tumor. Vasculature of the perfused brain was contrasted black against the brain parenchyma, and blood vessels that coursed through the tumor were clearly defined. In addition, a small accumulation of India Ink was localized to the main tumor (arrowhead in Fig. 19d), suggesting leakiness of blood vessels confined to the main tumor. On the basis of this result, we hypothesize that regions distant from the tumor edge likely have an intact blood–brain barrier, suggesting that the PTPµ probe may be able to cross the blood–brain barrier to detect dispersing tumor cells.
The probe labeled brains were then subjected to cryo-imaging analysis (Fig. 20) as previously described [6]. The cryo-imaging system obtained 2D, microscopic brightfield anatomical images, including vasculature, and multispectral fluorescence Case Western Reserve University Page 99 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
images of tumor and probe. We refined our previous method [29] to segment and visualize the vasculature, main tumor mass and dispersing cells to highlight the PTPµ probe labeling of only the dispersive glioma cells and cell clusters (refer to Material and
Methods section). Three-dimensional volumes were created for the main tumor mass
(pseudo-colored green) and vasculature (pseudo-colored red). Glioma cells no longer physically connected in any dimension to the main tumor were pseudo-colored yellow to indicate tumor cell dispersal. To specifically visualize the PTPµ probe overlay in 3D, the population of dispersing cells that was colabeled with the PTPµ probe (Cy5 fluorescence) was pseudo-colored pink by the computer program.
The rat CNS-1 glioma cell line rapidly disperses to distances of several millimeters from the main tumor [6]. A gradient of PTPµ probe fluorescence (Figs. 20c and 20g) was observed that encompasses the wave of GFP-positive cell dispersal from the tumor (Figs. 20b, 20f and 20j). Streams of CNS-1-GFP cells migrating along a defined structure were highlighted against the background tissue by the PTPµ probe
(Figs. 20g, 20h, 20k and 20l).
We have previously shown by immunoblot that the dispersive edge region of human glioblastoma tumors accumulates the PTPµ extracellular fragment [6], even though the tumor edge is comprised of normal brain cells and some dispersed tumor cells.
In our study, the PTPµ probe fluorescent signal was often brighter in the region of lower
GFP fluorescence, which corresponds to the area of tumor cell dispersal, than it was in the main tumor, which has higher GFP fluorescence (Figs. 20b–20d and 20f–20h). This Case Western Reserve University Page 100 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells result suggests that the dispersed cells may have more cleaved PTPµ extracellular fragment deposited in their adjacent microenvironment than in the main tumor, as recognized by the PTPµ probe. A magnified version of that same brain at the midline region of the frontal pole (asterisk) shows that GFP dispersing cells (Fig. 20f) are labeled with the PTPµ probe (Cy5: Figs. 20g and 20h) in 2D block face images.
Case Western Reserve University Page 101 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 20. The PTPµ probe specifically labels CNS-1 glioma cells that have dispersed from the main tumor. Unfixed mouse brains containing xenografts of GFP-expressing CNS-1 cells were cryo-imaged after in vivo labeling with PTPµ probe (a–h). Two- dimensional block face images are shown for brightfield (a and e), GFP
Case Western Reserve University Page 102 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
fluorescence (tumor; b and f) and Cy5 fluorescence (PTPµ probe; c and g). An overlay of GFP and Cy5 fluorescence demonstrates extensive labeling of the dispersed glioma cells with the PTPµ probe (d and h). A high magnification image demonstrates that the PTPµ probe labels a stream of dispersing tumor cells (f) at the midline (asterisk) (g and h). Three-dimensional reconstruction of the tumor labeled with PTPµ probe is illustrated as main tumor (green), and dispersing cells (yellow) in a magnified view (j). Complete vasculature for this brain specimen is also shown (i). Dispersed cells colabeled with PTPµ probe are pseudo-colored pink (k). (l) A magnified view of the tumor shown in (k) illustrates colabeled cells along the midline, which correspond with the labeled midline cells in the 2D overlay image (h). Scale bar in (i) represents 1 mm for panels (a–d and i). Scale bar in (k) represents 500 µm for panels (j and k). Scale bar in (l) represents 500 µm for panels (e–h and l).
Three-dimensional reconstruction of the tumors illustrates a GFP-positive main
tumor (pseudo-colored green) with a large population of GFP-labeled cells that disperse in many directions (pseudo-colored yellow) (Figs. 20i and 20j). Analysis of the PTPµ probe labeling of GFP-positive dispersed cells (pseudo-colored yellow) with the binary
Cy5 volume indicates that greater than 99% of the total dispersed cells were colabeled
with the PTPµ probe (pseudo-colored pink) (Figs. 20k and 20l and 22). Evaluation of
additional CNS-1-GFP intracranial tumors labeled with the PTPµ probe demonstrated
that the probe was highly effective at labeling the dispersed cell population (Fig. 21). In
the two examples shown, a substantial amount of cell dispersal is evident over the brain
vasculature in many directions (Figs. 21c and 21h).
Case Western Reserve University Page 103 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 21. Three-dimensional views of dispersed cells from CNS-1 intracranial tumors labeled by the PTPµ probe. Unfixed mouse brains containing xenografts of GFP-expressing CNS-1 cells were cryo-imaged and reconstructed in three dimensions after in vivo labeling with the PTPµ probe. Two-dimensional block face images are shown for brightfield (a and f), and 2D block face zoomed overlay of GFP (tumor) and Cy5 fluorescence (PTPµ probe; b and g) for two brain tumors. Three- dimensional reconstructions of the same tumor specimens showing the main tumor mass (pseudo-colored green), dispersed tumor cells (pseudo-colored yellow) and vasculature (pseudo-colored red) illustrate that the dispersing cells often migrate on blood vessels (c and h). The total dispersing cell population is extensively colabeled with the PTPµ probe (d and i), as shown pseudo-colored pink. Magnified views from (d and i) illustrate that PTPµ-colabeled cells are detected several millimeters from the main tumor (e and j). Scale bar in f represents 1 mm for panels (a and f) and scale bar in g represents 1 mm for panels (b and g). Scale bar in i represents 500 µm for panels (c, d, h and i) and the scale bar in j represents 500 µm for panels (e and j).
Further analysis of the PTPµ probe-labeled CNS-1 intracranial tumors was
performed to determine how far away from the main tumor mass dispersed cells could be
labeled with the probe. Four CNS-1 intracranial tumors were analyzed to determine the total number of the dispersing cell population that was colabeled with the PTPµ probe.
The data were segmented into incremental distances of 500 µm from the tumor edge and
Case Western Reserve University Page 104 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells then averaged from the four brains. Based on the algorithms, the results show that the
PTPµ probe colabels 99% of the dispersed cells to a maximum distance of 3.5 mm from the tumor edge (Fig. 22). In fact, the PTPµ probe is able to label even the distant dispersed tumor cells in multiple directions many millimeters from the main tumor mass
(Figs. 21d, 21e, 21i and 21j as well as 22).
Figure 22. Histogram of the average number of GFP-positive dispersing cells colabeled with the PTPµ probe per unit distance from the main tumor, ± standard error (n = 4 tumors analyzed).
LN-229-GFP intracranial tumors require a minimum of 4 weeks growth to exhibit appreciable tumor cell dispersal [6]. Labeling of LN-229-GFP tumors with the PTPµ
Case Western Reserve University Page 105 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
probe resulted in a bright Cy5 fluorescence signal within the main GFP-positive tumor, and labeling of greater than 99% of the dispersed cell population (Figs. 23a–23c).
Quantitation of Cy5 fluorescence from the LN-229 main tumors and dispersed cell populations labeled with the PTPµ probe was 114.7 ± 25.9 (n = 5 brains) compared to 5.8
(n = 1 brain) for the scrambled probe. In rare instances, glioma cells were deposited in close proximity to the lateral ventricle, resulting in extensive spread of cells along the ventricular walls, leptomeningeal regions, within the brainstem and in brain parenchyma at the cerebral/cerebellar junction (3D reconstruction shown in Figs. 23j–23l).
Fluorescent signal from the PTPµ probe colocalized with much of this dispersed cell population (Figs. 23d–23l).
Case Western Reserve University Page 106 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 23. Dispersing cells from LN-229 intracranial tumors are specifically labeled by the PTPµ probe. Unfixed mouse brains containing xenografts of GFP-expressing LN-229 cells were cryo-imaged and reconstructed in three dimensions after in vivo labeling with the PTPµ probe. Three-dimensional reconstruction of an LN229 tumor (a) shows the main tumor mass (pseudo-colored green), dispersed tumor cells (pseudo-colored yellow) and vasculature (pseudo-colored red) (b). PTPµ probe colabeling of dispersing cells (pseudo-colored pink) was observed at great distances from the main tumor (c). A second tumor example where the LN229 tumor cells spread through the ventricles of the brain is shown (d–l). Two- dimensional block face images are shown for brightfield (d and g), GFP fluorescence (tumor; e and h) and Cy5 fluorescence (PTPµ probe; f and i). Comparison of GFP (e) and Cy5 (f) fluorescence in zoomed images (h and i) demonstrates extensive overlap in signal in this specimen. Three-dimensional reconstruction of the same tumor specimen is shown (j–l). In addition to the main tumor, the total dispersing cell population is extensively colabeled with the PTPµ probe, as shown pseudo-colored pink (l). Scale bar in (l) represents 500 µm for panels (a, d–f and j–l). Scale bar in (c) represents 500 µm for panels (b and c). Scale bar in (i) represents 500 µm for panels (g–i).
Case Western Reserve University Page 107 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Results from these experiments suggest that the PTPµ probe crosses the blood– brain barrier to label not only the main tumor mass but also the vast majority of the dispersed tumor cells at a range up to 3.5 mm away from the main tumor mass (Fig. 23).
In contrast, the scrambled-Cy5 probe did not label tumors (Figs. 19a–19c). The PTPµ probe preferentially labeled the PTPµ extracellular fragment deposited in the adjacent tumor microenvironment of the dispersed cells (Figs. 19g and 19k). These results suggest that the PTPµ probe is a marker of the microenvironment of the main tumor, tumor edge and dispersing cells, and could be utilized for tumor imaging, a more complete surgical resection or could be a viable molecular targeting agent to deliver therapeutics.
4.4 Discussion Better detection tools for tumor imaging are needed. Molecular recognition of tumor cells would facilitate guided surgical resection. To achieve this goal, targeted imaging tools must specifically label tumor cells, not only in the main tumor but also along the edge of the tumor and in the small tumor cell clusters that disperse throughout the body. To further improve patient survival, probes for the targeted delivery of therapeutics must be developed. The use of the cryo-imaging system allows us to determine whether the PTPµ probe is able to label individual dispersing tumor cells at a great distance from the main tumor within the complex 3D environment of the brain. We demonstrate that the PTPµ fluorescent probe is a useful reagent for labeling the main tumor mass and migrating cells, offering a visible means of identifying the dispersed cell population thought to be responsible for tumor recurrence. The kinetics of the PTPµ probe binding to tumors after
Case Western Reserve University Page 108 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
tail-vein injection demonstrate that it binds within minutes of administration and lasts for
as long as 3 hr [78], thus providing a window of time for surgery. The PTPµ probe has a
high tumor contrast compared to background, as shown in the cryo-images acquired with
a low power objective. Microscopes used in surgical suites typically use higher power
objectives with a greater numerical aperture. As there is a direct relationship between
numerical aperture and fluorescence brightness, the signal-to-noise ratio of the SBK2-
Cy5 probe would be further improved in the surgical setting. Clinical safety of the probe would need to be evaluated, but our preliminary findings show no toxicity of the probe.
However, we have not performed any formal toxicology studies with the probe to date.
Greater than 50% of tumor volume resection has been shown to be achievable through fluorescence-guided resection of tumor tissue using the Aminolevulinic Acid
(ALA)-Protoporphyrin IX (PpIX) system [154], which visualizes fluorescently detectable
PpIX, produced as a result of exogenously introduced ALA [168]. In addition to the
ALA-PpIX system, other fluorescent labeling agents have been developed to label the main tumor (for review, see Ref. [169]). Targeted fluorophores against the EGF receptor type II (EGFR2/HER2) and VEGF receptors as well as folate receptor α have been used to intraoperatively label xenograft models of breast, ovarian and gastric cancers (in the case of HER2 and VEGFR)[170] and ovarian cancer (in the case of the folate receptor
α)[171]. Protease activatable fluorophores have also been used to detect tumor cells
[169]. For example, the protease cathepsin is enriched in the tumor microenvironment. A cathepsin-activated near-infrared fluorescent probe is effective at intraoperatively
Case Western Reserve University Page 109 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
labeling breast cancer cells in a rat tumor model [172]. An activatable cell-penetrating peptide (ACPP) conjugated to Cy5 is taken up by cancer cells only after proteolytic cleavage by MMPs 2 or 9 [173, 174]. The use of the ACPP-Cy5 probe as an intraoperative guide for surgery in mouse tumor models resulted in fewer cancer cells being left behind and better tumor-free and overall survival than in mice that underwent surgery without the fluorescent probe as a guide [173].
The ALA-PpIX system has significant clinical utility in surgical resection of
GBM, but its use results in only 50.2% of patients having complete tumor resection[156].
It is unclear why only 50% of the patients respond. This may be the limit for improving
upon surgical resection of brain tumors, but we hypothesize that better detection of
dispersive cells in GBM would enhance the ability to completely resect a tumor.
Although the ACPP-Cy5 probe is promising for tumor labeling outside of the brain [173],
the addition of a large molecular weight carrier to the ACPP (ACPP-D) to facilitate tumor uptake and reduce background fluorescence produces a molecule that may be too large,
28 kDa [175], to cross the blood–brain barrier. We hypothesize that the PTPµ probe could∼ be an improvement upon existing intraoperative labeling technologies because of its small size ( 3 kDa) and its ability to label virtually all of the highly dispersive cells that are characteris∼ tic of GBM, the very cells that are responsible for recurrence and the terminal nature of the disease.
Probes designed to label molecules that accumulate in the tumor microenvironment may also be advantageous as therapeutic targeting agents, as they can Case Western Reserve University Page 110 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells identify both the main tumor cell population and areas with infiltrating cells that contribute to tumor recurrence. The PTPµ probe labels the shed PTPµ extracellular fragment that accumulates in the tumor microenvironment, but is not present in normal brain [78]. For this reason, the PTPµ probe could also be used to deliver therapeutics to the tumor microenvironment, including chemotherapy, directly to all tumor and any infiltrating stromal cells. The animals were sacrificed 90 min after probe injection, a time interval too short to observe any biological effects of the probe. It is currently unknown whether use of the PTPµ probe affects tumor cell dispersal/invasion. Future studies will address the long-term clearance and biological effects of the probes in vivo. The ability to directly target the tumor microenvironment would increase both the specificity and sensitivity of current treatments, therefore reducing nonspecific side effects of chemotherapeutics that affect all cells of the body.
