University of Florida Thesis Or Dissertation Formatting

EXAMINING THE NEXUS AMONGST JOINT DEGENERATION, DISABILITY, AND CHRONIC PAIN USING A RODENT MODEL OF POST-TRAUMATIC KNEE OSTEOARTHRITIS

HEIDI ELISE KLOEFKORN

A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2016

To my husband

ACKNOWLEDGMENTS

While a doctoral degree is only conferred to one person, the degree is earned due to the efforts of many.

I would like to first thank my husband, Spencer Adams, and my parents, Randy and Leigh Ann Kloefkorn. Their support, love, and patience are invaluable.

I would like to thank my chair, Kyle Allen, for seeing my potential, encouraging me to succeed, and teaching me that failure is always an option. With his guidance (and a healthy dose of snark), I have grown as a scientist and academic with a solid foundation to continue my career. I am most grateful that, when it came to his efforts to improve my writing skills, failure was NOT an option.

I would also like to thank my lab mates for their comradery and expertise. They have unfailingly provided help, perspective, advice, and laughter at just the right moment.

TABLE OF CONTENTS

page

ACKNOWLEDGMENTS ...... 4

LIST OF TABLES ...... 8

LIST OF FIGURES ...... 9

ABSTRACT ...... 11

CHAPTER

1 INTRODUCTION ...... 13

Defining Osteoarthritis ...... 13 Cartilage Changes in OA ...... 14 Bony Changes in OA ...... 16 Synovial Changes in OA ...... 17 Pain and OA ...... 17 Preclinical Models of OA Degeneration and Pain ...... 19 Rodent Behavioral Tests ...... 21 Rodent Gait Analysis ...... 22 Spatiotemporal Video-Based Rodent Gait Assays ...... 23 Spatiotemporal Gait Parameters ...... 26 Using Histology to Assess Tissue Degeneration ...... 27 Early Histological Grading Schemes for OA ...... 28 Modern Histological Grading Systems for OA ...... 29 Aim of This Dissertation ...... 31

2 SPATIOTEMPORAL GAIT COMPENSATIONS FOLLOWING MEDIAL COLLATERAL LIGAMENT AND MEDIAL MENISCUS INJURY IN THE RAT: CORRELATING GAIT PATTERNS TO JOINT DAMAGE ...... 37

Experimental Design and Animal Surgery ...... 38 Spatiotemporal Gait Testing ...... 40 Mechanical Sensitivity...... 41 Histology ...... 42 Statistical Analysis ...... 43 Gait Patterns ...... 43 Evidence of Antalgic and Shuffling Gait Compensations ...... 45 Mechanical Sensitivity is Altered ...... 47 Histology Shows Progressive Degeneration ...... 47 Univariate Correlations ...... 48 Future Directions to Correlate Symptoms and Behavior ...... 50

3 AUTOMATED GAIT ANALYSIS THROUGH HUES AND AREAS (AGATHA): A METHOD TO CHARACTERIZE THE SPATIOTEMPORAL PATTERN OF RODENT GAIT ...... 68

Automated Gait Analysis Through Hues and Areas (AGATHA) ...... 69 Testing Approval ...... 72 Gait Testing ...... 72 Validating AGATHA to Manual Digitization and Exploring the Effect of Video Frame Rate ...... 73 Validating AGATHA Using an Orthopaedic Injury Model ...... 74 Validating AGATHA Using a Cervical Spinal Cord Hemisection Injury Model ...... 76 Validating AGATHA Using a Cervical Spinal Cord Contusion Injury Model ...... 77 AGATHA Comparison to Manual Digitization ...... 79 Orthopaedic Injury Model ...... 80 Spinal Cord Injury Model ...... 81 Future Directions to Improve Gait Analysis ...... 81

4 A GUI FOR THE EVALUATION OF KNEE OA (GEKO); AN OPEN-SOURCE TOOL FOR RAPID GRADING OF RODENT KNEE OA ...... 96

A GUI for the Evaluation of Knee OA (GEKO) ...... 97 Modeling OA and Preparing Histological Images ...... 100 Validating GEKO to Manual Grading ...... 101 Exploring Grader Variation ...... 101 Exploring Grader Repeatability and Reproducibility ...... 101 GEKO Reproduced Comparable or Better ICCs than Manual Grading ...... 102 Multiple Graders Reproduced Manual Grading Results with High ICCs ...... 103 Inter-grader ICCs were Relatively Low ...... 103 Intra-session Reproducibility was Lower than Inter-session Reproducibility ...... 104 Discussing GEKO ...... 104

5 HISTOLOGICAL CHANGES IN THE SUBCHONDRAL BONE AND SYNOVIUM CORRELATE TO RODENT BEHAVIOR IN A MODEL OF POST-TRAUMATIC KNEE OA ...... 116

Experimental Design ...... 119 Osteoarthritis Model ...... 119 Behavioral Tests ...... 120 Histology Results ...... 121 Preparing Histological Images ...... 121 OARSI Grading ...... 121 Measuring Bony Changes ...... 122 Measuring Trabecular Bone Area ...... 123 Quantifying Subintimal Synovium Changes ...... 124 Statistical Analysis ...... 125 Subchondral Bone Changes ...... 125 Synovial Changes ...... 126

Femoral Cartilage Thickening ...... 128 Correlations to Mechanical Sensitivity ...... 128 New Histology Measures Correlate with Spatiotemporal Gait ...... 128 Future Directions in Histology ...... 129

6 CONCLUSION ...... 141

LIST OF REFERENCES ...... 144

BIOGRAPHICAL SKETCH ...... 156

LIST OF TABLES

Table page

2-1 Correlations between histological evidence of joint damage and pain-related behaviors following MCLT sham and MCLT+MMT surgery in the rat ...... 67

3-1 Detection sensitivity of AGATHA and manual digitization methods...... 95

4-1 Summary of OARSI histological measures ...... 113

4-2 Cartilage degeneration score...... 113

4-3 ICCs of GEKO Validation ...... 114

4-4 ICCs for grader variation ...... 115

5-1 Correlations of histological measures and behavior...... 140

LIST OF FIGURES

Figure page

1-1 Histological sections of medial knee compartments stained with Toluidine blue...... 35

1-2 Representative Hildebrand plots of common gait changes ...... 35

1-3 Graphical representations of common gait arena Arrangements...... 36

1-4 Two planes of view can optimize spatial and temporal gait resolution ...... 36

2-1 Summary of spatiotemporal gait...... 57

2-2 Calculation of gait data residuals ...... 58

2-3 Spatiotemporal gait measures...... 59

2-4 Representation of spatial gait changes...... 60

2-5 Temporal gait measures ...... 61

2-6 Hildebrand temporal gait charts ...... 62

2-7 Summary of mechanical sensitivity ...... 63

2-8 Summary of histological measures ...... 64

2-9 Representative histology slides and measures ...... 65

2-10 Histology and behavior correlations...... 66

3-1 How AGATHA derives temporal measures...... 88

3-2 How AGATHA derives spatial measures ...... 89

3-3 Paired graphs comparing AGATHA with manual digitization...... 90

3-4 Effect of video frame rate on AGATHA and manual digitization...... 91

3-5 Effect of video frame rate on gait event accuracy ...... 92

3-6 Summary of AGATHA digitized orthopaedic model ...... 93

3-7 Summary of AGATHA digitized spinal cord injury models ...... 94

4-1 Representative Image of GEKO Interface ...... 109

4-2 Description of GEKO Process ...... 110

4-3 Paired raw data graphs of GEKO vs manual grading part 1...... 111

4-4 Paired raw data graphs of GEKO vs manual grading part 2 ...... 112

5-1 Summary of user-defined measures ...... 133

5-2 Calculating trabecular bone area ...... 134

5-3 Calculating synovium measures ...... 135

5-4 Subchondral bone changes ...... 136

5-5 Subintimal cell morphology...... 139

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

EXAMINING THE NEXUS AMONGST JOINT DEGENERATION, DISABILITY, AND CHRONIC PAIN USING A RODENT MODEL OF POST-TRAUMATIC KNEE OSTEOARTHRITIS

Heidi Elise Kloefkorn

August 2016

Chair: Kyle D. Allen Major: Biomedical Engineering

Osteoarthritis (OA) is a pervasive chronic, degenerative disease that prevents patients from performing many simple daily activities. OA patients tend to seek treatment only after significant knee pain has developed. However, since the severity of pain does not always correlate with radiographic evidence of joint destruction, OA patients can present across a wide-range of degenerative scales. While knee pain does tend to be more severe during activity, the specific source of the knee pain is often unknown. In addition, since OA is usually not diagnosed until moderate to severe stages of disease, the etiology of OA is often unknown. Multiple factors have been suggested to contribute to OA pathogenesis, including genetics, repetitive loading, trauma, and obesity; but regardless of the etiology, OA progression is challenging to alter or stop.

Understanding the relationship between OA-related degeneration and OA-related pain will help improve diagnosis and treatment of OA.

The global aim of this dissertation is to help bridge the gap between tissue degeneration and symptomatic pain and disability using a rodent post-traumatic model of knee OA. The data presented in this dissertation 1) correlate behavioral and

degenerative profiles in early, middle, and late stages of OA, 2) introduce a new automated gait analysis method, 3) describe a semi-automated graphic user interface

(GUI) for the evaluation of knee OA in rodents, and 4) introduce quantitative histological measures describing changes in bone and synovium that correlate with behavior. By better understanding OA pathogenesis and the relationship between OA pain and joint degeneration, new treatments and preventative measures can be developed for OA.

CHAPTER 1 INTRODUCTION

Defining Osteoarthritis

Osteoarthritis (OA) etiology is difficult to define but has been shown to arise from aging, genetic factors, mechanical imbalance, obesity, repetitive occupational movements, joint or limb morphology, and injury34,73,84. Just as there can be multiple risk factors/etiologies associated with OA, there can also be a wide range of joint degeneration and pain-related symptoms.

Joint degeneration can be difficult to detect clinically. Degeneration occurs in all tissues of the joint – cartilage, bone, synovium, ligaments, and menisci; but not all detection methods can describe changes in all joint tissues. The most common diagnostic indicators of OA are joint space narrowing, subchondral sclerosis, and presence of osteophytes which, can be observed through radiography, ultrasound, computerized tomography (CT), or magnetic resonance imaging (MRI)31,58,111,134.

However, these remodeled joint features typically appear in advanced stages of OA.

Early tissue changes in OA can include cartilage fibrillation and erosion, patellar fat pad hypertrophy, and synovial thickening, fibrosis, and inflammatory cell recruitment.

Radiographically, detecting many of these early changes are impossible. Ultrasound and MRI are more sensitive and can detect early articular cartilage, subchondral bone, synovium, and fat pad changes, but ultrasound techniques require a skilled technician and MRI is relatively expensive and not routinely used to diagnose OA. Ultimately, however, these imaging techniques cannot directly assess the biochemical environment of the joint, which has been shown to provide detailed descriptions of tissue damage, inflammation, and joint pain118.

Pain-related symptoms for OA are similarly complex and varied. Generally, OA pain tends to begin acutely and progress to a chronic, possibly neuropathic, state21,35,36,124,136,147. Pain and pain-related symptoms can vary greatly across individuals and across OA pathogenesis22,24,84. For example, an ankle sprain injury will result in proportional pain and disability, but for OA, there seems to be no predictable experience of pain or disability associated with the degree of tissue damage. Patients reporting severe joint pain may exhibit no joint degeneration, while patients with drastically damaged joints may have no pain at all22,36,124. However, once enough tissue damage has occurred, painful and disabling symptoms will eventually appear, though this point can be within a wide range of joint degeneration.

Though changes occur in all tissues within the joint, OA can be diagnosed when any subset of degenerative or painful conditions appear. Moreover, OA can be diagnosed both without any outward symptoms (asymptomatic OA) and, conversely, with only patient-reported symptoms (symptomatic OA)66,118,121,136. This unpredictable mix of tissue degeneration and pain confounds OA diagnosis and treatment. A deeper understanding of OA pathogenesis and OA-related pain and disability is needed to improve OA diagnosis, treatment, and prevention.

Cartilage Changes in OA

Within diarthroidial joints, articular cartilage serves to reduce joint friction and disperse loads. Articular cartilage extracellular matrix (ECM) is composed of water, type

II collagen, and proteoglycans which are divided into four zones: superficial, middle, deep, and calcified. Making up 10-20% of the cartilage thickness, the superficial zone is exposed to the joint space and designed to reduce friction and shear stress with tightly packed, aligned collagen oriented parallel to the cartilage surface. The middle zone

makes up the bulk 40%-60% of cartilage thickness and disperses some compressive forces with proteoglycans and thick collagen fibers. The deeps zone resists more intense compressive forces due to orthogonally aligned collagen fibers, high proteoglycan content, and lowest water concentration. Lastly, calcified cartilage protects the subchondral bone from the highest compressive forces. Cartilage ECM functions to dissipate loads to the subchondral bone, with chondrocytes maintaining and repairing the ECM. Constituting 2% of articular cartilage mass, chondrocytes are widely dispersed and typically only interact with the surrounding ECM, though they can also respond to growth factors and pressure131.

Cartilage degeneration results from a combination of altered joint catabolism and mechanical forces. Though it is unclear whether catabolic changes follow mechanical imbalance or vice versa, both are involved in cartilage degradation. One of the first changes in OA cartilage degeneration is increased proliferation of chondrocytes, chondrocyte clustering, and increased production of ECM proteins. Proteoglycan and type II collagen loss follow due to increased presence of catabolic proteins, including matrix metalloproteases (MMPs) and aggrecanases. As the cartilage degrades, production of regulatory proteins, cartilage components, stress markers, and transcription factors increase46. Additionally, both pro- (interleukins 6, 1β, 15, 18, and tumor necrosis factor α) and anti-inflammatory ( interleukin 2 and 10) cytokines are upregulated in early and late stages of OA46,47,91. Chemokines (such as vascular endothelial growth factor and interleukins 7 and 8) have also been shown to influence

OA inflammation and pain82,91. While these cascades serve as a wound healing response, the cartilage damage eventually becomes too severe for the body to heal and

the imbalance of destructive and constructive processes ultimately degrade cartilage.

As the cartilage matrix degrades molecularly, the overall structure weakens and breaks apart resulting in surface fibrillation (Figure 1-1). As the cartilage degenerates, the mechanical and catabolic environments continue to change eventually forming fissures and lesions through the full depth of the articular cartilage46,52,90.

The catabolic cascades degenerating cartilage become cyclic and loss of cartilage alters joint mechanics, with catabolism and dysfunctional mechanics both leading to a permanent break in homeostasis of the whole joint. Furthermore, articular cartilage contains no vasculature, so healing is relatively limited. Up to this point, clinical treatments to preserve or replace cartilage have had limited long-term effects33,74.

Articular cartilage is a vital tissue within the joint and there is a need to better understand how cartilage degeneration progresses and affects other joint tissues.

Bony Changes in OA

Structure and support are the primary functions of bone, but bone is also a dynamic tissue. Though previously thought to be molecularly isolated from articular cartilage, new OA studies suggest subchondral bone and articular cartilage communicate through diffusion of several cytokines (interleukin 6, tumor necrosis factor

α, vascular endothelial growth factor) and MMPs (MMP-13)89. Bone also structurally responds to joint changes. As OA progresses, cysts form and lesions develop in bone56,93 (Figure 1-1). Changes in subchondral bone remodeling (notably subchondral bone sclerosis) have been observed radiographically in many OA patients31,118,134. Bone responds to changes in its environment and the bony changes observed in OA are largely in response to altered joint forces and exposure to the joint space. When the subchondral bone is exposed to the joint space, cysts and infection can occur which can

be painful to the patient. At the joint margins, osteophytes can form and appear even without significant cartilage damage. There is a need to better understand how bone structurally and molecularly responds to joint changes.

Synovial Changes in OA

The synovium (joint capsule) isolates the joint space from the rest of the body, controls synovial fluid composition and volume, maintains cartilage lubricants in the synovial fluid, and monitors nutrition diffusion into the joint space (Figure 1-1). The synovium can be divided into an intimal layer (synovial lining) only a few cells thick facing the joint space and a subintimal layer which provides the majority of synovium mass and structure129. As OA develops, the synovium chronically thickens

(hyperplasia), toughens, and becomes infiltrated with inflammatory macrophages and lymphocytes46,87. In some instances, cartilage fragments can be deposited in the synovium and incite a local foreign body response, adding to the synovium’s overall inflammatory response46. Because the synovium is both vascularized and innervated, the synovium may be one source of OA pain, but more exploration is needed to understand the synovium’s role in OA pain.

Pain and OA

OA disease sequelae include joint pain, joint dysfunction, and overall decreased quality of life. These symptoms are assessed through physical exams, self-report, questionnaires, and sometimes task-based tests22,24,67,73.

The pain experience is highly complex and can be characterized as neuropathic, psychogenic, and nociceptive35,124,147. Neuropathic pain occurs as a result of nerve damage. Psychogenic pain results from psychosocial conditions, such as emotional or physical stress. Nociceptive pain results from thermal, mechanical, or chemical tissue

damage. Nociceptive pain can be further subcategorized as allodynic or hyperalgesic.

Allodynia is the perception of pain from a typically non-painful stimulus, such as the brush of clothing. Hyperalgesia is an elevation in intensity of pain perceived from a noxious stimulus that is disproportionate to what should be felt124. Nociceptive and neuropathic pain are transmitted along Aδ-fibers and C-fibers, but exhibit different genetic and molecular profiles within neural tissue during chronic states35,147. Studies using both clinical OA patients and preclinical OA models have suggested that nociceptive pain caused by inflammation is initially the bulk of pain felt by OA patients4,36. This might be because early OA patients can sometimes feel pain as the joint is loaded or articulated, but not when the limb is resting. However, movement- evoked OA pain eventually escalates and is experienced even at rest. Moreover, chronic stimulation of nociceptive neurons may ultimately damage neural pathways, increasing the contribution of neuropathic pain124,147. Chronic stimulation also recruits other non-nociceptive neurons to sense and signal pain, which is responsible for an increased pain field in the affected limb22,24.

Though pain and pain-related symptoms contribute largely to an OA patient’s experience, the origin of their pain is unclear. Many opinions hold that cartilage is the first tissue to degenerate, and since articular cartilage is not vascularized or innervated, this may explain why pain is sometimes not experienced in early stages of OA. However in some patients, pain is experienced in early OA and cannot be explained using cartilage degeneration alone. As such, other joint tissues may be the sources of pain, such as the synovium and subchondral bone, as in other pathologies, these tissues have been shown to be sources of pain7. As the joint degenerates in OA, the synovium

often becomes inflamed, and this inflammation may be one source of OA pain.

Likewise, as the joint degenerates, subchondral bone remodels and, in some cases, becomes exposed to the joint space. When this happens, edemas can form in the subchondral bone which have been shown to be very painful35,37. It is also possible for the pain to originate outside the joint space (muscle and neuropathic pain). Just as paths of joint degeneration can vary, it is possible the source of OA pain can also vary.

Preclinical Models of OA Degeneration and Pain

While patients can describe their OA symptoms in detail, assessing joint damage is difficult. Moreover, each patient and disease pathology is unique, precluding experimental designs with proper controls. However, more controlled experiments can be designed using animals. Moreover, the methods available to describe tissue degeneration in detail are an added benefit of animal models, and these benefits may produce a better understanding of OA pathogenesis and symptoms.

OA can be modeled in rodents using genetics, intra-articular injection of a chemical, or through surgical destabilization of the joint6,20,79,94. OA can develop spontaneously in genetically altered mice and in the Hartley guinea pigs, but the resulting OA is unpredictable, and inconsistent in severity and varies in rate of progression. Moreover, while these models are useful and similar to clinical OA, appropriate controls are difficult to incorporate into these spontaneous models relative to other animal models. OA does not develop in every animal, and, specifically with the

Hartley guinea pig, the underlying mechanism responsible for spontaneous OA might not be the same as in humans (possibly a secondary comorbidity)20.

Intra-articular injection of a chemical can partially mimic joint destruction and pain6. The most common chemical models of joint pain use carrageenan,

monoiodoacetate, complete freund’s adjuvant, and papain injected into mice, rats, rabbits, dogs, and horses. While these injections incite a severe inflammatory response, the tissue changes do not replicate degeneration observed clinically. Molecularly, these models also differ from human OA because different catabolic cascades (proteolysis, extracellular glycosidase, and cell apoptosis) are present or missing20. Moreover, the inflammation caused by these chemicals is often significantly more severe than what is seen clinically. However, chemically-induced OA models are widely established, easy to conduct, and have marked pain responses, which is an advantage, but these models may not be the most appropriate model to study OA pathogenesis6,20,38,69.

Joint changes can also be modelled through joint immobilization. While non- invasive, degeneration observed in these models does not reproduce the same molecular changes seen in clinical OA and might better represent degenerative changes associated with joint contracture. These models additionally incite secondary muscle atrophy different from clinical OA20.

Surgically-induced OA models simulate joint remodeling after a traumatic knee injury, such as ACL rupture or meniscal tear38,45,94. The disadvantages of surgical models include creating tissue trauma that is not seen clinically and failing to repair the ligament or meniscal injury that initiates the pathology. However, surgical OA models have the advantages of deriving from a known clinical etiology and having a slow, progressive timeline for joint degeneration that is more similar to clinical OA progression than other models. These models have become increasingly more prevalent in the last decade, with destabilization of the medial meniscus (DMM) in the mouse, anterior cruciate ligament transection (ACLT) in the rat, and medial meniscus transection (MMT)

in the rat being the most prominent surgical models for post-traumatic knee OA94. In both the ACLT and MMT models, full-thickness cartilage lesions develop in 4-6 weeks.

