Image-Based Characterization of the Mechanical Behaviour of Healthy and Metastatically-Involved Vertebrae

Transcription

1 Image-Based Characterization of the Mechanical Behaviour of Healthy and Metastatically-Involved Vertebrae By Chetan Choudhari A thesis submitted in conformity with the requirements for the degree of Masters of Applied Science Institute of Biomaterials and Biomedical Engineering, University of Toronto Copyright by Chetan Choudhari 2014

2 Image-Based Characterization of the Mechanical Behaviour of Healthy and Metastatically-Involved Vertebrae Abstract Chetan Choudhari Master of Applied Science Institute of Biomaterials and Biomedical Engineering University of Toronto 2014 Skeletal metastasis leads to changes in bone architecture, quality and strength, including microdamage accumulation. This dissertation aims to combine image-based computational and experimental techniques to study trabecular bone microdamage in healthy and metastatic whole bones. Deformable image registration was used to demonstrate proof of concept that post-euthanasia strain analysis of µct images represents in vivo quasi static mechanical behavior of whole rat vertebrae. The ability to concurrently identify microdamage in whole vertebrae using histologic techniques (calcein and fuschin) and contrast enhanced BaSO4 µct imaging was demonstrated and compared to stresses and strains calculated through micro finite element analysis. Significantly higher stresses and strains were found in regions of trabecular microdamage compared to undamaged regions, and in metastatic compared to healthy vertebrae. The techniques and knowledge developed through this work improve understanding of trabecular bone microdamage and form a solid platform for modeling the material and structural behaviour of skeletal tissue. ii

3 Acknowledgements First and foremost, I would like to express my special appreciation and gratitude to my advisor Dr. Cari Whyne. I attribute my Master s degree to her encouragement and effort. Without Cari this thesis, too, would not have been completed or written. Cari has a tremendous passion for science and a contagious enthusiasm for research. She has supported me throughout the duration of my project with her knowledge and insight, whilst patiently allowing me to think and work independently. One simply could not wish for a better and friendlier supervisor. I would like to thank my committee members Dr. Radovan Zdero, Dr. Albert Yee and Dr. Tom Willett for providing meaningful suggestions and insights, which contributed significantly to the overall success of this project. Working in Orthopedic Biomechanics Lab (OBL) has been a great learning experience for me. The incredible work ethic followed by each and every member of the lab was inspiring. The environment in the lab was full of energy and intellectually stimulating. I would like to thank the members of the OBL, who also share the credit for this work. I thank Dr. Margarete Akens for arranging the samples used in this project. Much of my understanding about animal models and experimental techniques used in this project can be attributed to her. I would also like to acknowledge Stew, Hamid, Zoryana, Mikhael and Edwin for making this project a thoroughly enjoyable ride. Additionally, I want would like to appreciate Katelyn for her work on the project over the summer. Lastly, I am forever grateful to my family and friends for their selfless love and support, which have allowed me to overcome the challenges I faced over the duration of the project. iii

4 Contents List of Tables... viii List of Figures... ix List of Appendices... xii Chapter 1: Introduction and Thesis layout Layout of the Thesis... 3 Chapter 2: Literature review Bone structure and composition Bone remodeling Anatomy of Vertebral column Spinal metastasis Fracture risk Clinical interventions for spinal metastasis Pre-clinical models of spinal metastasis Trabecular biomechanics of the vertebral body Trabecular microarchitecture in vertebral body Microdamage in trabecular bone Mechanical properties of trabecular bone Biomechanical analysis of bone Mechanical testing Microdamage analysis through staining Finite element analysis Image-based strain analysis Summary iv

5 Chapter 3: Post euthanasia micro-ct based strain analysis is able to represent quasi-static in vivo behavior of whole vertebrae Abstract Introduction Methods Deformable image registration algorithm Animal model Loading Imaging Strain calculations and data analysis Results Discussion Conclusion Chapter 4: Barium sulfate contrast enhanced μct imaging to identify microdamage in whole rat vertebrae Abstract Introduction Methods Staining protocols BaSO4 staining of rat vertebrae Histological validation of BaSO4 staining Results BaSO4 staining protocol for rat vertebrae BaSO4 and Calcein/Fuchsin compatibility Discussion Conclusion v

6 Chapter 5: Evaluation of tissue level stresses and strains under uniaxial compression of whole healthy and osteolytic rat spines Abstract Introduction Methods Animal models: Microdamage Evaluation using calcein/fuchsin staining and contrast enhanced μct Loading Strain fields and boundary conditions Alignment of histology slides Creating µfe models Statistics and data analysis Results Microdamage identification using histology Image registration to determine strain fields and boundary conditions Alignment of histology slides with unloaded scans µfe modeling of healthy and metastatic spines Determining tissue-level stresses and strains in histologically damaged and undamaged regions Stresses and strains in damage regions identified by BaSO Discussion Microdamage analysis using sequential staining Alignment of histology slides Trabecular stresses and strains in histologically identified microdamage Microdamage identification using BaSO4 contrast enhanced imaging vi

7 5.5.5 µfe modeling Conclusion Chapter 6: Concluding remarks Chapter 7: References Appendix 1: Copyright Permissions vii

8 List of Tables Table 2.1: Local compressive yield stresses and strains for trabecular bone.. 15 Table 3.1: Strains (µm/µm) obtained from the comparisons of loaded-unloaded images..27 Table 3.2: Strains (µm/µm) obtained from the comparisons of images under similar loading conditions Table 4.1: BaSO4 staining parameters used for various bone types till date...35 Table 5.1: Trabecular stresses and strains at locations of microdamage determined under axial load by previous studies Table 5.2: Average strains (µm/µm) obtained from the comparisons of loaded/unloaded images of healthy and osteolytic spines...59 Table 5.3: Volumetric concurrencies for healthy spines Table 5.4: Volumetric concurrencies for osteolytic spines Table 5.5: Average stress and strain from damaged and undamaged regions (healthy...70 Table 5.6: Average stress and strain from damaged and undamaged regions (metastatic). 71 Table 5.7: Comparison of local stresses and strains in healthy and osteolytic models...72 Table 5.8: Local stresses and strains in BaSO4 contrast enhanced damaged sites (healthy)...73 Table 5.9: Local stresses and strains in BaSO4 contrast enhanced damaged sites (metastatic)..74 Table 5.10: Comparison of local stresses and strains in healthy and osteolytic models (BaSO4) Table 5.11: Comparison of local stresses and strains within regions of damage identified by fuchsin and BaSO4 in healthy models Table 5.12: Comparison of local stresses and strains within regions of damage identified by fuchsin and BaSO4 in metastatic models viii

9 List of Figures Figure 2.1: Difference between the structures of cortical and trabecular bones. The inner trabecular bone is much more porous than the outer compact bone. (Public domain image from Wikimedia Commons) Figure 2.2: Anatomy of human spine; (left) all levels of the vertebral column, (centre) an individual vertebra with vertebral body and posterior elements, (right) series of 3 vertebrae displaying facet joints and intervertebral discs (Public domain image from Wikimedia commons)..8 Figure 2.3: Various types of spinal metastasis. A: Osteolytic, B: Osteoblastic, C: Mixed Focal, D: Mixed diffuse (Skrinskas 2009)... 9 Figure 2.4: µct image slice showing cross section (frontal plane) of a rat L1 vertebral body, with a cortical shell surrounding the trabecular centrum Figure 2.5: High resolution micro-ct image of trabecular specimen (left), converted into a voxel based micro finite element mesh (right) (Keaveny 2001).19 Figure 2.6: Deformable image registration algorithm determines strain by aligning and comparing scans of the same sample with and without load (Hojjat 2011) 20 Figure 3.1: Micro-CT compatible loading jig used to apply an axial compressive load to the 6th caudal vertebra in the rat tail Figure 3.2: Mean axial strains (µm/µm) obtained from deformable registration (* represents p-value<0.016). a) Strains obtained from all loaded-unloaded and all loadedloaded/unloaded-unloaded comparisons. b) Comparison of live and dead strains. c) Live and dead strains for scans under equivalent loading conditions Figure 3.3: Strain patterns in loaded-unloaded strain registrations. The strain is concentrated mostly around the endplate regions. a) Dead-loaded to dead-unloaded comparison b) Liveloaded to dead-unloaded comparison.. 28 Figure 3.4: Strain patterns generated by comparison of scans under equivalent loading conditions. a) Live-loaded to dead loaded b) Dead-loaded to dead-loaded c) Dead-unloaded to dead-unloaded Figure 3.5: Deformation maps generated form for live-dead images (a vertical, b lateral) and dead-dead images (c vertical, d lateral) Figure 4.1: Loading device used to load samples inside the micro-ct scanner Figure 4.2: Micro-CT images of unloaded rat vertebrae stained for 1 day (left), 2 days (centre) and 3 days (right). The bright spots within the vertebral body represent BaSO4. It can be observed that staining for more than 1 day causes overstaining (arrows indicate regions of overstaining) ix

10 Figure 4.3: BaSO4 is able to stain microdamage in osteolytic spine, without pooling. Arrows indicate damaged regions of high intensity, consisting of BaSO Figure 4.4: Fluorescence (a, b), brightfield (c) and µct (d) images of a pre-existing microdamage site highlighted by fuchsin (a, c), calcein (b) and BaSO4 (d) staining. Arrow indicates a region of preexisting microdamage observed in fluorescent images, but not on the bright field image Figure 4.5: Fluorescence (a, b), brightfield (c) and µct (d) images of a load induced microdamage site highlighted by fuchsin (a, c), and BaSO4 (d) staining, but not by calcein (b) Figure 5.1: Experimental design.. 48 Figure 5.2: Coronal histology slide from a healthy vertebral body imaged under fluorescence to identify calcein stained pre-existing damage (a), and under plain light to detect fuchsin stained load induced damage (b) Figure 5.3: a) Bright field image of a histology slide from a healthy sample. b) Fuchsin stained load induced damage on a trabecula at 20x magnification Figure 5.4: a) Pre-existing damage labelled by both calcein and fuchsin b) Load induced microdamage stained only by fuchsin, and not calcein...57 Figure 5.5: a) Coronal histology slide of osteolytic spine demonstrating reduced trabecular number and increased fuchsin accumulation in the osteolytic regions. b) Multiple microdamaged sites observed near osteolytic tumor tissue Figure 5.6: Coronal slices demonstrating strain fields obtained for healthy (a) and metastatic (b) whole vertebrae. Red and blue areas experience high and low strains respectively. Arrow denotes area osteolytic destruction under high strain..59 Figure 5.7: Displacement vectors generated using deformable image registration, represented by blue arrows generated for healthy (a) and metastatic (b) vertebrae...60 Figure 5.8: Bright field image of a histology slide from a healthy slide (a) and surface generated from µct image of the same slide (b)...61 Figure 5.9: Surface generated from registered µct scans of the block (posterior elements) and three slides superimposed on unloaded µct scan from a healthy sample Figure 5.10: Segmented unloaded µct images of the same healthy sample, each containing a histology slide identified as a separate material Figure 5.11: µct scan of a slide from healthy sample (yellow), along with the region of intersection between the slide and the unloaded scans (blue). The VC for this slide was 66% x

11 Figure 5.12: µct scan of a slide from metastatic sample (yellow), along with the region of intersection between the slide and the unloaded scan (green). The VC for this slide was 60% Figure 5.13: µfe grid generated form whole bone µct scan of a metastatic sample with elements corresponding to a histology slide highlighted in red. One such model was generated from each histology slide Figure 5.14: a) µct scan of a spinal motion segment under load. b) Surfaces on the endplates and facet joints were selected as loading surfaces for the middle vertebra (highlighted in pink) Figure 5.15: a) Bright field image of a coronal histology slide from a healthy sample stained with fuchsin to identify load induced microdamage b) Results showing tissue level maximum principal stress distribution in the elements within the µfe model corresponding to the histology slide. Arrows show correspondence between the histology slide and the model 68 Figure 5.16: a) Axial compressive load induced microdamage identified by fuchsin staining. b) Elements corresponding to the damaged site selected in the undeformed µfe model. c) Completed µfea demonstrates elevated maximum principal stress in the region of microdamage...69 Figure 5.17: Comparison of stresses (a) and strains (b) obtained from damaged and undamaged regions in healthy samples. * represents significant differences (pvalue<0.016) Figure 5.18: Comparison of stresses (a) and strains (b) obtained from damaged and undamaged regions in metastatic samples. * represents significant differences (pvalue<0.016) 71 Figure 5.19: Comparison of stresses (a) and strains (b) obtained from damaged regions in healthy versus metastatic samples. * represents significant differences (p-value<0.016)..72 xi