Cleavage of many cell surface proteins by ADAM and/or MMPs occurs in tumor tissue [176, 177] and is likely due to the increased expression of proteases in the adjacent tumor microenvironment, especially at tumor margins [178]. The techniques described for the development of fluorescently labeled PTPµ probes may also be used to develop probes to other proteins that are cleaved in GBM and other tumor types [177].
Case Western Reserve University Page 111 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
5 Chapter 5: Validation of Targeting for MRI Molecular Imaging Agents
5.1 Introduction Medical imaging is one of the most powerful clinical tools that is used for cancer
detection, therapy response, surgery guidance, and staging of the disease [49, 50]. Widely used imaging modalities for tumor detection include PET, MRI, CT, and mammography.
Each of these modalities has its own advantages and disadvantages. For example, PET
lacks the spatial resolution due to the partial volume effect which limits the resolution,
CT lacks soft tissue contrast, and although MRI provides a good contrast of soft tissue with a relatively excellent spatial resolution, sensitivity and/or resolution might be still an issue for micro metastases detection. Hence, current medical imaging research efforts are focused to improve the sensitivity and specificity of the medical imaging modalities for tumor detection [51, 52] by introducing molecular targeted imaging agents, improving imaging protocols, and/or combining imaging modalities to perform multimodal imaging, etc. We focus in this chapter on the preclinical evaluation of the newly developed MRI molecular imaging agents.
Preclinical evaluation of new molecular imaging agents that are targeted to metastases requires detection of whole mouse metastases in order to evaluate the efficiency and limitations of targeting at the whole body level to address the clinical translation of these agents. However, detection of whole body micrometastases is an extremely difficult task, since current preclinical imaging modalities lack the sensitivity, resolution and/or field of view to cover whole body metastases. Optical imaging is widely Case Western Reserve University Page 112 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
used in evaluating tumor models and targeted imaging agents due to its ability to image reporter gene and/or providing multispectral fluorescence imaging. Optical modalities include bioluminescence (BLI), and multispectral in vivo fluorescent imaging (such as
Maestro™ EX, PerkinElmer), these modalities are very sensitive and provide good contrast for labeled tumors. However, the use of these systems in whole body imaging is
limited by light scattering and absorption which limits the resolution and signal depth.
Because of this, previous attempts to image whole mouse GFP expressing metastases
were only efficient at shallow levels from the surface [92-95]. Fluorescence molecular
tomography (FMT, Visen) tried to resolve this by using near infrared fluorescent labels
which has better tissue penetration and less scattering and absorption. However,
resolution of FMT is limited due to the size of the mouse and is limited by the variety of
fluorophores that can be used since it uses near infrared excitation lasers. Two photon
and confocal microscopy have improved to image deeper sections with less photo
bleaching by using near infrared lasers [96], but are still limited in the field of view.
Although multiphoton intravital microscopy provide unique in vivo high resolution
optical images that can be used to monitor tumor progression, invasion and response to
therapy [97] , it is limited to small optical windows and inaccessible to all tissues in the
mouse [66] .
Cryo-imaging (BioInVision Inc., Cleveland, OH) resolves the limitations of the
aforementioned imaging modalities, and provides whole mouse imaging with single cell
resolution for fluorescently labeled cells [6, 29, 124-128, 179]. This makes the system
Case Western Reserve University Page 113 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
uniquely suited for whole mouse characterization of tumor models, micro-metastases, and targeted imaging agents as compared to other small animal imaging modalities. By acquiring brightfield and multispectral fluorescence data using cryo-imaging, we can efficiently evaluate any combination of tumor models and targeted imaging agents at single cell resolution across an entire mouse. However, this required building a platform methodology consisting of imaging protocols and image analysis algorithms to combine the information acquired from cryo-imaging with other modalities to evaluate the targeting of newly developed imaging agents. In this report, we specifically focus on combining cryo data with MRI data to evaluate targeted imaging agents designed for
MRI. Figure 24 shows a general diagram of the analysis platform. Basically, we start by creating a fluorescent tumor model and targeted imaging agent for MRI. We acquire MRI and cryo image data, and combine the information using 3D image registration to evaluate targeting at the single cell level. We also acquire histology, which can be acquired during cryo-imaging, and register the information back to the cryo images to provide assessment at the molecular level. This unique preclinical evaluation of imaging agents can provide more information about the clinical feasibility of these agents. This platform methodology can be employed in the same way to evaluate any combination of tumor model and imaging agent or theranostics.
Case Western Reserve University Page 114 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 24. Pre-clinical platform methodology for metastases targeting analysis. Starting with a metastatic tumor model and targeted imaging agent for in vivo imaging such as MRI, we propose a methodology to assess the targeting of imaging agents/theranostics at molecular level. The methodology employs cryo- imaging system, with sensitivity to resolve single cells, and histology that can be acquired during cryo-imaging. The goal is to evaluate targeting at molecular level. Through image registration and visualization software, we can provide answers to targeting efficiency and clinical feasibility of imaging agents in the clinical arena.
To build our platform for the whole mouse assessment of targeting to metastases, we employed a newly developed targeted imaging agent for MRI, (CREKA-Tris-(Gd-
DOTA)3) [84]. CREKA is a tumor-homing pentapeptide (Cys-Arg-Glu-Lys-Ala) [83], which homes to blood clotting proteins, specifically, fibrin-fibronectin complexes that are highly expressed in the tumor microenvironment. To evaluate the targeting of this peptide, we built imaging protocols and analysis algorithms using cryo-imaging and MRI data. We used a metastatic breast cancer model that expresses GFP to fluorescently identify all metastases using cryo-imaging. For the targeted peptide, we used two
Case Western Reserve University Page 115 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells counterparts of the peptide in our evaluations, the MRI counterpart (CREKA-Tris-(Gd-
DOTA)3) and the fluorescence counterpart Cy5.0-PEG-CREKA for further assessment using multispectral fluorescence cryo-imaging. By performing 3D multimodality registration, we analyzed GFP tumor location, and checked the corresponding locations in Cy5 and MRI to provide targeting assessment at single cell level and even at the molecular level by adding histology.
This study is part of a larger effort at Case Western Reserve University to develop new targeting imaging agents for MRI to improve contrast and enhance detection of micro-metastases. Following is the experimental design and overview of image analysis algorithms.
5.2 Materials and Methods A summary of the experimental design to evaluate the targeting of CREKA peptide is presented in Figure 25. We created a metastatic breast cancer mouse model by orthotopic injection of 4T1-luc2-GFP cell line in the fat pad. These cells express GFP, which can be detected in the multispectral cryo-imaging. After 30 days, we surgically removed the main tumor mass and allowed metastases to grow for 10 more days. Then, we injected
Cy5.0-PEG-CREKA (CREKA-Cy5, for short) in the tale vein and allowed 3 hours for tissue clearance. We followed that by the injection of CREKA-Tris-Gd(DOTA)3,
(CREKA-Gd, for short), in the tail vein. During the MRI imaging we captured two volumes, an in vivo scan using a 3D fat suppression sequence optimized for Gadolinium, then another high resolution sequence after animal death to be used in 3D registration.
Case Western Reserve University Page 116 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
After MRI, we prepared the mice for cryo-imaging by embedding the mice in O.C.T
(optimum cutting temperature, Tissue-Tek®) and flash freezing them using liquid nitrogen. In cryo-imaging, we acquired color brightfield, GFP green tumor, and Cy5 red peptide images. During cryo-imaging, we also acquired some histological slides using a special adhesive film and performed H&E staining. More details about the experimental design are described in section 2.
Figure 25. Experimental design for evaluation of CREKA peptide as MR imaging agent. Metastatic breast cancer tumor model was created using 4T1 cell line which express GFP as a label. Tumor cells were injected in the fat pad, then after 30 days, main tumor mass was surgically removed to allow metastases to grow. 10 days later, CREKA-Cy5 was injected in the tail vein for fluorescence evaluation of the peptide, after 3 hours of tissue clearance time, CREKA-Gd was injected for MRI evaluation. During MRI we captured in vivo volume and Ex-vivo high resolution volume for use in 3D image registration with cryo- imaging volumes. After embedding in O.C.T and snap freezing, we captured color anatomy brightfield, green GFP tumor, and red Cy5 peptide. Throughout the experiment, we acquired some histology slides using special adhesive tape.
Given the acquired volumes, we performed 3D multimodality image registration
between MRI high resolution volume and cryo brightfield, then registered histology to Case Western Reserve University Page 117 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
both. We designed image analysis tools to visually and quantitatively assess labeling. In
the following section, we explain our image registration and analysis algorithms.
5.2.1 Image Processing and analysis
3D registration of mouse to molecule data
We used the brightfield cryo-image volume as the reference volume and
registered all other image data (3D MRI and 2D histology) to it, Of course, the GFP and
Cy5 image volumes were already perfectly registered to the brightfield data as they were
all acquired concurrently. We now describe gray-scale, rigid, affine and non-rigid registration of the moving MRI volume to the reference 3D brightfiled. We used the
Nifty software package implementation for the algorithms described below [180-183].
There were important preprocessing steps. Color brightfield images were converted to gray scale using (Gray = 0.2126 × R + 0.7152 × G + 0.0722 × B). The very
high resolution gray-scale brightfield image volume was downsampled to 0.1 × 0.1 × 0.1
mm using anti-aliasing Lanczos filter. This was done to make voxels isotropic and to
bring the voxel size closer to MRI voxel size. High resolution MRI image volumes
(typically 0.1172 × 0.0977 × 0.1406 mm) were used. We did not resample the MRI
volume since the voxel size was close to the resampled cryo volume, thus avoiding an
extra interpolation step. Next, the cryo and MRI volumes were both linearly mapped to
an 8-bit (0-255) grayscale range. For cryo data, which are already 8-bit, window/level
was used to adjust the values in the histogram to increase the contrast. For MRI data,
which are typically 16 bit images (with values ranging from 0-32766), we linearly
Case Western Reserve University Page 118 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
mapped the range 0-15000 to the 8-bit range, as most of tissue information data were within that range. This mapping increased the contrast and was the easiest way to ensure that bins for the joint histogram in mutual information calculations were appropriately scaled. Image volumes were cropped using rectangular cropping just outside the tissue to remove extra background. We roughly registered the two volumes prior to the start of automated registration. As long as there was good overlap and volumes were similarly oriented, automated registration was deemed successful.
We employed the algorithm described in [181, 182] for the rigid and affine registration step. Basically, the algorithm perform rigid and affine registration based on a block matching strategy where local displacement is estimated based on matching cubes and using normalized cross correlation as similarity measure. Then, a global estimation for the transformation is calculated from these local displacements using least trimmed squares to provide a robust estimation for the global transform. To find the best global transformation, only a percentage of the blocks are used to estimate the final global transform. This is done by choosing the highest intensity variance blocks first. Multilevel approach is used and images at each level are subsampled by 2 by using a Gaussian filter.
Although this algorithm uses normalized cross correlation as similarity measure as opposed to mutual information, it proved its efficiency in multimodal image registration especially since normalized cross correlation is estimated locally using small blocks to find the match of high variance features such as edges [181, 182]. The block matching approach in this algorithm generated a better transform than other algorithms that use
Case Western Reserve University Page 119 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells global mutual information to generate a rigid and affine transform when we compared the results. Hence, we chose this algorithm to serve as initial registration for deformable registration. There are multiple parameters to optimize in this algorithm including: block size, neighborhood search size, and spacing between cubes.
Non-rigid registration was applied using the free form deformation (FFD) algorithm [180]. FFD uses a lattice of control points superimposed on the reference volume to estimate a transformation T(.) between the reference and the floating volumes.
The deformation at each voxel in the transformed floating image is calculated using a cubic B-spline interpolation of each of the coordinates in the surrounding 4x4x4 control points. Mathematically, each voxel in the floating image x= [x1, x2, x3], is mapped to the reference position location x = [x123 ,x ,x ] in the reference image by T(x) where T(x) =
[T1(x), T2(x), T3(x)] and is defined by
333 T(x) = ∑∑∑Bai (r ) B bj (r ) B ck (r )Ca,b,c (1) abc=000 = =
where:
Ba, Bb, Bc : 1D B-spline basis functions of order 3
Ca,b,c : the mesh of the control points locations
r= [ri, rj, rk] : Relative position of a voxel with respect to the closest control point and is found using
xxii dr= −1, i = −+(d 1) spacing spacing Case Western Reserve University Page 120 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Basically, the 3D transform T(x) at any voxel location is the results of the tensor
product of the 1D B-spline basis functions (Ba, Bb, Bc) which are of order 3 spline
polynomials. For each x= [x1, x2, x3], the relative position r= [ri, rj, rk] is found with respect to closest control point. Then the value of the transformation in each direction is calculated from the contribution of the 4 control points (2 below and 2 above) using the basis function at the voxel location interval.
Cost function
A cost function is iteratively optimized by varying displacements of the control points.
The total cost function (Ctotal) is given below where the w’s are weights to be empirically
determined. In our application, we found it necessary to use all three terms described
next.
Ctotal =(1-w1 -w 2 )NMI-(w1 )BE-(w2 )JL (2)
The cost function consists of a normalized mutual information and regularization terms. We used normalized mutual information (NMI) because of its ability to register multi-modality images [184] and its improved characteristics with non-overlapping
tissues as compared to standard mutual information [122].
H(I12 ) + H(I ) NMI(I12 , I ) = (3) H(I12 , I )
H(I1) and H(I2) are the marginal entropies of I1 and I2, and H(I1,I2) is the joint entropy. Case Western Reserve University Page 121 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
There is a bending energy (BE) smoothness constraint that can penalize any non-
smooth deformation that might represent unrealistic deformations. The general form of
this function was described in [180, 185]. Basically, this penalty term is based on the
second spatial derivative of the deformation field as shown below. Since the second
derivative measures curvature, high values indicate bending of the deformation field. By adding this penalty, a smooth deformation can be guaranteed.
2 2 3 ∂∂22Τ(x) Τ(x) ΒΕ = [ + ] ∑∑2 2 ∑ (4) 3 ∂ ∂∂ x∈R i=1,xi ij xxij ij≠
The determinant of the Jacobian is used to measure any change in local “volume.”
A value greater or less than one means expansion or compression, respectively. A penalty
term is added which penalizes both compression and expansion equally [186, 187]. It is
the logarithm of the absolute value of the determinant of the Jacobian (JL) as given below.
JL= ∑ log(| JT (x) |) (5) x∈R3
where
∂∂∂T(xxx ) T( ) T( ) 111 ∂∂∂xxx123 ∂∂∂T(xxx ) T( ) T( ) J (x) = det 222 T ∂∂∂xxx123 ∂∂∂T(333xxx ) T( ) T( ) ∂∂∂xxx123
Case Western Reserve University Page 122 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Optimization was done using a 3-level, multi-scale approach where the spacing between control points and image resolution was subsampled by 2 at each level. We designed an experiment to find the optimal weights (w1 and w2) in the cost function. The experiment is described in section 2.
Since we acquired the histological sections using a special adhesive film, the geometry of these slices was preserved. Hence, an algorithm to register theses slices to the cryo-imaging volume was not necessary. Instead we manually registered these slice to the corresponding slice in the cryo volume and inspected the results visually. By registering both MRI volume and histology to cryo volume, registration of MRI to histology is complete.