In ACLT, joint movements are no longer constrained by the ACL, resulting in shearing of the cartilage surface. In MMT, joint movements are more constrained than in ACLT, but compressive forces in the medial compartment become concentrated in the medial aspects of the medial compartment. The MMT model also provides the advantage of a sham surgery control, medial collateral ligament transection (MCLT) surgery, which does not develop cartilage lesions but does have some degree of joint destabilization3,76, which should allow behavioral effects caused by joint destabilization to be partly separated from behavioral effects caused by joint remodeling. As such, the

MMT model has a known etiology, a sham procedure, can be consistently produced in the rat, and mimics many aspects of clinical OA degeneration, making it a useful model to study post-traumatic knee OA.

Rodent Behavioral Tests

Clinically, OA pain and disability is assessed through self-report and a physical exam24,35. Defining pain and disability in rodents is more challenging, though a significant history of rodent behavioral phenotyping exists23,143. Several behavioral assessments have been used to quantify pain- and disability-related behaviors in rodents, such as tests for endurance (Rotorod), strength (grip strength meter), weight- bearing (incapacitance meter), mechanical sensitivity (von Frey / Randall-Selitto), and thermal sensitivity (thermal plantar test)15,38,94,116. Often, these behavioral metrics were created to observe neurological or psychological impairments and have been absorbed into the orthopaedic field as a behavioral standard. However, most traditional behavioral

assays are unable to detect or distinguish subtle behavioral changes observed in post- traumatic OA models.

Additionally, because rodents are prey animals, it is not always obvious when an animal is in pain. Rodents tend to hide injuries and discomfort when under stress, which can obscure results from many behavioral assays 71,94,105. Since, neurological and psychological models have larger behavioral changes, these behavior assays may not be sensitive enough for detecting changes in certain orthopaedic models. Moreover, these behavioral assays measure behavioral changes that may not appear in some orthopaedic models (such as thermal hypersensitivity).

While some common rodent behavioral assays may not be ideal for assessing changes in rodent models of OA, gait assays have been used to detect very subtle behaviors associated with orthopaedic injuries in rodents3,10,19,71,76,92,103. Open arena gait assays allow the animal to freely explore. This reduces animal stress and encourages the animal to walk at a self-selected speed – an important variable removed when using treadmill-based assays. Moreover, gait analysis is a technique used to assess OA in the clinic. While clinical gait testing is more sophisticated than rodent assays, rodent gait analysis offers a translatable platform to assess similar behaviors10,71,94.

Rodent Gait Analysis

Rodent gait has long been observed as a metric indicating health. Early gait observations were used to study musculoskeletal disorders common at the turn of the

20th century, though these measures were largely qualitative42,65. The first technical advancement in rodent gait analysis was to track footprints by covering the rodent’s paws with ink and directing them to run across a ream of paper64,108,120. Though

simplistic, this method allowed researchers to begin quantifying spatial gait changes and many of the fundamental spatial gait variables used today (stride length, step length, step width, toe angle, toe spread) were developed using this technique60–62. Ink pads are still used today, though other assays have improved upon the ink and paper method to incorporate temporal measures into gait analysis.

Some temporal gait descriptors have been measured by incorporating seismometers into an arena16, but eventually video cameras became more sensitive and cost effective to use in animal research14,60,64,138. This enabled the full capture of temporal gait characteristics in addition to spatial variables. Famous for his temporal plots describing quadrupedal gait, Milton Hildebrand pioneered how temporal patterns of quadrupedal gait were measured and classified60–62 (Figure 1-2). In his work,

Hildebrand focused specifically on ground contact times and sequences to describe gait patterns. Using Hildebrand’s definitions, rodents use a near trot as their preferred gait pattern. Similar to human walking, trotting is a synchronous sequence in which opposite limbs of the fore and hind limb pairs will move together. For quadrupeds, this means the left hind limb and right fore limb will move synchronously and out of phase with the contralateral hind limb and fore limb. Moreover, once temporal and spatial gait parameters were combined, it became apparent that certain temporal parameters (such as velocity) influence some spatial measures (such as stride length). By combining temporal and spatial (spatiotemporal) gait measurements, gait changes can now be described better than using either metric alone.

Spatiotemporal Video-Based Rodent Gait Assays

Quantifying rodent gait is challenging due to the animal’s small size and rapidity of motion. Though rodents walk in the same counter-synchronous style as humans,

gait-specific assays must be proportionally sensitive, both in spatial and temporal resolution, to detect rodent gait changes. As such, methods that have been used to describe human gait are difficult to scale to the rodent level.

Several video-based rodent gait systems are available commercially11,13,41,80,97.

These gait systems measure a standardized style of rodent gait, making them distinct from open field activity monitors that measure a different exploratory behavior. Video- based gait assays typically use a fixed camera and a translucent floor to record gait

(Figure 1-3). The CatWalk and GaitScan systems use light refraction to track paw contact with the arena floor53,145. Both systems use a high-speed camera (100 frames per second (fps)) to record rodent gait from the ventral view. The Digigait and

TreadScan systems incorporate a treadmill and inclines into their gait assay11,13. Using a clear treadmill design, the fixed high-speed camera (146 fps and 100 fps, respectively) observes the animal ventrally to estimate gait.

Hardware sensitivity is one limitation of current commercial gait systems. In more subtle orthopaedic models, temporal gait changes can be as short as 0.005 seconds and commercial gait systems can only record at video speeds up to 150 frames per second (fps). The detection limit in these systems is a change larger than 0.013 seconds (based on the Nyquist sampling theorem in which your detection frequency should be at least twice that of the observed effect). Another hardware limitation is related to animal views. Some systems calculate spatiotemporal measures only using one camera view, typically from below the animal. This approach can resolve spatial measures with high sensitivity, but can have significant error when resolving temporal measures (Figure 1-4).

As mentioned before, animal stress can also affect gait changes. Because they are prey animals, rodents tend to suppress gait changes under stress. Non-operant assays force the animal to participate in a given activity and have been shown to cause stress71,105. Treadmills, such as in the Digigait, force the animal to walk within a restricted area and range of speed. This approach can be a stressful environment for rodents, which can affect their gait changes71. Operant assays allow the animal to freely participate or avoid a given behavior, and in these environments, animals are less likely to mask their pain.

The operant, high-speed video recorded gait assay developed in our lab

(explained in detail in chapter 1) has been shown to detect temporal sequence shifts as short as 0.008 seconds in rats1,3. This sensitivity can be used to detect the subtle behavioral changes that may appear in early OA and has already been able to detect gait changes in late stages of OA. Moreover, by incorporating a second animal view using a mirror, both spatial and temporal variables can be resolved with optimal sensitivity.

Though this arena can detect subtle gait changes in rodents, analysis is performed manually to date and can take several months to calculate results. While analysis does not need to be quite as arduous as taking a rule to a paper inked with footprints, gait data sets are often large and manually calculating spatiotemporal gait variables can take significant time and effort even using digital methods. Spatiotemporal measures are derived from two time variables (foot-strike and toe-off) and a location (x and y coordinates) associated with each step. Manually digitizing these variables requires the digitizer to go through every frame of a video to identify each step. Once

foot-strike, toe-off, and paw location are recorded for each step, these data are then used to calculate several gait variables (explained in detail later in this chapter). While software such as MATLAB or excel can aid calculation of gait variables, summarizing parameters from a single animal can end up taking several hours. Furthermore, manual digitization is tedious and fatigue can reduce the accuracy of foot-strike, toe-off, and paw location estimations. Commercial gait platforms often provide automated gait analysis software, but these systems are relatively expensive, proprietary, and unalterable. There is a need for open-source freeware to automatically analyze gait while not limiting the user to a specific style of gait platform.

Spatiotemporal Gait Parameters

During walking, a limb alternates between ground contact (stance) and an aerial phase (swing). The stance phase begins with foot-strike when the animal makes contact with the ground and ends with toe-off when the animal breaks contact with the ground, and vice versa for the swing phase. Identifying the time and spatial location of these gait events is the foundation of the Hildebrand classification of quadrupedal gait sequences62. For optimal sensitivity, foot-strike and toe-off events should be determined from the sagittal view of the animal, while spatial parameters, such as stride length and step width, should be calculated from the ventral views of the animal.

Temporal variables summarize several time relationships in the gait sequence61,62. Percentage stance time (%ST), also known as duty factor, is stance time divided by stance plus swing. This variable represents the stance phase duration for a given limb in the gait cycle and tends to follow a more linear correlation to velocity than stance time alone71. Typically, animals spend equal time on the left and right limb of a limb pair (balanced). However, injury can alter a limb’s duty factor in the gait cycle.

Similarly, temporal symmetry measures the synchronicity of the gait sequence by describing when a limb's foot-strike occurs in time relative to the previous and following foot-strike of the opposite limb. During normal walking, a left limb foot-strike will occur roughly halfway (50%) in time between two right limb foot-strikes. In unilateral injuries, limping (antalgic) syncopations can occur, shifting symmetry away from 50%. Finally, stride frequency is a derivative temporal measure (velocity/stride length) that estimates how often a stride occurs as the animal walks. This variable is directly related to the commonly reported cadence variable (foot-strikes per second).

Spatial variables represent the geometric aspects of rodent gait62,71. Stride length is the most direct distance between two consecutive foot-strikes of a single limb. Step length is a similar measure, but instead takes the distance along the direction of travel between two consecutive foot-strikes of opposite limbs of a limb pair. Step width is the orthogonal distance between a contralateral foot-strike and two ipsilateral foot-strikes.

Using these common spatiotemporal variables, limp-like (antalgic) and shuffle-like gait compensations can be assessed in rodents.

Using Histology to Assess Tissue Degeneration

OA-related tissue degeneration can be viewed in detail on histological slides

(Figure 1-1). Using this technique, the following can be readily observed: cartilage

(lesion formation, chondrocyte cloning, proteoglycan loss), bone (formation of osteophytes, thickening of growth plate), and synovium (thickening, immune cell recruitment, angiogenesis) can be readily observed.

Histological evaluation of tissue is one benefit of rodent OA models. Histological methods are described in detail later in this dissertation, but the general principle is to stain a very thin slice of fixed tissue (called a section). Once stained, the section of

tissue can be viewed in detail under a microscope (Figure 1-1). By taking many sections, different stains can be used to assess structural, cellular, or molecular features of tissue. This versatility is advantageous when studying multiple aspects of a disease, but quantifying histological changes is challenging. Additionally, even within the field of OA, stains and measures used for assessing cartilage damage may not be standardized or optimized for assessing bony changes. However, in the last 40 years, there has been a movement to improve and unite histological grading resulting into ever improving methods.

Early Histological Grading Schemes for OA

The majority of modern histological grading methods are based on a macroscopic method developed by Collins114,117 and the microscopic Histologic

Histochemical Grading System (HHGS) developed by Mankin et al 43,95,117. These methods were both developed in the mid-20th century using human cadaver tissue or tissue removed during arthroplasty. Collins’ work focused primarily on the distal femur and qualitatively described gross cartilage surface structure, cartilage lesion size, and osteophytes. While no direct lineage traces Collins' method to more current methods, his general approach influenced how modern researchers assess tissue changes in

OA114.

The HHGS (also known as the “Mankin score”) developed by Henry Mankin is still widely used in clinical and preclinical OA research95. Mankin's method focuses solely on cartilage degeneration. The qualitative scoring system assigns a numeric value to four features of cartilage: structure, cells, Safranin-O stain saturation, and cartilage tidemark integrity. For these grades, a score of 0 indicates normal features and larger numbers indicate degenerative features, with the end score being a sum of all the

numbers from each feature. While this system was a vast improvement over previous grading practices and a pivotal first step to standardize and quantify histological grading, it also contains inherent flaws43,117,128. First, the HHGS is affected by the grader’s expertise, interpretation of the tissue changes, and skill at staining tissue with

Safranin-O. Safranin-O is ideal to observe proteoglycan loss indicative of cartilage health, but the stain is difficult to control and can return misleading results with large batch to batch variability. Furthermore, once the sections are stained and graded, the

HHGS results can be mathematically misleading. Each cartilage features contains 2 to 7 scoring options, so the score which contains 7 grades is proportionally more influential of the final score than the score with only 2 grades. Furthermore, only qualitative relationships exist between the numeric values of the scores, making it very difficult to determine where along pathogenesis the tissue has degenerated. And lastly, because the cartilage features are summed, it is not clear what type of damage the tissue has undergone. This is problematic because 2 identical scores may exhibit very different degeneration profiles. Though the HHGS has several fundamental flaws, it is easy and fast to perform and researchers have improved upon it to create more detailed modified

Mankin scoring systems used widely today.

Modern Histological Grading Systems for OA

In 2006, the Osteoarthritis Society International (OARSI) assembled a working group to develop a standardized grading system applicable for the entire OA research community. The result of this endeavor was the Cartilage Histopathology Assessment

System (OOCHAS), a cartilage-based scoring system in which each end score describes unique tissue damage117. Like Mankin’s HHGS, the OOCHAS also grades normal tissue as 0 and increasing scores associate with increasing degeneration.

However, the OOCHAS contains 13 distinct cartilage conditions between 0 and 6.5 and the grader assigns the entire joint a single value within that range. Though an improvement over the many different modified Mankin scores used throughout the field, this scoring method was unable to eliminate effects from grader subjectivity or ambiguous stain quality. Moreover, this system does not provide sub-scores that describe joint tissue changes observed by OA researchers throughout the joint.

A field-wide shift in OA histology occurred when OARSI published the histopathology initiative's updated recommendation in 201043. This mass of literature describes semi-quantitative measurements of cartilage, bone, and synovium changes in several clinical and preclinical OA models. This method is a marked improvement over previous grading systems not only because it is tailored to specific models and technology (tissue histology to clinical imaging), but tissue degeneration is also quantified in the whole joint. For the rat, the OARSI recommendations include measurements of the cartilage lesion width at different cartilage depths, qualitative cartilage damage scores for each 1/3 of the medial compartment tibial plateau, widths of qualitative cartilage damage affecting both the surface and deeper cartilage zones, ratios of missing cartilage relative to the full cartilage thickness for each 1/3 of the medial compartment tibial plateau, a qualitative score reflecting the state of the osteochondral interface for the medial tibial compartment, growth plate thicknesses, thickness of the synovium, and thickness of osteophytes. Though 85% of the measures still define cartilage damage, joint inflammation is measured indirectly via synovium thickness and three measures describe bone growth plate and osteophyte changes.

However, there is still need for improvement. Because OA is a disease of the whole

joint, more tissues need to be included in analysis. Cartilage damage is an obvious structural change in OA, but little correlation between cartilage damage and OA symptoms have been found4,76,135. Since joint pain may not originate with cartilage damage, other tissues should be explored.

Though other tissues should be analyzed, quantifying changes in tissue is difficult because of its structural complexity and variability. This diversity is caused, in part, by natural morphological variation between animals and by the harsh processes required to prepare a tissue sample for histology. Often tissue is slightly deformed from being dehydrated or infiltrated with paraffin. Even in a pristine tissue sample, joint tissues are not homogenous. Bone, cartilage, and the synovium all contain layers of unique cell patterns that often transition seamlessly. Moreover, these patterns can vary greatly between sections making the natural "biological noise” much greater than a given measure. Finding new, quantitative measures within this noise is needed to fully understand joint degeneration and communicate findings effectively.

Aim of This Dissertation

The global aim of this dissertation is to help bridge the gap between tissue degeneration and symptomatic pain and disability using a rodent post-traumatic model of knee OA. In Chapter 2, the behavioral and degenerative profiles of a unilateral medial meniscus transection (MMT) model of knee OA are evaluated at 1, 2, 4, and 6 weeks after surgical initiation. Behaviors assessed are mechanical sensitivity of the ipsilateral paw and spatiotemporal gait patterns. At each time point, animals were euthanized and the knees processed for histology. Each joint was sectioned and graded using the current OARSI histopathology recommendation. MMT animals exhibited progressive tissue degeneration and gait compensations culminating in several correlations between

histological tissue damage and behavior. Interestingly, the sham group (which received a medial collateral ligament transection (MCLT)) also exhibited some gait changes, but limited histological evidence of tissue damage. This chapter demonstrates that while gait modifications and histological changes are coincident, they are not necessarily correlated, even in a preclinical model of knee OA.

Chapter 3 introduces a new automated gait analysis method. This open source method for automated gait analysis through hues and areas (AGATHA) significantly reduces time spent digitizing and analyzing spatiotemporal gait data. AGATHA is used to explore the effects of video frame rate on the sensitivity and accuracy of gait data collection. Furthermore, AGATHA exhibits its utility in both orthopaedic and spinal cord injury models. Video frame rate significantly affected analysis accuracy and precision.

Of the spinal cord injury models, the C2-hemisection group walked with a pronounced limp of the hind feet. As measured by AGATHA, MMT animals exhibited the same trends seen in Chapter 2. Moreover, through the development of the AGATHA method, our techniques can now be more easily distributed to the OA community, hopefully allowing for more broad use of gait assessments in OA preclinical models.

Developed for many of the same reasons as AGATHA, Chapter 4 describes a semi-automated GUI for the evaluation of knee OA (GEKO) which grades histological sections according to the most current OARSI histopathology recommendations. Using intra class correlation coefficients (ICCs), GEKO is validated against manual grading using histological samples from 1, 2, 4, and 6 weeks after animals received the MMT surgery. GEKO is also used to explore inter-grader variation and as a teaching tool for new graders. When compared to manual grading, GEKO produced ICCs above 0.97 for

nearly all histological grades. GEKO also exhibited better consistency than manual grading when assessing repeated images within a single grading session. While inter- grader ICCs were relatively low for many histological measures, intra-grader ICCs were much higher. Moreover, graders produced ICCs generally above 0.95 when assessing intra-grader variation across multiple grading sessions. The same histological grades tended to have low ICCs for each comparison paradigm. These histological measures were largely near the asymptotes of grading ranges in which a small change in grader input can result in a large measurement change. Overall, GEKO reduced histological grading time by more than a factor of 8 without any appreciable loss in accuracy or precision. While the findings in Chapter 4 highlight some unknown or underappreciated sources of methodological grading variation in histology, GEKO also represents an opportunity to further standardize histological grading in the OA field.

In Chapter 5, quantitative histological measures describing changes in bone and synovium for MMT and MCLT animals at 1, 2, 4, and 6 weeks of OA development are described and correlated with behavior. Measures in the subchondral bone (ossification of articular cartilage, subchondral bone density, edemae) and synovium (cell population density, alignment of cells, cell aspect ratio) correlated with mechanical sensitivity and spatiotemporal gait patterns. Moreover, progressive tissue changes were also found in the MCLT sham group. Unlike cartilage-centric histological changes, these bony and synovial changes did show stronger correlations to gait compensations in the rat, indicating bone and synovium may be more likely contributors to early OA pain than cartilage.

General summary discussion and conclusions are provided in Chapter 6.

Furthermore, potential future directions are proposed, including commentary on sham procedures for post-traumatic OA, additional capabilities to incorporate into automated gait analysis and histological grading platforms, and other joint tissues that may contribute to OA pain.

Figure 1-1. Histological sections of medial knee compartments stained with Toluidine blue. Photos courtesy of author.

Figure 1-2. Representative Hildebrand plots of common gait changes (Jacobs 2014).

Figure 1-3. Graphical representations of common gait arena arrangements. . Photos courtesy of author, Noldus, and Mouse Specifics.

Figure 1-4. Two planes of view can optimize spatial and temporal gait resolution. Photos courtesy of author.

CHAPTER 2 SPATIOTEMPORAL GAIT COMPENSATIONS FOLLOWING MEDIAL COLLATERAL LIGAMENT AND MEDIAL MENISCUS INJURY IN THE RAT: CORRELATING GAIT PATTERNS TO JOINT DAMAGE

Since ligamentous and meniscal injuries significantly increase the risk for developing osteoarthritis (OA)32,88, surgical simulation of a ligament and/or meniscal injury is commonly used to model OA development in rodents57,72,142. In the rat knee, focal cartilage lesions form in the medial compartment several weeks after a surgically- simulated radial tear of the medial meniscus or rupture of the anterior cruciate ligament15,57,72. Using histology, detailed assessment of joint remodeling is possible in these rodent OA models43; however, quantifying the symptomatic and functional consequences of joint injury can be difficult in rodent OA models.

Following a combined surgical transection of the medial collateral ligament

(MCLT) and radial transection of the medial meniscus (MMT) in rats, fibrillation of the articular surface is seen at 1-2 weeks post-surgery, followed by the development of full- thickness cartilage lesions at 2-6 weeks post-surgery. Similarly, osteophytes form along the medial margin of the joint at 4-6 weeks post-surgery3,15,72. However, rats that received a sham surgery (MCLT alone) do not show evidence of cartilage loss or osteophyte formation over a similar time scale15,72. Our group has previously described gait abnormalities in rats receiving MCLT+MMT surgery, where rats receiving

MCLT+MMT surgery spent unequal time on their hind limbs (imbalanced stance time) and used asymmetric foot-strike sequences after surgery (limping)3. Similar compensations were identified in animals in the MCLT sham surgery; however, the magnitude of the compensation tended to vary between groups. As such, comparison of

MCLT to MCLT+MMT in the rat may provide an opportunity to investigate a joint trauma

that results in cartilage damage (MCLT+MMT) relative to a joint trauma that does not result in cartilage damage (MCLT sham), and how the specific effects of cartilage loss alter an animal’s behavior and gait.