12 List of Appendices Appendix 1: Copyright Permissions 99 xii

13 1 Chapter 1: Introduction and Thesis layout Skeletal metastasis is commonly diagnosed in breast, prostate and lung cancer patients. The vertebral column is the most common bone affected. Metastatic disease in the bony spine can be bone destroying (osteolytic), bone forming (osteoblastic) or a mixture of both (mixed). Spinal metastasis results in severe pain and poses an increased risk of fracture, impacting physical mobility. Compromising the mechanical stability of the spine can significantly affect patients quality of life. Advanced cases of spinal metastasis can lead to pathologic fracture and neurologic complications. Development of more advanced techniques is imperative to identify patients at an elevated risk of vertebral fracture, recognizing when there is a need for intervention and how decisions on intervention can be optimized. The decreased quality and architecture of trabecular bone in the metastatic spine leads to a rapid accumulation of unrepaired microdamage, culminating in an elevated risk of bone failure. Concentrations of local stresses and strains in response to the load applied stimulate the initiation and propagation of microdamage. The distribution of these stresses and strains within the bone matrix is governed by the local microarchitecture and bone quality. A compromised bone microstructure may lead abnormal stress/strain concentrations leading to bone microdamage. However, the exact relationship between trabecular level damage events and local stresses and strains is not well characterized. Quantifying the tissue level structural behavior of healthy and metastatically-involved vertebrae will enhance our understanding of the local mechanical environment at microdamage initiation, leading to better diagnostic criteria for fracture risk assessment. Biomechanical analyses of the metastatic spine have included computational, image-based and experimental techniques. Specifically, finite element (FE) modeling has demonstrated success in predicting fracture patterns and failure loads in bony structures including bones with metastatic defects. The overall objective of this project was to develop imaging methods and computational models that can accurately identify microdamage and quantify microstructural stresses and strains in whole healthy and osteolytic rat spines.

14 2 The overall objective was subdivided into the following three specific aims: Aim 1 Motivation: To date, strain based assessment of bone has been primarily limited to ex vivo specimens. In translating findings from ex vivo strain based studies, it is hypothesized that ex vivo µct-image-based strain analysis represents the in vivo quasi static behavior of whole bones. Research question: Does ex vivo analysis represent the in vivo mechanical behavior in the spine? Hypothesis: Ex vivo µ-image-based strain analysis represents in vivo quasi static vertebral behavior. Specific aim: Use deformable image registration to demonstrate the ability of ex vivo modeling to represent the in vivo behavior of the vertebral column through image-based strain analysis. Aim 2 Motivation: Contrast enhanced μct of whole bone provides a 3D alternative to traditional destructive 2D histologic microdamage analysis. However, demonstration of the ability and accuracy of this method to highlight microdamage in whole vertebrae is essential prior to further utilization of this technique. Research question: How can we best visualize/quantify microdamage in whole bone? Hypothesis: μct contrast enhanced (BaSO4) imaging accurately represents vertebral microdamage identified by whole bone calcein/fuchsin staining. Specific aim: Generate a robust protocol for μct contrast enhanced (BaSO4) imaging and calcein/fuchsin staining, which allows the direct comparison of these techniques to represent load induced damage within whole vertebrae Aim 3 Motivation: µfe modeling is an important technique which shows promise to accurately represent the structural integrity of vertebrae at the trabecular level within the spine.

15 3 Generation of robust models and analysis techniques may allow for development of a better understanding of the stability of healthy and osteolytic vertebrae. Research question: How can µfe analysis represent microdamage in whole healthy and osteolytic vertebrae? Hypothesis: µfe analysis generates computational models that yield consistent damage initiation thresholds in healthy and osteolytic vertebrae from athymic rats. Specific aim: Generate µfe models that accurately represent damage initiation and failure of whole healthy and osteolytic vertebrae based on histological, strain based and contrast enhanced μct damage quantification. Determine thresholds for damage initiation based on these models. 1.1 Layout of the Thesis The second chapter of this document includes a background of key concepts as well as current research relevant to this dissertation. The third chapter investigates the ability of ex vivo modelling to represent the in vivo behavior using image-based strain analysis (aim 1). It is modelled after a typical scientific paper, with an abstract, a brief introduction, methods, results and discussion, including comparisons with similar studies, strengths and weaknesses and significance of the findings. The fourth chapter again follows a paper based format and presents a technical evaluation of microdamage visualization techniques in whole bone (aim 2). The fifth chapter uses the protocol developed in chapter 4 to evaluate the ability of micro finite element modeling to represent microdamage in healthy and metastatic vertebrae (aim 3). The sixth chapter summarizes the most important conclusions of the project as a whole and provides recommendations for future research.

16 4 Chapter 2: Literature review 2.1 Bone structure and composition Bone is the primary structural component of the human skeletal system. In addition to supporting the body, bone serves many important functions, such as protection of vital organs, providing framework for movement and flexibility, sheltering the bone marrow and acting as a calcium reservoir. The strength of bone and its ability to resist fracture depends upon its quality, which is determined by tissue level material properties and architecture. The bone matrix is composed of organic and inorganic (mineral) phases, which include hydroxyapatite, collagen and small amounts of proteoglycans, non-collagenous proteins and water (Olszta 2007). The exact composition of bone varies with age, sex, type of bone and pathological conditions (Doblare 2004). The organic phase, the matrix of the bone tissue, is composed of 90% type 1 collagen along with other proteins and proteoglycans in addition to bone cells, growth factors and cytokines. The inorganic phase is predominantly made up of hydroxyapatite, precipitated on the organic bone matrix (Post 2010). The organic matrix and the inorganic phases together give the bone its characteristic strength and toughness. Based on its structural properties, bone can be characterized into two types cortical and trabecular. Cortical bone is dense with a very low porosity (less than 10%). High cortical bone density provides resistance to compression, bending and torsion. Cortical bone is arranged in layers of lamellae (Figure 2.1). In larger animals, concentric lamellae surrounding Haversian canals form cylindrical osteons, which are the structural and functional units of cortical bone in the diaphysis of long bones. Each osteon is surrounded cement line, made primarily of minerals, which resists fracture progression in cortical bone. The osteons are densely packed together along the longitudinal axis of the cortical bone (Olszta 2007). Such an arrangement makes cortical bones stronger along the longitudinal axis, as compared to the transverse axis. Cortical bone has also been found to be stronger in compression than in tension. Cortical bone demonstrates viscoelastic behavior under load and is stiffer and stronger at higher loading rates (Olszta 2007).

17 5 Trabecular bone is a three dimensional matrix comprised of interconnecting plates and rods, which are individually termed as trabeculae (Donnell 2006). Trabecular bone is located at the ends of long bones (tibia, femur, etc.) and within irregular bones (spine and pelvis). The highly porous trabecular matrix houses vascularized and nutrient rich bone marrow (Figure 2.1), where blood cells undergo differentiation. Trabecular bone has lamellar structure similar to cortical bone, with the lamellae running parallel to the trabeculae. The porous nature of trabecular bone yields an optimized weight to strength ratio for bone (Keaveny, 2001). The high porosity of the trabecular bone allows large plastic deformations under compressive loads. Owing to its porous matrix filled with liquid marrow, trabecular bone demonstrates poroelastic behavior under load and is stiffer at higher strain rates (Ochoa 1991). The mechanical properties of trabecular bone depend on the bone quality and morphology, and vary with the location within the body (Keaveny, 2001). Figure 2.1: Difference between the structures of cortical and trabecular bones. The inner trabecular bone is much more porous than the outer compact bone. (Public domain image from Wikimedia Commons).

18 6 2.2 Bone remodeling The structure and composition of bone is sensitive to its mechanical environment. Bone is continuously in a state of flux, being remodeled by a process of absorption and formation. According to the Wolff s law, bone remodels itself over time to adapt to biomechanical stimuli (Frost 1996, Wolff 1892). This process is accomplished by the coupling of three types of specialized cells which live within the bone matrix (Clarke 2008). Osteocytes: reside within the matrix to regulate bone remodeling Osteoblasts: bone synthesizing cells Osteoclasts: bone dissolving cells Osteocytes are terminally differentiated osteoblast cells, which become embedded in the bone matrix as it is being formed. These cells play an important role in maintenance and regulation of the bone microenvironment. Osteocytes convert the mechanical sensations of stress or bone damage to electro-chemical signals through mechanotransduction, which recruit mononuclear osteoclast precursors from circulation, to start remodeling of the bone. These precursor cells combine to form large multi-nucleated osteocytes, which digest old bone matrix using enzymes. For the subsequent bone formation, regulatory factors initiate the differentiation of mesenchymal progenitor cells to osteoblast. Osteoblasts are then recruited to fill the cavities with new bone minerals and matrix proteins. Osteoblasts buried within the newly formed matrix differentiate to become osteocytes which are connected to bone surface lining cells and other osteocytes through an extensive canalicular network (Clarke 2008, Manolagas 2000). This network plays an essential role in monitoring the bone tissue and initiating bone remodeling. The perpetual remodeling of the bone preserves the structural and mechanical integrity of bone throughout life. Factors such as age and pathological conditions (i.e. osteoporosis or metastasis) can interfere with natural bone remodeling process to alter the structure and mineral composition of bone, thus affecting its mechanical integrity.

19 7 2.3 Anatomy of Vertebral column The vertebral column (spine) is one of the most important sections of the human skeletal system. The series of vertebrae, connected through soft tissue, ligaments and intervertebral discs together form the flexible vertebral column, which houses the spinal cord, provides structural support for the maintenance of proper posture and enables flexible motion of the upper body (Seeley 2006). Being a primary load bearing component of the body, the vertebral column has been a key area of research in field of Biomechanics. The human spine is divided into three segments cervical, thoracic and lumbar. The cervical section, consisting of 7 vertebrae (C1-C7), lies in the neck region and is mainly responsible for the protection of the brain stem and support and motion of the head. Beneath the cervical section, lies the thoracic spine made up of 12 vertebrae (T1-T12). The main function of this section is to protect heart and lungs with the help of the rib cage. The 5 lumbar vertebrae (L1-L5) below the thoracic region are the greatest weight bearing components of the spine (Figure 2.1). Although the individual vertebrae of the spine vary in shape and size, the overall structure is similar. The primary load bearing component of a vertebra is the vertebral body. The load bearing ability of the vertebral body depends on its size. The size of the vertebral body progressively increases downward, with average strength of 2000N in the cervical spine to 8000N in the lumbar segment (Izzo 2013). The posterior elements of the vertebra form the vertebral arch, which consists of pedicles, laminae and three types of processes. The vertebral body, together with the pedicles and the laminae form a canal that protects the spinal cord. The articular processes connect with those of adjacent vertebra to form facet joints. Between two vertebrae lies the intervertebral disc which acts like a shock absorber. The disc is made up of made up of two parts. The central portion of the disc, the nucleus pulposus, is a gel like elastic substance made up of water, collagen (predominantly type 2) and proteoglycans. Surrounding the nucleus pulposus is a layered composite structure called the annulus fibrosis, made up of collagen fibers (predominantly type 1). The facet joints and the intervertebral discs together allow for the flexible movement and load transfer in the spine (Figure 2.2) (Seeley 2006).