Tumor and targeting analysis
Following registration, we created tools to analyze metastases and assess targeting visually and quantitatively. Visually, we created a quick targeting assessment tool to put all registered volumes in a common visualization framework. We examined each GFP tumor and the corresponding CREKA-Cy5 and CREKA-Gd locations, along with the brightfield color images for anatomical orientation. The tool allows digitally slice the volumes and compare all data at the same plane. Below is shown quantitative measures for the imaging agent targeting.
We segmented GFP tumors manually to visualize tumors in 3D and provide quantitative results on measurements. Segmentation was done by experts in cryo images to avoid segmenting any auto-fluorescence structures. Results of segmentation were also
Case Western Reserve University Page 123 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
compared to a control mouse with no tumors to make sure all segmented tumors are not
other auto fluorescence structures. In cases where tumor GFP signal was low, we used the brightfield volume to segment some of the bigger necrotic tumors since they appeared as a distinctive white spot in the brightfield images. Then, we performed connected component analyses with 26 neighborhood connectivity. We used the final segmentation results to measure metastases sizes, observe distribution in organs, create visualizations, and compare corresponding locations in Cy5 and MRI.
To relate the size of metastatic tumors with peptide average intensity in MRI and whether this intensity is detectable by a human observer, we developed a tool based on the Rose criterion of detection to find the number of detectable metastases in MR per organ. Basically, the Rose criterion provides an intensity threshold for an object detectability by a human observer based size, noise level in the background, and contrast
[188, 189]. The equation used to calculate Rose CNR for each object is as follows.
CA CNR = (6) Rose σ
Where C is the average contrast of a tumor calculated by subtracting the average
background intensity from the average tumor intensity. A is the area of tumor and σ is the
standard deviation of the background.
Rose CNR of 3-5 is the widely accepted threshold in literature for an object to be
detectable by a human observer [189]. However, the exact threshold value is affected by
different criteria including observer experience and the shape of the object to be detected. Case Western Reserve University Page 124 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
We chose a value of 4 to design our tool to provide a moderate choice between an
experienced user and detectability.
Since different organs in MRI have different background and noise, our tool
enables a user to select some background areas in each organ of interest. From these
areas, we can determine standard deviation and average intensity of the background tissue. Then, a bounding box around each GFP tumor is mapped to its corresponding location in the registered MRI. From this mapped location, we measured MRI average
intensity level at each slice through that tumor inside the bounding box. By using the
Rose CNR equation, we calculated the Rose value for that tumor at each slice and then
assigned the maximum value to that tumor. Our choice of assigning the maximum value
is based on the fact that a human observer will detect a tumor at its maximum contrast.
Moreover, analysts usually use a maximum intensity projection to visualize tumors. The area of that metastasis is calculated by using the area of the bounding box in the x-y
direction, since we usually evaluate targeting and calculated tumor diameters for our
images in that direction. Then, following analysis of all GFP tumors, we represent the
result in scatter plots where the size of the tumor (area) is on the x-axis and the Rose
CNR on the y-axis. On that plot, we can draw a curve that represents the threshold of
detection for that specific organ based on a Rose minimum CNR value of 4. Using data
from such a plot, we can related tumor size with detectability rate and determine the
percentage of detectable tumor in MRI for each organ and for every tumor size.
Case Western Reserve University Page 125 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
5.2.2 Experimental materials
Following is an overview of our tumor model and imaging protocols.
Tumor model and CREKA injection
We used six-week-old female BALB/c mice (Charles River) to create our tumor model.
The animals were kept at the Animal Core Facility of Case Western Reserve University
according to animal protocols approved by the CWRU Institutional Animal Care and Use
Committee. We used the 4T1-GFP-Luc2 (Caliper Life Sciences) cell line to create the
metastatic breast cancer mouse model. We orthotopically injected (1×106 in 40 µL PBS)
cells in the mammary fat pad of mice. Since this cell line expresses the luciferase gene,
we monitored tumor growth by D-luciferin injection (200 mL in PBS, 15 mg/mL) using
the Xenogen IVIS (Caliper Life Sciences, Hopkinton, MA) to perform Bioluminescence
imaging every 1-2 weeks. Patients with metastatic breast cancer usually have their
primary tumors surgically removed. To model this situation, the primary tumors were
surgically excised 30 days post inoculation when metastases have been established and
metastases were allowed to grow for another 10 days before study. Cy5.0-PEG-CREKA was injected (0.3 µmol/kg body weight) into mice bearing 4T1-luc2-GFP. After 3 hours of clearance, we placed the mice in a special mold we created in order to reduce deformation between MRI and cryo-imaging procedures. The mold is simply a cylindrical test tube with a size big enough to fit a mouse while still fitting inside the
MRI cylindrical chamber during imaging. We cut our tube in half longitudinally and placed the mouse in one half. We then placed the mouse and mold inside the MRI
Case Western Reserve University Page 126 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells machine. During the MRI procedure, mice were anesthetized with a 2% isoflurane- oxygen mixture in an isoflurane induction chamber and catheterized in the tail vein with a
30 gauge needle connected with 1.6m tubing filled with heparinized saline. For the MRI scans, we used a Bruker Biospec 7TMRI scanner (Bruker Corp.,Billerica, MA, USA).
During the MRI scan, we captured a preinjection volume, and then we injected (0.1 mmol of Gd/kg) by flushing with 80 µL of saline using the attached catheter to the tail vein.
After that, we acquired a post injection in vivo scan. Following the in vivo scan, we euthanized the mice by injecting 50 μL ketamine-xylazine. Then, a high resolution MRI scan was acquired. After MR imaging, we replaced the second half of the cylindrical test tube mold on the top of the first half, and sealed the mold using tape. We then filled the tube with O.C.T (optimum cutting temperature, Tissue-Tek®) freezing medium, and flash froze it using liquid nitrogen. Prior to cryo-imaging, we detached tour mold from the frozen mouse. Then, we placed the frozen mouse in the cryo-microtome chamber to perform cryo-imaging
Imaging Settings for MRI and Cryo-imaging of Samples
Preinjection and post injection scans were performed using the following sequence: fat suppression three-dimensional (3D) FLASH sequence with respiratory gating [repetition time (TR) = 25 ms, echo time (TE) = 2.8 ms, average = 3, 15° flip angle, in-plane field of view (FOV) = 6 cm, 18-mm slab thickness. Resolution: 0.1172 × 0.09766 × 0.562 mm, scan duration: 10m14s.
Case Western Reserve University Page 127 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Since cryo-images are high resolution, we acquired another high resolution dead
scan to perform image registration. The following sequence was used for this scan: fat
suppression three-dimensional (3D) FLASH sequence [repetition time (TR) = 80 ms,
echo time (TE) = 2.8 ms, average = 3, 15° flip angle, in-plane field of view (FOV) = 6
cm, 18-mm slab thickness. Resolution: 0.1172 × 0.09766 × 0.1406 mm, scan duration:
2h11m
In cryo-imaging, frozen mice were imaged at 10.472 × 10.472 µm in-plane
resolution, and sectioned at 50 µm using CryoViz (Bioinvision Inc, Cleveland Ohio).
Brightfield, GFP, and Cy5 images were acquired using a liquid crystal RGB filter and a monochrome camera. Fluorescence images were acquired using a dual band FITC/Cy5
fluorescence filter (Exciter: 51008x, Dichroic: 51008bs, Emitter: 51008m, Chroma,
Rockingham, VT). Typically, bright field, green, and red fluorescence volumes are about
120 GB for each mouse.
We used the method described in [190] to acquire some histological slides from each mouse during cryo-imaging. Basically, the method placing a special adhesive tape, called cryoFilmTM, on the blockface of the sample being imaged. When the blockface is
sectioned, the tissue section is left attached to the cryoFilmTM . Then the tape with the
tissue, can be stained, and placed on a slide with a coverslip for optical scanning
(Olympus VS120). In this work, we show the results of H&E staining. Using this tape
method, it is easier to take a section of the whole mouse. Moreover, it has the advantage
Case Western Reserve University Page 128 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
of preserving the geometry of the section, so it is easy to register the slice back to the
corresponding cryo slice simply using manual alignment.
Measuring volume change using CT
We designed a separate experiment to estimate the volume change in mice due to freezing. This measurement was done to confirm that there is indeed a volume change after freezing. Additionally, it would indicate whether the logarithm of the Jacobian
penalty is needed in our deformable registration cost function to regularize the volume
change. We created a simple experiment based on CT Hounsfield number change before
and after freezing. In the experiment, we replicated the conditions of MRI-cryo
experiment. We first acquired a whole mouse CT volume after death, then we covered the
animal with O.C.T and flash froze it in liquid nitrogen. We then acquired a second CT
volume. Before the experiment, we scanned a phantom of water to make sure that the
system was calibrated and the values were zero. We compared the Hounsfield numbers in
the two volumes to estimate the volume change in selected organs such as the liver,
kidneys, and brain. Since CT Hounsfield units are defined by:
µ( x, y) − µwater CT( x, y) = 1000 (7) µwater
where
CT(x,y): the Hounsfield unit at location x,y
µ(x,y): the linear attenuation coefficient
µwater: the attenuation coefficient of water Case Western Reserve University Page 129 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
We related the Hounsfield number change to the volume change, by rewriting equation (after dividing µ(x,y) by density ρ to get the mass attenuation coefficient (µ/ρ).
µ( x,y)ρ −µwater ρ CT( x, y) = 1000 (8) µwater
Since (ρ = mass / volume) and the overall mass does not change during freezing, the relative volume change can be found using [191]
Vafter 1000 + CT (before freezing) = (9) Vbefore 1000+CT (after freezing)
Finding optimal parameters for regularization
We designed an experiment to optimize the weights of the penalty/regularization terms (w1 and w2). The experiment was designed to optimize registration quality in the lung region, which is typically the most difficult region to register because of deformation between the MRI and the frozen cryo-volume. We noticed that registration quality in the lungs was more sensitive to the weights than other organs, due to the elasticity of the lung volume. The precise choice of bending energy and Jacobian weights will predict the exact deformation that happened during animal freezing. It is worth noting that registration quality in other organs did not change noticeably when varying
Case Western Reserve University Page 130 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells the weights. Since optimized values for the lungs tend to give good registration quality throughout the mouse, we designed experiments to optimize penalty weights for the lungs.
To assess registration quality, we computed overlap of the segmented lungs. We manually segmented lungs in the cryo reference volume and the MRI volume after rigid and affine registration. Segmentation was done by a person experienced in analyzing both
MRI and cryo-images. Segmentation was done once and used in all experiments. We then performed FFD registration using different combinations of weights, w1 and w2. In each trial, we registered the entire mouse, quantitatively evaluated registration quality in the lungs, and qualitatively assessed registration elsewhere. Pseudo code for the process is shown in Table 2.
Table 2. Pseudo code to find optimal parameters for cost function using lungs Input volumes: MRI Volume (after rigid and affine), cryo Volume Segment Lungs in both the cryo and MRI volumes for BE= 0: 0.005:0.1 for JL= 0:0.005:0.1 Perform whole mouse deformable registration using Ctotal as the cost function Extract the deformation field for the above transform Apply the transform on segmented MRI lungs Save the deformed lungs as a binary volume Evaluate the volume similarity between the deformed MRI lungs and the reference cryo lungs end end
We made multiple quantitative assessments from the segmented lungs. We measured the volume difference, VD [192].
Case Western Reserve University Page 131 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
| Cryo | - | MRI | VD = 2 | Cryo | + | MRI |
Additionally, we used Dice to assess overlap between the two volumes [193].
Cryo∪ MRI Dice = 2 | Cryo | + | MRI |
We also measured the average 3D surface distance between the registered MRI lungs and the reference cryo lungs. The surfaces were converted to a set of points, then the minimum distance from one boundary point to the other boundary was calculated.
The average distance was also calculated from all the points’ distances. The same was done for the other direction. Then the average distance of both was calculated. The distance in one direction is defined as [194]:
1 N SD =∑ min(dist(Cryoboundary(n) , MRI boundary )) N n=1
Evaluation of registration accuracy
We evaluated the quality of registration both visually and quantitatively. Visually, we created a checkerboard view in 3 planes, and we inspected the quality of registration of several features, such as the edges of organs and the bifurcations of major blood vessels.
Case Western Reserve University Page 132 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Quantitatively, besides the measure of lung registration quality parameters, we created a tool that enables the selection of points of interest in GFP volume and MRI volume. The goal is to independently measure the registration accuracy of metastases by using the
GFP rather than the bright field volume which was used in the registration. For smaller metastases, we measured the 3D Euclidian distance between the center of mass of a GFP metastasis and the corresponding metastasis in MRI. The selection of the center points was done manually at locations believed to be the center of each metastasis. For bigger metastases, we employed the shape irregularity in tumors to make accuracy measurements. These shape measurements include tumor edges and tips of protrusions from the surface of tumors. Selection of these points was done manually as well, and 3D distance between the selected points in GFP and MRI tumors was calculated. For each mouse we evaluated several small and big metastases in different organs. Results were reported as mean and standard deviation of the distance.
5.3 Results We optimized imaging parameters for both MRI and cryo imaging, and designed analysis and visualization software to help assess targeting of the CREKA peptide (Figure 2).
Results are shown from three representative mouse experiments.
For rigid and affine registration, we varied parameters consisting of block size, search size, and spacing between cubes [181], and visually inspected results. For resampled image volumes, we found that a block size of 20x20x20 voxels, a search neighborhood of 30x30x30 voxels, a block overlap of 3, and 30 iterations worked very
Case Western Reserve University Page 133 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells well. We used a multiscale version of the algorithm with 3 levels, where at each level, the block is subsampled by 2 for faster computation and more robust estimation of the rigid and affine registration. Fifty percent of cubes with the highest variance were used to estimate the final global transformation using trimmed least square regression. Algorithm performance was more sensitive to the block size than others parameters, and this block size worked very well for all our mice with no need to optimize the parameters for each mouse.
FFD registration, however, was more sensitive to its parameters. Following preliminary experiments, we settled on a grid point spacing of 5x5x5 voxels, 3 levels multiscale approach, and 500 iterations for each level with a tolerance of 0.001 in the cost function. The registration stops when either the number of iterations or the tolerance is reached. To optimize weights for the penalty terms, w1 and w2, we performed 421 image registrations between the cryo brightfield reference volume and the floating MRI volume following affine registration. After each registration, we applied the deformation field to the segmented MRI lungs obtained after affine registration. Registration quality measures are presented in (Figure 26 and 27) for a representative mouse. One can see that the optimal region for the weights (w1 and w2) is is a wide region towards the the center and away from the edges. In (Figure 27) we show the registration quality parameters for the same mouse for the cases where: no penalty terms are used, 3 other cases where registration was not optimal, and 4 cases (bold) where registration results are acceptable.