Due to the longitudinal design of our past experiment3, histological grades were only available at the final time point; and as such, correlative relationships between the detected gait abnormalities and osteoarthritic changes within the joint could not be constructed. In this study, the correlative relationships between gait compensations and osteoarthritic remodeling within the joint are investigated in rats with MCLT alone

(sham) and rats with MCLT+MMT. First, a detailed characterization of the spatiotemporal gait pattern and tactile sensitivity is provided for rats receiving MCLT sham or MCLT+MMT surgery at 1, 2, 4 and 6 weeks after surgery. Following this detailed characterization of rodent gait, osteoarthritic remodeling within the joint is described using the quantitative OARSI histological grading system for the rat43. Finally, correlative relationships between gait compensations and joint remodeling are investigated for MCLT sham and MCLT+MMT. Our data demonstrate that, while correlations exist between histological scores of joint damage and behavioral changes in the MCLT+MMT group, similar behavioral changes could be found without significant cartilage damage or osteophyte growth in the MCLT sham group. Combined, these data indicate cartilage damage and behavioral changes in the rodent may be coincidental in the MCLT+MMT model of OA and suggest non-cartilage mechanisms may be involved in the development of gait compensations following joint injury in the rat.

Experimental Design and Animal Surgery

64 male Lewis rats (~250 g) were obtained from Charles Rivers Laboratories and acclimated to the University of Florida housing facilities for one week prior to

investigation. After acclimation, 32 rats received MCLT+MMT surgery, and 32 rats received MCLT alone (sham), as previously described3,72; surgeries were performed in four 16 animal cohorts, with 8 MCLT+MMT surgeries and 8 MCLT sham surgeries performed for each surgical day. Briefly, animals were anesthetized in a 4% isoflurane sleep box, prepped for aseptic surgery, and transferred to a sterile field with anesthesia maintained by mask inhalation of 2% isoflurane. A medial midline skin incision was made on the right hind limb, and the medial collateral ligament (MCL) was exposed via blunt dissection and transected. At this point, the wounds of animals receiving MCLT alone (sham) were closed as described below. For animals receiving MCLT+MMT, the joint was placed in a valgus orientation to expose the central portion of the medial meniscus; then, a complete radial transection was performed in the central portion of the medial meniscus using an 11 blade scalpel. For animals tested at 2, 4, and 6 weeks after surgery, the surgical site was closed with 9 mm wound clips, which were removed

10-14 days after surgery, and thus were not present during behavioral testing. For animals tested at 1 week, 5-0 vicryl sutures were used to close the incision site; these sutures did remain in place during behavioral testing at 1 week. Spatiotemporal gait pattern and mechanical sensitivity testing was performed in 8 MCLT sham and 8

MCLT+MMT animals at 1, 2, 4, and 6 weeks after surgery, as described below (n=8 per treatment per time point). Surgical cohorts were conducted in the following order: Week

4, Week 1, Week 2, and Week 6. After behavioral testing, animals were humanely euthanized via exsanguination under deep anesthesia with knee joints collected for histological grading. All methods were approved by the University of Florida IACUC and conform to AAALAC recommendations on animal research.

Spatiotemporal Gait Testing

Animals were acclimated to the gait arena over a period of 3 days prior to the start of the experiment. Spatiotemporal gait parameters were then measured in all animals at their respective time point, as previously described and reviewed2–4,71.

Briefly, animals were placed in a 60”x18” open gait arena and allowed to freely explore without an external stimulus or food enticement. As animals voluntarily explored the gait arena, five videos of rodents walking were acquired with a high-speed camera

(RedLake, M3, 250 fps). A mirror set at a 45° angle under the transparent arena floor allowed for simultaneous recording from the side and beneath the animal. Videos that included a minimum of 4 complete gait cycles across a consistent walking speed were digitized by-hand using the DLTdataviewer subroutine in MATLAB59; digitizers were blinded to the treatment groups during processing. Digitized data were processed to calculate the median value of the following parameters for each trial: velocity, stance time, swing time, stride time, stride length, and step width. Using these parameters, spatial symmetry, temporal symmetry, percentage stance time, stance time balance, and the percentage of the gait cycle dedicated to single limb support were calculated as previously reviewed71 and described in Figure 2-1. Based on past data demonstrating limited changes to the fore limb spatiotemporal characteristics of rodents due to a hind limb injury1–5,68,71, this work focuses on spatiotemporal changes in the hind limbs only.

Stride length, step width, percentage stance time, and the single limb support phase are known to be strongly correlated to an animal’s walking velocity and weight, and failure to account for these sources of variation can reduce the sensitivity of subsequent statistical analyses71. To account for the effects of animal size and walking velocity, stride length, step width, percentage stance time, and the single limb support

phase were normalized to the predicted gait profile of weight- and velocity-matched naïve Lewis rats. This normalization process is visually described in Figure 2-2 using the percentage of the gait cycle dedicated to single limb support; our database on the gait characteristics of naïve Lewis rats is available for download at bme.ufl.edu/labs/allen and represents 280 gait trials collected in 28 different naïve

Lewis rats at 49 different weights over a period of 8 years. In brief, gait trials collected on naïve Lewis rats were used to predict the stride length, step width, percentage stance time, and the single limb support phase for a weight- and velocity-matched Lewis rat. With this prediction, velocity and weight independent residuals of stride length, step width, percentage stance time, and the single limb support phase can be calculated by subtracting the predicted value for a weight- and velocity-matched Lewis rat from the measured value for a Lewis rat with MCLT sham or MCLT+MMT surgery. Once data were transformed into velocity- and weight-independent parameters, data for each rat were averaged, such that the average gait profile for an animal directly corresponded to single histological profile in the correlation analyses described below. To be clear, the left-most graphs in Figure 2-2 visually describe the correlation between velocity and single limb support only; however, please note that residual data in the remainder of

Figure 2-2 and the figures presented in the results are normalized to both weight and velocity.

Mechanical Sensitivity

Mechanical sensitivity was assessed using Chaplan’s up-down protocol for von

Frey filaments in the experimental groups listed above and in eight littermate naïve

Lewis rats (control)18. The littermate-matched naïve control rats used for the von Frey analysis were also included in the naïve rat gait database; however because these

animals were not able to fully replicate the weights and velocities observed in our study, the full database was used for the gait analysis. These naïve rats were eventually used in a separate experiment unrelated to the present study.

During von Frey testing, experimenters were blinded to the animal surgical group during the test. Briefly, animals were acclimated to a wire mesh-floored cage for 15 minutes prior to the application of the 4.0 g von Frey filament to the plantar region of each hind foot. Using a von Frey filament series (0.6, 1.4, 2, 4, 6, 8, 15, and 26 grams), a withdrawal-tolerance sequence was constructed, wherein a less stiff filament was applied following a paw withdrawal and more stiff filament was applied following filament tolerance. Using these data, the force where withdrawal and tolerance are equally likely can be approximated through Chaplan’s approximation (50% paw withdrawal threshold)18.

Histology

Following behavioral testing, animals were humanely euthanized under deep anesthesia. Operated and contralateral knees were dissected, fixed in 10% neutral buffered formalin for 48 hours at room temperature, decalcified in Cal-Ex reagent for 3 weeks at 4°C (Fisher Scientific), then embedded in paraffin wax using vacuum infiltration. Sequential frontal sections (10µm) were acquired on a rotary microtome, taking at least 1 section every 100 µm through the central regions of the knee. The central region was defined as sections past the anterior horn of the medial meniscus through to the posterior horn of the meniscus. Toluidine blue staining was conducted on central sections, with the section that represented the most severe degeneration on the tibial plateau selected for grading. Sections were graded using the OARSI histopathology scheme for the rat43 ; when called for by the OARSI histopathology

scheme for the rat, pixels were converted to geometric distances using a calibrated digital recital. This scoring system evaluates cartilage damage, osteophyte formation, synovial membrane inflammation, and growth plate changes through semi-quantitative grading relative to sample images117 and by measuring physical changes in cartilage and bone in the 2-D histological images43.

Statistical Analysis

Dunnett’s test was first used to compare gait parameters to expected values or tactile sensitivity to pre-operative controls, correcting for compounding type I errors caused by 8 or 16 comparisons to control. Differences between surgical groups and across time were investigated using a 2-way analysis of variance test followed by a

Tukey’s honestly significant difference (HSD) post-hoc test when indicated. Correlative relationships between histological measures and behavioral measures were constructed for MCLT sham and MCLT+MMT animals using univariate linear models. In order to obtain data that spans the range of histological damage and behavioral changes, correlation models were constructed across time points for both MCLT sham and

MCLT+MMT animals (n=32).

Gait Patterns

Spatiotemporal gait abnormalities were detected in both rats with MCLT sham surgery and rats with MCLT+MMT. Spatial compensations are graphically summarized in Figure 2-3 and visually summarized in Figure 2-4. While walking velocity varied between time points, no velocity differences were found between MCLT sham and

MCLT+MMT surgery within a given time point (Figure 2-3A); however, the velocity differences between time points and the strong correlation between most gait

parameters and velocity highlight the importance of standardizing spatiotemporal gait parameters to the animal’s selected velocity (Figure 2-2).

Spatial symmetry investigates the geometric symmetry of the foot-strike pattern, where a value near 0.5 indicates a right hind limb footprint is approximately halfway between two left hind limb footprints along the direction of travel (Figure 2-1). While gaits were spatially symmetric at 1 week post-surgery, animals with MCLT+MMT surgery were spatially asymmetric at 2 and 6 weeks post-surgery (p=0.005 and p<0.001, respectively) and tended to be spatially asymmetric at 4 weeks post-surgery

(p=0.06, Figure 2-3B). MCLT sham caused spatially asymmetric gaits at 2 weeks post- surgery (p=0.01) and tended to be spatially asymmetric at 4 weeks (p=0.10), but not at

6 weeks post-surgery.

Step widths were narrower than expected in the MCLT sham at 1, 2, and 4 weeks post-surgery (p<0.001), but not at 6 weeks post-surgery. MCLT+MMT resulted in step widths that were comparable to historical controls at 1 week post-surgery, but were narrower at 2, 4, and 6 weeks post-surgery (p<0.001, Figure 2-3C). Narrower hind limb step widths may indicate that animals are primarily using their forelimbs for balance.

Stride length residuals were reduced with MCLT+MMT at 4 and 6 weeks (p=0.04 and p<0.001, respectively) and with MCLT sham at 6 weeks (p<0.001, Figure 2-3D).

Temporal changes are graphically summarized in Figure 2-5 and visually summarized in Figure 2-6. Please note that stance time imbalance and single limb support residuals are presented as percentage of the gait cycle, and this should not be confused with percentage change. For example, single limb support phase for a hind limb at walking velocities must be between 1-50% of the gait cycle; limb swing times

over 50% of the gait cycle are defined as running gaits. Based on historical data collected on naïve Lewis rats, the range of single limb support is 18.7-50% of the gait cycle. Thus, a 2% imbalance or 3% single limb support residual in expected single limb support is a significant magnitude relative to the range of possible values for temporal gait parameters (≈32% of gait cycle).

As with spatial symmetry, temporal symmetry investigates the symmetry of the foot-strike pattern in time, where a temporal symmetry value near 0.5 indicates an animal’s right foot-strike occurs approximately halfway between two left limb foot-strikes in time (Figure 2-1). Temporal asymmetries were detected in the MCLT+MMT group on week 1 and week 4 (p<0.001, p=0.03, respectively, Figure 2-5A). Similarly, stance time imbalance occurs when an animal spends more time on one limb relative to the contralateral limb (Figure 2-1). In our animals, stance time imbalance was observed in the MCLT+MMT group at week 1 and week 6 (p=0.003, p=0.048, respectively Figure 2-

5B).

Evidence of Antalgic and Shuffling Gait Compensations

Antalgic compensations occur when the contralateral limb compensates for an injured limb. Conceptually, this is similar to a limp associated with a sprain or muscle strain. Antalgic compensations are primarily identified by temporal asymmetry and stance time imbalance and to a lesser degree by spatially asymmetric footprint patterns71. The strongest evidence of antalgic compensation is seen in the MCLT+MMT group at week 1, including a combined temporal asymmetry and stance time imbalance.

At week 2, both MCLT sham and MCLT+MMT animals used spatially asymmetric patterns that were temporally symmetric and balanced, which is indicative of altered foot placement but not limb disuse. At week 4, MCLT sham and MCLT+MMT animals used

balanced gaits, with a tendency toward spatially asymmetric foot placement;

MCLT+MMT also had temporally asymmetric gait patterns. Overall, this difference in the week 2 and week 1 gait patterns of MCLT+MMT rats may represent a transition from antalgic to shuffle compensations (below). By week 6, the gait patterns of both MCLT sham and MCLT+MMT animals were temporally symmetric, with some evidence of stance time imbalance and spatial asymmetry in MCLT+MMT animals.

While the uninjured contralateral limb can compensate for an injured limb through antalgic compensations, shuffling gait compensations can also reduce the period of time where a limb must bear weight without contralateral limb support71. Conceptually, this protective gait is similar to the way one might walk the day after a hard workout at the gym (where muscles are sore and tight in both limbs). While conceptualizing shuffle compensations as ‘the day after a hard workout,’ these gait sequences are also protective. By ‘shuffle stepping’, stance times are increased on both limbs, effectively raising the periods of where both hind limbs are in ground contract (double limb support) and reducing periods where only one hind limb is in ground contact (single limb support). As such, these gait sequences are often described by reduced periods of single limb support and decreased stride lengths in both limbs of a limb system71. While there is some evidence of antalgic compensations in our animals, there is strong evidence of shuffling gait compensations in both MCLT sham at 6 weeks and

MCLT+MMT animals at 4 and 6 weeks post-surgery, indicated by reduced stride lengths and reduced single limb support phases on both hind limbs (p<0.001 in all groups,

Figure 2-5C).

To be clear, while shuffling compensations are often more difficult to identify in rodents than antalgic compensations, shuffling compensations should not be considered less significant in magnitude. Both antalgic and shuffling compensations protect an injured limb from loading by decreasing the period where a limb must bear weight without contralateral limb support. This protection, whether through an antalgic or shuffling compensation, is easiest to observe through reduced periods of single limb support. As an example, MCLT+MMT animals show evidence of antalgic compensation at week 1 by a reduced period of single limb support on the injured limb only; at week 4 and week 6, both MCLT sham and MCLT+MMT animals show evidence of shuffling compensation by reduced periods of single limb support in both limbs (Figure 2-2).

Mechanical Sensitivity is Altered

Animals in the MCLT group had heightened sensitivity to tactile stimuli at week 1 and week 4 (p<0.001 and p=0.009, respectively, Figure 2-7), but returned to near naïve control levels by week 6. Animals with MCLT+MMT had heightened sensitivity to tactile stimuli at week 1, 4, and 6 (p<0.001 at all time points) and tended to have heightened sensitivity at week 2 (p=0.07). In addition, the tactile sensitivity of animals with

MCLT+MMT was significantly different from MCLT sham at week 6 (p=0.002).

Histology Shows Progressive Degeneration

Histological changes are graphically summarized in Figure 2-8 with sample images from each time point provided in Figure 2-9. As expected, significant cartilage damage was observed in the MCLT+MMT group, but not in the MCLT sham group. The percentage of the articular cartilage surface that showed evidence of fibrillation was significantly higher in the MCLT+MMT group relative to the MCLT sham at all time points (p<0.04), with the severity of the lesion increasing over time (Figure 2-8A). In

general, the cartilage lesion was located in the central or medial aspects of the medial compartment, progressing from a maximum depth of 50% at 1 week to near-full thickness lesions at 6 weeks (Figure 2-8B). Coinciding with cartilage damage, significant osteophytes and calcified cartilage damage were found in the MCLT+MMT group at 4 and 6 weeks post-surgery (Figure 2-8C&D). The MCLT+MMT group had a significantly thicker joint capsule at week 1 (p=0.04).

Univariate Correlations

Correlations between histological changes in the joint and behavior are shown where the Pearson correlation coefficient R is shown on top and the p-value for the slope term in univariate model is shown on bottom (Table 2-1). Due to the lack of variance in histologic measures, correlations could not be constructed for the MCLT sham group other than for joint capsule thickness; however, a positive correlation was seen between synovial capsule thickness and hind limb sensitivity in the MCLT sham

(Figure 2-10). This correlation was not repeated in the MCLT+MMT group; instead, a negative correlation is identified between the amount of the cartilage surface affected and the 50% withdrawal threshold.

A positive correlation between joint damage and the 50% paw withdrawal threshold is somewhat confusing, as many would postulate that a thicker synovial lining would occur due to synovial inflammation and therefore should associate with a lower

50% withdrawal threshold (heightened sensitivity). If this relationship occurs, the

Pearson correlation coefficient should be negative, as found for the 50% withdrawal threshold and amount of the cartilage surface affected in the MCLT+MMT group.

Because the synovial capsule was damaged with MCLT+MMT surgery, but not with

MCLT sham, we speculate the thickening of the synovial lining may not occur until a

later time point with MCLT sham. Thus, the correlation identified in the MCLT sham group may be an arbitrary relationship that results from the recovery of limb sensitivity to baseline levels following surgery and the delayed development of synovial damage following MCLT sham; however, a skin incision sham (which was not included in this study) would be necessary to verify this hypothesis.

Multiple significant correlations between joint histological measures and animal behavior were identified in the MCLT+MMT group. First, stance time imbalance was negatively correlated to the cartilage lesion width and depth in MCLT+MMT animals. A limb imbalance greater than 0 indicates more time is spent on the left limb

(contralateral) than the right limb (injured), a compensation that is only observed at week 1. Again, most OA researchers would postulate this correlation would be positive; as the lesion size increases, the gait sequence becomes more imbalanced. Thus, like medial joint capsule thickness, correlations between limb imbalance and joint histology could be driven by arbitrary correlations resulting from the recovery of imbalance parameters to baseline levels following the MCLT+MMT surgery.

Negative correlations were observed between multiple histological measures and the stride length residual and single limb support residuals. Conceptually, these correlations follow the anticipated relationship; as the joint damage increases in severity, stride lengths reduce and periods of single limb support reduce. However, these same gait compensations were identified with MCLT sham, despite a lack of damage identified through the OARSI histological grading system (Figure 2-10). When

MCLT sham and MCLT+MMT data sets are assessed in conjunction, the correlations in

the MCLT+MMT group also appear to be coincidental, despite following the predicted pattern.

Future Directions to Correlate Symptoms and Behavior

While non-life-threatening, osteoarthritis (OA) is incurable and ultimately results in chronic debilitating symptoms. Complicating clinical OA treatment is the common finding that the severity of joint degeneration does not necessarily correlate to the symptomatic consequences of OA75. Clinically, debilitating symptoms can appear across a broad spectrum of joint degeneration, where patients may experience intense pain or joint dysfunction with little or no evidence of tissue damage or minimal symptoms with severe tissue degeneration12,55. The lack of a unifying relationship between cartilage damage and symptoms in OA patients may be attributed to psychosocial conditions and a person’s coping capabilities75,130. Stress and environmental factors can clearly affect OA pain experiences; however, a potential exists that cartilage loss simply does not explain a significant portion of OA symptoms, even if environmental factors are well controlled.

Our data in this study provide an example of this conundrum using a rodent OA model, where environmental factors are well-controlled between the OA and non-OA group. This experiment was specifically designed to compare animals receiving an MCL sham surgery to animals receiving an MCLT+MMT surgery. Rats began the experiment as littermates, were co-housed, tested side-by-side on the same day, and were on the same diet throughout the experiment. While correlations could be identified between cartilage damage and changes in animal behavior in the OA cohort (MCLT+MMT group), the same behavioral changes were found in the non-OA cohort (MCLT sham) despite the lack of significant cartilage damage. In combination, these data demonstrate

the inherent limits of correlative relationships identified in our OA cohort. When taken in context with MCLT sham results, correlations in the MCLT+MMT group appear to be coincidental or a relatively minor contributor to the behavioral phenotype in the OA group. Moreover, these data pose a fundamental question of the rat MCLT+MMT model of knee OA: Will therapeutics that prevent or reverse cartilage degeneration following a simulated meniscal injury have an efficacy on OA-related symptoms and disability?

Our experiment was designed to compare MCLT sham to MCLT+MMT surgery at multiple post-surgical time points, thereby allowing us to compare the behavioral changes in an OA cohort to a non-OA cohort at different levels of joint damage. While

MCLT sham and MCLT+MMT animals were treated identically at each time point, a historical database on weight- and velocity-matched naïve animals were used as controls. This database represents 280 gait trials collected in 28 different naïve Lewis rats at 49 different weights in 6 separate experiments conducted over a period of 8 years at 2 different research institutions (some rats were tested longitudinally across time). Because of the strong correlations of most gait variables to animal weight and walking velocity (Figure 2-2), weight- and velocity-matched historical controls are advantageous relative to pre-operative controls. Walking velocities can vary markedly between testing days and between trials, and most rats used in OA research gain 10-

50% body weight in the weeks after surgery. Failure to account for these covariates in the statistical analysis markedly reduces the sensitivity of the gait analysis71. The control database used in this study represents multiple experiments across a wide range of weights (307-425 grams) and walking velocities (15.4-76.3 cm/sec), allowing for the humane reduction of research animals by eliminating the need to replicate naïve data

collected in prior work. Nonetheless, historical controls are not without limitations, as environmental factors can vary between the experimental animals and the control database. However, it is also worth noting that stride length and single limb support time residuals consistently shifted down with post-surgical time, even though these data were collected from the cohorts in a random order (Week 4, Week 1, Week 2, Week 6); this indicates the downward trend is unlikely due to a temporal change in the environment.