20 8 Figure 2.2: Anatomy of human spine; (left) all levels of the vertebral column, (centre) an individual vertebra with vertebral body and posterior elements, (right) series of 3 vertebrae displaying facet joints and intervertebral discs (Public domain image from Wikimedia commons) 2.4 Spinal metastasis Metastasis is the migration of cancer from its origin to a new location in the body. Bone is one of the most frequent sites of metastasis. Breast, lung, prostrate and renal cancers are the most common types of cancers to metastasize to bone. The spine is the most frequent site of metastasis in the skeleton owing to its excessive and nutrient rich trabecular network. Approximately 1/3 of all cancer patients are diagnosed with spinal metastases (Naishadham 2012). Such patients suffer from significant consequences in terms of morbidity and pain originating from microfractures, burst fractures, nerve root infiltration, bone distortion and collapse (Raele 2001). Metastatic involvement in the spine results in the decoupling of healthy osteolytic and osteoblastic cell interaction, which not only alters the bone turnover, but also affects the bone density and architecture (Guise 2001). The disturbance of the intricate balance of remodeling can lead to alterations in the structural and material properties of bones, which can severely affect their load bearing properties. Osteolytic tumors up-regulate the production of

21 9 Figure 2.3: Various types of spinal metastasis. A: Osteolytic, B: Osteoblastic, C: Mixed Focal, D: Mixed diffuse (Skrinskas 2009) osteoclasts, which leads to an increase in bone resorption. This results in increased porosity, reduced bone density and the replacement of mineralized bone by soft tumor tissue, with a consequent increase in the risk of fracture. Osteoblastic tumors increase the genesis of osteoblasts, which leads to excessive bone deposition. Although osteoblastic tumor growth leads to increased amounts of bone, abnormal bone quality and micro structure leads to lower yield strength (Coleman 2001). Osteolytic tumors are characteristic to breast and lung cancer metastasis while osteoblastic lesions are frequently diagnosed in prostate cancer metastasis (Coleman 2001). However, many lesions are mixed exhibiting both osteolytic and osteoblastic components, with both diffuse and focal damage (Figure 2.3) Fracture risk Skeletal related events (SRE s) is a collective term for describing conditions such as hypercalcemia, pathological fracture, severe pain, spinal cord compression and bone instability, which occur as a result of metastatic involvement in bone (Von Moos, 2013). Close to two-thirds of patients with bone metastases develop at least one SRE (Lipton 2000). Recent advancement in cancer diagnostics and treatments has increased the life expectancy of cancer patients. However, this also leads to increased chances of complications from the metastatic involvement (Von Moos, 2013). Given the importance and high incidence of SREs, strategies to reduce the burden of SREs, and in particular fractures, are imperative. The most common fracture type resulting from osteolytic disease in the spine is a compression or wedge fracture, which results in the collapse of the anterior wall of the vertebral body (Wong 2013). Burst fractures, which can occur under high impact loading in

22 10 normal spines, cause the collapse of the posterior wall of the vertebral body. Burst fractures can also occur under normal physiologic loading conditions in osteolytic vertebrae, and may lead to neurologic complications arising from bone fragments or tumor tissue penetrating into the spinal canal (Whyne 2003). From 5-10% of all cancer patients suffer from spinal cord and nerve root injuries originating from spinal metastasis (Constans 1983) Clinical interventions for spinal metastasis Currently, bone metastasis is treated by utilizing a multimodal approach that may include radiation therapy, chemotherapy, surgical removal and systemic treatment with bisphosphonates (BP) (Bilsky 2005, Rades 2010). Historically most breast cancer metastases were diagnosed to be osteolytic in nature; however, with the introduction of modern cancer therapies (i.e. BP s) the relative incidence of osteolytic, osteoblastic, and mixed vertebral lesions is changing (Curtis 2007). The implications of new systemic therapies, on the pattern of disease in the metastatic spine are important in focusing new initiatives aimed at structural analysis and fracture risk assessment to ensure relevancy in today s patients with metastatic breast and other types of cancers. Vertebroplasty and kyphoplasty are performed to provide stabilization to the diseased spine to prevent fractures. These techniques include percutaneous insertion of polymethylmethacrylate in the vertebral body, which hardens to alleviate pain and provide structural support to the compromised spine (Bhatt 2013). Strength assessment and fracture risk prediction in the metastatic spine are of significant clinical importance as prevention of SREs in high risk patients may be possible through use of external bracing or surgical stabilization. Bone mineral density (BMD) is the standard clinical measure of mechanical strength of the bone. BMD is primarily measured using dual X-ray absorptiometry (DXA), and more recently using quantitative Computed Tomography (qct) (Engelke 2012). In recent years, advanced image-based techniques such as finite element analysis (see section 2.6), hip structural analysis and trabecular bone score have been developed, which account for bone quality and architecture, and are used in combination with BMD to provide more accurate biomechanical analysis of the metastatic spine (Engelke 2012).

23 Pre-clinical models of spinal metastasis Numerous pre-clinical in vivo and in vitro models have been developed to study spinal metastasis. Researchers have used human and animal cadaveric spine samples with simulated focal lesions to represent spinal metastasis (Dimar 1998, Ebihara 2004). Pre-clinical animal models of skeletal metastases may be utilized to represent pathological changes in bone, but these do not fully represent human anatomy or pathology (Blouin 2005, Cossigny 2012, Goldstein 2010, Singh 2005, Yoneda 1999). These models, however, can yield close to realistic estimates of the impact of tumor burden in the skeleton. Since the behavior of different tumor cells varies widely, the use of growing human cancer cells in animals may best represent the behavior of the particular cells in the human body and their subsequent response to treatment. Recent work in our group has successfully employed bioluminescence transfected HELA cells (previously thought to be human breast cancer cells) to produce osteolytic metastasis in rnu/rnu nude athymic rats. This model, along with an ACE 1 canine prostate cancer model for osteolytic and mixed metastatic disease, has been used in our laboratory to examine novel treatments and associated effects on fracture risk (Hojjat 2011, Herblum 2013, Hojjat 2012, Lo 2012). 2.5 Trabecular biomechanics of the vertebral body Being the primary load bearing component, the vertebral body is the biggest and the most important section of the vertebra (Izzo 2013). The vertebral body is composed of a thin cortical shell, wrapped around a trabecular meshwork, called trabecular centrum (Figure 2.4). The loads on the vertebral body are distributed between the cortical shell and the trabecular centrum. Although various factors (age, disease, shape of the vertebrae, etc.) dictate the load sharing between the shell and the centrum, recent studies have shown that cortical shell accounts for only about 10% of vertebral strength (Prakash 2007, Silva 1997). Microarchitecture of trabecular centrum has been shown to play an important role in mechanical resistance, especially in the spine (Cortet, 2001). In a study by Fields et al., the coefficient of determination of vertebral strength between finite element model and biomechanical testing improved from r 2 = 0.57 to r 2 = 0.85, when accounted for both BMD

24 12 and microarchitecture, as opposed just BMD (Fields, 2009). This underlines the importance of the trabecular morphology to the load bearing capacity of the vertebral body. Although multiple studies have shown the importance of microarchitecture, its exact contribution is not very well understood due to variability in vertebral shape, size and bone mass. Figure 2.4: µct image slice showing Cross section (frontal plane) of a rat L1 vertebral body, with a cortical shell surrounding the trabecular centrum Trabecular microarchitecture in vertebral body The vertebral trabecular centrum is comprised of interconnected rod-like and plate-like structures. The ratio of rods to plates in the trabecular bone is site specific, and influences local mechanical properties and failure mechanisms (Keaveny 2001). Failure in rod-like structures occurs mainly due to bending and buckling, followed by collapse. On the other hand, failure can occurs instantaneously in plate-like structures, without bending or buckling (Muller, 1998). Human vertebral trabecular bone is more prone to bending and buckling, given its higher concentration of rod-like trabecular elements (Hildebrand, 1999). The trabecular structure of the human vertebral body is anisotropic, and has a heterogeneous

25 13 architecture and density in anterior-posterior and vertical-transverse directions (Banse 2001), although some left-right symmetry is observed in the lumbar vertebrae (Simpson 2001). The trabeculae are thinner and more densely packed near the end-plate regions, compared to the central region of the vertebral body (Simpson 2001). These features can also be generally observed in rat vertebral bodies (Figure 2.4). Various quantitative parameters have been developed to study the structural morphology of the bone. Microarchitectural measurements include the width, number, separation of trabeculae, in addition to their spatial organization (Carbonare 2005). Most common morphological measurements include trabecular bone volume, trabecular thickness, trabecular number, trabecular spacing and degree of anisotropy (Donnell 2007). These stereological parameters have been extensively studied through the use of 2-D histomorphometry (Carbonare 2005). Alternatively, non-invasive imaging techniques such as high resolution micro-ct and micro-mr imaging allow non-invasive measurement of the trabecular structure in 3D (Rizzoli 2010). In a recent investigation, Hojjat et al. implemented automated algorithm on micro-ct images to quantify and compare microarchitectural parameters in healthy and osteolytic vertebrae. Significant decrease in trabecular bone volume, trabecular thickness and trabecular number and significant increase in trabecular spacing were observed (Hojjat 2011), demonstrating compromised microarchitecture in metastatic vertebrae. The trabeculae are preferentially aligned along the axis of loading, to adapt to the load applied. Consequently, human vertebral body consists of higher number of vertically aligned trabeculae (Keaveny, 2001). The axial compressive loads on the vertebral body are first accepted by the vertical struts, which transmit loads between the end-plates. The horizontal trabeculae allow dispersion of the loads, and prevent the buckling of vertical ones. However, in the presence of osteolytic metastasis, the vertical columns are progressively thinned as well as elongated due to the resorption of the horizontal lamellae (Hojjat 2011). The resistance of a column decreases by the square of increasing length and by the square of decreasing cross section (Izzo, 2013). Similarly, bone loss also leads to a decreased connectivity and increased trabecular spacing. This causes an overall degradation of the trabecular microarchitecture, leading to an increased risk of bone failure. Osteoblastic tumors

26 14 lead to increased bone deposition and bone density; however, decreasing bone material properties and abnormal microstructure result in weaker structural properties (Hojjat 2012) Microdamage in trabecular bone Under sufficient loads, damage to the trabecular bone can be observed in the form of microdamage or microfractures. Microdamage can present as linear (parallel cracks along lamellar surface or cross-hatched patterns across the lamellae) or diffuse damage (Fyhrie, 1994). Microfractures, or complete fractures of trabeculae, occur much less frequently and are the end result of the accumulating microdamage spanning across the trabecular thickness (Yeh, 2001). Linear microdamage is caused by compressive stress, while diffuse damage is generally a result of tensile loading (Vashishth, 2000; Wenzel, 1996). The occurrence and repair of microdamage is a part of the healthy bone remodeling process. However, unrepaired bone microdamage has been demonstrated to be a contributing factor to skeletal fragility and the accumulation of microdamage with increasing age or disease condition (metastasis or osteoporosis) causes weakening of the bone and an increased risk of fracture even during normal physical activities (Frost, 1960). According to a study by Yeh et al., widespread accumulation of microdamage within trabeculae, and not microfractures, is a more likely explanation for the reduction of apparent strength and stiffness of the trabecular structure after an isolated overload (Yeh, 2001). Burr et al. have also demonstrated that diffuse damage correlated linearly with modulus loss; whereas linear microcracks had a quadratic relationship with modulus loss (Burr, 1998). Clinically, microdamage accumulation has also been identified as a major risk factor for bone fracture (Iwata 2014). In spite of the significance of trabecular microdamage to the mechanical health of the bone, much is still to be understood about the relationship between local damage events and microstructural stresses and strains Mechanical properties of trabecular bone Apparent or continuum level analyses take into account whole bones or specimens with multiple trabeculae (Morgan 2005). From an engineering perspective, continuum trabecular bone forms a composite cellular solid with viscoelastic properties. The density and structure of the trabecular bone varies across age, species and anatomical site, leading to heterogeneity

27 15 in mechanical properties (Keaveny 2001). The elastic modulus and yield stress of the trabecular bone are directly proportional to the apparent bone density. Other important factors include trabecular orientation and anisotropy ratio (Keaveny 2001). Strain, being the ratio of yield stress and elastic modulus, forms a much simpler criterion for bone failure. Bone in general has been shown to yield consistently at an apparent strain of approximately 1% (Keaveny 2001). As a result, strain based analysis is very important for biomechanical study of bones. The mechanical behavior at trabecular or tissue level is different from the apparent level in the absence of architectural considerations. The tissue level properties can be calculated using various techniques such as nanoindentation (Zysset 1999), atomic force microscopy (Kinney 2000) and back calculation (Rietbergen 1995). The local compressive yield stresses and strains for trabecular bone are included as table 2.1. Various micro-mechanical studies have found the tissue modulus for bone to lie between GPa (Keaveny 2001). Table 2.1: Local compressive yield stresses and strains for trabecular bone Authors Bone type Yield stress (MPa) Yield strain (%) Bayraktar 2004 Niebur 2000 Verhulp 2008 Bevill 2006 Bevill 2008 Human femoral trabecular bone Trabecular bone cores from steers Trabecular bone from bovine tibia Human femoral trabecular bone Human Vertebral trabecular bone Biomechanical analysis of bone Biomechanical investigation of diseased spines from the pre-clinical animal models includes experimental, image-based and computational analyses (Dimar 1998, Ebihara, 2004, Hong 2004, Snyder 2006).