Other mice showed similar regions with slight differences in the optimal region. Hence,
Case Western Reserve University Page 134 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells we found an optimal set based on the overlapped region from all mice. This set of weights can be used across all mice for the deformable registration. Based on the graphs and tables, we chose a bending energy weight (w1) of 0.045 and Jacobian weight (w2) of
0.055 to be the parameter of registration. After finding optimized parameters for rigid, affine, and deformable registration, we applied the whole registration as one step, where the transformation matrix from the rigid affine step is used to initialize the control points’ positions in the deformable registration, and then we interpolate once at the end to obtain the final registration results.
Figure 26. Registration quality measures for lungs. We varied the weights (w1 and w2) of penalty terms bending energy and logarithm of Jacobin determinant, respectively. The range is between 0-0.1 for each. Left is a mesh plot of the volume difference measurement between segmented cryo lungs and MRI lungs after deformable registration. Middle is the Dice measurement and right is the mean surface distance. w1 is the left axis and w2 is the right.
Case Western Reserve University Page 135 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 27. Registration quality measures for selected (w1 and w2). We show the case for no penalty is used, 3 other cases where registration was not optimal, and 4 cases (bold) where registration results were acceptable.
We examined the volume change due to freezing in the CT experiment, and
compared that to the volume change obtained after deformable registration. We manually
aligned the mouse CT volumes (before and after freezing) to avoid intensity change,
which might affect our volume change calculations. Figure 28 shows a slice through a
mouse before and after freezing. Both of these slices were windowed by a window size of
400 and center of 0. From the figure, one can see that there is indeed an expansion as
reflected visually by physical volume change, but there is also a gray scale value change between the two images, where it is brighter before freezing. The decreased values of
Hounsfield numbers reflect an overall volume expansion in most organs due to freezing.
We selected similar regions of interest in both volumes, and measured the mean intensity
values, then we estimated the volume change using equation (7). Results showed that
Case Western Reserve University Page 136 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells most organs expanded by 2% - 10%. Specific numbers for selected organs are: liver volume increased by (8.3 ± 1.5) %, left kidney by (9.5 ± 2) %, right kidney by (7.5 ± 2)
%, and brain by (4 ± 1.2) %.
Figure 28. Volume change due to freezing using CT. On the left is a slice through a mouse before freezing, on the right is a similar slices after freezing. One can visually see the volume change physically and by the intensity values change which represent Hounsfield numbers. Images shown have the same window width and center.
Case Western Reserve University Page 137 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
On the other hand, we examined the determinant of Jacobian maps for the transform obtained after deformable registration. In Figure 29 we show the XZ and XY views of the results of registration along with the Jacobian determinant map at that specific slice for a representative mouse. The map is windowed between 0.85 and 1.25.
Brighter values reflect compression (Jacobian determinant > 1), and darker values reflect expansion (Jacobian determinant < 1). We examined the value of the determinant at several regions of interest. The values showed that indeed the tissue expanded in most organs, except for the lungs, where the tissue compressed. Values of the determinant range from 0.9 – 0.98 for areas outside the lungs. However, the values average 1.18 within the lung region, indicating shrinkage. These value are consistent with the CT experiment, and show that a volume change was needed to perform the registration.
Hence, the Jacobian penalty is needed to regularize these changes and avoid tissue folding.
Case Western Reserve University Page 138 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 29. Determinant of Jacobian maps. We estimated the local volume change after applying deformable registration by calculating the determinant of the Jacobian of the total transform. We show an example of such maps for a representative mouse in XY view and XZ view. Upper image is the gray scale cryo reference image. Image below is the result of rigid only registration shown for comparison. Image below is the result after affine and deformable registration. Image at the bottom is the determinant of Jacobian map for that specific slice. We selected some region of interest across different organs of the mouse and calculated the mean and standard deviation to find the local volume change across the mouse. In XY view, from left to right, the value of Jacobian determinant in the selected region is, muscles 0.93 +/- 0.03, kidney 0.94 +/- 0.02 , spleen 0.95 +/- 0.02. In the XZ view, the values of the determinant, from left to right, are spinal cord 0.91 +/- 0.03, liver 0.92 +/- 0.01, lungs 1.18 +/- 0.05 and brain 0.97 +/- 0.01.
We examined registration quality both visually and quantitatively. Visually, we examined checkerboard views before and after registration in 3 planes. In Figure 30, we show checkerboard views for the results of registration for a representative mouse. As
Case Western Reserve University Page 139 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells this figure shows, although the parameters were optimized for lungs, other features outside the lungs are registered very well. Arrows point at some features in both volumes to assess registration quality visually. Features include blood vessels and organ edges.
For example, one can see the continuity of lungs, liver, and blood vessels edges when transitioning from cryo to MRI and vice versa.
Case Western Reserve University Page 140 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 30. Visual evaluation of registration quality for whole mouse. Figure shows multi-planer checker board view of registration results. In (A) arrows point at edges of lungs and liver in XY view. In (B) arrows point at edges of lungs, liver and spinal cord in XZ view, and in (C) a blood vessel in liver is shown in YZ view. (D) snap shot from (movie1) to visually assess registration accuracy in 3D. Arrow point at liver edge, one can see in the movie the edge continuity in 3D. Clearly, registration results showed a very good registration in organs of interest.
For quantitative assessment of registration accuracy, the registration quality measures for the lungs that we used to determine the optimized weights for penalty terms, provided assessment of lungs registration accuracy. However, we measured the registration accuracy of several metastases in mice since our goal is to analyze these tumors for targeted imaging agents. We picked 20 tumors from the GFP cryo volumes Case Western Reserve University Page 141 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
from all over the mouse with different sizes ranging from 0.2 - 4 mm in diameter. For
bigger tumors we calculated the 3D Euclidian distance between the edges of the tumors
in both modalities and some features such as the tip of protrusion from tumor surfaces.
For smaller tumors we just calculated the distance between the centers of theses tumors.
Results showed that the average tumor distance was 0.29 mm with a standard deviation
of 0.16 mm. Since our image voxel sizes are 0.1x0.1x0.1, this distance corresponds to a registration average accuracy of about 3 voxels. The average registration accuracy for all
other mice were also in the same range of about 3 voxels.
Following registration, we now show results of tumor analysis and targeted
imaging agent analysis. Expert image analysts manually segmented GFP tumors within
the cryo-imaging data. In some cases, GFP images alone were sufficient for identifying
metastases. In other cases, experts used both brightfield and GFP to segment metastases.
In Figure 31, we show a 3D visualization for metastases in the whole mouse, along with
2D views of fused GFP and brightfield images of selected organs. It is clearly seen from
the figure that this tumor model metastasizes to everywhere in the mouse. Metastases
were found in the lungs, liver, bone marrow and even in the brain. In the figure, we
colored the metastases based on size: yellow for metastases < 0.5 mm in diameter, red for
metastases in the range 0.5 - 2 mm, and green for metastases > 2 mm. we performed
connected component analyses to find the number of metastases in each mouse. Results
show that the average number of metastases in the mice we imaged is 156 ranging in
diameter from 0.1 - 6 mm. For this particular mouse shown in the figure, there are 166
Case Western Reserve University Page 142 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells metastases with 92 metastases < 0.5 mm, 63 metastases in size range of 0.5 - 2 mm, and
11 metastases > 2 mm. An average of about 65 % of the total number of metastases were found in the lungs.
Figure 31. Metastases distribution in mouse. Metastases were segmented manually using GFP and color brightfield volume. Segmentation results were visualized using volume rendering with different colors. Yellow for metastases diameter <0.5 mm, red for metastases in the range 0.5-0.2 mm an green for metastases > 2mm. in the mouse shown, there are 166 metastases with 92 in yellow, 63 in red and 11 in green. Fused GFP and bright field show anatomical locations of metastases. In the figure we show examples of metastases in (a) lungs, (b) vertebra, (c) liver and (d) brain.
Following tumor segmentation, we visually examined peptide targeting within the
4 registered volumes: brightfield, tumor GFP, CREKA-Cy5, and CREKA-Gd. For each metastasis that we detected in the brightfield and GFP volumes, we examined the
Case Western Reserve University Page 143 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
corresponding location in Cy5 and MRI to assess signal intensity and labeling potential.
In Figure 32, we show an example of a slice through a mouse in all 4 registered volumes.
Arrows in the figure point at 3 big metastases that were labeled by CREKA-Cy5 and
CREKA-Gd. As shown in the figure, we noticed that the peptide tends to label the edges of bigger metastases, and covers entire smaller metastases. Tumor diameters in the figure shown are: 7.5 mm, 6.3 mm and 2.5 mm, respectively.
Figure 32. Multi-modality view of a slice through mouse after registration. Color anatomy brightfield, green GFP tumor, red CREKA-Cy5 peptide and CREKA-Gd MRI are shown next to each other. One can look at bright field and GFP images to locate metastases across slice and evaluate the signal of imaging agent in Cy5 and MRI images. Arrows point at some 3 big metastases. We masked the area around outside theses tumor in GFP and Cy5. Peptide tends to label the edges of bigger metastases and label the whole area of smaller ones. The average signal to background at the edges of tumor on the left is 1.7, 2.1 for
Case Western Reserve University Page 144 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
MRI and 1.9, 2.3 for Cy5, respectively. Average signal to background on the tumor on the right is 3.6 for Cy5 3.4 for MRI. Tumor diameters are 7.5 mm, 6.3 mm and 2.5 mm, respectively. Given the four registered volumes, we inspected peptide targeting to micro metastases (0.2 – 2 mm). In Figure 33, we show multiple examples of micro metastases at different locations in a mouse and the corresponding CREKA-Cy5 and CREKA-Gd images. Examples of micrometastases were found in bone marrow, vertebra, adrenal gland and lungs. In Figure 33a, a micro metastasis in the bone marrow of the upper shoulder is shown. Its diameter is 0.3 mm. Figure 33,b shows a micrometastasis in the adrenal gland with a diameter of 0.1 mm. The tumor cells formed a distinctive ring inside the organ. Many micrometastases were also detected by both CREKA-Gd and CREKA-
Cy5 in the lungs. Figure 5c shows an example of one such micrometastasis, with a diameter of 2 mm. In general, CREKA-Cy5 detected more metastases than CREKA-Gd.
Case Western Reserve University Page 145 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 33. Multi-modality evaluation of micro-metastases targeting. Examples of micrometasteses were found in bone marrow (A), adrenal glands (B) and lungs (C). The diameter of these metastases are 0.3, 0.1 and 2 mm respectively. Bar is 1mm.
To help provide a quantitative assessment of the number of metastases that were targeted by CREKA-Gd, we employed the Rose criterion of detection to estimate the number of metastases that have a sufficient MRI signal above background and can be distinguished by an observer. The Rose threshold for detection should be calculated for each organ individually because the background and noise levels are different for each tissue. Since most of the metastases are in the lungs, we show the results of such an analysis in the lungs of a representative mouse. The red curve in Figure 34 represents the
Case Western Reserve University Page 146 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
minimum Rose CNR required for detecting a specific metastasis calculated based on
lungs background and noise levels. For each GFP metastasis in the lungs, we calculated
the Rose CNR at each slice through that metastasis and found the maximum value. Then
we plotted the maximum area of that metastasis versus the calculated Rose CNR. Since
all larger (> 2 mm) metastases were easily distinguished from background, we only show
the result of metastases with diameters 0.2 - 2 mm (area 0.04 – 4 mm2). From the graph,
all metastases that are above 0.3 mm2 where detectable, and about 36 % of smaller
micrometastases (< 0.3 mm2) were detectable. In the figure, we show two examples of the detected metastases marked in red on the graph. We also show an example where the
CREKA-Gd signal was not distinguishable from the background (Figure 34c). Other mice showed a similar detectability rate for metastases in the lungs. The same analysis can be performed for CREKA-Cy5. Results showed a higher detectability rate for
CREKA-Cy5 with an average of around 90 %.
Case Western Reserve University Page 147 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 34. Rose Criterion of detection for lungs MRI metastases. Scatter plot of metastases in lungs after applying Rose CNR for each metastasis. Results shown are for one mouse, similar results obtained from other mice. Red curve in graph represent Rose threshold of detection calculated based on lungs tissue background and noise level in MRI images with a threshold of 4. All metastases above the curve can be detected by a human observe based on this criterion. All micrometastases above 0.3 mm2 where detectable (metastases in the green area). Detectability of micrometastases that are less than 0.3 mm2 was around 36% (metastases in the gray area).We selected some metastases in the graph and inspected the results (red colored points A-C). In (A) and (B), tumor metastases of area 0.25 and 1 mm2 were detected by CREKA-Gd. In (C) a metastases area of 0.2 mm2 was not detected (Arrows). Bar is 1 mm.
Case Western Reserve University Page 148 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
By adding histology to our analysis platform, we can provide additional information about the tumor model and targeting of the probe at molecular level. We acquired some histology sections following cryo-imaging of the block face. Since we use the tape acquisition method to acquire the sections, the morphology of the sections is preserved, and they correspond exactly to the acquired cryo image from the block face.
We simply performed manual rigid body alignment to register the two images.
Information obtained from histology enabled further understanding of the targeting of the peptide at the molecular level and distribution of the label. Also, histology can provide information about tumor cell differentiation, inflammation, and necrosis. In Figure 35, we show an example of an acquired H&E section. In the tumor shown, the targeting of the peptide was inhomogeneous in both Cy5 and MRI, as evidenced by the higher intensity at the edge of the tumor. By adding histology, one can see that the high intensity labels were closer to the blood vessel, as seen in the cryo brightfield and confirmed by histology.
Case Western Reserve University Page 149 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 35. Histology and cryo imaging example. We acquired some histology slices and stained with H&E to further study tumors and labeling at molecular level. Here we show an example of a tumor were the targeted image agent signal was inhomogeneous. In the figure the location of tumor is shown using bright field (A). Images (B-E) show the tumor in GFP, Cy5, MRI and H&E, respectively. Image in (F) is zoomed view for the left selected area in (E) where a high signal for the peptide in Cy5 and MRI is observed. One can see in F that indeed there are tumor cells and they are close to blood vessel. (G) is zoomed view for the right selected area in (E) where signal intensity of the peptide was low, this area shows less tumor cells and blood vessels explaining the low intensity of the peptide.
Case Western Reserve University Page 150 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
5.4 Discussion The need to target tumor cells at the molecular level led the development of molecular
targeted agents. Validation of these developed agents is required in order to assess their
efficiency in targeting all tumor cells and at a whole body level. Evaluation of these
agents in the preclinical stage requires a gold standard. Hence, we demonstrated a platform methodology, consisting of imaging protocols and image analysis algorithms, for the assessment of emerging targeted imaging agents and theranostics. The key for this platform is cryo-imaging (CryoVizTM), which provides microscopic color anatomy and
molecular fluorescence images, with single tumor cell sensitivity of an entire mouse. In
this report, we evaluated the ability of (CREKA-Tris-(Gd-DOTA)3), a new MR imaging
agent, and its red fluorescent counterpart, Cy5.0-PEG-CREKA, to detect even the
smallest of tumors in the whole mouse, metastatic breast cancer model. The ultimate goal
is that these targeted agents, once proven efficient, can be used not only for noninvasive
diagnostics but also as tool to aid in surgical resection [158]. Later they can be attached to nanoparticles to create theranostic vehicles for both diagnosis and therapy delivery.