Unfortunately, the causes of gait abnormalities following MCLT sham and

MCLT+MMT in the rat remain uncertain. At the outset of this experiment, some differences in the gait patterns of MCLT sham and MCLT+MMT animals were anticipated. In addition, the lack of the skin incision sham group presents the potential that the act of cutting the skin, and not the injuries to the joint through either MCLT or

MCLT+MMT, is causing the gait compensations and tactile sensitivity changes over the

6 weeks experiment. Nonetheless, our primary conclusion is that, despite the development of full-thickness cartilage defects, calcified cartilage damage, and osteophyte formation in the MCLT+MMT group, there is no discernable gait pattern difference between MCLT sham and MCLT+MMT and differences in tactile sensitivity were limited to the 6 week time point. The lack of association between these histologic parameters and rodent gait compensations would still hold even if the skin incision is the root cause of the gait abnormality, and this lack of association between cartilage damage and behavioral changes in a rodent model of OA highlights the lack of known unifying relationships between OA pathogenesis and the development of OA disease sequelae.

Another limit of this experiment includes the lack of pre-operative data for von

Frey testing; instead, naïve littermate controls were used as control data. While it would have been useful to understand how the tactile sensitivity of our animals changed over the experiment, the primary intention of this experiment was to create a data set with behavioral profiles paired to a histological profile, such that correlations between behavior and histology could be constructed. More sophisticated statistical correlation models may assist in identifying relationships between joint damage and behavioral changes in the future, and assessment of change in sensitivity, rather than raw sensitivity, may improve these correlations in the future.

Clearly, mechanical destabilization of the joint due to ligamentous injury of the

MCL may be causing a mechanical dysfunction that ultimately manifests in changes to the spatiotemporal gait pattern. However, it is somewhat unusual that the mechanical dysfunction caused by MCLT would develop temporally over several weeks. If destabilization of the joint due to MCLT was the primary factor, gait compensations would be expected immediately after transection and for some dysfunction to occur consistently across the post-surgery time points. Instead, our data indicate a progressive development of shuffling gait compensations over time in both the MCLT sham and MCLT+MMT groups. This temporal shift seems to indicate mechanisms other than mechanical loss of the MCL are involved in the development of the gait compensation found after these simulated joint injuries.

For correlation analyses, the OARSI histopathology assessment scheme for the rat was advantageous because many of the parameters measured are real numbers, rather than the ordinal ranks typical of many histologic grading schemes. Nonetheless,

the OARSI histopathology scheme still tends to be cartilage-centric and largely focused on structural changes in the joint. This approach to grading joint damage may neglect the continuum of changes happening throughout the OA joint, including cellular and molecular level changes. First and foremost among molecular level changes, pro- inflammatory and catabolic mediators are chronically up-regulated in OA15,38,67, and OA pain is often considered to be inflammatory. Similarly, inflammation plays a critical role in MCL injury and repair after injury17,83, and up-regulation of inflammatory mediators can affect muscle function. As such, behavioral changes associated with the MCL sham and MCLT+MMT surgery may be more closely linked to local inflammation at the site of each injury. While joint inflammation was not directly assessed in this study, synovial capsule thickness (an indirect assessment of synovitis) did not appear to explain either gait or tactile sensitivity changes in the MCLT+MMT group. However, direct assessment of inflammatory cytokines and chemokines in the MCL, synovial fluid, synovial lining, or fat pad could be used in the future to more thoroughly evaluate the correlation between joint inflammation and behavioral changes.

In addition to inflammatory pain components, strong evidence has emerged that

OA pain has neuropathic pain components119,136, including evidence of damage to nociceptive fibers in the periphery of the joint98. Joint innervation actively responds to the OA environment and joint injury39,94, and damaged sensory nerves can release neuropeptides that stimulate other nerve endings. Similarly, damage to the neuromuscular system can occur with MCL rupture83,132. Whether within the joint or around the MCL, chronic damage to the peripheral nervous system can increase the excitability of neurons in the dorsal horn of the spinal cord (central sensitization),

allowing non-painful stimuli to be processed and interpreted as painful. Since rats with

MCLT sham and MCLT+MMT surgery both experience chronic musculoskeletal injuries, behavioral changes observed in both models may be more indicative of nervous system remodeling both in the periphery and centrally, rather than structural changes to the cartilage and bone within the joint. Again, direct assessment of changes in joint innervation and the associated dorsal root ganglia and dorsal horn of the spinal cord could be used in the future to more thoroughly evaluate the correlation between neuronal and behavioral changes in this model of OA.

It is worth noting the von Frey test examines tactile sensitivity in the hind paw, and since the surgically-simulated injury is at the knee for both MCLT sham and

MCLT+MMT, the von Frey test is detecting secondary (or referred) hypersensitivity.

Secondary hypersensitivity may be more indicative of neural damage or remodeling of the central nervous system than changes in the knee. Knee bend or application of pressure to the knee could allow for more direct assessment of primary hypersensitivity in these models; however, it should be noted these methods require animal restraint that may affect the behavioral measure. An advantage of gait analysis is that limb use and ‘movement-evoked’ changes can be assessed with minimal researcher interaction and reduced animal stress. Of course, it remains difficult to assess the relative contribution of hypersensitivity, mechanical dysfunction, proprioceptive changes, and dysesthesia for a given gait compensation.

Movement-evoked pain is an early characteristic of OA, and as a result, a person or animal may modify their gait pattern to protect an injured limb from loading and motion. If protective patterns are repeated over time, muscles surrounding the joint will

adapt and a fear of specific movement patterns may develop. For chronic diseases like knee OA, it is not yet clear if long-term use of a protective gait sequence promotes or prevents future joint degeneration; it is also not yet clear if the use of protective gaits or limb guarding will propagate OA-related disability. Thus, the gait abnormalities identified in this experiment are possibly learned behaviors that develop from the prior protection an injured limb. Future experiments blocking joint afferents at the time of injury or after the onset of symptoms could begin to more thoroughly examine this hypothesis.

Figure 2-1. Summary of spatiotemporal gait. . Shows the hind limb footprint pattern and temporal gait sequence of a rat; the fore limb footprints and temporal gait sequence have been omitted for clarity. Common spatial characteristics of the rat include stride length, step width, and step length; however, since stride length and step length are strongly associated, spatial symmetry can be used to describe the placement of the right footprint relative to two left footprints. The temporal characteristics of a single limb include stride time, stance time, and swing time. Stance time is typically normalized to stride time, as percentage stance time (also known as duty factor) follows a more linear relationship to an animal’s walking velocity. Similarly, the single limb support phase for a given limb is frequently assessed as relative to the gait cycle. The synchronicity of the gait cycle can be evaluated through stance time balance and temporal symmetry. The gait cycle is balanced when an animal spends equal time on its left or right limb, represented by a stance time balance of 0. A gait cycle is temporally symmetric when the right foot-strike occurs halfway in time between two left foot-strikes, represented by a temporal symmetry of 0.5.

Figure 2-2. Calculation of gait data residuals. Stride length, step width, percentage stance time, and the single limb support phase are known to strongly correlate with an animal’s walking velocity and size. Failure to account for the effects of velocity and weight in the analysis of a rodent’s gait will reduce the sensitivity of subsequent statistical analyses due to an increase in unexplained variance (or mean squared error). To account for the effects of animal size and walking velocity, stride length, step width, percentage stance time, and the single limb support phase were normalized to the predicted gait profile of velocity- and weight-matched naïve Lewis rats. This normalization process is shown for single limb percentage stance time at 1 week and 6 weeks post-surgery. The control line is based upon historical data on the gait characteristics of naïve Lewis rats (solid line with 99% confidence bands). This database is available for download at bme.ufl.edu/labs/allen and represents 280 gait trials collected in 28 different naïve Lewis rats at 49 different weights over a period of 8 years. The control line is used to predict the stride length, step width, percentage stance time, and the single limb support phase for a size- and velocity-matched Lewis rat. With this prediction, velocity- and weight-independent residuals of stride length, step width, percentage stance time, and the single limb support phase can be calculated by subtracting the predicted value for a velocity- and weight-matched Lewis rat. Once data are transformed into velocity- and weight-independent residuals , multiple trials of a rat are averaged, such that the average gait profile for an animal directly corresponds to single histological profile in the correlation analyses. These 8 values for each group-time point were used to construct the data presented in Figure 2-3 and Figure 2-5.

Figure 2-3. Spatiotemporal gait measures. Both MCLT sham and MCLT+MMT surgery results in altered spatial gait parameters in rats. Velocity was significantly faster at week 4 and 6 relative to week 1 and 2 (p<0.026). Rats with MCLT+MMT surgery used spatially asymmetric foot-strike patterns at week 2 and week 6, where the right step length was longer than expected (spatial symmetry > 0.5; p=0.005 and p<0.001, respectively). On week 4, the spatial pattern of the MCLT+MMT group tended to be asymmetric, where the right step length was shorter than expected (p=0.06). Spatial asymmetry indicative of longer right step lengths were also seen in the MCLT sham at week 2 (p=0.01) and tended to occur on week 4 (p=0.06). Step widths were narrower than expected in both groups, indicated by a step width residual less than 0.0. The MCLT sham used narrower step widths at week 1, 2, and 4 (p<0.001), but not at week 6. The MCLT+MMT used narrower step widths at week 2, 4, and 6, but not at week 1 (p<0.001). Stride lengths were shorter than expected with MCLT+MMT at 4 and 6 weeks (p=0.04 and p<0.001, respectively) and with MCLT sham at 6 weeks (p<0.001), indicated by a stride length residual of less than 0.0. No significant differences were identified between the MCLT sham and MCLT+MMT group within a specific time point. Data are presented as mean ± standard error of the mean.

Figure 2-4. Representation of spatial gait changes. Using data for a naïve rat walking at an average velocity as a representative spatial pattern, shifts in the spatial pattern caused by MCLT sham or MCLT+MMT are plotted for a rat of the same weight walking at the same velocity. At week 1, the patterns are similar between groups. At week 2, step widths begin to narrow in both MCLT sham or MCLT+MMT animals. At week 4, stride length begins to shorten in the MCLT+MMT group (compare 2nd left footprint between groups). At week 6, stride lengths are significantly reduced in both the MCLT sham and MCLT+MMT group (compare 2nd left footprint between groups). Unfortunately, due to the left justified nature of this plot, spatial asymmetries (left to right step length divided by stride length) are difficult to visualize due to the concurrent changes in stride length.

Figure 2-5. Temporal gait measures. Temporal symmetry investigates the synchronicity of the foot-strike pattern in time, where a temporal symmetry near 0.5 indicates the foot-strike of the right limb temporally occurs halfway between two left limb foot-strikes. At week 1 and week 4, temporal asymmetries were greater than 0.5 in the MCLT+MMT group (p<0.001, p=0.03), indicating the time to transition from left to right foot-strike was longer than the time to transition from right to left foot-strike. Stance time imbalance occurs when more time is spent on one limb relative to its contralateral limb. An imbalance greater than 0.0 was observed at week 1 in the MCLT+MMT group (p=0.003), indicating more time was spent on the left limb than right limb. Conversely, an imbalance less than 0.0 was observed at week 6 in the MCLT+MMT group (p=0.048), indicating more time was spent on the right limb than the left limb. Temporal gait compensations generally reduce the single limb support phase spent on the injured limb. In the case of unilateral injuries, a reduced single limb support phase is observed on the injured limb only. In conjunction with stance time imbalance findings in Panel B, reduced right limb single limb support phases, but not left limb, were found in the MCLT+MMT group at week 1 (p=0.03). In shuffling compensations, single limb support is reduced on both limbs of a limb pair; a shuffling compensation can be observed through reduced single limb support phases in both limbs of both the MCLT sham and MCLT+MMT group at week 4 and week 6 (p<0.001 in all groups. Data are presented as mean ± standard error of the mean.

Figure 2-6. Hildebrand temporal gait charts. As in Figure 2-4, data for an average size naïve rat walking at an average velocity were used to construct a representative hind limb foot-strike sequence for a walking rat using a modified Hildebrand chart (top). In this chart, the left hind limb foot-strikes occur at the same time in each group. Using this presentation, temporal asymmetries can be observed by the shift of right hind limb foot-strike later in time for the MCLT+MMT group at week 1 and week 4. Since the right hind limb single limb support phase (Right Limb SLS) occurs during the swing phase of the left hind limb, reduced right hind limb single limb support can be observed via a delayed toe-off in the left hind limb in the MCLT+MMT group at week 1 and in both MCLT sham and MCLT+MMT at week 4 and week 6. Conversely, left hind limb single limb support (Left Limb SLS) occurs during the swing phase of the right hind limb. As with spatial symmetry in Figure 2-4, alterations in left hind limb single limb support are more difficult to see with due to the left foot-strike standardization

Figure 2-7. Summary of mechanical sensitivity. MCLT sham and MCLT+MMT cause heightened sensitivity in the operated limb at 1 week, indicated by a reduction in the paw withdrawal threshold relative to littermate naïve control levels (p<0.001). While animals with MCLT sham surgery recovered to control levels by week 6, animals with MCLT+MMT tended to maintain a heightened sensitivity across the time points, with significant reductions in the paw withdrawal threshold at week 4 and week 6 (p<0.001 at all time points) and tendency for heightened sensitivity at week 2 (p=0.07). In addition, the paw withdrawal thresholds for animals with MCLT+MMT were significantly different from MCLT sham at week 6 (p=0.002). Data are presented as mean ± standard error of the mean.

Figure 2-8. Summary of histological measures. Cartilage lesions were identified in the MCLT+MMT groups, but not the MCLT group. The width of the cartilage lesion was significantly larger in the MCLT+MMT group than the MCLT group at the cartilage surface at each time point (#, p<0.04), and the loss width at 50% depth and 95% depth was wider than the MCLT group at week 6 (#, p<0.002). Moreover, the percentage of the surface affected increased over time in the MCLT+MMT group (1,2,4, p<0.008). The cartilage lesions were primarily located in the central and medial aspect of the medial compartment,

with the depth of articular cartilage damage being significantly greater in the MCLT+MMT group than the MCLT sham group in at least 1 location at all time points (#, p<0.002). Moreover, the depth of the lesion progressed from near 50% at week 1 to full thickness by week 6. Significant growth of osteophytes was identified in the MCLT+MMT group at week 4 and week 6 relative to MCLT control (#, p<0.001. Moreover, the size of the osteophyte was larger at week 4 and week 6 compared to week 1 and week 1 in the MCLT+MMT group (1,2, p<0.001). In addition, significant damage to the calcified cartilage was observed in the MCLT+MMT group relative to the MCLT group at week 4 and week 6 (#, p<0.001), with the damage score in the MCLT+MMT at week 6 significantly higher than at week 1 and week 2 (1,2, p<0.001). Differences in the medial capsule thickness between MCLT sham and MCLT+MMT were only observed on week 1 (#, p=0.04). Data are presented as mean ± standard error of the mean.

Figure 2-9. Representative histology slides and measures. Representative histological images are shown for each surgical group at each time point, with a representative image for a healthy control from the contralateral joint shown in the bottom left. The cartilage lesion measurements described in Figure 2-8 are visually described in the bottom row. Significant damage to the articular cartilage can be seen in the MCLT+MMT group, including cartilage loss, osteophyte growth, and damage to the calcified cartilage region. Very little cartilage damage or osteophyte growth is seen in the MCLT group. Photos courtesy of author.

Figure 2-10. Histology and behavior correlations. While medial joint capsule repair correlated to tactile sensitivity in the MCLT sham group, medial joint capsule repair did not associate with tactile sensitivity in the MCLT+MMT group (Top Panel). Instead, cartilage matrix loss width at the surface correlated to tactile sensitivity in the MCLT+MMT group, despite evidence of cartilage damage in the MCLT sham group. Also, while correlations between stride length residual and injured limb (right) single limb support phase could be identified in the MCLT+MMT group, similar gait changes could be identified in the MCLT sham despite evidence of joint damage (Middle and Bottom Panel, respectively).

Table 2-1. Correlations between histological evidence of joint damage and pain-related behaviors following MCLT sham and MCLT+MMT surgery in the rat

CHAPTER 3 AUTOMATED GAIT ANALYSIS THROUGH HUES AND AREAS (AGATHA): A METHOD TO CHARACTERIZE THE SPATIOTEMPORAL PATTERN OF RODENT GAIT

Locomotion is a natural behavior and similar in many species; thus, gait analysis has been widely used to explore the behavioral consequences of diseases. Moreover, because gait analysis is non-invasive, techniques have been developed for both clinical and preclinical research. When combined with other comprehensive biological assays, gait analysis can describe the behavioral phenotype of a disease model, thereby providing a more thorough understanding of the functional and symptomatic consequences of different diseases states109,133,137.

For many researchers who wish to add gait analysis to their behavioral repertoire, a commercial gait system is purchased, such as the CatWalk, Digigait, or

Treadscan platforms11,13,54. These systems can be used immediately and have automated software to calculate gait variables11,13,41,49,97,112,146. However, commercial gait analysis systems are relatively expensive and their analysis software is typically proprietary and unalterable. Furthermore, these systems use different detection techniques which may not be suitable for all rodent models28,71. Moreover, many of these gait systems have not been validated to manual calculation of the gait pattern, hence the methodological errors in these techniques are not necessarily well understood. Alternatively, if purchasing a gait system is not possible, building a gait arena from scratch can be cost effective, but analysis is often performed manually.

Adaptable, open-source gait analysis software can provide a third option for researchers to build their own arena and tailor an open-source gait analysis specifically to their field and disease model. Moreover, by releasing the digitization code to the user base,

methodological errors can be assessed in the recording method, lighting balance, and digitization parameters.

The goal of this study was to develop an adaptable, open-source, gait analysis method that can also be used to study rodent gait patterns. An open-source, automated gait analysis method is presented, for the first time, for the assessment of rodent spatiotemporal gait patterns, called Automated Gait Analysis Through Hues and Areas

(AGATHA, available for free download at https://github.com/OrthoBME/AGATHA). This manuscript begins by describing how AGATHA utilizes two views of rodent walking to detect foot-strike and toe-off events; then, expands to describe how AGATHA uses these data to calculate spatiotemporal gait parameters. AGATHA was then validated relative to manual digitization using rodent gait videos sampled at 1000, 500, 250, 125,

60, and 30 frames per second (fps). Once AGATHA was validated against manual digitization, AGATHA’s utility was explored using gait videos from rat models of post- traumatic osteoarthritis, C2 spinal cord hemisection, and C3/C4 lateralized cervical spinal cord contusion. Overall, AGATHA required less effort and produced nearly identical results to manual digitization at high frame rates. For frame rates above 125 fps, AGATHA was able to detect and differentiate the gait profile of each injury model.

Combined, these findings validate AGATHA relative to manual digitization and demonstrate its ability to detect spatial and temporal gait changes in rodents.

Automated Gait Analysis Through Hues and Areas (AGATHA)

A custom MATLAB code, named Automated Gait Analysis Through Hues and

Areas (AGATHA), was created to identify the time and positional coordinates of foot- strike and toe-off events in videos of walking rodents. This code is available for free download at https://github.com/OrthoBME/AGATHA. With these time and position

coordinates identified, AGATHA then calculates several spatiotemporal gait parameters explained in detail below.

AGATHA first isolates the sagittal view of the animal and locates the silhouette of the animal in each frame of the video (Figure 3-1A). This is achieved by subtracting a background image where the animal is not present in the arena, transforming the frame into a HSV (Hues, Saturation, Value) image, and using the hue value to convert the

HSV image into a binomial silhouette (Figure 3-1B). Next, AGATHA locates the row of pixels representing the interface between the rat and the floor. In the binomial silhouette, the single row of pixels representing the floor interface is composed of background pixels (indicated by a value of 0, black) and rat contact points with the floor

(indicated by a value of 1, white)(Figure 3-1C). Because a walking rat always has at least two limbs in ground contact during locomotion, this interface can be identified in every frame; however, please note that AGATHA may not accurately locate the rat-floor interface if the animal moves with a gait pattern containing a completely aerial phase

(running trot or bounding). Second, AGATHA excludes the majority of nose and tail contacts with the floor from the analysis by comparing the contact point (white pixels) to the animal’s center of area in the sagittal view, eliminating contacts that are proportionally too far from the center of area to be considered a paw strike.

By stacking the rat/floor interface across multiple frames, foot contact with the ground can be visualized over time (Figure 3-1D). AGATHA uses these stacked pixel rows of the rat-floor interface to determine the earliest frame associated with the paw entering the rat-floor interface (foot-strike) and last frame associated with the paw leaving the rat-floor interface pixel row (toe-off). Within the paw contact object seen in

Figure 3-1D, foot-strike is defined as the earliest frame of the paw contact object (blue triangles and circles), while the toe-off is defined as the latest frame of the paw contact object (red triangles and circles). These temporal events can then be used to calculate temporal symmetry of the hind limbs (the percentage of the gait cycle representing when the right foot-strike occurs between two sequential left foot-strikes), duty factor

(the percentage of the gait cycle that the limb is in contact with the ground) of the hind limbs, duty factor of the fore limbs, duty factor imbalance of the hind limb, step frequency, and limb phase between the fore and hind limbs (variables explained in detail later in the manuscript). In this manuscript, data analysis is focused solely on the hind limb parameters.