28 Mechanical testing Mechanical testing has been widely implemented for structural analysis of healthy and diseased spines (pre and post-treatment). Over the course of mechanical testing, forcedisplacement curves are obtained, which are then used to calculate mechanical parameters such as ultimate force, stress and stiffness (Ebihara, 2004, Hong 2004, Lo 2012). Compressive loading through the intervertebral discs better represents the healthy physiologic loading scenario in the spine and allows for accurate load distribution to be applied to the vertebral endplates. Vertebral motion segments containing three vertebrae and intact posterior elements can be used to generate physiological loading in the middle vertebra. Mechanical testing has shown trabecular bone of the vertebrae to be more prone to initial failure as compared to cortical bone, with the regions near the endplates exhibiting the highest risk (Eswaran 2007) Microdamage analysis through staining A variety of techniques have been developed to detect microdamage in bone (Lee 2003). Healthy bone contains pre-existing microdamage, originating from normal physiological loading. As such, sequential staining of two site specific stains is required to differentiate pre-existing and test induced microdamage (Lee 2000). Microdamage in bone cleaves bonds in the bone matrix, exposing new surfaces with charged ions, 55% of which are Ca 2+ ions (Lee 2000). Various stains can act as chelating agents, which can bind with Ca 2+ to form stable rings. Stains such as calcein, xylenol orange and alizarin complexone can bind with Ca 2+ ions to label the micro cracks. Bone samples can be labeled with different stains before and after mechanical loading to detect load induced damage. Sequentially stained bone samples can be fixed and sectioned followed by analysis with bright field and fluorescent microscopy. These sequential techniques have been shown to successfully label new microdamage (Lee 2000). However, use of multiple chelating agents for sequential staining has been shown to cause dye replacement (Lee 2000). En bloc staining of bone specimens using basic fuchsin hydrochloride has demonstrated success in microdamage identification (Burr 1995). Sequential staining with a chelating agent and fuchsin staining can circumvent

29 17 the problem of stain substitution. Herblum et al. has illustrated the use of calcein green and fuchsin staining to differentiate between pre and post loading damage (Herblum 2013). Although robust, such staining techniques are two dimensional and destructive. Nondestructive 3-D alternatives would enable spatial correlation of damage within whole samples. However, micro cracks on their own cannot be resolved by commercially available micro-computed Tomography (µct) systems (Landrigan 2011). Barium sulfate (BaSO4) contrast enhanced µct imaging has been demonstrated to detect accumulation of microdamage in trabecular and cortical bone (Landrigan 2011, Turnbull 2011, Wang 2007). Samples are immersed in barium chloride followed by sodium sulfate under vacuum for a fixed amount of time. Barium and sulfate ions diffuse and collect in void spaces (microcracks), where they precipitate to form BaSO4. Measurements of contrast enhanced microdamage in bone have been validated using scanning electron microscopy (Wang 2007) and basic fuchsin staining/histology (Landrigan 2011, Turnbull 2011). However in this, separate specimens were used to compare BaSO4 and fuchsin staining (the two techniques were not applied to the same samples). Compared to histology, BaSO4 demonstrated greater variability in staining micro cracks, as it was also found to be collected in voids and on free surfaces (Turnbull 2011). Optimization of parameters such as staining times and concentrations can minimize such non-specific staining. Contrast enhanced µct does provide a non-destructive and 3D alternative to conventional histology Finite element analysis Numerical solutions to very complex problems in structural mechanics can be obtained using finite element (FE) analysis. FE analysis involves discretization of a complex model into components or elements with simple geometry. The response of the overall mathematical model is then approximated by summation of the responses obtained at each individual element. Generation of FE models requires proper mesh geometry, material properties and boundary conditions (Logan 2012). The FE method finds wide applications in skeletal biomechanics, as it allows parametric representation and analysis of complex geometric and material property distributions. FE analysis has also been used successfully to predict fracture and damage in healthy and diseased bone matrix (Lotz 1991, Whyne 2003).

30 18 Conventional continuum models consider the bone sample as a whole. Tissue properties may be assigned to be homogeneous or as a function of image-based density. However, continuum models ignore the trabecular level microarchitecture. This limits the ability of such models to resolve the mechanical behavior of individual trabeculae in healthy and metastatic regions of a diseased bone. Micro-FE (µfe) models can be generated by directly converting the voxels of µct data sets into finite elements representing trabecular architecture (Figure 2.5). This allows incorporation of the actual trabecular morphology into the FE model. However, this leads to a tremendous increase in memory requirements and computational time. As a result, µfe has most often been applied to assess the mechanical properties of bone cores (Gong 2007, Kim 2007, Neiber 2000, Zauel 2006). A few authors have applied this technique to whole human (Ito 2007) and rat bones (Herblum 2013). Models have utilized isotropic elasticity to represent the bone tissue with marrow space modeled as void to generate meshes with 8- noded hexahedral elements (Herblum 2013, Nagaraja 2007, O Neal 2010, Rhee 2009, Verhulp 2008). An element size of 1/4 th of the mean trabecular thickness has been shown to yield acceptable balance between accuracy and computational cost (Guldberg 1998, Ladd 1998, Neiber 1999). Simplification of loading and boundary conditions to minimize computation expense has been accomplished in spinal motion segments using image registration algorithms based on comparisons of loaded and unloaded μct scans. These algorithms can generate vector fields which can be utilized as boundary conditions at the end plates and facet joints (Herblum 2013, Nagaraja 2005, 2007).

31 19 Figure 2.5: High resolution micro-ct image of trabecular specimen (left), converted into a voxel based micro finite element mesh (right) (Keaveny 2001) The output parameters from µfe models typically include von Mises and principal stresses and strains. According to Von Mises failure criterion, the material under multi-axial loading will yield when the distortional energy is equal to or greater than the critical value for the material. However, von Mises cannot differentiate between tensile and compressive stresses and works best under multi-axial loading (Nagaraja 2005). Principal stress and strain have been used extensively in µfe studies based on axial compressive loading. A measure of principal strain in bone is of particular importance owing to its independence with respect to bone density Image-based strain analysis Deformable image registration of micro imaging datasets (such as µct) is another 3D alternative for measurements of experimental strain and displacement fields in bone samples. Such digital volume correlation methods track the deformation of microstructural features and patterns within unloaded and loaded image volumes to yield full field strain measurements (Roberts 2014). This approach has found applications in studying strains and

32 20 strain patterns in metal foams (Smith 2002), agarose gel (Franck 2007), rocks (Lenoir 2007) and collagen (Roeder 2004). Hardisty et al. extended this technique to study strain distributions within whole rat vertebrae. The multi-resolution algorithm compared small subsets of image data from the loaded and unloaded images to generate displacement fields, which were then used to calculate strain (Figure 2.6) (Hardisty 2009). Image registration has several advantages over commonly used strain analysis techniques such as strain gauges and FE modeling. Strain gauges are limited to measurement of surface strains (Roberts 2014) and application of strain gauges to small bones and curved surfaces, such as rat vertebrae, presents a challenge. In contrast, image registration provides continuum level strain fields throughout a volume and can be used to analyze small samples. In addition, unlike FE analysis, image registration does not require definition of a mesh, material properties assumptions or boundary conditions to determine strain (Hardisty 2009). However, image-based strain registration requires an extensive amount of computational time and memory. The limited resolution of such algorithms also makes it challenging to resolve strains at an individual trabecular level. Future studies directed towards improved computing capabilities and developing more sophisticated feature tracking algorithms may permit the employment of digital volume correlation to study the mechanics of bone at a trabecular level. Figure 2.6: Deformable image registration algorithm determines strain by aligning and comparing scans of the same sample with and without load (Hojjat 2011).

33 21 These techniques have been used extensively and with great success in studying the ex vivo biomechanics of healthy and diseased bones. However, the ability of such techniques to accurately reflect the in vivo behavior still needs to be verified. Such a verification will justify the use of the ex vivo analyses to study the physiological behavior of bone. 2.7 Summary Osteolytic metastasis in the spine leads to a decreased bone content and deteriorated microarchitecture, which results in an increased risk of microdamage accumulation and fracture initiation. While the effects of bone mineral density and architecture have been studied extensively with respect to fracture risk in the metastatic spine, the micromechanics of yielding and failure (damage) have received less attention, representing an attractive area of research bridging the current knowledge gap. A better understanding of local damage and failure properties in the affected vertebrae is vital for the improvement of fracture risk assessment. A combination of mechanical loading, histological damage labeling, µct imaging based techniques and µfe methods, which incorporate the trabecular level morphology of the bone, have a potential to provide more robust models and be able to better represent the complex mechanics of the metastatic spine at the tissue level. Quantifying the structural behavior of healthy and metastatically-involved vertebrae through advanced experimental, computational and imaging based methods forms the central theme of the current project. This will enhance our understanding of the potential impact of such techniques and their potential to describe the microstructural behavior of the metastatic spine. Together these methods may elucidate the microstructural behavior of trabeculae within healthy and metastatic environments, which will be helpful in guiding the future use of computational analysis in structural assessment of vertebral strength.

34 22 Chapter 3: Post euthanasia micro-ct based strain analysis is able to represent quasi-static in vivo behavior of whole vertebrae. 3.1 Abstract 3D strain measurement in whole bones allows representation of physiological, albeit quasistatic, loading conditions, however such work to date has solely been performed on specimens post mortem. The main purpose of this study is to verify the efficacy of post euthanasia strain based analysis to characterize the in vivo mechanical behavior of rat vertebrae. A μct compatible custom loading device was used to apply 75N load to a 3-level rat tail motion segment of a healthy rat. Multiple loaded and unloaded μct scans were acquired before and after sacrificing the rat. A 3-D volume correlation method which employs registration of 2 µct images of the same specimen under similar or different loading conditions was used to calculate strains in live and post mortem vertebrae. No significant difference was found in the in vivo strains (-0.011±0.001) and ex vivo strains (-0.012±0.001) obtained from the comparisons of loaded and unloaded images (p=0.3). Comparisons between unloaded-unloaded and loaded-loaded scans yielded significantly lower axial strains as expected. Qualitatively, high strains were observed adjacent to growth plate regions in comparing the loaded and unloaded images. Strain patterns in the loadedloaded and unloaded-unloaded scans were inconsistent as would be expected in representing noise. Overall, live and dead loaded to unloaded comparisons yielded similar strain patterns and magnitudes. This study demonstrated a proof of concept, suggesting that post euthanasia µct based strain analysis is able to represent the in vivo quasi static behavior of rat tail vertebrae.

35 Introduction Strain measurement has been extensively employed for ex vivo assessment of bone failure (Doblare, 2004; Kopperdahl, 1998). Experimentally, strain has been primarily measured via strain gauges attached to bone samples undergoing biomechanical testing (Cristofolini, 2013) yielding measurements of local surface strains. Image registration is an alternative nondestructive approach which can spatially resolve full strain fields in three-dimensions. 3D image registration of micro-imaging data sets, such as µct, has been utilized to calculate strain in excised cortical bone and trabecular bone cores (Bay, 1999; Christen, 2012; Nagaraja, 2011; Lynch; 2004; Waarsing, 2004). Such image registration algorithms have also been validated to accurately measure and spatially resolve strain in 3D in rat whole bones using µct imaging (Hardisty, 2009). In a recent study, the efficacy of applying image registration to whole human vertebrae was also investigated (Hussain, 2012). These algorithms work under the premise that strain fields can be computed by aligning and comparing images of a sample acquired under unloaded and loaded configurations (Figure 2.5). During registration, the unloaded image can be defined as a grid of nodes, each of which has a spatial region defined around it. These nodes are then mapped to the corresponding nodes in the loaded images by matching intensity of the spatial regions around the nodes. The nodal displacements are then used to obtain the strain field (Roberts, 2014). Prior researchers have used this approach to measure the effect of the growth plate stiffness (Hardisty, 2010), metastasis (Hardisty, 2012) and treatment (Hojjat, 2011) on bone strain in whole ex vivo rat vertebrae. However, such algorithms have primarily been utilized to study cortical and trabecular bone specimens. Application to whole bones allows the representation of more physiological, albeit quasi static, loading conditions. Also, most biomechanical analyses to date have solely been performed on specimens after sacrifice. The aim of this study is to demonstrate the ability of post euthanasia modeling to represent the in vivo quasi static behavior of whole vertebrae through image-based strain analysis. It is hypothesized that post euthanasia µct image-based strain analysis represents the in vivo quasi static behavior of whole bones.