Here we discuss some points regarding our experimental design and analysis algorithms.
Since high resolution MRI scans lasted for 2 hours after the animal death, we were unsure whether integrity of the GFP signal would be compromised. We performed an experiment to check the intensity change in GFP signal using the Maestro™
multispectral fluorescence system. We measured the intensity of GFP of some tumors
before and 2 hours after animal death. Within this time window, we did not notice any
significant change in intensity values which made us proceed with capturing high Case Western Reserve University Page 151 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells resolution MRI for registration purposes. In fact, after 2 hours of death and before freezing, rigor mortis occurred, which stiffened the animal and made organ sliding minimal. We confirmed this by simply pressing against the mouse and found it became stiffer after the MRI high resolution scan. Increased stiffness made it easier to perform registration without the need for organ sliding algorithms or other special registrations that require manual segmentation and might result in edges artifacts.
In the registration algorithms, we compared the results of the block matching algorithm with other mutual information based rigid affine algorithms. For our volumes, the results were better than other algorithms. Although this method uses normalized cross correlation as a similarity measure, it performed very well if a small block is used for matching. We were interested in finding a good rigid and affine algorithm since, from our experience, this will lead to better results during the deformable registration step. We compared different block sizes until we found that 20 is the best for our images, as smaller cubes were more susceptible to noise, and bigger cubes resulted in a less accurate estimation of the transformation matrix.
During the FFD registration, we focused on lungs registration to find optimized weights for the regularization terms. We focused on this area since this mouse model of breast cancer is highly metastatic to lungs, and most of metastases (an average of 65%) appeared in lungs. Second, the lungs are very elastic and deform the most among other organs, as we showed using the determinant Jacobian maps, where lungs shrank by an average of 18% after death. Finally, registration accuracy of all other organs were less
Case Western Reserve University Page 152 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells sensitive to the combination of weights. We found that as long as some set of parameter worked for lungs, it worked for other organs.
We performed 421 registrations with different combinations of penalty terms.
Registration was performed on the whole mouse rather than a cropped region for the following reasons. First, this ensured that the combination of weights is suitable for the whole mouse not only the lungs. We visually inspected the registration accuracy for each candidate combination of weights parameters, and we made sure no tissue folding or other artifacts happen at other locations in the mouse for that specific combination.
Second, our goal is to register the whole mouse, not only the lungs, since tumor metastases also occur outside the lungs. Third, using piecewise registration to obtain the whole mouse registration can lead to artifacts, especially at the edges of the cropped regions.
The overall registration accuracy of the whole mouse was very good, especially in the lungs and liver, where most of the metastases are present. Registration accuracy in the stomach, spleen and parts of kidneys was less optimal. It seems that if we add a sliding organ algorithm, or if even we just cropped these regions and registered just these organs, this might improve results. However in these regions there are very few metastases in this tumor model. Hence we did not deem it necessary to optimize parameters or try different algorithms to improve the registration in these regions.
We resampled cryo images to 0.1 x 0.1 x 0.1 mm, a resolution closer to the acquired high resolution MRI images, to make voxels isotropic, avoid interpolation
Case Western Reserve University Page 153 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
artifact after each iteration of registration, reduce computation time, and reduce memory
usage. Tumor registration accuracy of 2 - 3 voxels or (0.2-0.3 mm) was good enough to
remove ambiguity about mapping the imaging agent to GFP tumors. However, going
back to the native resolution of the acquired cryo data (0.0104 x 0.0104 x 0.05 mm) after registration is possible, especially, when visually assessing micrometastases targeting. In fact, in generating some of the figures above, and after we performed the registration, we used images from the high resolution cryo volume in order to visually see even single metastatic cells. However, this step is not necessary for the quantitative measures.
Results of CT experiment showed that there is indeed a volume change in organs, as organs tend to expand after freezing. We showed specific results for the kidneys, liver, and brain. We did not measure the volume change in lungs based on this experiment, since air was still in the lungs when we acquired the first volume before freezing, and this can affect the volume change measurements as air goes out after death and freezing.
However, CT predicted an overall volume expansion, hence we included the logarithm of the determinant of the Jacobian to regularize the volume change and avoid tissue folding.
Results of the determinant of the Jacobian of the final transform showed that indeed most
organs expanded as indicated by a determinant value less than one, while lungs shrank, as
indicated by a determinant greater than 1. This is not our usual approach in freezing the
mice, as we used a plastic mold in this experiment to limit mouse deformation. This
added layer slowed the freezing process, which resulted in a volume change. Faster
freezing might result in less volume change, and if we were to perfuse the animal by a
Case Western Reserve University Page 154 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
cryo protectant, this might result in less volume change. However, it may wash out the
probe.
Although the parameters of rigid, affine, and deformable registration that resulted
from cryo brightfield and MRI high resolution registration can be applied to the post
injection in vivo MRI volume. This can be done by first registering the in vivo and the
high resolution MRI scans, and then applying the registration parameters from cryo and
high resolution MRI scan to the in vivo registration volume. We found that this step is not necessary for analyzing imaging agent targeting. As we show in (Figure 36), even after death, CREKA-Gd signal is still evident and tumors can be still distinguished from background.
Case Western Reserve University Page 155 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 36. Metastases targeting by CREKA-Gd over multiple scans. (A) The preinjection image where no tumors can be distinguished from background. (B) CREKA-Gd post injection in vivo image. In (C) is the high resolution scan obtained over 2 hours after animal death. One can still distinguish tumor from background as arrows show.
During the manual segmentation of metastases, we examined the cryo images and found that bigger metastases can also be distinguished in the brighfield images, in addition to the fluorescent images, since they appear as distinct white spots compared to the background tissue. Hence, brighfield was used to segment big metastases especially if
Case Western Reserve University Page 156 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
they were necrotic and lack GFP signal. For smaller metastases we segmented using GFP
images, however brightfield was used to give us anatomical details for better orientation.
Micrometastases detectability depends on multiple factors, including the number of targeted molecules needed to generate sufficient contrast, and the sensitivity of the imaging modalities. In our case, since CREKA targets fibrin-fibronectin complexes in blood clots in the tumor microenvironment, detectability of the tumor depends on the number of theses complexes that exist in the microenvironment. We noticed that some
GFP micro metastases were not detectable in both Cy5 and MRI. One reason for that is when we further analyze these metastases, we find they did not have enough blood vessels around them to form blood clots and deposit enough fibrin-fibronectin complexes.
We also noticed that the number of micrometastases detected by CREKA-Cy5 is higher than the number detected by CREKA-Gd. One reason for that might be the number of molecules required to generate sufficient contrast was not reached for MRI, and since fluorescence imaging is more sensitive than MRI, the number of imaging agent molecules required to generate a sufficient contrast is lower. Another reason might be the size of molecules. CREKA-Gd is attached to 3 gadolinium molecules which make its size bigger and result in less diffusion in tissues than its fluorescence counterpart.
By adding histological analysis to our platform, we can further analyze the tumor model by studying labeling inhomogeneity and further understanding the tumor
microenvironment. We showed an example of H&E staining to further explain the
imaging agent targeting pattern. Immunohistochemical staining can be performed to
Case Western Reserve University Page 157 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells further study the microenvironment. For example, sections can be stained by antibodies for fibrin-fibronectin complexes to further assess the correlation of the targeted molecule and the imaging agent.
The tape transfer method preserves the morphology of the sections acquired. We typically capture histology sections at 10 µm in thickness between two successive cryo- imaging sections. The registration accuracy depends on the section thickness used in the cryo-imaging experiment. During our experiments, mice were sectioned at 50 µm to acquire cryo volumes. This results in a registration accuracy of ≥25 µm between the cryo section and the histology section. Since registration accuracy between MRI and cryo volumes is within 0.2 - 0.3 mm, by adding histology to MRI, this results in registration accuracy within 0.23 - 0.33 mm.
Finally, Rose analysis was done in an automated fashion. This tool is very useful to give an idea about the targeting efficiency of the MRI probe. Since registration is accurate within 3 voxels (0.3 mm), manual mapping of very small micrometastases might improve the results of detectability. However, for our analysis we focus on showing results of metastases bigger than 0.2 mm. Also, this analysis does not consider false positive signals in MRI images and masking these results might result in even more accurate results. Extensions are foreseen for a more complicated human observer based model.
Results of the developed platform enables feedback necessary to improve MRI probe efficacy. For example, since CREKA-Cy5 detected more micrometastases than the
Case Western Reserve University Page 158 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
MRI counterpart, this suggests a need to either: modify the acquisition sequence for MRI to improve sensitivity, or increase the number of chelates (in this case gadolinium) attached to each peptide, to improve detectability rates in the MRI. The platform will be extended to compare different MRI targeted peptides and theronastics to further compare targeting efficiency.
Conclusion
We have shown very promising results for a unique platform methodology which can uniquely characterize targeting of imaging agents to metastatic tumors that can be of great use to many cancer researchers. The goal of this paper is to show analysis and imaging protocols for this platform. This platform should be readily applicable to different tumor types, other diseases, peptides, nanoparticle designs, etc. Extensions are foreseen, such as in examining the efficacy (e.g., silencing of siRNA) of theranostic nanoparticles.
Case Western Reserve University Page 159 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
6 Chapter 6: Discussion and Future work
The outcome of this dissertation provides a unique preclinical platform methodology of great use to many cancer researchers. The main goal was to develop algorithms and imaging protocols for cryo volumes to further evaluate the targeting of imaging agents to tumor cells. With the unique capability of cryo imaging in resolving single fluorescently labeled tumors cell, this dissertation provided a comprehensive analysis methodology for evaluating imaging agent targeting to dispersing cells and micrometastases. Moreover, by employing 3D multimodality registration, targeting of MRI imaging agents to micrometastases was validated, a step further toward translation to the clinic.
In Chapter 2-3, I presented a unique methodology to analyze cell dispersal in
GBM mouse tumor models. This unique analysis revealed for the first time the pattern of
GBM cell dispersal in the brain along major dispersal pathways by imaging the whole brain at single cell resolution. Multiple GBM cell lines were assessed to determine those showing dispersal characteristics similar to human disease, as cell dispersal characterization is an important step to understand the human disease, and both an accurate tumor model and sensitive imaging modality can lead to better understanding of targeting of the imaging agent.
This analysis was extended, in Chapter 4, to evaluate targeting of fluorescently labeled SBK2 peptide visually and quantitatively at single cell resolution and at whole brain scale. Imaging agent with and without targeting peptides was employed to assess enhanced permeability and retention (EPR) versus targeting. Results of targeting were
Case Western Reserve University Page 160 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
very promising as the fluorescence peptide showed labeling capability of more than 99%
of dispersing cells. Once toxicology studies are performed, this peptide can be used a
surgical tool for tumor resection. Moreover, currently the peptide is being tested for MRI
imaging as a molecular contrast agent [82] to assess its feasibility toward clinical
applications.
Extensions of the analyses I presented in these chapters are possible. For example,
blood vessel detection can be extended to detected even smaller blood vessels. The goal
was to create a quick visualization tool for blood vessels detection to study dispersal.
This tool depends on edge detection filters and the resolution of the captured bright field
images. Hence, the diameter of blood vessels that were reliably detected were in the
range of 30 µm which was satisfactory for our analyses. However, if one is interested in
detecting even smaller vessels, higher resolution bright field is required. Also, since
filtering depends on the contrast of blood vessels, one can fluorescently label the blood vessel which will improve the contrast and enhance detection. Labeling can be done either by cardiac perfusion by injecting dyes such as lipophilic carbocyanine dye (Dil)
[195], or by using vascular specific transgenic animal models that express GFP [196], for example. Moreover, more sophisticated algorithms that measure the vesselness can be employed to improve that detection of very small blood vessels such as the ones that compute the Hessian matrix and extract Eigen values of the Hessian matrix to obtain a vessel-feature filter feature filter [197-199].
Case Western Reserve University Page 161 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Although the primary goal was to detect dispersing cells and assess targeting, one
can take advantage of the unique microscopic resolution of cryo-imaging to further study
GBM mouse models and extend the analyses to include even more cell lines. For
examples, mathematical models about main tumor mass growth and size change can help
understand the disease further and predict its growth over time. Also, mathematical
models about dispersal formation over time can be helpful in predicting the time frame of
GBM invasion. All these analyses can be done by imaging different brains at several time
points after tumor implantation.
In Chapter 5, we further extended the analyses from one organ to study metastases
of different sizes, in a whole mouse model of metastatic breast cancer. Moreover,
registration between MRI and cryo was implemented to assess targeting of an MRI imaging agent at molecular level. The cryo volumes provide the microscopic resolution needed to identify tumors in GFP, and by mapping to MRI, evaluation of the MRI imaging agent was performed along with the targeted peptide fluorescence counterpart.
Also, by including histology, evaluations at molecular level were also possible. Software that helps in the visual and quantification assessment of targeting was also created. This is very significant as we provide a unique capability identify whole mouse metastases at cellular level resolution from cryo volume and evaluate targeting to all metastases of the
MRI imaging agent by registration. Registration accuracy in the regions of interest, where most of the metastases are, such as lungs and liver, was very adequate to analyze micrometastases in these regions. However, in other regions, such as spleen and around
Case Western Reserve University Page 162 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells the stomach, the registration can be further improved by incorporating sliding organs algorithms [200-204], or even just registering a cropped region around these organs if micro metastases in these regions were of interest in future mouse tumor models.
Image registration accuracy depends on obtaining good quality MRI volumes with good resolution and contrast in order to achieve an accurate registration with the high resolution cryo volumes. Hence, we acquired a higher resolution MRI volume to be used for registration purpose. However, for conventional high field MRI, acquiring high resolution volumes and optimizing the sequence parameters might take long time to achieve a compromise between resolution and contrast. Alternatively, extension of this work might include better techniques to acquire the MRI volumes such as integrating
FISP (Fast Imaging with Steady-state Precession) sequence which improves speed, contrast and SNR [205]. Since fluorescent targeted peptide detected more micrometastases than the MRI peptide counterpart, this means MRI enhancing techniques can result in a better micrometastases detection rate. Hence, employing the newly developed MR finger printing technique, which has the advantage of recording both T1 and T2 changes due to Fe or Gd, and it accounts for motion, partial volume, and B0/B1 inhomogeneity, can result in better sensitivity/specificity to improve the detection rate in
MRI.
Currently, we used our experience in segmenting whole mouse GFP metastases and avoiding autofluorescence structures. There are different techniques that can be implemented to reduce autofluorescence in the cryo imaging system for more efficient
Case Western Reserve University Page 163 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
and faster segmentation results. These techniques include using excitation filtering where
one can use two excitation filters, one in the range of the GFP wavelength, and another
filter in the range below the GFP to capture two fluorescence image. Then by performing
image subtraction the final image will have a significant reduction in the auto fluorescence. This technique is currently under implementation by my colleagues in Dr.