In addition to the frame associated with foot-strike and toe-off, the sagittal image provides an estimation of the spatial foot-strike location along the axis of travel (Figure

3-2A). Using this approximate location, the precise spatial location of foot-strike can be determined in the ventral view of the animal (Figure 3-2B). To achieve this for white rodents, each paw print is evaluated on a frame just after foot-strike and on a second frame just prior to toe-off. In these two images, the pixels representing paw contact with the ground remain nearly constant while other pixels change, allowing the paw print to be isolated using iterative color thresholds that examine pixel change between the two video frames (Figure 3-2C). The centroid of the paw print is then identified and used to represent the paw’s spatial location during stance (Figure 3-2D). These spatial events are used to calculate stride length, step widths, and step length symmetry. In addition, the centroid of the rat in the ventral view is determined for every frame and used to calculate velocity.

In total, AGATHA calculates and exports the following gait variables: trial velocity, temporal symmetry of the hind limbs, duty factor of the hind limbs, duty factor of the fore limbs, duty factor imbalance of the hind limb, stride length, hind limb step width, step frequency, limb phase between the fore and hind limbs, and number of steps in the video61,62. These parameters are based on the classic Hildebrand gait diagram to describe quadrupedal gait sequences60–62. While AGATHA is able to acquire fore limb data, we have elected to compare AGATHA to manual digitization using only hind limbs; this was selected to reduce the cost and time of manual digitization.

Testing Approval

All methods and testing were approved by the University of Florida Institutional

Animal Care and Use Committee (IACUC) and conform to the Association for

Assessment and Accreditation of Laboratory Animal Care (AAALAC) recommendations on animal research and National Institutes of Health (NIH) guidelines.

Gait Testing

Rats voluntarily explored an overground arena while videos were collected using a high-speed camera (RedLake, M3), as previously described 2,3,5. Briefly, animals were placed into a clear acrylic arena (6"x12"x72") with a mirror mounted below the floor set at a 45˚ angle, such that the sagittal and ventral planes of the animal could be viewed simultaneously with a single camera placed approximately 2 meters from the arena.

Though not required for AGATHA to function, the back wall and ceiling in the central 48” of the arena were covered with green vinyl to provide a contrasting background for the camera views of the animal. Only white rats have been used in this study, though

AGATHA routines are available to describe gait parameters of white, black, and nude rodents (both mice and rats) against a range of background colors. The remaining 12”

at both ends of the arena were covered with black vinyl to provide a "safe zone" for the animal. Halogen lights were positioned to provide indirect lighting from below and from the side of the animal. Prior to each recording session, the arena was leveled and images of a calibration grid were acquired. During recording, animals explored the arena for 20 minutes without experimenter interaction. Walking trials with overall constant velocity (linear fit of position data vs. time of R2>0.95) and a minimum of three full gait cycles (minimum of 4 foot-strike events per foot) were recorded at 1000 fps when the animal passed through the center of the arena, with a minimum of five trials per animal collected. The calculation of each gait variable conforms with the standard definitions of the Hildebrand gait diagram61,62; these equations can be found in prior work or in our methodological review71.

Validating AGATHA to Manual Digitization and Exploring the Effect of Video Frame Rate

To validate the accuracy and precision of AGATHA and to explore the effects of video frame rate, 25 videos of four naïve Lewis rats were collected at 1000 fps as described above. These 1000 fps videos were then re-sampled every 33 frames (30 fps), 17 frames (60 fps), 8 frames (125 fps), 4 frames (250 fps), and 2 frames (500 fps) to create lower frame rate doppelganger videos. Then, the 25 root videos and 150 doppelganger videos were digitized with AGATHA or via manual digitization using the

DLTdataviewer subroutine for MATLAB59. Briefly, for manual digitization, each video was loaded into the DLTdataviewer subroutine and displayed frame by frame, with a person marking the hind limb foot-strike or toe-off event on the ventral view of the rat in the frame where the event first occurs. DLTdataviewer then logs the pixel coordinate and frame number of the event. Velocity was calculated by tracking the nose position

every 0.2 seconds, equating to every 100th frame for 500 fps, every 50th frame for 250 fps, every 25th frame for 125 fps, every 10th frame for 125 fps, and every 5th frame for 30 fps. Using these data, trial velocity, temporal symmetry, duty factor, duty factor imbalance (left duty factor-right duty factor), stride length, and step width are calculated for AGATHA and manual digitization.

To compare AGATHA to manual digitization, the mean of each gait parameter was compared using a paired t-test at each frame rate. Similarly, to compare the precision of the digitization methods, the standard deviation of each gait measure was compared using an F-test at each frame rate. For the paired t-tests and F-tests, corrections for compounding type 1 error were not applied because the null hypothesis aligns with the desired outcome (AGATHA is equivalent to manual digitization). To evaluate the effect of frame rate on the digitization results, nested ANOVA were conducted with post-hoc Tukey’s HSD tests, when indicated. Finally, based upon the variability observed within our naïve data set and effect sizes reflecting the sensitivity required to discern subtle differences between experimental groups in previous publications3,5,71, a power analysis was conducted to identify the number of animals needed for each frame rate (beta=0.80, alpha=0.05, assuming normalization to a weight and velocity matched standard and a minimum of 5 trials per animal).

Validating AGATHA Using an Orthopaedic Injury Model

To evaluate AGATHA’s ability to detect gait compensations in an orthopaedic injury model, 18 Lewis rats (3 months, 200-250 grams, male) were obtained from

Charles Rivers Laboratories and allowed to acclimate to the University of Florida housing facilities for two weeks prior to the investigation. After acclimation, 8 rats received a surgical medial meniscus transection injury (MMT) and 5 rats received a

medial collateral ligament transection injury (MCLT sham), as previously described3,72.

The 5 remaining rats were left as naïve controls.

Briefly, animals were anesthetized in a 4% isoflurane sleep box, prepped for aseptic surgery, and transferred to a sterile field with anesthesia maintained by mask inhalation of 2% isoflurane. A medial midline skin incision was made on the right hind limb, and the medial collateral ligament was exposed via blunt dissection and transected. At this point, the wounds of animals receiving MCLT alone were closed. For animals receiving MMT, the joint was then placed in a valgus orientation to expose the central portion of the medial meniscus; then, a complete radial transection was performed in the central portion of the medial meniscus using an 11 blade scalpel.

Surgical sites were closed with 9 mm wound clips, which were removed 10-14 days after surgery. Upon surgical recovery, rats received 48 hours of buprenorphine (0.03 mg/kg) to reduce post-surgical discomfort.

Animals were gait tested at four weeks after surgery, as described above.

Because duty factor, stride length, and step width are dependent on the animal’s walking velocity, data from the naïve animals were used to calculate the residual distance for each trial (measured gait parameter – predicted value based on naïve cohort), as previously described76. Data over multiple trials for a given animal were averaged, followed by a 1-way analysis of variance with a post-hoc Tukey’s HSD test, when indicated.

At the conclusion of the experiment, animals were transferred to an unrelated

IACUC protocol.

Validating AGATHA Using a Cervical Spinal Cord Hemisection Injury Model

The ability of the AGATHA method to detect changes in gait following a unilateral high cervical SCI was examined using a spinal cord hemisection model48,123. Adult male

Sprague-Dawley rats (N=9) were obtained from Harlan Scientific and housed at the

McKnight Brain Institute Animal Care Facility at the University of Florida. Anesthesia and injury methods have been previously described13,27,29,40. Briefly, five rats were anesthetized by isoflurane, induced in a closed chamber at 3-4% in O2 and maintained in a surgical plane at 1-2% in O2 via nose cone for the duration of the surgery. A 1-inch dorsal midline incision was made from the base of the skull extending caudally to approximately the fourth cervical segment (C4). A laminectomy was performed at the second cervical segment (C2) to expose the cervical spinal cord. A small incision was made in the dura and a lateral hemisection performed on the left side of the spinal cord using a microscalpel, followed by gentle aspiration. Using this approach, the completeness of the lesion was readily visible and the extent of the lesion was reproducible. After lesion extent was visually confirmed, the dura was closed with interrupted 9-0 sutures, the overlying muscle sutured (with 4-0 sutures), and the skin closed with stainless steel surgical wound clips. Following completion of the surgical procedure, rats were placed on supplemental O2 via nose cone until they recovered from anesthesia and were awake and moving. Post-surgical care included administration of buprenorphine at 12 hr intervals (0.03 mg/kg, s.q.) for the initial 48 hours post-injury and delivery of lactated Ringers solution (5ml/12 hr, s.q.) and oral

Nutri-cal supplements (1-3 ml, Webster Veterinary, MA, USA) until adequate volitional drinking and eating resumed. Four naïve (spinal intact), gender matched rats were used as controls.

Injured animals were gait tested nine weeks after surgery, as described above.

Again, because duty factor, stride length, and step width are dependent on the animal’s walking velocity, the gender-matched naïve animals were used to form a control line; then, the residual distance from the control line was calculated for each trial (measured gait parameter – predicted value based on naïve cohort), as previously described76.

Data over multiple trials for a given animal were averaged, followed by a 1-way analysis of variance with a post-hoc Tukey’s HSD test, when indicated.

Validating AGATHA Using a Cervical Spinal Cord Contusion Injury Model

Further validation of the AGATHA method was accomplished using a spinal cord contusion model that more closely mimics the most common SCIs occurring in humans81. For these studies, 12 adult female Sprague–Dawley rats were obtained from

Harlan Scientific and housed at the McKnight Brain Institute Animal Care Facility at the

University of Florida. Three females received no surgical treatments, and were studied as a control group for comparison with the contused animals.

Six animals served as a sham surgery control group and received cervical laminectomy but not contusion injury. Rats were anesthetized by isoflurane, induced in a closed chamber at 3-4% in O2 and maintained in a surgical plane at 1-2% in O2 via nose cone for the duration of the surgery. A 1-inch dorsal midline incision was made from approximately the second to fifth cervical segment (C2–5). A laminectomy was performed at the third and fourth cervical segments (C3-4) and the overlying muscles were closed with sterile 4-0 Vicryl suture. Skin incision was closed using sterile wound clips. Following completion of the surgical procedure, rats were placed on supplemental

O2 via nose cone until they had recovered from anesthesia and were awake and moving. The same post-surgical care was administered as described above.

Three animals received a lateralized C3/C4 contusion. Rats were anesthetized by isoflurane, induced in a closed chamber at 3-4% in O2 and maintained in a surgical plane at 1-2% in O2 via nose cone for the duration of the surgery. A dorsal midline incision was made from approximately the second to fifth cervical segment (C2–5).

Following laminectomy at the C3/4 level, a left lateralized spinal contusion was made using the Infinite Horizon pneumatic impactor (Precision Systems & Instrumentation,

Lexington, KY)125. Once the spinal cord was exposed, the impactor probe (2.5 mm diameter tip) was positioned between midline and the left lateral edge of the spinal cord at C3/C4, and was raised approximately 5 mm above the intact dura. The cord was contused at a pre-set nominal force of 200 kilodynes (dwell time = 0). Upon completion of the injury procedure, overlying muscles were closed in layers with sterile 4–0 Vicryl suture and the skin incision was closed using sterile wound clips. Following completion of the surgical procedure, rats were placed on supplemental O2 via nose cone until they had recovered from anesthesia and were awake and moving. Post-surgical care included administration of buprenorphine at 12 hr intervals (0.03 mg/kg, s.q.) for the initial 48 hours post-injury and delivery of lactated Ringers solution (5ml/12 hr, s.q.) and oral Nutri-cal supplements (1-3 ml, Webster Veterinary, MA, USA) until adequate volitional drinking and eating resumed.

Laminectomy animals were gait tested 10 days after surgery and the contused animals were gait tested 4 weeks after surgery, as described above. Again, because duty factor, stride length, and step width are dependent on the animal’s walking velocity, the gender-matched naïve animals were used to form a control line; then, the residual distance from the control line calculated for each trial (measured gait parameter –

predicted value based on naïve cohort), as previously described76. Data over multiple trials for a given animal were averaged, followed by a 1-way analysis of variance with a post-hoc Tukey’s HSD test, when indicated.

AGATHA Comparison to Manual Digitization

In the 1000 fps root videos, AGATHA reproduced manual digitization results for temporal symmetry, duty factor imbalance, and stride length; however, AGATHA showed differences relative to manual digitization for hind limb duty factor, velocity, and step width (Figure 3-3). On average, velocities were 8.5 cm/sec faster, hind limb duty factors were 1.5% lower, and step widths were 0.2 cm wider with AGATHA relative to manual digitization of the same trial. However, as can be observed in Figure 3-3, the inter-trial variability was much greater than the variability introduced by these methodological variables, and neither method had an advantage in the precision of the gait measurements at 1000 fps.

Video frame affected the measurement of hind limb duty factors (Figure 3-4 A-B), but did not affect the average measurement of any other gait parameters (Figure 3-4 C-

F). For hind limb duty factors, frame rates below 125 fps were significantly different from frame rates above 125 fps for both digitization methods (p<0.03). To further investigate this effect, the time associated with foot-strike and toe-off events was examined relative to the 1000 fps video (Figure 3-5). Here, the video frame rate affected the accuracy associated with the identification of both the foot-strike and toe-off events relative to the

1000 fps video for both digitization methods (p<0.001). This finding is shown in Figure

3-5 using foot-strike and toe-off time residuals (time of a gait event in the doppelganger minus the time of the same gait event in the 1000 fps root video). At 30 fps, both foot- strike and toe-off time residuals were lower with AGATHA than for manual digitization

(p<0.001). At 60 fps, only the foot-strike time residual was significantly different between

AGATHA and manual digitization (p<0.001). However, these errors converged to near- zero values at recording rates over 125 fps.

Increasing the frame rate also reduced the variability of imbalance and symmetry

(p<0.03, Figure 3-4). In addition, for both AGATHA and manual digitization methods, the identification of foot-strike and toe-off events became more precise with increasing frame rate (p<0.002, Figure 3-5). In general, the precision of foot-strike and toe-off events in AGATHA was similar to manual digitization at half the video frame rate.

Finally, a power analysis was conducted to assess the number of animals per group (n) required for the detection of a given effect for each digitization method at each video frame rate (Table 1). At low video frame rates, a larger number of animals are required to achieve similar detection sensitivities to higher frame rates, regardless of the digitization method. Since the variability of imbalance and symmetry reduces with frame rate, this result was expected. However, the n required for other gait parameters also decreased with sampling rate, indicating that, while the increased precision with frame rate may not be statistically significant, the increased frame rates may still markedly affect the measurable effect size and animal numbers in an animal experiment.

Orthopaedic Injury Model

AGATHA detected gait abnormalities in MMT animals, including increased left and right hind limb duty factors and decreased stride lengths (Figure 3-6). The shifts are represented in the top of Figure 3-6 via a velocity-corrected image of the footprint pattern and a velocity-corrected Hildebrand plot of the foot-strike and toe-off events.

Please note that the sizes of the ellipses representing the footprint are representative of

1 standard deviation of the stride length residual (width) and 1 standard deviation of the

step width residual (height). This “shuffle step” gait pattern (higher stance times with shorter stride lengths) is representative of a parallel gait compensation, where the periods of the gait cycle when a hind limb must support weight without contralateral support are reduced by increasing the relative stance time on both limbs (white bars lower in the Hildebrand plot are smaller in MMT animals relative to other groups).

Spinal Cord Injury Model

AGATHA also detected gait abnormalities in models of cervical spinal cord injury models (Figure 3-7). For animals with cervical contusion, significant gait compensations were limited to the reduced stride lengths; however, the gait sequences also tended to be asymmetric (symmetry < 0.5, p=0.093). For animals with a C2 hemisection, significant imbalances were observed in the hind limb duty factors, with more relative time spent on the left (injured side) hind limb (imbalance > 0). Moreover, stride lengths were markedly shortened by approximately 4 cm and hind limb step widths were narrowed by approximately 1 cm. This “limp-like” gait pattern was present in both spinal cord injury models, and is likely representative of an antalgic gait compensation in which one limb of a limb pair tends to preferentially compensate for an injured limb.

Future Directions to Improve Gait Analysis

This study introduces, for the first time, a new digitization method for the characterization of rodent spatiotemporal gait patterns, AGATHA. Using AGATHA, detectable rodent gait compensations were quantified in an orthopaedic injury model and in two spinal cord injury models. Moreover, AGATHA's accuracy and precision were validated relative to manual digitization of the same videos, with AGATHA yielding a similar calculation to manual digitization for videos recorded above 125 fps. Using

AGATHA, spatiotemporal gait profiles can be acquired in lieu of the laborious effort of

manual digitization in significantly less time. As an open-source gait analysis platform,

AGATHA can assess rodent gait using a simple arena that is adaptable to different video recording speeds, multiple colors of animals, and has been optimized for both mice and rats.

Prior work has shown that significant methodological and material differences in gait analysis systems can yield inconsistent results when applied to identical rodent videos28. To be clear, methodological advantages in our approach were not examined relative to commercial systems, nor do we claim any advantage over these packages.

AGATHA, however, does provide an open-source platform which has been validated to manual digitization, and through which these methodological sources of variability can be examined. Unfortunately, little literature is devoted to assessing these methodological sources of error in rodent gait analysis platforms28,71,97,144. The limited number of system comparison studies is partly due to lack of gait system access, proprietary software protection, and testing bias. Because commercial gait platforms are relatively expensive, it is also unlikely for multiple platforms to exist within a single laboratory or university. Even if access were not a problem, directly comparing gait results can be difficult due to different hardware (such as different video frame capture rates). Moreover, commercial software packages tend to protect the algorithms for calculating gait variables as proprietary, making the source of differences difficult to pinpoint.

Currently, few free gait analysis software are available. MouseWalker is freeware that can analyze gait collected on a light-refracting, overground arena, similar to the set- up of the commercial Catwalk system41,104. MouseWalker is fully automated and can

process both color and grayscale videos. However, the equations used to calculate gait variables have been released as a MATLAB executable; as such, these equations are not easily manipulated to test methodological changes in this script. Nonetheless, the

MouseWalker method may be scaled to larger animals and different fur colors, though it has currently only been used to describe gait for mice.

Another open-source analysis package is Locomouse92. Using simultaneous sagittal and ventral views, Locomouse is a MATLAB program that automatically tracks body and limb movement of black mice from grayscale videos. While Locomouse is automated and supports full access to the MATLAB code, it currently calculates only a few gait variables used by other systems. However, Locomouse uses a manually- defined reference library to choose how to track paws. While this is not a formal validation to manual digitization, the software attempts to match human results directly rather than relying on a completely independent, un-validated algorithm. With more open-source options available, researchers may combine the strengths of each method specifically to their needs. Moreover, because different digitization methods use different markers to calculate gait variables, gait variables can have slightly different values when calculated by different methods. Freely providing the code used to generate the gait data will help disambiguate gait variables within the field and improve comparison across studies.

AGATHA did tend to calculate a lower trial velocity than manual digitization. This finding is most likely driven by differences in the measurement from each digitization method. Recall, AGATHA measures velocity by estimating the animal’s centroid in every frame within a video, while historically, our lab has estimated velocity by noting

the nose location in frames separated by 0.2 secs intervals. Mathematically, AGATHA’s approach should be more accurate due to a higher sampling frequency. Moreover,

AGATHA is likely more consistent due to the variability of the animal’s head position during walking, and this increase in head sway may cause an over estimate of velocity during manual digitization.

For temporal measures, AGATHA tended to under estimate hind limb duty factor relative to manual digitization. This under estimation is most likely driven by the identification of toe-off events. AGATHA identifies toe-off as the frame where the paw is no longer in contact with the rat-floor interface. However, at toe-off, the rodent’s toe and foot rotate slightly, such that the toe can clear the floor. Manual digitizers tend to click a frame that is slightly after this toe rotation. As such, AGATHA may provide a less biased estimation of toe-off, though this effect would need to be verified in a model where altered toe-clearance motions could be expected.

While manual digitization was generally more precise, AGATHA was able to achieve comparable precision at frame rates over 125 fps. As expected, human digitizers are generally more precise than digitization algorithms63,115, especially for experimental designs where fatigue of the digitizer does not play a significant role. This is well known for radiological scans and segmentation algorithms. Thus, it is critical to validate any digitization algorithm to manual methods, a validation criterion that has not necessarily been presented for some commercially available systems. Validating

AGATHA against manual digitization, the estimated n per group at 1000 fps was generally only 1 or 2 animals per group different; at 125 fps, the estimated n per group was approximately 2 to 3 animals per group different. Of course, actual number of

animals per group for a given experiment will depend on the experimental design, variability of the animal model, and the desired detectable effect size; but, these data do indicate AGATHA can yield a viable alternative to manual digitization.

The apparent transition in accuracy and precision around 125 fps may be explained by sampling frequency theory. To optimize sensitivity and minimize error, the

Nyquist-Shannon rule states an effective sampling frequency should be greater than twice the frequency of the fastest factor being measured141. When applied to gait analysis systems, the fastest factor to be measured is the anticipated gait abnormality, which has been shown to range between 0.001 and 0.025 seconds in rats5. Sampling theory posits the smallest, accurately measurable gait change in a video taken at 125 fps is 62.5 Hz (0.016 seconds), which is near the mid-point of reported gait abnormalities. As such, any changes shorter than 0.016 seconds cannot be confidently detected at 125 fps, and this detection limit will only increase at lower video frame rates

(e.g. at 30 fps, detectable changes must be larger than 0.067 seconds). Again, since

AGATHA can be used across multiple frame rates, this flexibility allows the user to adjust the sampling rate to the preclinical model, while potentially achieving a sensitivity that is greater than most commercial systems that record at 150 fps or less.