36 Methods Deformable image registration algorithm An intensity matching deformable image registration routine was previously developed and validated (Hardisty, 2009). This method uses registration of loaded and unloaded µct images of the same sample to calculate strain under an applied load. The algorithm was coded in C++ using the Insight Toolkit (ITK) and was implemented as a plug-in to AmiraDEV (FEI Visualization Science Group, Burlington, USA). In the first step of the algorithm, the unloaded and the loaded scans are registered using affine mapping, which facilitates transformation using rotation, scaling, shearing and translation (12 degrees of freedom). The initial deformable registration proceeds iteratively, optimizing the fit based upon normalized mutual information metric. The unloaded scan is then split along the three axes, yielding 8 pieces. Each piece is individually registered to the loaded scan, using the affine transform from the previous registration as the initial guess. The division and registration of sub-pieces then continues until a user defined maximum level is reached. After the final level of registration, the affine transform is used to calculate the displacement of the center of each registered sub-region. Continuum level displacement and strain fields are then calculated using the following set of equations. A(x,y,z) = TP(x,y,z,) P(x,y,z) e = ½ ( A T + A) where e is the strain matrix, A is the displacement vector, T is the affine transform found from the registration, and P is the voxel location Animal model One healthy female Sprague Dawley rat (17 weeks old, 300g) was used for this study. In vivo loaded μct imaging was performed on the live rat under general anesthesia (4% isoflurane/oxygen). After euthanasia (via intercardiac injection of 120mg/kg Euthanyl), dead loaded and unloaded μct images were acquired. All animal work was performed under institutional animal care committee approval (University Health Network, Toronto, Canada).

37 Loading A μct compatible custom loading device was used to apply live and dead axial compressive load of 75 N to a 3-level vertebral motion segment in the rat tail. This load was previously demonstrated to induce detectable strain in the vertebrae without fracture (Hojjat 2011). A radiolucent jig, made from polycarbonate tube (2.5cm diameter and 20cm long), was attached to the 5 th to 7 th caudal vertebrae in the anaesthetized rat via percutaneous pins (2 per level) and attached to loading rings (Figure 3.1). The loading rings fit within a pre-calibrated spring based loading device. The device was calibrated by loading select specimens in line with an ELFM-T2M, 500 N capacity load cell (Entran Devices Inc., Fairfield, USA) and via calibration within a material testing machine (MTS Bionix 858, Eden Prairie, USA) instrumented with a 250 lb (113.4 kg). load cell in a specialized calibration set up. Load was applied to the rat tail manually through the threaded tube. The loading rate was not continuous, but rather, simulated as quasi-static. Prior to the load application, preconditioning was performed using increasing load amounts, until the desired load (75N) was stabilized. After every load application or removal, the load on the tail was allowed to stabilize prior to scanning Figure 3.1: Micro-CT compatible loading jig used to apply an axial compressive load to the 6th caudal vertebra in the rat tail.

38 Imaging With the device in place, μct scans of the vertebral motion segment were acquired adjacent to bone density phantoms, with an average scan time of 90 minutes (X-ray source 80kV, 13.3μm isotropic voxel size, Inveon CT, Siemens Healthcare, Erlangen, Germany). A total of 6 μct scans were acquired in the given sequence and configurations: live under load (1), dead under load (2, 3), dead unloaded (4, 5) and dead under load (6). Post euthanasia scans were acquired after sacrifice Strain calculations and data analysis The μct images were reconstructed and the intensities rescaled based on the bone phantoms. The middle vertebra from each scan was cropped and the bone regions segmented using an intensity threshold. The resultant removal of background noise from the segmentation allowed for better registration. Each loaded scan was registered with each unloaded scan to determine the load induced strain. Registration of scans under similar load configurations (loaded-loaded and unloaded-unloaded) was also performed to determine the error in the method. Prior to registration each pair of scans was first manually aligned in 3D. A built-in module (AffineRegistration) in AmiraDEV was then applied to the data sets to achieve accurate alignment. Deformable registration was then implemented using four levels of subdivision. Axial (zz) strain (percent) was evaluated as the primary outcome variable. The axial strain distributions in the rat-tail vertebrae were characterized by calculating the mean strain, median strain and the 10 th (minimum) and 90th (maximum) percentile of the strain. T-tests were used to evaluate differences between mean, median and 10 th and 90 th percentile strain values in considering live to dead and dead to dead images. Since multiple T-tests were performed, Bonferroni correction was applied (significance level adjusted to α=0.016). Qualitative comparisons were also performed based on the generated strain contours within the vertebrae.

39 Results The mean, median, 10% and 90% strains obtained from the registration of the acquired images are presented in Tables 3.1 and 3.2. In comparing loaded-unloaded scan configurations (1-4, 1-5, 2-4, 2-5, 6-4 and 6-5) the applied load generated mean axial compressive strains with an average value of ± Comparisons between unloadedunloaded (4-5) and loaded-loaded (1-2, 1-3, 1-6, 2-3, 2-6 and 3-6) scans yielded significantly lower mean axial strains as expected, representative of the error in the method (Figure 3.2a). No significant difference was found in comparing live (1-4, 1-5) or dead (2-4, 2-5, 3-4, 3-5, 6-4 and 6-5) images with respect to the mean strains generated in the loaded vs. unloaded configurations (p=0.21) (Figure 3.2b). In comparing images acquired under equivalent loading conditions, (loaded-loaded and unloaded-unloaded) similar strain errors were found in assessing dead to dead (2-3, 3-6 and 2-6) images vs. dead to live (1-2, 1-3 and 1-6) images (Figure 3.2c). However, loaded-loaded comparisons yielded significantly higher mean strains than the unloaded-unloaded comparison. The median and 90% strain values were observed to follow the same pattern as the mean strains. The average 10% strains for all comparisons were not significantly different. Table 3.1: Strains (µm/µm) obtained from the comparisons of loaded-unloaded images Live-loaded to deadunloaded Dead-loaded to deadunloaded Mean strain 10% strain Median strain 90% strain ± ± ± ± ± ± ± ±0.004 Table 3.2: Strains (µm/µm) obtained from the comparisons of images under similar loading conditions Mean strain 10% strain Median strain 90% strain Live-loaded to deadloaded ± ± ± ±0.004 Dead-loaded to deadloaded ± ± ± ±0.003 Dead-unloaded to dead-unloaded

40 Mean Axial strain (µm/µm) Mean axial strain (µm/µm) Mean axial strain (µm/µm) Loaded-loaded and Unloadedunloaded * Live loaded-dead unloaded Dead loaded-dead unloaded Live to dead Dead to dead a) b) c) Figure 3.2: Mean axial strains (µm/µm) obtained from deformable registration (* represents p-value<0.016). a) Strains obtained from all loaded-unloaded and all loaded-loaded/unloaded-unloaded comparisons. b) Comparison of live and dead strains. c) Live and dead strains for scans under equivalent loading conditions Qualitatively, the strain patterns were similar in comparing live and dead loaded-unloaded images, with high strains adjacent to the inferior and superior growth plates (Figure 3.3). Strain patterns in the loaded-loaded and unloaded-unloaded scans were inconsistent as would be expected since it is strain error represented in these images (Figure 3.4). a) b) Figure 3.3: Axial strain patterns in loaded-unloaded strain registrations. The strain is concentrated mostly around the endplate regions. a) Dead-loaded to dead-unloaded comparison b) Live-loaded to dead-unloaded comparison

41 29 a) b) c) Figure 3.4: Axial strain patterns generated by comparison of scans under equivalent loading conditions. a) Live-loaded to dead loaded b) Dead-loaded to dead-loaded c) Dead-unloaded to dead-unloaded From the lateral vertical displacement map (3.5 b, d), it can be observed that there were minor horizontal deformations generated due to the load applied. However, since the load is primarily applied along the z-axis, the vertical deformation is more pronounced (3.5 a, c). As a result, subsequent analysis was focused on strain in the z direction.

42 30 a) b) c) d) Figure 3.5: Deformation maps generated form for live-dead images (a vertical, b lateral) and dead-dead images (c vertical, d lateral).

43 Discussion In vivo application of load to the rat tail vertebra was successfully accomplished within a live animal µct scanner allowing the comparison of live and post euthanasia strain measurement. No significant difference between live and dead comparisons was found in comparing the loaded-unloaded average axial strain measurements. The average strains were in the range of previously reported strains in rat tail vertebrae loaded in a similar configuration (Hardisty, 2010). The difference between the analyses of live and dead images was within the variance found within each group. The strain was primarily concentrated around the growth plate regions of the vertebra (Figure 3.3). Growth plates have previously been shown to absorb the majority of strain in rat tail vertebrae (Hardisty, 2010). Growth plates have much lower elastic modulus compared to the trabecular bone (Fuji, 2000; Keaveny, 2001), resulting in greater deformation under load. The stiffness of growth plate in rats increases with maturity, but was still relatively compliant in this 17-week old rat (Villemure, 2009). The strain obtained from the registration of loaded-loaded images (-0.003) was found to be 50 times higher as compared to that of unloaded-unloaded images ( ). Hardisty et al. found the error in strain calculation to increase, as the magnitude of strain applied increased. This may explain part of the strain differences observed in loaded-loaded and unloadedunloaded comparisons. Another factor may be actual changes occurring in the load distribution due to reapplication of the load. Changes in load between comparisons of loaded-loaded configurations were minimized by keeping the in vivo applied load fixed during euthanasia and for the repeated post euthanasia loaded scans (scans 2 and 3). However, reapplication of the load for the final scan (subsequent to unloading) did not alter the findings. Small amounts of stress relaxation in the tails in response to the applied load may have caused variations in the strain generated across the multiple loaded images. Nonetheless, the strains obtained from comparisons of images under similar loading conditions are in the range of standard deviations of the strains obtained from loadedunloaded comparisons. As well, measured error in strain calculation, represented by comparison of unloadedunloaded images, was found to be (accuracy), with a standard deviation of 0.004

44 32 (precision). Previous work utilizing this deformable registration routine was able to measure strain in whole ex vivo rat vertebrae with accuracy and precision of and respectively, based on an unloaded zero strain case (Hardisty, 2009). The average final window of registration used in this study was 22x22x30 pixels. The size of the region registered influences the accuracy and precision of strain calculations. As such, increase in the size of the registered region leads to increased precision in the strain field and a decreased accuracy (Hardisty, 2010). The differences of accuracy and precision in the two studies are owing to the differences in the size of the regions registered. The scale of the size of the individual regions registered lie between the apparent and trabecular level. The apparent strains determined in this study (~1.2%) are similar to apparent strain found previously in trabecular bone (Keaveny, 2001). The deformable image registration method used in this study has several advantages over some of the traditional methods for resolving bone strain. Biomechanical testing of whole bone samples can only resolve bulk strain and strain gage application is limited on small bone structures. Finite element analysis (FEA) has been used to determine strain patterns in whole vertebrae; however, physiologic loading of the spine (via intervertebral discs) greatly increases the complexity and computational expense of such models. Accurate knowledge of loading, geometry, material properties and boundary conditions is essential for a successful FEA (Tsafnat, 2011). Image registration requires no assumptions, can directly be applied to two sets of images to determine strain (Harsidty, 2009) and be used in combination with FEA to set bone boundary conditions without modeling the intervertebral discs (Herblum, 2013). A limitation of the study was lack of comparison of that loaded and unloaded scans with an in vivo unloaded scan. The initial experimental design did include this acquisition; however, the in vivo unloaded image was rendered unusable due to a motion artifact in the µct scan. As well, the study was focused on a single sample. The repeated measures analysis, however, required considerable computational effort. This work on a single specimen demonstrated a proof of concept and subsequent work can evaluate variability among multiple specimens and loading conditions. The image registration method used in this study relies on contrast between the bone and the marrow/soft tissue, to accurately register the images (Hardisty, 2009). Yet it is the structure

45 33 size required to accurately calculate the displacement field that in fact limits the resolution (Verhulp, 2004). The limited resolution cause the trabeculae adjacent to the relatively much more compliant growth plates to appear highly strained. The limited spatial resolution of the deformable registration algorithm results in measuring only the average strain within individually registered regions based on the pattern within the region. This smooths the strain field and yields a strain resolution that is much lower than the voxel size of the image. This limits the ability of the technique to accurately locate small regions of high strains at an elevated risk of damage. Efforts are currently being made in our group to improve the spatial resolution of the algorithm using feature based registration. 3.6 Conclusion The intensity based image registration module demonstrated that live and dead loaded to unloaded comparisons yielded similar strains, concentrated along the inferior and superior growth plates. Small errors in strain calculation were represented by loaded-loaded and unloaded-unloaded comparisons. This technical proof of concept study suggests that post euthanasia µct based strain analysis is able to represent the in vivo quasi static behaviour of rat tail vertebrae.