Wilson’s laboratory. Also, by using enhanced GFP tumor cells that express higher
amount of the protein can significantly reduce the exposure time resulting in reduced auto
fluorescence. Additionally, multispectral camera can be incorporated in the cryo-imaging machine, which uses deconvolution to remove the autofluorescence.
Whole body metastases were segmented manually. Extension to automate this process may include a machine learning approach to aid in tumor detection. For example,
GFP tumor features (such as color, texture, values of linear filters, and values of morphological filtering, etc.) can be used to build a classifier to perform automated tumor segmentation in GFP. Moreover, by mapping GFP tumor to MRI locations, one can use the features in these MRI locations to train a classifier for robust automated detection of
MRI metastases.
Future extension to include counting of the number of peptide molecules at each
micrometastasis is also possible. By integrating structured illumination microscopy (SIM)
into the cryo-imaging system, this will enable acquiring the fluorescence signal from just
a thin optical section removing all subsurface fluorescence that might affect
quantification. Then one can create calibration curves between intensity and a calibration
Case Western Reserve University Page 164 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
sample of known number of molecules. By creating calibration curves in units of
molecules/voxel, this will enables estimating the number of peptide molecules at each
metastasis. The results can be shown in histograms relating the number of probe
molecules to the size of metastases. This can help in further understanding the labeling, by studying how close the metastasis is to blood vessels and in which organs metastases
accumulate more targeted peptide. Moreover, by adding histology and
immunohistochemical labeling, one can compare the number of the imaging agent molecules to the number of the targeted molecules within the tumor microenvironment, and provides correlation studies to explain the labeling pattern of the imaging agents.
Building on the Rose criterion, one can employ more sophisticated human perception models, such as the Channelized Hotelling Observer [206-210], to better assess the detectability of a targeted imaging agent. These advanced models are more similar to human observers as they account for more spatial features than the Rose model.
They also account for statistical properties of both the signal and background noise.
These observers are under development by my colleagues in the laboratory. The use of an improved model observer will enable robust, clinically translatable detectability evaluations for the targeted imaging agent development process.
Registration of cryo data is also possible with other modalities such as PET. Since most PET scanners acquires CT volumes concurrently, one can use the structural information from CT for image registration, then apply registration parameters on the
PET. Such extension will enable further studying of imaging agents designed for PET
Case Western Reserve University Page 165 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
imaging, and evaluate their targeting efficiency at a molecular level.
Regardless of the tumor model used here in building the imaging protocols and
analysis algorithms, this platform should be readily applicable to different tumor types,
other diseases, theranostic nanoparticles, etc. Extensions of this project are foreseen and will be carried on by other colleagues in my laboratory. Especially extension of the algorithms and protocols to include PET image registration, incorporating more efficient
MRI sequences etc. Moreover, extensions will include the study of other tumor models and targeted peptides, studying tumor stem cells, and comparison between different targeted peptides and theranostics.
Case Western Reserve University Page 166 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
7 Appendix
MRI-cryo registration
In order to keep the mouse in the same posture and minimize deformation and mouse movement during MRI imaging and freezing, which will make image registration easier to do. We placed the mice in a cylindrical mold, figure below. The mold was created by cutting an appropriate size test tube in half longitudinally. Then, we placed the mouse in one half, and the mouse with the mold was placed in the MRI machine for imaging. All injection was done using a catheter attached to the tale vein with no need for mouse movement while injection. After MRI imaging, we replaced the second half of the tube and sealed with tape. Then we filled the tube with O.C.T (optimum cutting temperature,
Tissue-Tek®) and flash froze using liquid nitrogen. After freezing, we detached the tube placed the mice in cryo microtome for cryo-imaging.
Case Western Reserve University Page 167 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 37. The mold that was used in cryo-MRI registration to reduce mouse deformation during MRI and freezing. (A) shows the plastic tube used as a mold. (B) shows a mouse placed in one half for MRI imaging. After we performed MRI imaging we placed the second half (C) and filled with O.C.T then froze. Result of after freezing is shown in (D).
Tumor progression over time
We used 4T1 cell line to create a breast cancer metastatic tumor mouse model. The cells express luciferase and GFP. We used BLI to monitor tumor growth over time and confirm metastases formation as shown in the figure below.
Case Western Reserve University Page 168 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
Figure 38. Whole tumor growth and metastases formation over time using BLI. Images shown are for a representative mouse bearing 4T1- luc2-GFP tumor cells.
Case Western Reserve University Page 169 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
List of References
1. Ntouroupi, T.G., et al., Detection of circulating tumour cells in peripheral blood with an automated scanning fluorescence microscope. Br J Cancer, 2008. 99(5): p. 789-95.
2. Joyce, J.A. and J.W. Pollard, Microenvironmental regulation of metastasis. Nat Rev Cancer, 2009. 9(4): p. 239-52.
3. Chambers, A.F., A.C. Groom, and I.C. MacDonald, Dissemination and growth of cancer cells in metastatic sites. Nat Rev Cancer, 2002. 2(8): p. 563-72.
4. Nakada, M., et al., Molecular targets of glioma invasion. Cell Mol Life Sci, 2007. 64(4): p. 458-78.
5. Teodorczyk, M. and A. Martin-Villalba, Sensing invasion: cell surface receptors driving spreading of glioblastoma. J Cell Physiol, 2010. 222(1): p. 1-10.
6. Burden-Gulley, S.M., et al., Novel cryo-imaging of the glioma tumor microenvironment reveals migration and dispersal pathways in vivid three- dimensional detail. Cancer Res, 2011. 71(17): p. 5932-40.
7. Lengyel, E., Ovarian cancer development and metastasis. Am J Pathol, 2010. 177(3): p. 1053-64.
8. Paget, S., The distribution of secondary growths in cancer of the breast. Lancet 1889. 1: p. 571–573
9. Fidler, I.J., The pathogenesis of cancer metastasis: the 'seed and soil' hypothesis revisited. Nat Rev Cancer, 2003. 3(6): p. 453-8.
10. Nguyen, D.X., P.D. Bos, and J. Massague, Metastasis: from dissemination to organ-specific colonization. Nat Rev Cancer, 2009. 9(4): p. 274-84.
11. Wicha, M.S., Cancer stem cells and metastasis: lethal seeds. Clin Cancer Res, 2006. 12(19): p. 5606-7.
12. Beck, B. and C. Blanpain, Unravelling cancer stem cell potential. Nat Rev Cancer, 2013. 13(10): p. 727-38.
13. Kleffel, S. and T. Schatton, Tumor dormancy and cancer stem cells: two sides of the same coin? Adv Exp Med Biol, 2013. 734: p. 145-79.
Case Western Reserve University Page 170 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
14. Martin-Padura, I., et al., Residual dormant cancer stem-cell foci are responsible for tumor relapse after antiangiogenic metronomic therapy in hepatocellular carcinoma xenografts. Lab Invest, 2012. 92(7): p. 952-66.
15. Chen, K., Y.H. Huang, and J.L. Chen, Understanding and targeting cancer stem cells: therapeutic implications and challenges. Acta Pharmacol Sin, 2013. 34(6): p. 732-40.
16. Houthuijzen, J.M., et al., The role of mesenchymal stem cells in anti-cancer drug resistance and tumour progression. Br J Cancer, 2012. 106(12): p. 1901-6.
17. Dean, M., T. Fojo, and S. Bates, Tumour stem cells and drug resistance. Nat Rev Cancer, 2005. 5(4): p. 275-84.
18. Eyler, C.E. and J.N. Rich, Survival of the fittest: cancer stem cells in therapeutic resistance and angiogenesis. J Clin Oncol, 2008. 26(17): p. 2839-45.
19. Bao, S., et al., Glioma stem cells promote radioresistance by preferential activation of the DNA damage response. Nature, 2006. 444(7120): p. 756-60.
20. Fillmore, C.M. and C. Kuperwasser, Human breast cancer cell lines contain stem- like cells that self-renew, give rise to phenotypically diverse progeny and survive chemotherapy. Breast Cancer Res, 2008. 10(2): p. R25.
21. Baumann, M., M. Krause, and R. Hill, Exploring the role of cancer stem cells in radioresistance. Nat Rev Cancer, 2008. 8(7): p. 545-54.
22. Rich, J.N., Cancer stem cells in radiation resistance. Cancer Res, 2007. 67(19): p. 8980-4.
23. Pantel, K., C. Alix-Panabieres, and S. Riethdorf, Cancer micrometastases. Nat Rev Clin Oncol, 2009. 6(6): p. 339-51.
24. Weber, G.F., Why does cancer therapy lack effective anti-metastasis drugs? Cancer Lett, 2013. 328(2): p. 207-11.
25. Iacob, G. and E.B. Dinca, Current data and strategy in glioblastoma multiforme. J Med Life, 2009. 2(4): p. 386-93.
26. Kitange, G.J., K.L. Templeton, and R.B. Jenkins, Recent advances in the molecular genetics of primary gliomas. Curr Opin Oncol, 2003. 15(3): p. 197- 203.
Case Western Reserve University Page 171 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
27. Furnari, F.B., et al., Malignant astrocytic glioma: genetics, biology, and paths to treatment. Genes Dev, 2007. 21(21): p. 2683-710.
28. Ichimura, K., et al., Molecular pathogenesis of astrocytic tumours. J Neurooncol, 2004. 70(2): p. 137-60.
29. Qutaish, M.Q., et al., cryo-image analysis of tumor cell migration, invasion, and dispersal in a mouse xenograft model of human glioblastoma multiforme. Mol Imaging Biol, 2012. 14(5): p. 572-83.
30. Louis, D.N., et al., The 2007 WHO classification of tumours of the central nervous system. Acta Neuropathol, 2007. 114(2): p. 97-109.
31. Louis, D.N., Molecular pathology of malignant gliomas. Annu Rev Pathol, 2006. 1: p. 97-117.
32. Candolfi, M., et al., Intracranial glioblastoma models in preclinical neuro- oncology: neuropathological characterization and tumor progression. J Neurooncol, 2007. 85(2): p. 133-48.
33. Siegel, R., et al., Cancer statistics, 2014. CA Cancer J Clin, 2014. 64(1): p. 9-29.
34. Weigelt, B., J.L. Peterse, and L.J. van 't Veer, Breast cancer metastasis: markers and models. Nat Rev Cancer, 2005. 5(8): p. 591-602.
35. Siegel, R., D. Naishadham, and A. Jemal, Cancer statistics for Hispanics/Latinos, 2012. CA Cancer J Clin, 2012. 62(5): p. 283-98.
36. Siegel, R., D. Naishadham, and A. Jemal, Cancer statistics, 2012. CA Cancer J Clin, 2012. 62(1): p. 10-29.
37. Psaila, B., et al., Priming the 'soil' for breast cancer metastasis: the pre- metastatic niche. Breast Dis, 2006. 26: p. 65-74.
38. Obedian, E., D.B. Fischer, and B.G. Haffty, Second malignancies after treatment of early-stage breast cancer: lumpectomy and radiation therapy versus mastectomy. J Clin Oncol, 2000. 18(12): p. 2406-12.
39. Jatoi, I., et al., Significance of axillary lymph node metastasis in primary breast cancer. J Clin Oncol, 1999. 17(8): p. 2334-40.
40. Coleman, R.E., Skeletal complications of malignancy. Cancer, 1997. 80(8 Suppl): p. 1588-94.
Case Western Reserve University Page 172 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
41. Kozlow, W. and T.A. Guise, Breast cancer metastasis to bone: mechanisms of osteolysis and implications for therapy. J Mammary Gland Biol Neoplasia, 2005. 10(2): p. 169-80.
42. Selzner, M., et al., Liver metastases from breast cancer: long-term survival after curative resection. Surgery, 2000. 127(4): p. 383-9.
43. AJCC Cancer Staging Manual 2002(6th Edition).
44. Nieuwenhuis, E.J., et al., Assessment and clinical significance of micrometastases in lymph nodes of head and neck cancer patients detected by E48 (Ly-6D) quantitative reverse transcription-polymerase chain reaction. Lab Invest, 2003. 83(8): p. 1233-40.
45. Spychała, A., D. Murawa, and K. Korski, The clinical importance of micrometastases within the lymphatic system in patients after total gastrectomy. Reports of Practical Oncology & Radiotherapy, 2011. 16(6): p. 232-236.
46. Cox, C.E., et al., Significance of Sentinel Lymph Node Micrometastases in Human Breast Cancer. Journal of the American College of Surgeons. 206(2): p. 261-268.
47. Braun, S., et al., A pooled analysis of bone marrow micrometastasis in breast cancer. N Engl J Med, 2005. 353(8): p. 793-802.
48. Rampaul, R.S., et al., Pathological validation and significance of micrometastasis in sentinel nodes in primary breast cancer. Breast Cancer Res, 2001. 3(2): p. 113- 6.
49. Fass, L., Imaging and cancer: a review. Mol Oncol, 2008. 2(2): p. 115-52.
50. Mueller-Lisse, U.G., et al., Functional imaging in penile cancer: PET/computed tomography, MRI, and sentinel lymph node biopsy. Curr Opin Urol, 2008. 18(1): p. 105-10.
51. Robinson, P.J., Imaging liver metastases: current limitations and future prospects. Br J Radiol, 2000. 73(867): p. 234-41.
52. Williams, M.B., et al., Future directions in imaging of breast diseases. Radiology, 1998. 206(2): p. 297-300.
53. Soret, M., S.L. Bacharach, and I. Buvat, Partial-volume effect in PET tumor imaging. J Nucl Med, 2007. 48(6): p. 932-45.
Case Western Reserve University Page 173 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
54. Cooper, K.L., et al., Positron emission tomography (PET) and magnetic resonance imaging (MRI) for the assessment of axillary lymph node metastases in early breast cancer: systematic review and economic evaluation. Health Technol Assess, 2011. 15(4): p. iii-iv, 1-134.
55. Kuhl, C.K., et al., Mammography, Breast Ultrasound, and Magnetic Resonance Imaging for Surveillance of Women at High Familial Risk for Breast Cancer. Journal of Clinical Oncology, 2005. 23(33): p. 8469-8476.
56. Greish, K., Enhanced permeability and retention (EPR) effect for anticancer nanomedicine drug targeting. Methods Mol Biol, 2010. 624: p. 25-37.
57. Maeda, H., H. Nakamura, and J. Fang, The EPR effect for macromolecular drug delivery to solid tumors: Improvement of tumor uptake, lowering of systemic toxicity, and distinct tumor imaging in vivo. Adv Drug Deliv Rev, 2013. 65(1): p. 71-9.
58. Iyer, A.K., et al., Exploiting the enhanced permeability and retention effect for tumor targeting. Drug Discov Today, 2006. 11(17-18): p. 812-8.
59. Sardanelli, F., et al., Gadobenate dimeglumine-enhanced MR imaging breast vascular maps: association between invasive cancer and ipsilateral increased vascularity. Radiology, 2005. 235(3): p. 791-7.