It should also be noted that, while this study found manual digitization was more precise than AGATHA, manual digitization is a tedious, labor-intensive process. It is not uncommon for weeks or months of post-processing to occur for studies employing manual digitization1,3,76. As such, gait event criteria during manual digitization may shift due to fatigue or altered selection criteria across digitization days. One advantage of automated gait analysis is the consistency of the measure across trials. While these

fatigue effects were not studied in this work, automated systems have been found to produce more repeatable results than manual digitization methods in some other methodologies144. Moreover, when a single video is digitized multiple times, AGATHA will choose the same frames each time unless the video or code is altered, while manual digitization has the potential to choose a different frame for that same gait event when asked to digitize the same video more than once9,126,139,144. Thus, for larger studies, automated digitization may prove more repeatable relative to manual methods, though these sources of methodological error have not yet been examined for rodent gait analysis.

Importantly, AGATHA was able to detect both antalgic and shuffling gait abnormalities in rodents. In our prior work, shuffling changes were observed in both the

MMT and MCLT sham at 4 weeks76; here, only the MMT animals exhibited the shuffling compensation. However, it should be noted that, in our past study, the four week time point was the first time point to exhibit the shuffling compensation. Thus, it is possible this particular cohort of MCLT animals had not yet transitioned to a shuffling compensation. Nonetheless, AGATHA was able to detect the expected shuffling compensation in the MMT cohort, confirming AGATHA’s utility when applied to this model3,76. AGATHA also detected gait changes in the spinal cord injury models.

Interestingly, though the injuries were restricted to the cervical portion of the spine, hind limb gait changes were measurable and significant. Moreover, unlike the orthopaedic injury model, the C2 hemisection model exhibited graded, antalgic gait compensations in which the left side was favored. Similar spatial gait changes have been observed in comparable hemisection and contusion injury models30,54,78,96, though only one

publication was found that accounted for velocity covariates5. Nonetheless, these data indicate high-speed gait analysis via AGATHA was able to provide a detailed gait pattern analysis indicative of spinal cord injury.

In conclusion, AGATHA is a robust method for automatically analyzing and detecting small changes in rodent gait, which has been validated to manual digitization.

Moreover, frame rate is an important factor affecting the accuracy of gait analysis, whether employing AGATHA or a manual digitization method. While differences were found between AGATHA and manual digitization for some spatiotemporal variables, these errors were far smaller than the inter-trial variability. More importantly, these errors did not prevent the identification of shuffling gait compensations in an orthopaedic injury model or antalgic compensations in a spinal cord injury model. Combined, these data indicate AGATHA can be used to explore some methodological differences and characterize the spatiotemporal gait patterns in rodent injury models.

Figure 3-1. How AGATHA derives temporal measures. A) Original video frame. B) The transformed image frame in which the animal is isolated and shows as white, while the background is negated and shows as black. C) Graphical representation of the row of pixels identified as the floor in the transformed image. Within this row of pixels, portions of the animal in contact with the floor show up as white. D) Consecutive pixel rows representing the floor are stacked and the paw print objects can be represented 2-dimensionally for calculation of foot-strike and toe-off. Red shapes identify foot-strike and blue

shapes represent toe-off. Triangles are used for fore-limbs and circles are used for hind-limbs. Photos courtesy of author.

Figure 3-2. How AGATHA derives spatial measures. A) Original video frame. B) AGATHA’s isolated paw image from the ventral view. C) AGATHA’s filtered representation of the paw print. D) A sequence of hind paw prints from a full trial. Photos courtesy of author.

Figure 3-3. Paired graphs comparing AGATHA with manual digitization. Paired scatterplots of raw data from the 1000 fps root videos of the naïve Lewis rat cohort. AGATHA reproduces data digitized manually for symmetry, duty factor imbalance, and stride length, but was slightly different for velocity, hind limb duty factor, and step width. Still, differences between manual digitization and AGATHA were much smaller than inter-trial variability.

Figure 3-4. Effect of video frame rate on AGATHA and manual digitization. Raw data scatter plots of each gait variable for each digitization method at all video frame rates. Low video frame rate appears to affect AGATHA's accuracy more than manual digitization, especially for temporal variables. Spatial variables seem largely unaffected by video frame rate.

Figure 3-5. Effect of video frame rate on gait event accuracy. A) The temporal residual of foot-strike. B) the temporal residual of toe-off. The foot-strike and toe-off residuals are in reference to the matching gait event from each video’s respective 1000 fps root video. To directly compare each video frame rate, the residuals are also represented in raw time (seconds) instead of frames.

Figure 3-6. Summary of AGATHA digitized orthopaedic model. A) Overall, MCLT sham animals walked faster than naïve animals. B) A graphical representation of spatial variable residuals by surgical group. Normalized to the left hind foot placement, each ellipse center is located at its group's average stride length and step width. The major and minor axes of each ellipse correspond with one standard deviation of the respective residual (in this case, horizontal axes represent stride length residual deviations while vertical axes represent step width residual deviations). C) A Hildebrand graph summarizing the residualized temporal differences between groups during a step cycle. The left side of each colored box indicates foot-strike, the horizontal length of each colored box indicates duration of stance, and the right side of each colored box indicates toe-off. The white space in between represents the duration and temporal location during the step cycle of the swing phase for that limb. D)-I) Animal average scatter plots of gait results. *: indicates p<0.05.

Figure 3-7. Summary of AGATHA digitized spinal cord injury models. A) Overall, there were no velocity differences between laminectomy animals, cervical

contusion animals, and their respective naïve group, but the C2-hemisection animals walked significantly slower than their respective naïve group. B) A graphical representation of spatial variable residuals by surgical group for each spinal cord injury model. Normalized to the left hind foot placement, each ellipse center is located at its group's average stride length and step width. The major and minor axes of each ellipse correspond with one standard deviation of the respective residual (in this case, horizontal axes represent stride length residual deviations while vertical axes represent step width residual deviations). C) Hildebrand graphs summarizing the residualized temporal differences between spinal cord injury groups during a step cycle. The left side of each colored box indicates foot-strike, the horizontal length of each colored box indicates duration of stance, and the right side of each colored box indicates toe-off. The white space in between represents the duration and temporal location during the step cycle of the swing phase for that limb. D)-I) Animal average scatter plots of gait results. *: indicates p<0.05.

Table 3-1. Detection sensitivity of AGATHA and manual digitization methods. The data represented in this table derives from the 1000 fps root videos of the Lewis naïve rat cohort. The detection sensitivity was set at a value comparable to the smallest detectable difference in means between two groups reported in the literature3,76. This detection sensitivity was used to calculate the required animals per group (n) to achieve this sensitivity at a statistical power of 0.8 and a significant level of 0/05. The red and green colors in the boxes are a qualitative scale highlighting larger n (red) and lower n (green).

CHAPTER 4 A GUI FOR THE EVALUATION OF KNEE OA (GEKO); AN OPEN-SOURCE TOOL FOR RAPID GRADING OF RODENT KNEE OA

In rodent models of osteoarthritis (OA), degeneration is assessed primarily using histology. Histology can be used to assess structural, cellular, and molecular features of joint damage in multiple tissues throughout the joint. This approach provides a detailed platform to study OA pathogenesis, but quantifying tissue changes can be challenging.

Several histological grading methods are currently used to assess OA-related joint degeneration. The histologic histochemical grading system (HHGS) developed by

Mankin et al95 with several modified versions56,110, assigns a numeric value to different cartilage features to qualitatively assess joint damage. However, this approach focuses primarily on cartilaginous changes. Moreover, the ordinal nature of the HHGS and modified Mankin schemes can produce identical values for different degenerative profiles, which can be misleading. To provide a more quantitative assessment of multiple OA-related joint tissue changes in rodents, the Osteoarthritis Research Society

International (OARSI) recently recommended a scheme measuring changes in cartilage, bone, and the synovium43. However, this more quantitative approach has not yet been widely implemented, partly due to the time and tedious nature of histological grading with this system. By reducing the time and effort of histological grading using the OARSI histology scheme for the rat, more researchers would be able to perform the more quantitative measures of joint damage included in this scheme.

In this paper, a graphic user interface (GUI) for the evaluation of knee OA

(GEKO) is introduced for the first time. GEKO is a MATLAB-based tool which helps graders perform the OARSI recommended grading scheme for the rat. First, GEKO loads a histological image and guides the grader to identify several tissue features.

Then, using these user inputs, GEKO calculates medial tibial plateau width, cartilage degeneration widths at three cartilage depths, cartilage degeneration scores, total and significant cartilage degeneration widths, cartilage zonal depth ratios, osteophyte size, medial joint capsule repair, and growth plate thicknesses at two locations. GEKO was first validated against manual grading for reproducibility (ability to produce identical results from the same data using two different methods or graders) using inter-class correlation coefficients (ICCs). GEKO reduced grading time by a factor of 8 while still reproducing manual grading results with ICCs above 0.94 for many histological variables. Once GEKO was validated, variation between graders was also explored.

GEKO produced more consistent grades for repeated images than manual grading

(repeating previous results from the same grader or method), but found that graders may have different interpretations of certain variables. Moreover, certain histological measures, such as osteophyte size, are well described and produced high ICCs for every comparison and every grader, while other histological measures inherently contained more room for interpretation. These data indicate GEKO is a useful tool to help graders produce high quality OARSI histological variables in much less time and effort than grading manually. Moreover, because GEKO is quick and easy to use, factors affecting grader variation can be explored in detail.

A GUI for the Evaluation of Knee OA (GEKO)

The GUI for the evaluation of knee OA (GEKO) is a MATLAB-based program written to help a grader assess histological images of rodent OA using the OARSI recommended grading scheme for rodents43. The suggested grades are well-defined in a previous publication43 and briefly defined here in Table 4-1. GEKO guides the grader

to identify several tissue features, then uses these inputs to calculate variables from the

OARSI histopathology initiative for the rat.

GEKO first loads the histological image to be graded, then provides written instructions and a reference image to identify each tissue feature (Figure 4-1). The features identified by the grader are: 1) medial tibial plateau width, 2) synovial thickness, 3) medial growth plate thickness, 4) lateral growth plate thickness, 5) osteochondral interface, 6) total cartilage degeneration width, 7) significant cartilage degeneration width, 8) cartilage lesion outline, and 9) osteophyte size. To identify these features, the grader is prompted to click endpoints defining each measure. The endpoints identified by the grader represent coordinates on the image and are used to calculate linear distances in pixels. For some features, such as the osteochondral interface and cartilage lesion, the grader is prompted to trace the feature using a series of clicks.

Once the grader has defined the appropriate tissue features, 9 other measures can be calculated without further user input: cartilage matrix loss width at three cartilage depths, cartilage degeneration score for each zone, and zonal depth ratio for each zone. To calculate these measures, the coordinates defining the tibial plateau, cartilage lesion, and osteochondral interface are used. First, GEKO calculates the angle between horizontal and the line representing the tibial plateau. Using this angle, the coordinates defining the tibial plateau, cartilage lesion, and osteochondral interface are rotated around the lateral most point defining the tibial plateau using a rotation matrix.

To calculate the cartilage matrix loss widths, the cartilage lesion is traced with a series of user-input points and transformed into a filled object within a matrix (Figure 4-

2A). This matrix is bounded above by the cartilage surface and below by the average of the coordinates of the osteochondral interface within the cartilage lesion width (Figure 4-

2B). The matrix is bounded on the sides by the left and right most coordinates of the lesion. Within this bounding box, the cartilage lesion object has a value of 1, while everything else has a value of 0. Because the cartilage lesion object does not always have squared edges, the width of the lesion at the cartilage surface is calculated from the top 8% of pixel rows. For a 4x magnification image of 900x1200 pixels, this is approximately 3 pixel rows and sufficient to overcome most object roughness noise. Of the selected pixel rows, the row with the maximum value is chosen to represent the cartilage matrix loss width at the cartilage surface. Cartilage matrix loss width is also calculated at 50% cartilage depth and the cartilage bone interface using the central 8% of rows and the bottom 8% of rows, respectively.

To calculate the cartilage degeneration score for each zone, the tibial plateau is divided into three equal zones: zone 1 (medial), zone 2 (central), and zone 3 (lateral)

(Figure 4-2C). Then, the maximum width of cartilage lesion in each zone is measured from the top 8% of pixel rows of the cartilage lesion object. This maximum lesion width is transformed into a percentage by dividing by the width of the zone (1/3 of the tibial plateau length). Using the percentage of the zone covered by the lesion, an ordinal rank is assigned for each zone (Table 4-2).

Zonal depth ratios from the OARSI scheme for the rat measure the percentage of damaged cartilage relative to the full cartilage thickness at the center of each zone. In these cases, damaged cartilage is defined as loss of chondrocytes or proteoglycans, not only missing cartilage. However, because GEKO currently only accounts for the

presence of cartilage and does not distinguish between damaged and undamaged cartilage, the zonal depth ratio calculated by GEKO measures the percentage of missing cartilage relative to the full cartilage width at the center of each zone (Figure 4-

2D). To measure GEKO’s zonal depth ratio, the vertical column of pixels representing the center of the zone is isolated. Then, the number of pixels corresponding with missing cartilage (cartilage surface to the depth of the cartilage lesion) is divided by the number of pixels representing the full cartilage thickness (cartilage surface to osteochondral interface).

Modeling OA and Preparing Histological Images

All methods and testing were performed with the approval from the University of

Florida Institutional Animal Care and Use Committee (IACUC).

Histological images representing a range of post-traumatic knee OA, a sham control, and naïve joints were taken from a previous study76. In that study, post- traumatic knee OA was modeled surgically in Lewis rats (3 months) through a medial collateral ligament transection, with exposure of the joint space, followed by a simulated radial transection of the medial meniscus (MMT)72,76. The sham controls received a medial collateral ligament transection procedure with exposure of the joint space without injuring the meniscus (MCLT sham). Tissue degeneration was allowed to develop in

MMT and MCLT sham animals for 1, 2, 4, and 6 weeks after surgery. After euthanasia, knees were collected, processed for paraffin embedding using vacuum infiltration, sectioned frontally at 10 µm, and stained with Toluidine blue, as suggested by the

OARSI histopathology initiative for the rat43. For this study, the section representing the most severe damage was imaged (EVOS XL, 4x magnification) for 42 joints (MMT n=6 per time point, MCLT sham n=3 per time point, naïve n=6).

100

Validating GEKO to Manual Grading

To validate GEKO, 42 histological images were randomized and graded by one trained grader (Grader 1) manually and using GEKO (Figure 4-1). Manual grading was performed by loading each image into Adobe Illustrator CS6 and placing lines over the image to identify each measure. Using a constant derived from a calibration image, illustrator line lengths were converted into microns and manually entered into Excel.

For numerical measures, significance between manual grading and GEKO results was determined using paired t tests grouped by surgery and time point. For ordinal data (cartilage degeneration score), significance was determined using a series of Mann-Whitney U tests. Additionally, GEKO’s reproducibility was assessed using inter-class correlation coefficients (ICCs). Taking the average ICC value, alpha model

ICCs with two-way random compensation were calculated using SPSS statistical software8,101,127.

Exploring Grader Variation

Once GEKO was validated to manual grading, the set of 42 histological images described above was graded by 5 additional trained graders (Graders 2-6). For each grader, the image set was randomized and presented as a batch to be graded in a single session.

Exploring Grader Repeatability and Reproducibility

To assess repeatability within GEKO and within manual methods (repeating results using the same method), three images were randomly repeated in the set of manual and GEKO images (creating sets totaling 48 images each). These data sets were graded by Grader 1 both manually and using GEKO. Additionally, using

101

randomized versions of these 48 image sets, grader repeatability within a single grading session was assessed by five other trained graders (Graders 2-6).

To assess grader reproducibility across several grading sessions, the 48 image set described above was graded by five graders (2-6) at four different times separated by 1 week.

To determine inter-rater reliability, ICCs were calculated as described above.

GEKO Reproduced Comparable or Better ICCs than Manual Grading

GEKO took less time to grade histological images than by using manual methods

(p<0.0001) (Figure 4-3 and 4-4). More importantly, GEKO reduced grading time without compromising quality. When comparing the histological variables for each surgical group and time point, GEKO produced different results for 12 of the 141 cases (Figure

4-3 and 4-4). Statistically, 7 false positives can be expected in 141 separate t-tests.

Moreover, of the 12 differences identified, 11 cases were between MMT comparisons and 1 was between a naïve group comparison for medial compartment medial growth plate (p<0.0497). In MMT animals, GEKO produced larger results than manual grading for tibial plateau width for weeks 2 and 4 (p<0.0296), cartilage matrix loss width at 50% cartilage depth at week 1 (p<0.0173), medial growth plate thickness at weeks 1 and 2

(p<0.0337), and lateral growth plate thickness at weeks 1 and 4 (p<0.0181). GEKO produced lower results than manual grading in MMT groups for cartilage matrix loss width at the cartilage bone interface (p<0.0425) and medial zonal depth ratio at weeks

1, 4, and 6 (p<0.0371).

When assessing GEKO’s reproducibility of manual results, GEKO produced highly correlative ICCs (Table 4-3). The ICCs were above 0.95 for 7 measures, above

0.90 for 3 measures, above 0.80 for 3 measures, and below 0.80 for only 1 measure

102

(cartilage matrix loss width at 100% cartilage depth). Because there was no variance in the data, zone 3 zonal depth ratio of lesions was unable to produce an ICC.

To assess Grader 1’s repeatability within a single grading session, GEKO produced higher ICCs than manual grading for nearly every histological measure

(Table 4-3). The majority of GEKO ICCs for single-session reliability were above 0.97.

For the cases where GEKO ICCs were not larger than manual grading ICCs, GEKO

ICCs were no more than 0.04 lower. For GEKO, zonal depth ratio of zone 1 and lateral growth plate thickness produced the two lowest ICC values of 0.81 and 0.83, respectively. For manual grading data, tibial plateau width and medial growth plate thickness produced the lowest ICC values of 0.619 and 0.626, respectively.

Multiple Graders Reproduced Manual Grading Results with High ICCs

When compared with manual grading, Graders 2-6 generally had high ICCs for cartilage matrix loss widths at 0% and 50% cartilage depth, cartilage degeneration score for zones 1 and 2, total and significant cartilage degeneration width, and osteophyte size (Table 4-4). Cartilage matrix loss width at 100% cartilage depth, cartilage degeneration score for zone 3, zonal depth ratios, and growth plate thicknesses all produces low ICCs for all graders.

Inter-grader ICCs were Relatively Low

Inter-grader reliability ICCs were typically lower than 0.85 (Table 4-4).

Osteophyte size had the largest inter-grader ICC at 0.977 with cartilage degeneration scores for zones 1 and 2 next at 0.975 and 0.954, respectively. Similar to Grader 1’s results, cartilage matrix loss width at 100% cartilage depth and zonal depth ratio of zone

3 were the lowest scores (0.195 and 0 respectively) with both growth plate thickness

ICCs following at 0.585 and 0.542.

103

Intra-session Reproducibility was Lower than Inter-session Reproducibility

Graders 2-6 produced mixed results when grading repeated images within a single grading session (Table 4-4). Typically, the lowest ICCs calculated for all graders were tibial plateau width, zonal depth ratio of zone 1, and both growth plate thicknesses ranged from ICCs of 0.29 to 0.907. Ranging from ICCs of 0.88 to 0.998, cartilage degeneration scores, total and significant cartilage degeneration widths, osteophyte size, and medial joint capsule repair were generally the highest ICCs for each grader.

Across the four grading sessions, Graders 2-6 produced slightly higher ICCs than their respective ICC values for repeated images within a single grading session (Table

4-4). Again, cartilage degeneration scores, cartilage degeneration widths, osteophyte size, and medial joint capsule repair produced the largest ICCs for each grader while tibial plateau width, zonal depth ratio for zone 1, and growth plate thicknesses produced lower ICC values. Additionally, for most graders, the ICC values for cartilage degeneration score for 100% cartilage depth were lower across sessions than the ICCs calculated for repeated images within a single session.

Discussing GEKO

Histological grading is tedious, fatiguing, and time-consuming. For the first time, a semi-automated GUI for the Evaluation of Knee OA (GEKO) was introduced, validated against manual grading, and used to explore variation within graders. GEKO produced similar histological grades as manual grading, but improved consistency relative to manual grading. Moreover, GEKO is easy to use and reduced grading time by more than a factor of 8 and enabled the testing of within session and between session

104

repeatability. In the future, these advantages over manual grading will allow researchers to explore unknown sources of error such as grader skill and fatigue.

Though GEKO and manual grading did not reproduce identical results, the variability of the changes is within the variability between sections taken at the same time point (inter-specimen variability). Unsurprisingly, the differences identified were nearly all in MMT comparisons, which also contain the largest overall variation for each measure. Because these differences were observed in both fully-user defined and

GEKO-calculated variables, natural grader variation may explain a portion of these differences between GEKO and manual grading. Because GEKO is semi-automated and relies on human inputs, GEKO has not removed grader error. However, the ICCs calculated for the repeated images are higher using GEKO than using manual methods, which indicates GEKO may improve a grader’s consistency.

A previous publication reports ICCs for some manual measures of the OARSI histopathology initiative for rat knee OA43. In that study, cartilage matrix loss widths, cartilage degeneration score for zones 1 and 2, and osteophyte size all produced ICCs above 0.9. GEKO produced comparable or better ICCs for each of those measures except cartilage matrix loss width at 100% cartilage depth, but this may be due to

GEKO’s strict guides for calculating histological variables without human intuition. For the other measures reported in the OARSI scheme for the rat (cartilage degeneration score for zone 3 and cartilage degeneration widths), GEKO produced much higher

ICCs, again this indicates that GEKO may help graders receive more consistent scores than manual grading.