46 34 Chapter 4: Barium sulfate contrast enhanced μct imaging to identify microdamage in whole rat vertebrae 4.1 Abstract Barium sulfate (BaSO4) acts as a contrast agent for µct imaging. Recent studies have exhibited the ability of BaSO4 to highlight microdamage regions within trabecular and cortical bone sections, as well as whole femurs. Microdamage identified by BaSO4 in µct images has been validated against computational modeling, histologic staining and scanning electron microscopy. BaSO4 staining provides a viable 3D and non-destructive alternative to the lengthy 2D process of histomorphometry. In this study a series of pilot experiments were performed to generate a robust protocol for μct contrast enhanced (BaSO4) imaging and histological staining, to allow the direct comparison of these techniques to represent load induced damage within whole vertebrae. These experiments yielded staining times for whole healthy vertebrae, established compatibility of BaSO4 with histological stains (calcein and fuchsin) as well as demonstrated the success BaSO4 staining on whole osteolytic vertebrae.

47 Introduction Microdamage accumulation has been identified as a risk factor for fracture initiation (Iwata 2014). A number of studies have focused their efforts towards understanding mechanisms of microdamage initiation and propagation (Herblum 2013, Nagaraja 2007, O Neal 2010). However, central to these studies is the capability to accurately identify pre-existing and load induced microdamage within the bone samples being analyzed. Histological staining has been the primary method to characterize and quantify regions of microdamage (Lee 2003). However, contrast enhanced µct imaging through barium sulfate (BaSO4) staining has recently been utilized as a 3D alternative to study microdamage accumulation in cortical bone (Leng 2008, Landrigan 2011), trabecular bone (Wang 2007) and whole femurs (Turnbull 2011). These studies have implemented various approaches, such as FE analysis (Leng 2008, Turnbull 2011), scanning electron microscopy (Wang 2007, Leng 2008, Turnbull 2011) and histomorphometry (Landrigan 2011, Wang 2007) to validate the ability of BaSO4 to detect microdamage. While BaSO4 microdamage staining has been proven effective, the staining parameters vary with the type of bone. The following table includes the staining concentrations and staining times used by these four studies. Table 4.1: BaSO4 staining parameters used for various bone types to date Authors Type of bone Staining Staining Leng 2008 Bovine Cortical bone 7 days 1 M Landrigan 2011 Human cortical bone (old) 3 days 0.5 M Turnbull 2011 Whole rat femur 3 days 0.5 M Wang 2007 Bovine trabecular bone 2 days 0.5 M Table 4.1 suggests that optimization of the BaSO4 staining times and concentration is required as they are dependent on the type of bone to being analyzed. The current study is aimed at performing a series of pilot experiments to adapt the BaSO4 staining protocol for accurate detection of microdamage build up in whole healthy and osteolytic rat spines. Calcein and Fuchsin based sequential staining (Herblum 2013) and histology will be used to qualitatively validate BaSO4 staining. BaSO4 is not seen in fluorescent or optical microscopy

48 36 (Landrigan 2011) and there is no chemical limitation to this process, yet it is not known if the physical presence of the BaSO4 may influence the calcein/fuchsin staining. As such, the current study also aims to investigate the compatibility of BaSo4 and calcein/fuchsin. 4.3 Methods Staining protocols Calcein staining: Calcein green staining was performed to identify pre-existing trabecular damage. 1% calcein green solution (J.T. Baker, Centre Valley, PA), prepared in distilled water with 0.9% NaCl and 2% NaHCO3, was used for staining (Herblum 2013). Vertebral samples were immersed in the calcein solution and placed under vacuum for 16 hours. Since calcein undergoes photo bleaching, the vials containing the stain and stained samples were wrapped in aluminum foil. Post staining, the samples were washed in distilled water for 1 hour on top of a shaker to remove excess stain. Excitation under blue light causes calcein to fluoresce green. Basic fuchsin staining: Mechanical loading induced microdamage was stained using Fuchsin stain. The samples were immersed in a 70% ethanol solution for a minimum of 24 hours before staining. The samples were stained in one hour sequential steps by 1% solution of basic fuchsin hydrochloride (J.T. Baker, Centre Valley, PA) in a series of graded ethyl alcohols (80%, 80%, 95%, 95%, 100% and 100%) under vacuum (Burr 1995). Bright field imaging can be used to analyze histology slide stained with fuchsin. Barium sulfate staining: For BaSO4 staining, the samples were first soaked in a solution of equal parts by volume of distilled water, acetone and BaCl2.2H2O (certified ACS crystal, Fisher Scientific, Fair Lawn, NJ) for the desired amount of staining time. The samples were then immersed in a solution containing equal parts by volume of PBS, acetone and Na2SO4 for the same time as the previous step (Leng 2008, Turnbull 2011, Wang 2007). Each step was followed by 1 hour washes in distilled water to remove excess ions.

49 BaSO4 staining of rat vertebrae Intact rat spines were salvaged from healthy rats. The initial staining concentration was selected to be 0.5M (Turnbull 2011). BaSO4 staining time was evaluated by staining 3 whole vertebrae for 1, 2 and 3 days each. The stained samples underwent µct scanning using an X- ray source of 55kV at 5 μm isotropic voxel size (µct-100, Scanco Medical, Brüttisellen, Switzerland). The original protocol also suggested use of phosphate bovine serum (PBS) while preparing BaCl2 solution. However, a precipitate was observed when preparing the solution, which suggested a possible reaction between barium and phosphate. PBS was replaced with water in order to solve this problem. The BaSO4 staining protocol was then applied to an osteolytic vertebra from an rnu/rnu rat. The whole spine sample was stained with BaSO4 at a concentration of 0.5M for a staining time of 1 day Histological validation of BaSO4 staining Spinal segments containing 3 vertebrae were excised from two rnu/rnu healthy rats. The segments were stained with calcein to mark pre-existing microdamage accumulation. A µct compatible loading device (Figure 4.1) (µct-100, Scanco Medical, Brüttisellen, Switzerland) was used to induce microdamage in the healthy and metastatically-involved spine samples. The unique design of the loading device allows for accurate control of loading parameters within the the µct scanner. The top and bottom vertebrae of each 3 level motion segment were potted with PMMA to prevent slipping and lateral movement of the sample while loading. A custom jig that fits inside the loading device was designed to allow potting of the spinal segments. The segments were preconditioned under uniaxial compression for 3 cycles at 40N at a strain rate of 3µm/s. Axial compressive loading of 100 N was then applied to the motion segments and held static for 3 hours (Herblum 2013).

50 38 Loading chamber: Samples potted in PMMA using a loading jig which fits inside the chamber Loading piston: Compressive load applied from the bottom Enclosure containing the load cell and the motor for the piston Interface with the scanning system Figure 4.1: Loading device used to load samples inside the micro-ct scanner After loading, the middle vertebrae were extracted from the motion segments, and were stained with BaSO4, followed by µct scan acquisition. The samples were then stained with fuchsin and embedded in methyl methacrylate for subsequent histological sectioning. Histological sections were scanned at 20x magnification to obtain bright field and green filtered fluorescence images using the Mirax slide scanning system (Hospital for Sick Children, Toronto). Red filtered fluorescence images were also obtained as fuchsin has been shown to fluoresce red (Lee 2003). µct scans of the histological slides were also obtained. Pre-existing microcracks (calcein stained) and load induced microdamage sites (stained with both calcein and fuchsin) were identified on the fluorescence and bright field images for direct comparison with of BaSO4 staining of corresponding regions in the µct scans.

51 Results BaSO4 staining protocol for rat vertebrae The µct images for vertebrae stained for 1, 2 and 3 days are included as Figure 4.2. Staining of the samples for more than 1 day resulted in nonspecific accumulation of BaSO4 in the rat vertebrae. Figure 4.2: Micro-CT images of unloaded rat vertebrae stained for 1 day (left), 2 days (centre) and 3 days (right). The bright spots within the vertebral body represent BaSO4. It can be observed that staining for more than 1 day causes overstaining (arrows indicate regions of overstaining). The µct image of the metastatic spine stained with BaSO4 is included as Figure 4.3. The bright spots within the trabecular lattice are the BaSO4 stained regions of damage. No anomalous buildup of BaSO4 was observed, confirming a successful staining.

52 40 Figure 4.3: BaSO4 is able to stain microdamage in osteolytic spine, without pooling. Arrows indicate damaged regions of high intensity, consisting of BaSO BaSO4 and Calcein/Fuchsin compatibility BaSO4 and calcein/fuchsin were found to overlap in the selected regions. This suggested an agreement and compatibility between BaSO4 and calcein/fuchsin staining. Figure 4.4 shows fluorescence, brightfield and µct images of a pre-existing damage region. Figure 4.5 includes similar images for a load induced damage region. BaSO4 can be observed to stain both the pre-existing and load induced damage, in conjunction with calcein and fuchsin.

53 41 a) b) c) d) Figure 4.4: Fluorescence (a, b), brightfield (c) and µct (d) images of a pre-existing microdamage site highlighted by fuchsin (a, c), calcein (b) and BaSO4 (d) staining. Arrow indicates a region of preexisting microdamage observed in fluorescent images, but not on the bright field image.

54 42 a) b) c) d) Figure 4.5: Fluorescence (a, b), brightfield (c) and µct (d) images of a load induced microdamage site highlighted by fuchsin (a, c), and BaSO4 (d) staining, but not by calcein (b) The µct images in the above figures have been created by superimposition of the thresholded scan on the regular scan of the histology slide. BaSO4, being a contrast agent has a higher threshold than bone, allowing the BaSO4 stained bone regions to be separated from the unstained regions. Such an overlay allows visualization of the damaged regions as well as the surrounding trabecular meshwork. 4.5 Discussion The selection of initial concentration of 0.5 M for staining whole vertebrae was based upon concentration used previously to stain whole femurs (Turnbull 2011). The staining times (1, 2, and 3 days) were based also on those used in the recent studies (Table 4.1). BaSO4 was found to reach the trabecular bone much faster in the vertebrae since the cortical bone of rat femur is much thicker compared to the cortical shell of the spine. Staining the spines for 1

55 43 day provided sufficient penetration of the stain to label the microdamage, without surplus pooling (Figure 4.2). Based on these results, it was decided that the rat spine samples in subsequent experiments would be stained in BaSO4 for 1 day at 0.5M concentration. Additionally, distilled water would replace PBS in preparing BaCl2 solution. The µct and the bright field images (Figure 4.4 and 4.5) show that the selected damage regions were highlighted by both BaSO4 and calcein/fuchsin sequential staining. However, it can be noticed that despite the general locations of the damaged regions on the µct and the bright field images are the same, the shapes and sizes of the regions of interest differ. Similar observations can be made with regard to the fluorescence images. These differences are not unexpected, since the type of stains and imaging methods differ. Bright field imaging only captures the information on the surface of the histology slide. In contrast, the µct scan provides 3D information about the damaged area. Higher resolution µct images can be used to increase the accuracy of spatial correlation of the damaged regions. An interesting observation can be made from Figure 4.4, in that a pre-existing microcrack (arrow), although stained with fuchsin, was not observed in the bright field image. Bright field imaging only captures the surface of the histology slide. Fluorescent imaging has better penetration, capturing sub-surface regions of microdamage. However, this also leads to an increased background signal, resulting in blurrier images. Based on these observations, both bright field and fluorescence images of the histology slides were used to identify fuchsin stained regions of microdamage. It can also be observed from Figures 4.4 and 4.5, that the µct images show multiple damaged regions surrounding the regions of interest, which were not stained by fuchsin. A possible reason for this discrepancy could be the different size of the stain molecules. The barium and sulfate ions are much smaller in size compared to the calcein and fuchsin molecules. This allows BaSO4 precipitation to take place in very small microdamaged areas, where calcein and fuchsin cannot enter. This may be an added advantage of using BaSO4 over such dyes and chelating agents. The compatibility of calcein, fuchsin and BaSO4 originates from variance in staining mechanisms. Calcein is a chelating agent, which combines with exposed Ca 2+ ions at the

56 44 damaged regions (Lee 2000). Fuchsin has been reported to stain the collagen fibers of the bone (Lee 2003). BaSO4, on the other hand, undergoes precipitation reaction to stain microdamage in bone (Wang 2007). Since all these reactions take place independently, the three stains are able to function without interference. 4.6 Conclusion No studies have previously utilized sequential staining (calcein and fuschin) to identify load induced microdamage in conjunction with BaSO4. A series of experiments were used to establish BaSO4 staining parameters for future use in the project. Overall, this study demonstrates that BaSO4 staining can be used in conjunction with sequential histology staining and may be a robust method for non-destructive and 3D evaluation of microdamage accumulation in whole vertebrae.