60. Aime, S., et al., High Relaxivity Contrast Agents for MRI and Molecular Imaging, in Molecular Imaging, A.A. Bogdanov, Jr. and K. Licha, Editors. 2005, Springer Berlin Heidelberg. p. 99-121.
61. Prabhakar, U., et al., Challenges and key considerations of the enhanced permeability and retention effect for nanomedicine drug delivery in oncology. Cancer Res, 2013. 73(8): p. 2412-7.
62. Gaeta, C.M., et al., Recurrent and metastatic breast cancer PET, PET/CT, PET/MRI: FDG and new biomarkers. Q J Nucl Med Mol Imaging, 2013. 57(4): p. 352-66.
63. Gore, J.C., et al., Magnetic resonance in the era of molecular imaging of cancer. Magn Reson Imaging, 2011. 29(5): p. 587-600.
64. Glunde, K., et al., Magnetic resonance spectroscopy in metabolic and molecular imaging and diagnosis of cancer. Chem Rev, 2010. 110(5): p. 3043-59.
Case Western Reserve University Page 174 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
65. Zhou, Z. and Z.R. Lu, Gadolinium-based contrast agents for magnetic resonance cancer imaging. Wiley Interdiscip Rev Nanomed Nanobiotechnol, 2013. 5(1): p. 1-18.
66. Conway, J.R., N.O. Carragher, and P. Timpson, Developments in preclinical cancer imaging: innovating the discovery of therapeutics. Nat Rev Cancer, 2014. 14(5): p. 314-28.
67. Zhang, J. and J.S. Liu, Tumor stroma as targets for cancer therapy. Pharmacology & Therapeutics, 2013. 137(2): p. 200-215.
68. Bournazou, E. and J. Bromberg, Targeting the tumor microenvironment: JAK- STAT3 signaling. JAKSTAT, 2013. 2(2): p. e23828.
69. Samples, J., M. Willis, and N. Klauber-Demore, Targeting angiogenesis and the tumor microenvironment. Surg Oncol Clin N Am, 2013. 22(4): p. 629-39.
70. Fang, H. and Y.A. Declerck, Targeting the tumor microenvironment: from understanding pathways to effective clinical trials. Cancer Res, 2013. 73(16): p. 4965-77.
71. Schroeder, A., Targeting nanotherapeutics to the tumor microenvironment: how accurately can we aim? Ther Deliv, 2013. 4(7): p. 771-3.
72. Huang, S., et al., Tumor targeting and microenvironment-responsive nanoparticles for gene delivery. Biomaterials, 2013. 34(21): p. 5294-302.
73. Sounni, N.E. and A. Noel, Targeting the tumor microenvironment for cancer therapy. Clin Chem, 2013. 59(1): p. 85-93.
74. Barar, J., Targeting tumor microenvironment: the key role of immune system. Bioimpacts, 2012. 2(1): p. 1-3.
75. Kenny, P.A., G.Y. Lee, and M.J. Bissell, Targeting the tumor microenvironment. Front Biosci, 2007. 12: p. 3468-74.
76. Joyce, J.A., Therapeutic targeting of the tumor microenvironment. Cancer Cell, 2005. 7(6): p. 513-20.
77. Clarke, J.M. and H.I. Hurwitz, Targeted inhibition of VEGF receptor 2: an update on ramucirumab. Expert Opinion on Biological Therapy, 2013. 13(8): p. 1187- 1196.
Case Western Reserve University Page 175 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
78. Burden-Gulley, S.M., et al., A novel molecular diagnostic of glioblastomas: detection of an extracellular fragment of protein tyrosine phosphatase mu. Neoplasia, 2010. 12(4): p. 305-16.
79. Ensslen-Craig, S.E. and S.M. Brady-Kalnay, Receptor protein tyrosine phosphatases regulate neural development and axon guidance. Dev Biol, 2004. 275(1): p. 12-22.
80. Burgoyne, A.M., et al., PTPmu suppresses glioma cell migration and dispersal. Neuro Oncol, 2009. 11(6): p. 767-78.
81. Burgoyne, A.M., et al., Proteolytic cleavage of protein tyrosine phosphatase mu regulates glioblastoma cell migration. Cancer Res, 2009. 69(17): p. 6960-8.
82. Burden-Gulley, S.M., et al., Molecular Magnetic Resonance Imaging of Tumors with a PTPmicro Targeted Contrast Agent. Transl Oncol, 2013. 6(3): p. 329-37.
83. Simberg, D., et al., Biomimetic amplification of nanoparticle homing to tumors. Proc Natl Acad Sci U S A, 2007. 104(3): p. 932-6.
84. Zhou, Z., et al., Peptide targeted tripod macrocyclic Gd(III) chelates for cancer molecular MRI. Biomaterials, 2013. 34(31): p. 7683-93.
85. Suva, L.J., R.J. Griffin, and I. Makhoul, Mechanisms of bone metastases of breast cancer. Endocr Relat Cancer, 2009. 16(3): p. 703-13.
86. Fehm, T., et al., Detection and characterization of circulating tumor cells in blood of primary breast cancer patients by RT-PCR and comparison to status of bone marrow disseminated cells. Breast Cancer Res, 2009. 11(4): p. R59.
87. Aslakson, C.J. and F.R. Miller, Selective events in the metastatic process defined by analysis of the sequential dissemination of subpopulations of a mouse mammary tumor. Cancer Res, 1992. 52(6): p. 1399-405.
88. Bibby, M.C., Orthotopic models of cancer for preclinical drug evaluation: advantages and disadvantages. Eur J Cancer, 2004. 40(6): p. 852-7.
89. Eccles, S.A., et al., Preclinical models for the evaluation of targeted therapies of metastatic disease. Cell Biophys, 1994. 24-25: p. 279-91.
90. Francia, G., et al., Mouse models of advanced spontaneous metastasis for experimental therapeutics. Nat Rev Cancer, 2011. 11(2): p. 135-41.
Case Western Reserve University Page 176 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
91. Humm, J.L., et al., A stereotactic method for the three-dimensional registration of multi-modality biologic images in animals: NMR, PET, histology, and autoradiography. Medical Physics, 2003. 30(9): p. 2303-2314.
92. Yang, M., et al., Whole-body optical imaging of green fluorescent protein- expressing tumors and metastases. Proc Natl Acad Sci U S A, 2000. 97(3): p. 1206-11.
93. Oyajobi, B.O., et al., Detection of myeloma in skeleton of mice by whole-body optical fluorescence imaging. Mol Cancer Ther, 2007. 6(6): p. 1701-8.
94. Kaijzel, E.L., G. van der Pluijm, and C.W. Lowik, Whole-body optical imaging in animal models to assess cancer development and progression. Clin Cancer Res, 2007. 13(12): p. 3490-7.
95. Hoffman, R.M., In vivo imaging of metastatic cancer with fluorescent proteins. Cell Death Differ, 2002. 9(8): p. 786-9.
96. Ishii, T. and M. Ishii, Intravital two-photon imaging: a versatile tool for dissecting the immune system. Annals of the Rheumatic Diseases, 2011. 70: p. I113-I115.
97. Alexander, S., et al., Preclinical intravital microscopy of the tumour-stroma interface: invasion, metastasis, and therapy response. Curr Opin Cell Biol, 2013. 25(5): p. 659-71.
98. Kaufman, M.H., et al., Computer-aided 3-D reconstruction of serially sectioned mouse embryos: its use in integrating anatomical organization. Int J Dev Biol, 1997. 41(2): p. 223-33.
99. Odgaard, A., et al., A direct method for fast three-dimensional serial reconstruction. J Microsc, 1990. 159(Pt 3): p. 335-42.
100. Kaufman, M.H., et al., Computer-generated three-dimensional reconstructions of serially sectioned mouse embryos. J Anat, 1998. 193 ( Pt 3): p. 323-36.
101. Lasnon, C., et al., Contrast-enhanced small-animal PET/CT in cancer research: strong improvement of diagnostic accuracy without significant alteration of quantitative accuracy and NEMA NU 4-2008 image quality parameters. EJNMMI Research, 2013. 3(1): p. 5.
102. Beyer, T., et al., A combined PET/CT scanner for clinical oncology. J Nucl Med, 2000. 41(8): p. 1369-79.
Case Western Reserve University Page 177 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
103. Xiong, C., et al., In Vivo Small-Animal PET/CT of EphB4 Receptors Using 64Cu- Labeled Peptide. Journal of Nuclear Medicine, 2011. 52(2): p. 241-248.
104. Zaidi, H., O. Mawlawi, and C.G. Orton, Simultaneous PET/MR will replace PET/CT as the molecular multimodality imaging platform of choice. Medical Physics, 2007. 34(5): p. 1525-1528.
105. Maseeh uz, Z., et al., Whole body simultaneous PET/MRI: one-stop-shop? J Pak Med Assoc, 2014. 64(2): p. 201-4.
106. Partovi, S., et al., Clinical oncologic applications of PET/MRI: a new horizon. Am J Nucl Med Mol Imaging, 2014. 4(2): p. 202-212.
107. Cho, H., et al., Noninvasive multimodality imaging of the tumor microenvironment: registered dynamic magnetic resonance imaging and positron emission tomography studies of a preclinical tumor model of tumor hypoxia. Neoplasia, 2009. 11(3): p. 247-59, 2p following 259.
108. Wahl, R.L., et al., "Anatometabolic" tumor imaging: fusion of FDG PET with CT or MRI to localize foci of increased activity. J Nucl Med, 1993. 34(7): p. 1190-7.
109. Veit-Haibach, P., et al., PET–MR imaging using a tri-modality PET/CT–MR system with a dedicated shuttle in clinical routine. Magnetic Resonance Materials in Physics, Biology and Medicine, 2013. 26(1): p. 25-35.
110. de Jong, G., et al., FDG-PET, CT and MRI for the detection of experimental colorectal liver metastases: An exploratory study. Society of Nuclear Medicine Annual Meeting Abstracts, 2011. 52(1_MeetingAbstracts): p. 312.
111. Deroose, C.M., et al., Multimodality imaging of tumor xenografts and metastases in mice with combined small-animal PET, small-animal CT, and bioluminescence imaging. J Nucl Med, 2007. 48(2): p. 295-303.
112. Pichler, B.J., M.S. Judenhofer, and C. Pfannenberg, Multimodal imaging approaches: PET/CT and PET/MRI. Handb Exp Pharmacol, 2008(185 Pt 1): p. 109-32.
113. Garofalakis, A., et al., Fusion of [(18)F]FDG PET with fluorescence diffuse optical tomography to improve validation of probes and tumor imaging. Mol Imaging Biol, 2013. 15(3): p. 316-25.
114. Chow, P.L., et al., A method of image registration for small animal, multi- modality imaging. Phys Med Biol, 2006. 51(2): p. 379-90.
Case Western Reserve University Page 178 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
115. Li, X., et al. Constrained non-rigid registration for whole body image registration: method and validation. 2007.
116. Xia, L., et al. Automatic registration of whole body serial micro CT images with a combination of point-based and intensity-based registration techniques. in Biomedical Imaging: Nano to Macro, 2006. 3rd IEEE International Symposium on. 2006.
117. Xia, Z., et al., Registration of 3-D CT and 2-D flat images of mouse via affine transformation. IEEE Trans Inf Technol Biomed, 2008. 12(5): p. 569-78.
118. Baiker, M., et al. FULLY AUTOMATED WHOLE-BODY REGISTRATION IN MICE USING AN ARTICULATED SKELETON ATLAS. in Biomedical Imaging: From Nano to Macro, 2007. ISBI 2007. 4th IEEE International Symposium on. 2007.
119. Xiao, D., et al. A non-rigid registration method for mouse whole body skeleton registration. 2010.
120. Wang, H. and F. Baowei. A Robust B-Splines-Based Point Match Method for Non-Rigid Surface Registration. in Bioinformatics and Biomedical Engineering, 2008. ICBBE 2008. The 2nd International Conference on. 2008.
121. Dogdas, B., et al., Digimouse: a 3D whole body mouse atlas from CT and cryosection data. Phys Med Biol, 2007. 52(3): p. 577-87.
122. Studholme, C., D.L.G. Hill, and D.J. Hawkes, An overlap invariant entropy measure of 3D medical image alignment. Pattern Recognition, 1999. 32(1): p. 71- 86.
123. Christensen, G.E., Consistent linear-elastic transformations for image matching. Information Processing in Medical Imaging, Proceedings, 1999. 1613: p. 224-237.
124. Gargesha, M., et al., Visualization of color anatomy and molecular fluorescence in whole-mouse cryo-imaging. Comput Med Imaging Graph, 2011. 35(3): p. 195- 205.
125. Krishnamurthi, G., et al., Removal of subsurface fluorescence in cryo-imaging using deconvolution. Opt Express, 2010. 18(21): p. 22324-38.
126. Roy, D., et al., 3D cryo-imaging: a very high-resolution view of the whole mouse. Anat Rec (Hoboken), 2009. 292(3): p. 342-51.
Case Western Reserve University Page 179 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
127. Steyer, G.J., et al., Removal of out-of-plane fluorescence for single cell visualization and quantification in cryo-imaging. Ann Biomed Eng, 2009. 37(8): p. 1613-28.
128. Wilson, D., et al., Whole Mouse cryo-Imaging. Proc Soc Photo Opt Instrum Eng, 2008. 6916: p. 69161I-69161I9.
129. Tao, K., et al., Imagable 4T1 model for the study of late stage breast cancer. BMC Cancer, 2008. 8: p. 228.
130. Li, J. and G.G. Sahagian, Demonstration of tumor suppression by mannose 6- phosphate/insulin-like growth factor 2 receptor. Oncogene, 2004. 23(58): p. 9359-68.
131. Zou, K.H., et al., Statistical validation of image segmentation quality based on a spatial overlap index. Acad Radiol, 2004. 11(2): p. 178-89.
132. Lim, J., Two-Dimensional Signal and Image Processing. Englewood Cliffs : Prentice Hall, 1990: p. 548.
133. Mendonca, A.M. and A. Campilho, Segmentation of retinal blood vessels by combining the detection of centerlines and morphological reconstruction. IEEE Trans Med Imaging, 2006. 25(9): p. 1200-13.
134. Ma, W.Y. and B.S. Manjunath, EdgeFlow: a technique for boundary detection and image segmentation. IEEE Trans Image Process, 2000. 9(8): p. 1375-88.
135. Johnson, D.H., et al., Improved fat-water reconstruction algorithm with graphics hardware acceleration. J Magn Reson Imaging, 2010. 31(2): p. 457-65.
136. Ding H, N.A., Gutmann DH, Guha A, A Review of Astrocytoma Models. Neurosurgical Focus, 2000. 8:1–8.
137. Barth, R.F. and B. Kaur, Rat brain tumor models in experimental neuro-oncology: the C6, 9L, T9, RG2, F98, BT4C, RT-2 and CNS-1 gliomas. J Neurooncol, 2009. 94(3): p. 299-312.