105

The most reproducible grades were a mix between fully user-defined measures

(such as osteophyte size) and semi-automated measures derived from user inputs

(such as cartilage degeneration score). Highly reproducible measures may be better defined, with effective tissue markers and little room for grader interpretation.

Conversely, grades that were not highly reproducible may have been interpreted differently by different graders. For example, growth plate thicknesses produced uniformly low ICCs for every comparison. These measures are fully-user defined using

GEKO; however, the edge defining the growth plate undulates naturally. This undulation causes natural noise within the calculation if a grader chooses even a slightly different location to measure. A better approach may be to measure the growth plate thickness at several locations and calculate an average, thereby reducing the natural anatomical variability of the growth plate.

The measures of cartilage matrix loss width at 100% depth, cartilage degeneration score in zone 3, and zonal depth ratio of zone 3 also produced relatively low ICCs for all comparisons. Because these three measures are all semi-automated calculations from user inputs, their value depends on accurate measures of the cartilage lesion, tibial plateau width and osteochondral interface. Slight inconsistencies in these root measures can propagate to the derivative measures. Specifically, small changes in defining the tibial plateau will change the size and location of each cartilage zone, altering what portions of the lesion are used to calculate the cartilage degeneration score. Crucially, the tibial plateau width measure produced relatively low ICCs for each comparison. Given the inconsistency of the tibial plateau width measure, the cartilage degeneration widths were surprisingly consistent in each comparison, resulting in high

106

ICCs in every comparison. However, the noise propagated from the tibial plateau width measure may have affected the accuracy and repeatability of the cartilage degeneration scores and zonal depth ratios, though these measures also rely on how the cartilage lesion and osteochondral interface are defined. Errors in the tibial plateau width, cartilage lesion, and osteochondral interface will change the location of the lesion and zones used to calculate zonal depth ratios. A better approach may be to outline the entire cartilage tissue to continuously define the relationships between the tibial plateau, cartilage lesion, and osteochondral interface. This approach would also allow for new measures of cartilage thickening in different parts of the joint. By measuring these three variables together, their spatial location and orientation relationship might be more consistent.

The grading inconsistencies above were also present in the manual grading results, suggesting the variation is primarily due to human error. That said, GEKO calculates histological grades solely on the grader inputs while humans are able to make logical decisions using many factors in each situation. For example, GEKO will only record a cartilage matrix loss width at 100% cartilage depth if the lesion falls into the bottom 8% of pixel rows defining the cartilage. Though potentially more independently biased than GEKO, a human may produce a different value when looking at the histological image and decide where 100% cartilage depth is located on each slide and whether the lesion is breaching that depth. Thus, GEKO will produce more repeatable calculations of histological grades from user inputs and this may reduce user bias during the grading session.

107

In conclusion, GEKO is a validated tool to improve histological grading of rodent knee OA. GEKO reduced overall grading time by more than a factor of 8 while preserving high quality of data. Due to speed, repeatability controls were easily introduced during a grading session and used to explore sources of grader variability.

Overall, GEKO is a robust tool to improve quantitative histological grading.

108

If present, please click the two endpoints defining the thickest diameter of the osteophyte from the chondro-osseous junction to the cartilage surface. If not present, just press enter

Figure 4-1. Representative Image of GEKO Interface. Photos courtesy of author.

109

Figure 4-2. Description of GEKO Process

110

Figure 4-3. Paired raw data graphs of GEKO vs manual grading part 1. *: indicates p<0.05.

111

Figure 4-4. Paired raw data graphs of GEKO vs manual grading part 2. *: indicates p<0.05.

112

Table 4-1. Summary of OARSI histological measures

Table 4-2. Cartilage degeneration score

113

Table 4-3. ICCs of GEKO Validation

114

Table 4-4. ICCs for grader variation

115

CHAPTER 5 HISTOLOGICAL CHANGES IN THE SUBCHONDRAL BONE AND SYNOVIUM CORRELATE TO RODENT BEHAVIOR IN A MODEL OF POST-TRAUMATIC KNEE OA

Osteoarthritis (OA) patients often first present in the clinic due to joint pain35,124.

However, the severity of tissue damage does not necessarily relate to the severity of joint pain. While there are many ways to clinically assess OA pain, joint damage in humans is typically indirectly measured through radiography, ultrasound, MRI, or biomarker analysis of serum or synovial fluid111,118,122. Rodent models provide a more detailed platform to describe tissue changes in OA, with surgical models of post- traumatic OA producing tissue degeneration similar to that seen clinically. Moreover, with the advancement of detailed rodent behavioral analysis, rodent models now also allow for the exploration of the relationship between OA symptoms and joint damage.

Traditionally, histological grading is used to describe tissue degeneration in OA rodent models. The most prevalent grading scales originate from the Mankin scheme.

However, this system principally focuses on qualitative ranks of cartilage damage and does not fully assess other joint changes44,117,128. In 2010, the Osteoarthritis Research

Society International (OARSI) recommended histological assessment methods for specific animals43,77,86,102. For rodent models, these recommendations emphasized quantitative measures of cartilage damage, with a few supporting qualitative ranks43.

Since maladaptive changes occur throughout the joint in OA, the inclusion of multiple joint tissues in the grading process has improved our understanding of OA progression.

For example, though cartilage lesions are a significant effect of OA, the alteration of mechanical forces in the joint also affects the remodeling of the subchondral bone50,56,93. Likewise, the breakdown of cartilage can incite an inflammatory response in

116

the synovium82. Since multiple tissues in a joint communicate and adapt to joint changes through complex feedback loops, one tissue cannot change in isolation. As such, new measures to quantify changes throughout multiple joint tissues may be able to complement the existing OARSI measures, providing a more complete histological picture of the degeneration process.

Overall, the updated OARSI scheme for the rat has expanded the scope of histological analysis and supported the movement from ordinal-based grading systems to continuous variables that quantify tissue changes. Unfortunately, our prior work demonstrated these cartilage-centric measures were only weakly correlated to behavioral changes in a rodent OA model76. While this lack of correlation between joint changes and pain-related symptoms mirrors the clinical literature24,25,71, a better understanding of the relationship between OA progression and OA pain may be attained by more fully defining joint tissue changes throughout the entire joint.

In our prior work, gait modifications and changes in limb sensitivity were observed in conjunction with progressive cartilage damage in animals with combined MCLT and medial meniscus injury76. However, these same behavioral changes were also observed with the MCL transection alone, despite minimal damage to the articular cartilage.

These findings have led us to hypothesize that bone and synovial changes may be more related to changes in rodent behavior.

Following the same quantitative histological grading philosophy described in the OARSI scheme for the rat, histological measures of bone and synovium were developed for the rat. In this study, several histological quantifications are presented and correlated with

117

previously reported behavior in a post-traumatic rodent model of knee OA76.

Additionally, classic bone histomorphometric measures are included to help describe changes in subchondral bone. By using the existing behavioral data and histological resources, bone and synovial histological measures could be developed to complement existing OARSI measures and retroactively compared across studies.

Our data show that, along with cartilage damage measures included in the OARSI grading system for the rat, trabecular bone area and ossification increased progressively in both MMT and MCLT sham animals relative to naïve. Moreover, this increase in trabecular bone area also correlated with mechanical sensitivity, temporal imbalance in hind limb duty factor, and a reduction in animal stride length. Femoral cartilage thickening only occurred in MMT animals, but became progressively thicker at later time points. Though femoral cartilage thickness only significantly changed in MMT animals, the effect was strong enough to correlate with mechanical sensitivity, gait velocity, and a reduction in animal stride length in the whole model. Lastly, cell morphology in the subintimal layer of the synovium was altered in both MMT and MCLT sham animals relative to naïve. These synovial changes also correlated with mechanical sensitivity, gait velocity, and a reduction in animal stance time. Combined, these findings suggest early gait modifications and heightened mechanical sensitivity subsequent to surgical destabilization of the joint in the rat may be more related to bone and synovial tissue than to articular cartilage damage. Moreover, incorporation of quantitative bone and synovial changes in histological grading can provide a more thorough understanding of OA as a disease of the entire joint.

118

Experimental Design

All procedures were performed according to the University of Florida Institutional Animal

Care and Use Committee (IACUC), the Association for Assessment and Accreditation of

Laboratory Animal Care (AAALAC), and National Institutes of Health (NIH) guidelines.

Surgical techniques, mechanical sensitivity testing, spatiotemporal gait analysis, and

OARSI recommended histopathology grading for the rat have been reported for the animals included in this study in a previous manuscript (chapter 2)76. The focus of this study is to identify previously unmeasured histological features in bone and synovium using a similar histological grading philosophy described in the OARSI scheme for the rat43. These measures are intended to complement the current OARSI scheme and provide additional quantitative measures to assess tissue degeneration associated with

OA. Additionally, these histological measures were correlated with previously reported mechanical sensitivity data and spatiotemporal gait patterns76.

Osteoarthritis Model

To model OA, 72 male Lewis rats (3 months, 250g) were obtained from Charles River

Laboratories and allowed to acclimate to the housing environment for at least one week prior to any testing. A medial collateral ligament (MCL) transection and medial meniscus transection (MMT) was performed on 32 animals, as described previously72. Briefly, a medial skin incision was made along the knee joint, and tissue was bluntly dissected to expose the MCL. Then, the MCL was transected (MCLT), and the joint was placed in a valgus orientation to expose the joint space. The meniscus was then transected radially prior to closing. In addition, 32 animals received the same procedure through the MCLT, but without meniscal injury (MCLT sham). Prior work by several groups has

119

demonstrated that MCLT alone does not result in cartilage damage over 4-6 weeks and is frequently referred to as the sham procedure for the MMT procedure3,45,76. In the 2, 4, and 6 week post-op groups, wounds were stapled closed using 9 mm wound clips, which were removed 10-14 days after surgery. Since staples interfere with gait testing, the wounds of animals being tested at week 1 post-surgery were sutured using interrupted 5-0 Vicryl sutures in place of wound clips. Spatiotemporal gait and mechanical sensitivity testing was performed at weeks 1, 2, 4, and 6 post-surgery (n=8 per group per time point). The 8 remaining animals served as naïve controls.

Behavioral Tests

Mechanical sensitivity was assessed with Chaplan’s up-down protocol18. The 50% paw withdrawal threshold (50% PWT) was calculated by applying a series of von Frey filaments to the distal plantar region of the ipsilateral limb18. Smaller 50% PWT values indicate heightened mechanical sensitivity.Spatiotemporal gait analysis was performed as described previously1,3,76. Briefly, animals were acclimated to an open arena (60”x6”) with a transparent floor, a mirror beneath the floor placed below at 45°, and a high- speed video camera (RedLake, M3, 250 frames per second) placed to record the animal’s sagittal and ventral views. Animals were allowed to explore the arena without interference for 20 minutes or until 5 trials were collected. Videos were digitized manually using DLTdataviewer59 and spatiotemporal gait variables were calculated, including velocity, stride length, step width, temporal symmetry, duty factor of each hind limb, and duty factor imbalance. These gait variables are well described in previous studies and our methodological review 62,71.

120

Histology Results

After behavioral testing, animals were humanely euthanized. Knee joints were dissected and fixed for 48 hours in 10% neutral buffered formalin (Fisher Scientific). Ranging from

18-25 days, joints were decalcified at 40 °F in Cal-Ex (Fisher Scientific), with the decalcifying agent exchanged every 48 hours until the bone was soft. The joints were then processed for paraffin embedded via vacuum infiltration and sectioned frontally on a rotary microtome (Leica Biosystems, RM 2255) at a thickness of 10 µm. The sections were collected in strips of 10 (representing a total thickness of 100 µm) with 3-5 sequential sections of tissue mounted onto a slide. Thus, each slide represents joint changes at every 100 µm.

Preparing Histological Images

Three sequential slides representing areas of cartilage damage were stained with

Toluidine blue, and the medial compartment of each section was imaged at 4x magnification on an EVOS XL Core (Thermo Fisher Scientific). A total of 9 medial compartment images were taken of three sequential sections from the three sequential slides, with the one representing the most significant cartilage damage (the “original histological image”) used for the OARSI grading (Figure 5-1). One additional 10x magnification image of the medial, proximal synovium was taken from the section representing the most significant cartilage damage.

OARSI Grading

As suggested by the OARSI histopathology scheme for the rat, the single image representing the most severe tissue damage was graded as previously described76.

121

Measuring Bony Changes

Using the single image representing the most severe cartilage damage, measurements of bony changes were assessed as follows: first, the image is loaded into MATLAB and the grader identifies the medial tibial plateau width, medial and lateral femoral cartilage thicknesses, cartilage ossification, and any edemae that are present using a MATLAB- based graphic user interface (Figure 5-1). The script then calculates ossification width, edema size, and femoral condyle cartilage thicknesses, as described below.

To quantify subchondral bone ossification, the grader first estimates the medial and lateral endpoints of the tibial osteochondral interface over the loading zone of the tibial plateau (Figure 5-1). Then, the grader traces the osteochondral interface of the ossified cartilage. The script uses these coordinates to estimate the width and height of subchondral bone ossification as ratios of the osteochondral interface width.

If present, subchondral bone edemae are identified and traced by the grader (Figure 5-

1). Then, the script calculates the edema’s size. Edemae were also tracked throughout the 9 total histological images of the medial compartment. If an edema was present on an image, the image was assigned a value of 1; otherwise, it was assigned a value of 0

(indicating no edema was present on that image). For each animal, a ratio was calculated to express the percentage of the 9 sections containing edemae.

For cartilage changes on the femoral condyle, the grader identifies one medial and one lateral cartilage thickness orthogonal to the condyle (similar to growth plate thickness measures in the OARSI histopathology scheme for the rat) (Figure 5-1). The script then uses these coordinates to calculate the cartilage thickness at both points. The femoral

122

condyle cartilage thickness ratio is calculated by dividing the medial cartilage thickness by the lateral cartilage thickness; thus, uniform cartilage will result in a value near 1, while changes in cartilage changes will result in deviations from 1.

Measuring Trabecular Bone Area

Trabecular bone area was calculated using the original recommended equations from the American Society for Bone and Mineral Research (ASBMR) standard and the 2012 updated ASBMR recommendations for 2-dimensional bone histomorphometry26,113. The

ASBMR standard describes each component of the recommended equations, but does not provide detailed guidelines to measure each component. For example, trabecular bone volume is defined as trabecular bone volume divided by total bone volume

(trabecular bone volume plus bone marrow voids), but no specific method was suggested to collect trabecular bone volume or total bone volume. In this manuscript, trabecular bone area was calculated using the 9 medial compartment histological images as described below.

First, each image was edited in Adobe Photoshop CS5.1 to isolate the subchondral bone. To do this, the image was first rotated so the tibial plateau was horizontal and cropped to eliminate unused portions of the image. At this point, all tissues in the image besides subchondral trabecular bone and bone marrow voids were erased and filled in as black (Figure 5-2A). It is worth noting this elimination of tissue included tibial articular cartilage and any pre-osteophytes – which were removed to avoid error associated with differences in cartilage morphology. To isolate the bone marrow voids from the trabecular bone, any points where the dehydrated bone marrow touched subchondral bone were erased (Figure 5-2A). Then, the images were loaded into MATLAB, which

123

automatically removed the dehydrated bone marrow using object and color filters

(Figure 5-2A). The script then calculates the number of bone marrow voids, size of bone marrow voids, and trabecular bone area over the entire subchondral bone area.

Trabecular bone area is calculated using equation 1. Additionally, the script subdivides the subchondral bone area into a 3x3 grid and calculates the trabecular bone area in each zone (Figure 5-2B).

푇표푡푎푙 퐵표푛푒 푀푎푟푟표푤 푉표𝑖푑 퐴푟푒푎 푇푟푎푏푒푐푢푙푎푟 퐵표푛푒 퐴푟푒푎 = (1 − ) × 100 (퐸푞. 5 − 1) 푇표푡푎푙 푆푢푏푐ℎ표푛푑푟푎푙 퐵표푛푒 퐴푟푒푎

Quantifying Subintimal Synovium Changes

Using the synovium image, subintimal cell density, aspect ratio, and alignment were quantified. The synovium image was cropped in Adobe Photoshop CS5.1 to remove all other tissue from the image, rotated to align with the synovium edge, then loaded into

MATLAB (Figure 5-3A). Since Toluidine Blue is metachromic and has an affinity for nucleic acids, subintimal cell bodies are stained dark blue, while the surrounding extracellular matrix stains a much lighter blue (though more positively charged tissue can also appear purple). This feature allows MATLAB to isolate subintimal cells from the background using color and hue/saturation filters, correcting for the natural variance in the stain by using a red and green channel ratio. Once the cells were isolated, the script automatically calculates the number of cells, average cell density (# cells/100 pixels2), average length of major and minor axes of the cells, average cell aspect ratio (major axis/minor axis), and % of aligned cells (within +/- 3 degrees from 0). Prior to analysis, residuals of the major and minor cell axes lengths were also calculated relative to the respective average value for naïves.

124

Statistical Analysis

For ossification width, subchondral bone edema size, and femoral condyle cartilage thicknesses, trabecular bone area, number of cells in the synovial subintima, average cell density in the synovial subintima, average length of major and minor axes of the cells in the synovial subintima, average cell aspect ratio in the synovial subintima, and

% of aligned cells in the synovial subintima, 1-way ANOVAs with Tukey’s HSD post-hoc tests were used to assess differences across time points and between surgical groups.

Following ANOVAs and post-hoc tests, univariate regression models were used to explore the relationship between individual histological changes and behavioral changes. Behavior changes in these animals are reported in prior work (chapter 2)76 and in provided supplementary data (Supplemental Table 1). Similarly, OARSI histopathology scores reported previously are provided in Supplementary Table 2.

Because major and minor cell axes were found to change independently, an additional multivariate linear regression model using the major axis length residual, time point, and surgery (with interactions) was used to predict minor axis length residual for MMT and

MCLT sham animals.

Subchondral Bone Changes

Significant subchondral bone changes occurred in both MMT and MCLT sham groups, but not in naïve animals. In MMT animals, significant ossification width of the tibial cartilage occurred as early as week 1 post-surgery and became progressively more severe relative to naïve animals (p<0.0001, Figure 5-4A). MCLT animals also exhibited progressively widening ossification, but ossification did not become significantly different from naïve animals until week 4 (p=0.048).

125

At later time points, trabecular bone area increased (subchondral bone sclerosis) in both MMT (p=0.031) and MCLT (p=0.040) sham animals relative to naïve animals

(Figure 5-4B). Moreover, in MMT animals, the increased trabecular bone area may be largely influenced by trabecular bone area in zones B1 and C1, which include the majority of ossified cartilage (Figure 5-4C). In MMT animals, zone C1 area is larger than the same area as naïve animals at all time points (p=0.043), and zone B1 is area larger than the same area in naïve animals at weeks 4 and 6 post-surgery (p=0.002). Though

MCLT sham animals did exhibit increased overall trabecular bone area, no significant zonal differences were found at any time points when compared to naïve. Though overall and zonal area changed with respect to surgery and time point, no differences were observed in number or size of bone marrow spaces for any comparison.

Edemae only appeared in MMT animals. Edema size was significantly larger than naïve and MCLT animals at week 6 post-surgery (p<0.0001). In MMT animals, the percentage of sections containing edemae was a more sensitive measure, progressing in prevalence at each time point post-surgery relative to naïve animals (p<0.0001) and

MCLT sham animals (p<0.0001) (Figure 5-4D). The prevalence of edema in MMT animals was also significantly larger at weeks 4 and 6 relative to week 1 (p<0.0001) and near significance at week 6 relative to week 2 post-surgery (p=0.058).

Synovial Changes

In healthy animals, subintimal fibroblasts qualitatively appear as spindles with strong population directionality (Figure 5-5). Cell bodies in the synovial stroma of MMT animals are more spherical and lack directionality. MCLT sham subintimal cell bodies are morphologically between naïve and MMT animals, being less spindled and less aligned

126

than naïve fibroblasts, but having a more spindled shape and more alignment than MMT fibroblasts.

Overall, cell aspect ratio (a quantification of cell shape) was larger in naïve animals relative to MMT (p<0.0001) and MCLT sham animals (p<0.0001), with MCLT sham animals having larger cell aspect ratios than MMT animals (p<0.0001). In MMT and

MCLT animals, cell aspect ratio at weeks 2 and 6 post-surgery were also smaller than week 1 values (p=0.030). As such, MMT and MCLT sham subintimal cell aspect ratios were different from naïve at each time point (p<0.0001), but showed no progressive changes nor differences at individual time points. However, when group averages of minor axes are plotted against major axis, MMT and MCLT sham animal values track in different linear fashions as time progresses (Figure 5-5A). Surgery, major axis, and the interaction between surgery and major axis all affect minor axis (p=0.019, p=0.002, and p<0.0001, respectively). Week did not affect minor axis length, but the interaction between week and major axis residual was near significant (p=0.078).

Fewer cells were aligned in MMT animals relative to naïve animals at all time points

(p<0.0001) and fewer cells were aligned in MCLT sham animals relative to naïve animals at weeks 2, 4, and 6 post-surgery (p<0.0001) (Figure 5-5B). Though fewer cells were aligned in both MMT and MCLT sham animals, the decrease was more severe in

MMT animals, becoming significantly different between MMT and MCLT sham animals at week 6 post-surgery (p=0.020).