57 45 Chapter 5: Evaluation of tissue level stresses and strains under uniaxial compression of whole healthy and osteolytic rat spines 5.1 Abstract In this study, µct imaging based µfe models were used to determine tissue level damage criteria in whole healthy and osteolytic vertebrae. T13-L2 spinal segments were excised from osteolytic (n=3) and healthy (n=3) female athymic rnu/rnu rats. Osteolytic metastasis was generated by intercardiac injection of HELA cancer cells. Micro-mechanical axial loading was applied to the spinal motion segments under μct imaging. Vertebral samples underwent BaSO4 staining and sequential calcein/fuchsin staining to identify load induced microdamage. μct imaging was used generate specimen specific μfe models of the healthy and metastatically-involved whole rat vertebrae. Model boundary conditions were generated through deformable image registration of loaded and unloaded scans. Elevated stresses and strains were detected in regions of microdamage identified through histological and BaSO4 staining within healthy and osteolytic vertebral models, as compared to undamaged regions. Additionally, damaged regions of metastatic vertebrae experienced significantly higher local stresses and strains than those in the damaged regions of healthy specimens. The range of maximum principal stresses and strains for microdamage initiation was 80±27Mpa and 0.71±0.17% in healthy samples and 127±61Mpa and 0.98±0.44%. The experimental, imagebased and computational techniques used in this study demonstrated success in identifying and characterizing local stresses and strains in the regions of trabecular microdamage. Knowledge gained in this work forms a strong platform for more advanced techniques for biomechanical analyses of healthy and diseased bones.

58 Introduction Microdamage formation within the skeleton during routine physiological loading serves as a stimulant for bone remodeling. However abnormal buildup of microdamage leads to skeletal fragility, especially in cancellous bone (Burr 1998). Accumulation of unrepaired microdamage is usually the consequence of damaged microstructure and inferior bone quality, resulting from age-related changes or skeletal pathology (i.e. osteoporosis or bone metastasis) (Iwata 2014). Despite its significance to the mechanical properties of the bone, the trabecular stresses and strains experienced at the initiation of microdamage are not well characterized. Evaluation of stresses and strains associated with microdamage initiation at the local level may offer a better platform for improvement of fracture risk assessment techniques and the development of therapeutic methods for treatment of skeletal fragility diseases such as metastasis. Microdamage sites can be experimentally identified through sequential staining and histomorphometry (Lee 2000). Microdamage staining along with high resolution micro-finite element (µfe) models has been utilized to study local damage initiation properties of trabecular bone. µfea (Herblum 2013, Nagaraja 2005) has the capability to model the trabecular morphology, allowing calculation of stresses and strains at histologically identified damage sites (Keaveny 2001). Nagaraja et al., in a series of experimental studies, used this approach to determine variations in microdamage initiation parameters in response to age related changes to human and bovine trabecular bones (Nagaraja 2005, 2007, 2011, Green 2011). However, these studies were performed in trabecular bone cores. Cores were essential due to the massive sizes of the µfe models and the associated computational time and memory requirements. µfe analysis of whole bone allows loading of the bone specimens through joints and/or soft tissues, simulating more physiological loading conditions. Herblum et al. recently demonstrated successful application of µfea to show elevated stresses and strains in regions containing mechanically induced microdamage within whole healthy rat vertebrae (Herblum 2013). A summary of local stresses and strains determined by these studies is included as table 5.1. µfea has also been employed to study local yield properties of cortical bone (Bayraktar 2004), as well as effects of therapeutics on

59 47 bone architecture (Boyd 2011). Although these two studies were not aimed at studying microdamage, the use of µfe modelling was demonstrated. Table 5.1: Trabecular stresses and strains at locations of microdamage determined under axial load by previous studies Authors Bone type Stress range (MPa) Strain range (%) Nagaraja 2005 Nagaraja 2007 Nagaraja 2011 Green 2011 Bovine trabecular bone cores Bovine trabecular bone cores Bovine trabecular bone cores Human trabecular bone cores (young) (old) (young) (old) (young) (old) (young) (old) (young) (old) (young) (old) Herblum 2013 Whole rat vertebrae Spinal metastasis progressively degrades the trabecular architecture of the vertebral body, leading to an increased risk of fracture (Kurth 2001). The initiation and propagation of unrepaired microdamage, which precede fracture, has not been quantified in metastatic spines. This study was aimed at the development µfe models to determine the thresholds for histologically identified damage within whole healthy and osteolytic vertebrae. The efficacy of BaSO4 contrast agent to highlight microdamage in 3D within healthy and osteolytic spines was also evaluated. 5.3 Methods The workflow of this study is included as Figure 5.1. Briefly, the healthy and osteolytic whole vertebrae were first stained with calcein to identify pre-existing microdamage. µct images of the unloaded samples were acquired. The samples were subjected to axial compressive load for 3 hours and were µct scanned while under load. The specimens were stained with BaSO4, µct scanned again and stained with fuchsin, followed by embedding and histologic sectioning. Load induced microdamage sites were identified on the histology slides through bright field, fluorescence and contrast enhanced imaging. µct images of the histology slides were aligned with the unloaded µct scans of the samples. These µct

60 48 datasets were converted directly into µfe models, which were analyzed using ABAQUS. Deformable image registration of loaded and unloaded µct scans was used to obtain boundary conditions for the µfea. Finally statistical analysis was performed to analyze the local stress/strain data obtained from these models. Figure 5.1: Experimental design

61 Animal models: A previously described rat tumor model was used to investigate the biomechanical implications of osteolytic metastasis on the trabecular bone within the spine (Hardisty 2011, Lo 2012). Luciferase transfected HELA human cancer cells (previously described as human MT1 breast cancer cells) were introduced in 3 rnu/rnu female nude rats (4-5 week old) (Harlan, Indianapolis, IL, USA) via intercardiac injection under general anesthesia. Two weeks later, bioluminescence imaging was performed to quantify tumour burden in all tumour cell injected rats under general inhalation anaesthesia. Bioluminescence signal was acquired using the IVIS Bioluminescent Imaging system (+ Corp., Alameda, CA, USA). The rats were euthanized via intercardiac injection (120mg/kg euthanyl) under general anesthesia 14 days post injection. The age and weight of the three rats were between 7-8 weeks and g respectively. 3 healthy rnu/rnu female rats of similar age and weight were also sacrificed. Spinal motion segments from T13-L2 were extracted from the healthy (n=3) and osteolytic (n=3) rats for further analysis, with L1 as the vertebra of interest Microdamage Evaluation using calcein/fuchsin staining and contrast enhanced μct Intact spinal motion segments were stained with calcein green (as described in section 4.3) to identify pre-existing microdamage. µct images of unloaded and loaded spinal motion segments were acquired at an isotopic voxel size of 11.4 microns at 55 KeV and 200µA (µct-100, Scanco Medical, Brüttisellen, Switzerland). Following load application (see next section), the middle vertebrae (L1) were excised and stained with BaSO4 (as described in section 4.3) to identify load induced damage, and additional μct scans were acquired. The specimens were then stained with 1% basic fuchsin (as described in section 4.3), to sequentially label de novo microdamage. The middle vertebra of each segment was stained as a whole bone (the bone marrow was not drained and end plates and the posterior elements were left intact). The stained samples were embedded in methylmethacrylate for histology. Coronal sections of 60-80μm thickness were prepared from each of the embedded samples for histology. The order of each slide was recorded, which was necessary for registration later in the study. All the histology work was performed in the Histology Lab at the Faculty

62 50 of Dentistry, University of Toronto. The histology slides were scanned at 20x magnification using bright field and green filtered fluorescence imaging using the Mirax digital slide scanning system at the Hospital for Sick Children, Toronto. Locations of linear microcracks were identified on the basic fuchsin histology slide images to indicate the presence load induced microdamage on individual trabeculae. Fuchsin stained damage regions also containing calcein were overlooked as load induced microdamage was the focus of the current study. Although diffuse damage could be visualized with fuchsin staining, only locations of linear microdamage were selected for analysis. Damage regions were identified prior to µfe model creation blinded to the individual conducting the FEA in order to prevent any bias in the results. µct scans of the histology slides were also acquired in order to validate BaSO4 staining against histological staining. The presence and absence of BaSO4 in the histologically identified damaged regions selected was examined. As well, the slides were examined to determine the presence of BaSO4 stained sites not containing fuchsin or calcein Loading The µct compatible loading device (Figure 4.1) was used to induce microdamage in the healthy and metastatically-involved spine samples. The unique design of the loading device allows for µct scanning of the samples while under load. The top and bottom vertebrae of each 3 level motion segment were potted in a custom jig with bone cement to prevent slipping and lateral movement of the sample while loading. The motion segments of the healthy and osteolytic rats were preconditioned under uniaxial compression for 3 cycles at 40N at a constant strain rate of 3µm/s. Axial compressive loading of 100 N for healthy rat vertebrae and 50 N for metastatic rat vertebrae were applied to the spinal motion segments for 3 hours. These loads were experimentally found to create microdamage in the respective types of samples without fracture (Herblum 2013, Hardisty 2011). The use of 3 motion segments along with posterior elements was selected to mimic physiological albeit quasistatic loading conditions.

63 Strain fields and boundary conditions The deformation registration algorithm (refer to section for more details) was used to extract strain patterns by aligning and comparing the loaded and unloaded scans of healthy and osteolytic spines (AmiraDEV 3.1) (Hardisty 2009). These strains were not directly comparable to µfea because of differences in resolution and hence were not used to validate the µfe models. A modified version of the registration algorithm (MultiResolutionRegistration) was used to determine boundary conditions for the FE models. Using the image registration of the loaded/unloaded scans, fifty-eighty thousand displacement boundary conditions were extracted for the FE analysis. The calculated vector displacement fields were used to assign FE displacement boundary conditions at surfaces of the endplates of the vertebral body and the facet joints under axial compressive load (Nagaraja 2005, Nagaraja 2007, Herblum 2013) Alignment of histology slides All µct scans were thresholded to segment the bone from surrounding non-bone areas. The 3D segmented µct images of the histology slides were registered to the respective unloaded µct scans of the whole samples. Additional µct images of the remaining blocks were acquired to facilitate accurate registration of histologic slides with unloaded µct data. The histology blocks were first aligned using a built in automated routine in AmiraDEV (Affine Registration). Having more volumetric data within the blockss (as compared to the slide sections) makes automated alignment of the blocks simpler to perform. Once the blocks were aligned, they revealed approximate locations of the corresponding slides (this information was recorded while the sections were cut). Using the blocks as a reference point, the slides were first aligned to the unloaded scan manually. Following manual alignment, Affine Registration was employed to improve the alignment of the slide. Once a slide was aligned, the region within the unloaded µct dataset corresponding to the hisology slide was identified as a separate material. The accuracy of the alignment was assessed by calculating the volumetric concurrency (VC) of the two scans. VC was evaluated as the ratio of bone overlap between the histology slide and the unloaded scan to the bone volume of the

64 52 histology slide. Post alignment, the segmented scans were downsampled to a voxel size of 35 µm. Registration of the slides allowed each histology slide to be identified within the unloaded scan prior to µfe model generation. This allowed extraction of µfea results for the elements corresponding to the histology slides, for direct comparison with histologically identified microdamage. To register the locations of BaSO4 staining in the µfe models, the µct scans of the unloaded and BaSO4 stained vertebrae were utilized. As above, extraction of µfea results for the elements corresponding to the BaSO4 staining was limited to the area contained in the excised slides, for direct comparison with identified microdamage Creating µfe models The µfe models were generated in AmiraDEV from the unloaded µct scans of the middle vertebra (with histology slides defined as separate element sets). A voxel based meshing algorithm (Voxelator) has been implemented in AmiraDEV 5.2.2, which was used to generate a mesh of 8-noded hexahedral elements from the segmented µct scans. Bone marrow and osteolytic tumor regions were left as void spaces in the models. Abaqusinputwriter module was then used to create an Abaqus input file (ABAQUS, Pawtucket, RI). In order to create loading surfaces, a custom built algorithm (GridToSurface) was used to create surfaces from the voxel based mesh. The end plates and facet joints were manually selected as loading serfaces. The vectors from the deformation field (see section for more details) corresponding to the selected surfaces were assigned as boundary conditions. DisplacementBC was used to assign the displacement vectors to the appropriate nodes within the model. Each final Abaqus input file included the generated mesh, loading surfaces and boundary conditions. Abaqus Standard 10-1 was used to run the finite element models. A Young s modulus and a Poisson s ratio of 12.5 GPa and 0.3 respectively were assigned as material properties for bone (Herblum 2013, Kinney 2000). Each of the model, containint over a million elements were executed as isotropic, homogeneous, linear, static models using the supercomputing facility at SciNet, University of Toronto.