138. Kruse, C.A., et al., A rat glioma model, CNS-1, with invasive characteristics similar to those of human gliomas: a comparison to 9L gliosarcoma. J Neurooncol, 1994. 22(3): p. 191-200.
139. Tyminski, E., et al., Brain tumor oncolysis with replication-conditional herpes simplex virus type 1 expressing the prodrug-activating genes, CYP2B1 and
Case Western Reserve University Page 180 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
secreted human intestinal carboxylesterase, in combination with cyclophosphamide and irinotecan. Cancer Res, 2005. 65(15): p. 6850-7.
140. Scherer, H.J., Structural Development in Gliomas. The American Journal of Cancer, 1938. 34(3): p. 333-351.
141. Burger PC, S.B., Vogel FS, Surgical pathology of the nervous system and its coverings. 2002(3rd ed, New York: Wiley).
142. Radaelli, E., et al., Immunohistopathological and neuroimaging characterization of murine orthotopic xenograft models of glioblastoma multiforme recapitulating the most salient features of human disease. Histol Histopathol, 2009. 24(7): p. 879-91.
143. Zagzag, D., et al., Hypoxia- and vascular endothelial growth factor-induced stromal cell-derived factor-1alpha/CXCR4 expression in glioblastomas: one plausible explanation of Scherer's structures. Am J Pathol, 2008. 173(2): p. 545- 60.
144. Kislin, K.L., et al., NHERF-1: modulator of glioblastoma cell migration and invasion. Neoplasia, 2009. 11(4): p. 377-87.
145. Nakada, M., et al., Ephrin-B3 Ligand Promotes Glioma Invasion through Activation of Rac1. Cancer Research, 2006. 66(17): p. 8492-8500.
146. Nakada, M., et al., The phosphorylation of EphB2 receptor regulates migration and invasion of human glioma cells. Cancer Res, 2004. 64(9): p. 3179-85.
147. Nakada, M., et al., The phosphorylation of ephrin-B2 ligand promotes glioma cell migration and invasion. Int J Cancer, 2010. 126(5): p. 1155-65.
148. Paulino, V.M., et al., TROY (TNFRSF19) Is Overexpressed in Advanced Glial Tumors and Promotes Glioblastoma Cell Invasion via Pyk2-Rac1 Signaling. Molecular Cancer Research, 2010. 8(11): p. 1558-1567.
149. Bagci, T., et al., Autocrine semaphorin 3A signaling promotes glioblastoma dispersal. Oncogene, 2009. 28(40): p. 3537-50.
150. Enloe, B.M. and D.G. Jay, Inhibition of Necl-5 (CD155/PVR) reduces glioblastoma dispersal and decreases MMP-2 expression and activity. J Neurooncol, 2011. 102(2): p. 225-35.
151. Molina, J.R., et al., Invasive glioblastoma cells acquire stemness and increased Akt activation. Neoplasia, 2010. 12(6): p. 453-63. Case Western Reserve University Page 181 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
152. Kohler, B.A., et al., Annual report to the nation on the status of cancer, 1975- 2007, featuring tumors of the brain and other nervous system. J Natl Cancer Inst, 2011. 103(9): p. 714-36.
153. Stupp, R., et al., Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N Engl J Med, 2005. 352(10): p. 987-96.
154. Pichlmeier, U., et al., Resection and survival in glioblastoma multiforme: an RTOG recursive partitioning analysis of ALA study patients. Neuro Oncol, 2008. 10(6): p. 1025-34.
155. Ewelt, C., et al., Glioblastoma multiforme of the elderly: the prognostic effect of resection on survival. J Neurooncol, 2011. 103(3): p. 611-8.
156. Stummer, W., et al., Extent of resection and survival in glioblastoma multiforme: identification of and adjustment for bias. Neurosurgery, 2008. 62(3): p. 564-76; discussion 564-76.
157. Gerstner, E.R., et al., Advances in neuroimaging techniques for the evaluation of tumor growth, vascular permeability, and angiogenesis in gliomas. Curr Opin Neurol, 2008. 21(6): p. 728-35.
158. Van Meir, E.G., et al., Exciting new advances in neuro-oncology: the avenue to a cure for malignant glioma. CA Cancer J Clin, 2010. 60(3): p. 166-93.
159. Aricescu, A.R., et al., Molecular analysis of receptor protein tyrosine phosphatase mu-mediated cell adhesion. EMBO J, 2006. 25(4): p. 701-12.
160. Aricescu, A.R. and E.Y. Jones, Immunoglobulin superfamily cell adhesion molecules: zippers and signals. Curr Opin Cell Biol, 2007. 19(5): p. 543-50.
161. Aricescu, A.R., C. Siebold, and E.Y. Jones, Receptor protein tyrosine phosphatase micro: measuring where to stick. Biochem Soc Trans, 2008. 36(Pt 2): p. 167-72.
162. Brady-Kalnay, S.M. and N.K. Tonks, Identification of the homophilic binding site of the receptor protein tyrosine phosphatase PTP mu. J Biol Chem, 1994. 269(45): p. 28472-7.
163. Cismasiu, V.B., et al., The MAM (meprin/A5-protein/PTPmu) domain is a homophilic binding site promoting the lateral dimerization of receptor-like protein-tyrosine phosphatase mu. J Biol Chem, 2004. 279(26): p. 26922-31.
Case Western Reserve University Page 182 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
164. Del Vecchio, R.L. and N.K. Tonks, The conserved immunoglobulin domain controls the subcellular localization of the homophilic adhesion receptor protein- tyrosine phosphatase mu. J Biol Chem, 2005. 280(2): p. 1603-12.
165. Zondag, G.C., et al., Homophilic interactions mediated by receptor tyrosine phosphatases mu and kappa. A critical role for the novel extracellular MAM domain. J Biol Chem, 1995. 270(24): p. 14247-50.
166. Gebbink, M.F., et al., Cloning, expression and chromosomal localization of a new putative receptor-like protein tyrosine phosphatase. FEBS Lett, 1991. 290(1-2): p. 123-30.
167. Gebbink, M.F., et al., Purification and characterization of the cytoplasmic domain of human receptor-like protein tyrosine phosphatase RPTP mu. Biochemistry, 1993. 32(49): p. 13516-22.
168. Pogue, B.W., et al., Review of Neurosurgical Fluorescence Imaging Methodologies. IEEE J Sel Top Quantum Electron, 2010. 16(3): p. 493-505.
169. Keereweer, S., et al., Optical image-guided surgery--where do we stand? Mol Imaging Biol, 2011. 13(2): p. 199-207.
170. Terwisscha van Scheltinga, A.G., et al., Intraoperative near-infrared fluorescence tumor imaging with vascular endothelial growth factor and human epidermal growth factor receptor 2 targeting antibodies. J Nucl Med, 2011. 52(11): p. 1778- 85.
171. van Dam, G.M., et al., Intraoperative tumor-specific fluorescence imaging in ovarian cancer by folate receptor-alpha targeting: first in-human results. Nat Med, 2011. 17(10): p. 1315-9.
172. Mieog, J.S., et al., Image-guided tumor resection using real-time near-infrared fluorescence in a syngeneic rat model of primary breast cancer. Breast Cancer Res Treat, 2011. 128(3): p. 679-89.
173. Nguyen, Q.T., et al., Surgery with molecular fluorescence imaging using activatable cell-penetrating peptides decreases residual cancer and improves survival. Proc Natl Acad Sci U S A, 2010. 107(9): p. 4317-22.
174. Jiang, T., et al., Tumor imaging by means of proteolytic activation of cell- penetrating peptides. Proc Natl Acad Sci U S A, 2004. 101(51): p. 17867-72.
Case Western Reserve University Page 183 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
175. Olson, E.S., et al., Activatable cell penetrating peptides linked to nanoparticles as dual probes for in vivo fluorescence and MR imaging of proteases. Proc Natl Acad Sci U S A, 2010. 107(9): p. 4311-6.
176. Craig, S.E. and S.M. Brady-Kalnay, Cancer cells cut homophilic cell adhesion molecules and run. Cancer Res, 2011. 71(2): p. 303-9.
177. Craig, S.E. and S.M. Brady-Kalnay, Tumor-derived extracellular fragments of receptor protein tyrosine phosphatases (RPTPs) as cancer molecular diagnostic tools. Anticancer Agents Med Chem, 2011. 11(1): p. 133-40.
178. Edwards, D.R., M.M. Handsley, and C.J. Pennington, The ADAM metalloproteinases. Mol Aspects Med, 2008. 29(5): p. 258-89.
179. Burden-Gulley, S.M., et al., Single cell molecular recognition of migrating and invading tumor cells using a targeted fluorescent probe to receptor PTPmu. Int J Cancer, 2013. 132(7): p. 1624-32.
180. Rueckert, D., et al., Nonrigid registration using free-form deformations: Application to breast MR images. Ieee Transactions on Medical Imaging, 1999. 18(8): p. 712-721.
181. Ourselin, S., et al., Reconstructing a 3D structure from serial histological sections. Image and Vision Computing, 2001. 19(1–2): p. 25-31.
182. Ourselin, S., R. Stefanescu, and X. Pennec, Robust Registration of Multi-modal Images: Towards Real-Time Clinical Applications, in Medical Image Computing and Computer-Assisted Intervention — MICCAI 2002, T. Dohi and R. Kikinis, Editors. 2002, Springer Berlin Heidelberg. p. 140-147.
183. Modat, M., et al., Fast free-form deformation using graphics processing units. Computer Methods and Programs in Biomedicine, 2010. 98(3): p. 278-284.
184. Pluim, J.P., J.B. Maintz, and M.A. Viergever, Mutual-information-based registration of medical images: a survey. IEEE Trans Med Imaging, 2003. 22(8): p. 986-1004.
185. Wahba, G., Spline Models for Observational Data. CBMS-NSF Regional Conference Series in Applied Mathematics. 1990: Society for Industrial and Applied Mathematics. 177.
186. Rohlfing, T., et al., Volume-preserving nonrigid registration of MR breast images using free-form deformation with an incompressibility constraint. IEEE Trans Med Imaging, 2003. 22(6): p. 730-41. Case Western Reserve University Page 184 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
187. Cachier, P. and D. Rey, Symmetrization of the Non-rigid Registration Problem Using Inversion-Invariant Energies: Application to Multiple Sclerosis, in Medical Image Computing and Computer-Assisted Intervention – MICCAI 2000, S. Delp, A. DiGoia, and B. Jaramaz, Editors. 2000, Springer Berlin Heidelberg. p. 472- 481.
188. Rose, A., The sensitivity performance of the human eye on an absolute scale. J Opt Soc Am, 1948. 38(2): p. 196-208.
189. Bushberg, J.T., The essential physics of medical imaging. 3rd ed. 2012, Philadelphia: Wolters Kluwer Health/Lippincott Williams & Wilkins. xii, 1030 p.
190. Kawamoto, T., Use of a new adhesive film for the preparation of multi-purpose fresh-frozen sections from hard tissues, whole-animals, insects and plants. Archives of histology and cytology, 2003. 66(2): p. 123-143.
191. Pech, P., et al., Attenuation Values, Volume Changes and Artifacts in Tissue Due to Freezing. Acta Radiologica, 1987. 28(6): p. 779-782.
192. Akbarzadeh, A., et al., Evaluation of whole-body MR to CT deformable image registration. J Appl Clin Med Phys, 2013. 14(4): p. 4163.
193. Dice, L.R., Measures of the Amount of Ecologic Association Between Species. Ecology, 1945. 26(3): p. 297-302.
194. Klein, A., et al., Evaluation of 14 nonlinear deformation algorithms applied to human brain MRI registration. Neuroimage, 2009. 46(3): p. 786-802.
195. Li, Y., et al., Direct labeling and visualization of blood vessels with lipophilic carbocyanine dye DiI. Nat Protoc, 2008. 3(11): p. 1703-8.
196. Lawson, N.D. and B.M. Weinstein, In vivo imaging of embryonic vascular development using transgenic zebrafish. Dev Biol, 2002. 248(2): p. 307-18.
197. Berlemont, S. and J.C. Olivo-Marin, Combining local filtering and multiscale analysis for edge, ridge, and curvilinear objects detection. IEEE Trans Image Process, 2010. 19(1): p. 74-84.
198. Chapman, B.E. and D.L. Parker, 3D multi-scale vessel enhancement filtering based on curvature measurements: application to time-of-flight MRA. Med Image Anal, 2005. 9(3): p. 191-208.
Case Western Reserve University Page 185 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
199. Du, Y.P. and D.L. Parker, Vessel enhancement filtering in three-dimensional MR angiograms using long-range signal correlation. J Magn Reson Imaging, 1997. 7(2): p. 447-50.
200. Pace, D.F., S.R. Aylward, and M. Niethammer, A locally adaptive regularization based on anisotropic diffusion for deformable image registration of sliding organs. IEEE Trans Med Imaging, 2013. 32(11): p. 2114-26.
201. Pace, D.F., et al., Deformable Image Registration of Sliding Organs Using Anisotropic Diffusive Regularization. Proc IEEE Int Symp Biomed Imaging, 2011: p. 407-413.
202. Delmon, V., et al., Registration of sliding objects using direction dependent B- splines decomposition. Phys Med Biol, 2013. 58(5): p. 1303-14.
203. Buerger, C., T. Schaeffter, and A.P. King, Hierarchical adaptive local affine registration for fast and robust respiratory motion estimation. Med Image Anal, 2011. 15(4): p. 551-64.
204. Vandemeulebroucke, J., et al., Automated segmentation of a motion mask to preserve sliding motion in deformable registration of thoracic CT. Med Phys, 2012. 39(2): p. 1006-15.
205. Duerk, J.L., et al., Remember true FISP? A high SNR, near 1-second imaging method for T2-like contrast in interventional MRI at .2 T. J Magn Reson Imaging, 1998. 8(1): p. 203-8.
206. Wunderlich, A. and F. Noo, New Theoretical Results on Channelized Hotelling Observer Performance Estimation with Known Difference of Class Means. IEEE Trans Nucl Sci, 2013. 60(1): p. 182-193.
207. Brankov, J.G., Evaluation of the channelized Hotelling observer with an internal- noise model in a train-test paradigm for cardiac SPECT defect detection. Phys Med Biol, 2013. 58(20): p. 7159-82.
208. Yu, L., et al., Prediction of human observer performance in a 2-alternative forced choice low-contrast detection task using channelized Hotelling observer: impact of radiation dose and reconstruction algorithms. Med Phys, 2013. 40(4): p. 041908.
209. Kulkarni, S., et al., A channelized Hotelling observer study of lesion detection in SPECT MAP reconstruction using anatomical priors. Phys Med Biol, 2007. 52(12): p. 3601-17.
Case Western Reserve University Page 186 of 187
Mohammed Qutaish, PhD Thesis: Cryo-Imaging Assessment of Imaging Agent Targeting to Dispersing and Metastatic Tumor Cells
210. Gifford, H.C., et al., Channelized hotelling and human observer correlation for lesion detection in hepatic SPECT imaging. J Nucl Med, 2000. 41(3): p. 514-21.
Case Western Reserve University Page 187 of 187