65 Statistics and data analysis The slices within the µfe models corresponding to the histology slides were isolated for comparison with histology. For quantitative analysis, regions (n = 20) of microdamage highlighted by the fuchsin were selected from both groups (healthy and metastatic). Surrounding undamaged regions (n = 20) were also selected for comparison, in accordance with previous studies (Nagaraja 2005, Nagaraja 2007, Herblum 2013). The finite elements corresponding to these regions were selected within the models. Stress (Von Mises and maximum principal) and strain (max principal) parameters from the damaged and undamaged regions from both healthy and osteolytic models were extracted for comparison. Normal distribution of this data was verified through One-Sample Kolmogorov-Smirnov Test. T-tests were then used to compare stress/strain values in the damaged and undamaged regions of the same group, as well as those in osteolytic and healthy bones. Ten additional microdamage sites stained only by BaSO4 (and not calcein and fuchsin) were selected for both healthy and metastatic models. Similar analysis was performed to compare the stresses and strains in the damaged and undamaged regions. Additional t-tests were performed to compare the local stresses and strains in the damaged regions stained by these two techniques. A significance level of α=0.016 was determined using Bonferroni correction, to adjust for multiple t-tests. 5.4 Results Microdamage identification using histology Healthy: Figure 5.2 shows green filtered fluorescence and bright field images of the same slide from a healthy rat. These images were visualized using the Mirax viewer, which allows zooming in and out on the high resolution microscopy images (Figure 5.3). Both calcein and fuchsin demonstrated the ability of staining microdamage in whole healthy vertebrae. Using these stains before and after loading, allowed the separation of pre-existing microdamage from damage accumulated from loading. Fuchsin was observed to stain the pre-existing as well as

66 54 the mechanically induced microdamage. However, since calcein only stained a priori defects, fuchsin stained regions not containing calcein were considered to be post mechanical loading damaged areas. This has been demonstrated in Figure 5.4. Twenty such microdamaged regions, containing only fuchsin, were selected for analysis. Fuchsin was also found to fluoresce red under green excitation. However damaged regions in fuchsin fluorescence images looked similar to the edges of trabeculae. As a result, fuchsin labelling was primarily analyzed in the bright field images. Fuchsin stain was found to pool in the bone marrow regions (Figure 5.2). Thicker sections increased buildup of fuchsin in the marrow; making it challenging to identify damaged regions near the endplate. Thin sections allow much better identification of bone damage. However, the thickness of the sections is necessary for automated alignment with the unloaded µct images. As found in previous studies, slices between 50-80µm thick allow proper damage visualization, while providing enough information for 3D alignment (Herblum 2013). However, since the bone samples are hard embedded in PMMA such accuracy in thickness is difficult to achieve, without advanced apparatus. The thickness of the slices acquired as a part of this project ranges between µm. The slide in Figure 5.2 is one of the thicker sections, while that in Figure 5.3 is one of the thinner sections. The amount of pooling, and the associated difficulty in recognizing microdamage with respect to section thickness is evident from these two images.

67 Figure 5.2: Coronal histology slide from a healthy vertebral body imaged under fluorescence (a) to identify calcein stained pre-existing damage, and under plain light (b) to detect fuchsin stained load induced damage 55

68 Figure 5.3: a) Bright field image of a histology slide from a healthy sample. b) Fuchsin stained load induced damage on a trabecula at 20x magnification 56

69 57 a) b) Figure 5.4: a) Pre-existing damage labelled by both calcein and fuchsin b) Load induced microdamage stained only by fuchsin, and not calcein. Metastatic: Microdamage identification in osteolytic samples was much simpler compared to the healthy vertebrae. Better stain penetration and increased removal of excess stain was achieved as a consequence of decreased bone density (Figure 5.5a). Lower trabecular number in osteolytic vertebrae also allowed better definition of trabecular bone near the endplate regions. A higher concentration of microdamaged trabeculae was observed near the endplate (Figure 5.5b), as observed in previous studies (Eswaran 2007). In general, the number and severity of damaged sites in the metastatic samples was much greater compared to the healthy samples, as expected. Basic fuchsin was not found to stain the tumor tissue itself, but rather accumulated in the surrounding regions. However, this did not affect the ability of fuchsin to highlight trabecular microdamage (Figure 5.5b). No pooling of calcein was observed.

70 58 a) Figure 5.5: a) Coronal histology slide of osteolytic spine demonstrating reduced trabecular number and increased fuchsin accumulation in the osteolytic regions. b) Multiple microdamaged sites observed near osteolytic tumor tissue. b) Image registration to determine strain fields and boundary conditions The deformable image registration described in chapter 3 of this document was used to determine the strain fields by comparing the loaded and unloaded µct images of healthy and osteolytic samples. Vertical strain patterns obtained in response to uniaxial compressive loading healthy and osteolytic samples are included as Figure 5.6. As before, the higher strains were primarily concentrated near the growth plate area for both samples (Hardisty 2010). However, higher strains were also observed in the osteolytic regions of low trabecular number (Figure 5.6b) well below the end plates, similar to Hojjat 2011.

71 59 a) b) Figure 5.6: Coronal slices demonstrating strain fields obtained for healthy (a) and metastatic whole vertebrae. Red and blue areas experience high and low strains respectively. Arrow denotes area osteolytic destruction under high strain. The axial strain distributions in the vertebrae were characterized by calculating the mean, median and the 10 th and 90th percentile strains. The average strain values for healthy and metastatic specimens are included as table 5.2. The strain values were compared using independent T-tests. No significant differences in strain values were observed. However, it should be noted that the osteolytic samples were loaded to 50N while the healthy vertebrae were loaded to 100N. Table 5.2: Average strains (µm/µm) obtained from the comparisons of loaded/unloaded images of healthy and osteolytic spines Mean Median 10% 90% Healthy ± ± ± ±0.01 Metastatic ± ± ± ±0.01 T-test p-value

72 60 A modified version of the algorithm, MultiResolutionRegistration, was implemented in AmiraDEV to determine the displacement fields to be used as boundary conditions for the µfe models. This algorithm also registers loaded and unloaded scans of a sample, to yield a vector deformation field. Deformation fields for axial compression healthy and osteolytic samples along the Z-direction are shown in Figure 5.6. The deformation field is represented by a set of arrows signifying the direction and magnitude of the displacement of the registered sub-region. Regions near the endplates show displacement primarily along the Z axis, as expected, since it was the principal direction of axial loading. The displacement vectors are also consistent with the axial strain patterns in Figure 5.6, where regions of high axial strains were observed around the growth plate area a) b) Figure 5.7: Displacement vectors generated using deformable image registration, represented by blue arrows generated for healthy (a) and metastatic (b) vertebrae

73 Alignment of histology slides with unloaded scans Healthy: µct images from the slides were registered to the unloaded scans using manual and rigid affine registration. Figure 5.8 shows a histologic image and a surface generated from µct scan of a slide from a healthy sample. Such surfaces were utilized during the registration process. A registration of the block together with the histology slides is demonstrated in Figure 5.9. The histology slides were approximately parallel to the block, making the alignment procedure much simpler once the block was registered. A separate model was created for each slide, to allow easy extraction of the data from the µfe models (Figure 5.10). a) b) Figure 5.8: Bright field image of a histology slide from a healthy slide (a) and surface generated from µct image of the same slide (b)

74 Figure 5.9: Surface generated from registered µct scans of the block (posterior elements) and three slides superimposed on unloaded µct scan from a healthy sample 62

75 63 Figure 5.10: Segmented unloaded µct images of the same healthy sample, each containing a histology slide identified as a separate material The effectiveness of the combined manual and automated loading was quantified using volumetric concurrency (VC). A built in AmiraDEV module (Quantification) was used to measure the volume of the histology slides and intersecting areas between the slides and the unloaded µct scan. Segmentation allows µct scans to be represented in binary units, where all bone voxels were assigned a value of 1 and all non-bone regions were assigned a value of 0. Multiplication of segmented aligned histology slide and unloaded scans yielded the intersecting regions as all the non-overlapping regions was assigned a value of 0. Volumetric concurrencies for all slides, calculated as a ratio of the overlapping volume over the volume of the histology slide, are included in table 5.3. The concurrencies averaged 68% for healthy rats, with a range from 63-77%. These values are in agreement with the amount of alignment previously seen using this technique (Herblum 2013). Figure 5.11 includes visuals of µct scan of the histology, along with the overlapping volume, to demonstrate successful alignment. Table 5.3: Volumetric concurrencies for healthy spines Sample Number of slides Average VC for the sample (%) Healthy Healthy Healthy

76 64 Figure 5.11: µct scan of a slide from healthy sample (yellow), along with the region of intersection between the slide and the unloaded scans (blue). The VC for this slide was 66% Metastatic: A similar process of registration was followed for the osteolytic samples. Alignment of the slides with the whole bone scans was more difficult in the metastatic samples. Lower bone volume hindered the ability of automated registration to accurately register the images. The volumetric concurrencies for metastatic slides are included in table 5.4. Compared to healthy samples, a lower VC of 64% was obtained, with a range from 59-79%. The overlap of a

77 65 representative histology slide and whole bone scan of a metastatic vertebra is presented in Figure Table 5.4: Volumetric concurrencies for osteolytic spines Sample Number of slides Average VC for the sample (%) Osteolytic Osteolytic Osteolytic Figure 5.12: µct scan of a slide from metastatic sample (yellow), along with the region of intersection between the slide and the unloaded scan (green). The VC for this slide was 60%

78 µfe modeling of healthy and metastatic spines A series of steps utilizing multiple custom built algorithms were implemented to process the healthy and osteolytic µct scans to create µfe grids with elements corresponding to the histology slices localized separately. A separate model was created for every histology slide. A single model incorporating all the slices can also be generated. However, individual models allow easier extraction and visualization of the µfe results for elements corresponding to the slide. Figure 5.13 demonstrates a µfe mesh containing the histology slide from a metastatic sample identified as a separate set of elements. The displacement vectors acquired through deformable registration (section 5.4.2) were used as displacement boundary conditions. Loading surfaces were manually selected at the regions of direct load transfer through the motion segment (Figure 5.14). Based on µct scans of the loaded samples (Figure 5.14a), facet joints and endplates were chosen to be the surfaces of direct contact between the middle and peripheral vertebrae (Figure 5.14b). Vector displacements overlapping the loading surfaces were inserted into the model to mimic the loading conditions. Figure 5.13: µfe grid generated form whole bone µct scan of a metastatic sample with elements corresponding to a histology slide highlighted in red. One such model was generated from each histology slide

79 67 b) a) Figure 5.14: a) µct scan of a spinal motion segment under load. b) Surfaces on the endplates and facet joints were selected as loading surfaces for the middle vertebra (highlighted in pink) µfe models thus generated were assigned material properties to the bone elements (Section 5.3.6). The models were run at the SciNet supercomputing facility. Each model consisted between million elements and thousand boundary conditions. 18 CPU s, each containing 2GB RAM, were used to run the models, for a runtime of 2-3 hours. Upon completion of a µfea, the elements representing the histology slides were extracted from the whole µfe model for further analysis. Figure 5.15 shows a microscopic image of a histology slide, along with corresponding region from the µfe model.

80 68 a) b) Figure 5.15: a) Bright field image of a coronal histology slide from a healthy sample stained with fuchsin to identify load induced microdamage b) Results showing tissue level maximum principal stress distribution in the elements within the µfe model corresponding to the histology slide. Arrows show correspondence between the histology slide and the model Determining tissue-level stresses and strains in histologically damaged and undamaged regions Healthy: Altogether 20 regions of load induced microdamage were selected from the fuchsin stained histology slides of healthy samples along with 20 adjacent undamaged areas (Herblum 2013, Nagaraja 2005). The elements corresponding to these regions of interest were identified within the µfe models (Figure 5.16). The figure shows the maximum principal stress

81 69 distribution across the selected elements and stress concentration within the elements pertaining to the damage site. a) b) c) Figure 5.16: a) Axial compressive load induced microdamage identified by fuchsin staining. b) Elements corresponding to the damaged site selected in the undeformed µfe model. c) Completed µfea demonstrates elevated maximum principal stress in the region of microdamage