Per the medial joint capsule repair measure recommended by the OARSI histopathology initiative, joint capsules thickened in MMT and MCLT sham animals

127

relative to naïve animals. As such, subintimal cell density of MMT and MCLT sham animals were substantially larger than naïve animals at all time points even after correcting for image size (p<0.0001) (Figure 5-5C).

Femoral Cartilage Thickening

Cartilage thickening was observed on the femoral condyle cartilage in MMT animals. In

MMT animals, the ratio between medial compartment medial femoral cartilage thickness to medial compartment lateral femoral cartilage thickness was larger than naïve values and MCLT sham values at weeks 4 and 6 post-surgery (p<0.0001). The femoral cartilage thickness ratio remained near 1 in all MCLT sham and naïve animals.

Correlations to Mechanical Sensitivity

Mechanical sensitivity, as measured by 50% PWT, was significantly correlated to zone

B1 trabecular bone area, zone C1 trabecular bone area, femoral cartilage thickness ratio, and the population of aligned subintimal cells (Table 2). Behavioral measures are provided in Supplemental Table 2 and in prior work76. Of these correlations, the strongest correlative relationship was between subintimal cell alignment and 50% PWT.

New Histology Measures Correlate with Spatiotemporal Gait

Edemae, zones B1 and C1 trabecular areas, femoral cartilage thickness ratio, and % aligned subintimal cells correlate with gait velocity (Table 2). Trabecular bone area, zones B1 and C2 trabecular areas, ossification width, and femoral cartilage thickness ratio correlate with stride length residual. While histological measures from multiple tissues correlated with velocity changes, nearly all histological measures correlating with stride length residual were from the subchondral bone. Temporal symmetry also correlated with ossification in the subchondral bone. Changes in left hind limb duty factor correlated with subintimal cell aspect ratio (p=0.0042). Subintimal cell aspect ratio

128

also tended to correlate for right hind limb duty factor (p=0.0819). Only duty factor imbalance and step width residual did not correlate with our histological measures of subchondral bone and synovial changes.

Future Directions in Histology

OA affects all tissues of the joint. Structural changes include cartilage lesion formation, increased (and in some cases decreased) subchondral bone density, formation of osteophytes along the joint margin, breakdown of synovial fluid, and fibrosis and thickening of the synovium. Additionally, inflammation chronically alters the joint at both the molecular and tissue levels, disrupting joint homeostasis and promoting catabolic disease processes. Most histological assessments of OA in rodents do not describe these changes, instead focusing on cartilage degradation. Exploring ways to quantify

OA-related changes in these other tissues can provide new insights into the totality of

OA pathogenesis.

The OARSI scheme recognizes the significance of synovial changes and dedicated one measure to its thickness. The subintimal cell morphology measures presented in this study adds detail to this OARSI measure by quantifying several more features in the synovium. These data indicated subintimal cells have a sensitive, graded response to injury and might provide a useful indicator of joint health. Not only were MMT and MCLT sham subintimal cells different from naïve, their morphologies were distinct from one another and continued to change at each time point. These changes also correlated with changes in behavior and mechanical sensitivity (50% PWT).

Though cell morphology changes were successfully measured using the Toluidine blue stain, the natural variance in the dye presented challenges. Toluidine blue can easily

129

appear purple when under exposed and to blue when over exposed, requiring unique color thresholds to be used for each image. This variability sometimes obscured accurate cell identification and resulted in large standard deviations in the data. This problem was partially addressed by using a red to green channel ratio (blue images tend to have a lower ratio while purple images tend to have a higher ratio), but better image analysis could increase the sensitivity of this measure. Additionally, other stains, such as hematoxylin & eosin, may increase precision; however, Toluidine blue is easy to perform and is recommended for other histological measures by the OARSI histopathology initiative for rodents. As such, to assess cartilage, bone, and synovial damage on the same section, Toluidine blue is advantageous.

Subchondral bone is a complex network of marrow voids and trabecular bone. On histological slides, little order can be identified in subchondral bone, with patterns changing drastically from slide to slide. Using CT scans, subchondral bone has been shown to become more dense (sclerosis) in humans with OA70,85,134,140. Subchondral bone density has also been observed in rodents using a micro-CT99,100,107 and ultrasound121,122. Using histology from a similar OA model, another study56 measured trabecular bone area according to the ASBMR histomorphometry nomenclature committee113. In that study, trabecular bone area was measured by imaging a 600x800

μm portion of subchondral bone from two slides located 100 μm apart. Trabecular bone volumes calculated from these two images were then combined to represent a single trabecular bone volume for each joint. This approach was not sensitive enough to detect increasing bone area in our model (data not shown). However, expanding this approach

130

to nine sections was enough to confirm subchondral bone changes in MMT and MCLT animals.

While analyzing subchondral bone area and volume using histology is not as precise as using μCT, the histological approach presented herein was still able to collect spatial and structural data from cartilage, bone, and synovium from the same section. This work was also limited to using slides near the cartilage lesion to assess subchondral bone changes, but could be expanded to other joint locations, as other studies have shown trabecular bone volume changes in other areas51,99,106.

During ossification, cartilage converts to bone; thus, ossification could be responsible for the increased bone area in MMT and MCLT sham groups relative to naïve. However, comparisons of trabecular bone area in zones B1 and C1 (containing the ossified cartilage), reveal no significant difference between MCLT sham and naïve animals at any time point (Figure 5-4C). This indicates changes over the entire medial subchondral bone compartment might contribute to the difference between MCLT sham and naïve animals. Moreover, the subchondral bone changes observed in MCLT sham animals result from mechanical instability due to loss of the MCL and damage to the synovium during surgery, and not due to damage to cartilaginous tissues. Furthermore, the behavioral changes observed in the MCLT sham group correlate with these bone and synovial changes.

In this study, strong correlations were observed between behavior and histological measures. These relationships were not isolated to a single tissue and most behaviors correlated with multiple histological measures. Moreover, these correlations were not

131

restricted to the MMT group, as they were previously76. In the OARSI scheme for rodents, undamaged tissue will grade as a zero for most measures. This presents a challenge when correlating histology to behavior because quantitative measures do not assign a zero value to normal behaviors. For example, naïve rodents will histologically grade zero for cartilage damage measures, but display a range of tolerances for mechanical sensitivity calculations. By adding histological measures that assign continuous numerical values to joint damage, as described in this study, correlations between behavior and tissue changes can be made for all subjects.

The bone and synovium measures presented in this work described joint damage not widely assessed in OA models. Moreover, these changes were quantitative, continuous, and correlated better with behavioral changes than in previous studies. Combined, these data confirm OA-related tissue changes should be assessed in all joint tissues.

132

Figure 5-1. Summary of user-defined measures. Users define osteochondral width, medial femoral cartilage thickness, and lateral femoral cartilage thickness by clicking the two end points defining the measure. The user uses a series of clicks to outline any edemae or ossification that are present. Photos courtesy of author.

133

Figure 5-2. Calculating trabecular bone area. A) shows how the image is prepared in photoshop. B) trabecular bone are is calculated over the entire subchondral bone area and also within 9 zones. Photos courtesy of author.

134

Figure 5-3. Calculating synovium measures. A) shows how the image is prepared in photoshop. B) subintimal cells are identified and several measures are taken from each cell. Photos courtesy of author.

135

Figure 5-4. Subchondral bone changes. A) Ossification occurs in both MMT and MCLT Sham animals and becomes progressively more severe. B) Subchondral bone area increases in MMT animals by week 4 post-surgery and in MCLT animals by week 6 post-surgery. C) Zonal trabecular bone area is shown

136

graphically for zones A1 (left), B1 (central), and C1 (right). MMT animals had significantly larger trabecular bone areas than both MCLT Sham and naïve animals. D) Edema appeared in some MMT animals as early as week 1 but occurred in all MMT animals by week 4. Data are shown as animal averages. *: denotes significance from naïve. ^: denotes significance from MCLT Sham values at respective time point. 1: denotes significance from respective group at week 1 post-surgery.

137

138

Figure 5-5. Subintimal cell morphology. A) Group averages and standard deviations of minor axis residuals plotted against major axis residuals. Both MMT have larger minor axis measures and smaller major axis measures relative to naïve animals. There is also a trend for the major axis to progressively reduce relative to naïve across time points. B) Progressively lower percentages of subintimal cell population are aligned in both MMT and MCLT Sham animals. C) Cell proliferation, indirectly measured as cell density, increased progressively in both MMT and MCLT Sham groups, though the effect in MMT animals was more significant than MCLT Sham animals by week 6 post- surgery. Data are shown as animal averages. *: denotes significance from naïve. ^: denotes significance from MCLT Sham values at respective time point. 1: denotes significance from respective group at week 1 post-surgery. 2: denotes significance from respective group at week 2 post-surgery.

139

Table 5-1. Correlations of histological measures and behavior. Bold denotes a significant correlation (p<0.05). Italics denote a tendency to be significant (p<0.10).

140

CHAPTER 6 CONCLUSION

In this thesis, behavioral changes associated with OA were correlated with joint degeneration in a rodent model of post-traumatic knee OA. Two new tools were introduced to improve behavioral analysis and histological grading. Finally, new histological measures of the bone and synovium were presented and correlated with behavioral OA symptoms.

In Chapter 2, gait changes occurred over the same timescale as histological changes in the MMT model. However, since these same gait modifications occur with

MCLT alone (without any sign of cartilage damage), it is unlikely the gait modifications are due to cartilage damage. Instead, the data presented in Chapter 5 indicate bone and synovial damage is occurring both with MCLT and with MMT, and these bony and synovial changes may be associated with gait modifications. Combined, these data indicate histological measures of bony and synovial changes may be a better predictor of gait compensations than cartilage alone.

The automated method for gait analysis introduced in this work (AGATHA) was shown to detect gait changes in rats with orthopaedic or spinal cord injuries. This approach has since been adapted to analyze gait changes in mice and process animals of different colors (hairless, black, and spotted). AGATHA’s greatest strengths are scalability and adaptability, but it does require some structure in arena lighting and set- up. The AGATHA algorithms are robust and further generalizations of its algorithms will allow AGATHA to be used for non-rodent quadrupedal disease models and in new environments, such as fluoroscopy or force-instrumented arenas.

141

GEKO tool was presented in Chapter 4 to assist in histological grading. While

GEKO was validated for performing histological grading according to the OARSI histopathology initiative, there is enormous potential to assess other histological changes for OA and in other diseases. In Chapter 5, a GEKO-like approach was adapted to explore bone and synovial changes in OA joints, but these algorithms can be adapted to other OARSI recommended grading initiatives, as well as be expanded to incorporate any other feature a user wishes to describe within a joint. That said, there is room to improve quantitative histological metrics for OA. While the OA field as a whole has shifted toward more quantitative histological measures, many variables retain ordinal features. For example, osteophyte size is a continuous measurement in microns, but there is a binary element for choosing whether an osteophyte is present in the first place. Thus, you have an arbitrary discontinuity between the value given for no presence of an osteophyte (0) and much larger values assigned to joints containing osteophytes. By measuring the cartilage thickness at the joint margin (where osteophytes appear) in all joints, this osteophyte metric could be transformed into a true continuous variable. While not all measures can be transformed into purely continuous or ordinal variables, designing future histological variables with an understanding of how they are to be analyzed will create more accurate and robust histological descriptors.

Moreover, GEKO might be the perfect platform for training researchers to perform histological grading. GEKO is quick and easy to use and user input information can be retained for comparison with histological grading experts. By creating a platform where new graders can refine their skills, histological grading may become more standardized across the OA field and even across different models of OA.

142

Cartilage, bone, and synovial damage are primary contributors to joint degeneration. However, the links between these metrics and OA pain did not fully describe the relationship between joint pain and degeneration in our data sets. For example, OA-related changes in neural tissue, both peripherally and centrally, were not explored in these data, nor were molecular markers of pain assessed in any joint tissues. These data suggest other tissues should be included when looking for the link between OA pain and OA degeneration.

The relationship between OA pain and OA degeneration is complex, variable, and involves multiple tissues. Rodents can be used to study both behavioral and degenerative symptomology of post-traumatic knee OA. By improving analysis techniques, a deeper understanding between joint pain and degeneration can be explored. Importantly, the tools presented herein are designed to be accessible to other researchers and help standardize common metrics used by the preclinical OA field. By standardizing field methodologies, published results can become more consistent, comparisons can be made between results from different labs, and researchers will be able to include high quality gait analysis and histological grading into their repertoire.

143

LIST OF REFERENCES

1. Allen, K. D., S. B. Adams, B. a. Mata, M. F. Shamji, E. Gouze, L. Jing, D. L. Nettles, L. D. Latt, and L. a. Setton. Gait and behavior in an IL1β-mediated model of rat knee arthritis and effects of an IL1 antagonist. J. Orthop. Res. 29:694–703, 2011.

2. Allen, K. D., T. M. Griffin, R. M. Rodriguiz, W. C. Wetsel, B. Kraus, J. L. Huebner, L. M. Boyd, and L. A. Setton. Decreased Physical Function and Increased Pain Sensitivity in Mice Deficient for Type IX Collagen. Arthritis Rheum. 60:2684–2693, 2010.

3. Allen, K. D., B. A. Mata, M. A. Gabr, J. L. Huebner, S. B. Adams, V. B. Kraus, D. O. Schmitt, and L. A. Setton. Kinematic and dynamic gait compensations resulting from knee instability in a rat model of osteoarthritis. Arthritis Res. Ther. 14:R78, 2012.

4. Allen, K. D., D. Ph, S. B. Adams, and L. A. Setton. Evaluating Intra-Articular Drug Delivery for the Treatment of Osteoarthritis in a Rat Model. Tissue Eng. Part B Rev. 16:81–92, 2010.

5. Allen, K. D., M. F. Shamji, B. A. Mata, M. A. Gabr, S. M. Sinclair, D. O. Schmitt, W. J. Richardson, and L. A. Setton. Kinematic and dynamic gait compensations in a rat model of lumbar radiculopathy and the effects of tumor necrosis factor-alpha antagonism. Arthritis Res. Ther. 13:R137, 2011.

6. Ameye, L. G., and M. F. Young. Animal models of osteoarthritis: lessons learned while seeking the “Holy Grail”. Curr. Opin. Rheumatol. 18:537–47, 2006.

7. Baker, K., A. Grainger, J. Niu, M. Clancy, A. Guermazi, M. Crema, L. Hughes, J. Buckwalter, A. Wooley, M. Nevitt, and D. T. Felson. Relation of synovitis to knee pain using contrast-enhanced MRIs. Ann. Rheum. Dis. 69:1779–83, 2010.

8. Bartko, J. J. On Various Intraclass Correlation Reliability Coefficients. Psychol. Bull. 83:762–765, 1976.

9. Bartlett, R., M. Bussey, and N. Flyger. Movement variability cannot be determined reliably from no-marker conditions. J. Biomech. 39:3076–9, 2006.

10. Bauman, J. M., and Y.-H. Chang. High-speed X-ray video demonstrates significant skin movement errors with standard optical kinematics during rat locomotion. J. Neurosci. Methods 186:18–24, 2010.

11. Beare, J. E., J. R. Morehouse, W. H. Devries, G. U. Enzmann, D. A. Burke, D. S. K. Magnuson, and S. R. Whittemore. Gait Analysis in Normal and Spinal Contused Mice Using the TreadScan System. J. Neurotrauma 2056:2045–2056, 2009.

144

12. Bedson, J., and P. R. Croft. The discordance between clinical and radiographic knee osteoarthritis: a systematic search and summary of the literature. BMC Musculoskelet. Disord. 9:116, 2008.

13. Berryman, E. R., R. L. Harris, M. Moalli, and C. M. Bagi. Digigait quantitation of gait dynamics in rat rheumatoid arthritis model. J. Musculoskelet. Neuronal Interact. 9:89–98, 2009.

14. Bonatz, A. E., H. Steiner, and J. P. Huston. Video image analysis of behavior by microcomputer: categorization of turning and locomotion after 6-OHDA injection into the substantia nigra. J. Neurosci. Methods 22:13–26, 1987.

15. Bove, S. E., K. D. Laemont, R. M. Brooker, M. N. Osborn, B. M. Sanchez, R. E. Guzman, K. E. Hook, P. L. Juneau, J. R. Connor, and K. S. Kilgore. Surgically induced osteoarthritis in the rat results in the development of both osteoarthritis- like joint pain and secondary hyperalgesia. Osteoarthritis Cartilage 14:1041–8, 2006.

16. Brown, J., and J. Hoopes. A Device for Recording a Rat’s Position in a Straight Alley. Am. J. Psychol. 78:130–133, 1965.

17. Chamberlain, C. S., E. M. Leiferman, K. E. Frisch, S. L. Brickson, W. L. Murphy, G. S. Baer, and R. Vanderby. Interleukin Expression after Injury and the Effects of Interleukin-1 Receptor Antagonist. PLoS One 8:e71631, 2013.

18. Chaplan, S. R., F. W. Bach, J. W. Pogrel, J. M. Chung, and T. L. Yaksh. Quantitative assessment of tactile allodynia in the rat paw. J. Neurosci. Methods 53:55–63, 1994.

19. Chen, Y.-J., F.-C. Cheng, M.-L. Sheu, H.-L. Su, C.-J. Chen, J. Sheehan, and H.- C. Pan. Detection of subtle neurological alterations by the Catwalk XT gait analysis system. J. Neuroeng. Rehabil. 11:62, 2014.

20. Christopher B. Little and Margaret M. Smith, Little CB and Smith MM, Christopher B. Little and Margaret M. Smith, and Little CB and Smith MM. Animal Models of Osteoarthritis. Curr. Rheumatol. Rev. 4:175–182, 2008.

21. Cicuttini, F. M., J. Baker, D. J. Hart, and T. D. Spector. Association of pain with radiological changes in different compartments and views of the knee joint. Osteoarthr. Cartil. 4:, 1996.

22. Courtney, C. a, M. A. O’Hearn, and T. G. Hornby. Neuromuscular function in painful knee osteoarthritis. Curr. Pain Headache Rep. 16:518–24, 2012.

23. Crawley, J. N. What’s Wrong With My Mouse?: Behavioral Phenotyping of Transgenic and Knockout Mice, 2nd Edition. Hoboken, New Jersey: John Wiley & Sons, Inc., 2007, 544 pp.

145

24. Cubukcu, D., A. Sarsan, and H. Alkan. Relationships between Pain, Function and Radiographic Findings in Osteoarthritis of the Knee: A Cross-Sectional Study. Arthritis 2012:984060, 2012.

25. Davis, M. A., W. H. Ettinger, J. M. Neuhaus, J. D. Barclay, and M. R. Segal. Correlates of knee pain among US adults with and without radiographic knee osteoarthritis. J Rheumatol 19:, 1992.

26. Dempster, D., J. Compston, M. Drezner, F. Glorieux, J. Kanis, H. Malluche, P. Meunier, S. Ott, R. Recker, and A. Parfitt. Standardized Nomenclature, Symbols, and Units for Bone Histomorphometry: A 2012 Update of the Report of the ASBMR Histomorphometry Nomenclature Committee. J. bone Miner. Res. 20:233–243, 2013.

27. Doperalski, N., M. Sandhu, R. Bavis, P. Reier, and D. Fuller. Ventilation and phrenic output following hi cervical spinal hemesection in male vs. female rats. Respir. Physiol. Neurobiol. 18:1199–1216, 2013.

28. Dorman, C. W., H. E. Krug, S. P. Frizelle, S. Funkenbusch, and M. L. Mahowald. A comparison of DigiGait TM and TreadScan TM imaging systems : assessment of pain using gait analysis in murine monoarthritis. J. Pain Res. 7:25–35, 2014.

29. Dougherty, B., K. Lee, E. Gonzalez-Rothi, M. Lane, P. Reier, and D. Fuller. Recover of inspiratory intercostal muscle activity following high cervical hemisection. Respir. Physiol. Neurobiol. 18:1199–1216, 2013.

30. Dunham, K. a, A. Siriphorn, S. Chompoopong, and C. L. Floyd. Characterization of a graded cervical hemicontusion spinal cord injury model in adult male rats. J. Neurotrauma 27:2091–2106, 2010.

31. Egloff, C., T. Hügle, and V. Valderrabano. Biomechanics and pathomechanisms of osteoarthritis. Swiss Med. Wkly. 142:w13583, 2012.

32. Englund, P., and L. Lohmander. Risk factors for symptomatic knee osteoarthritis fifteen to twenty-two years after meniscectomy. Arthritis Rheumatol. 50:2811– 2819, 2004.

33. Farr, J., B. Cole, A. Dhawan, J. Kercher, and S. Sherman. Clinical cartilage restoration: Evolution and overview. Clin. Orthop. Relat. Res. 469:2696–2705, 2011.

34. Felson, D. T. The epidemiology of knee osteoarthritis: Results from the Framingham Osteoarthritis study. Semin Arthritis Rheum 20:, 1991.

35. Felson, D. T. The sources of pain in knee osteoarthritis. Curr. Opin. Rheumatol. 17:624–8, 2005.

36. Felson, D. T. Developments in the clinical understanding of osteoarthritis. Arthritis

146

Res. Ther. 11:203, 2009.

37. Felson, D. T., C. E. Chaisson, C. L. Hill, S. M. S. Totterman, M. E. Gale, K. M. Skinner, L. Kazis, and D. R. Gale. The Association of Bone Marrow Lesions with Pain in Knee Osteoarthritis. Ann. Intern. Med. 134:541–549, 2001.

38. Fernihough, J., C. Gentry, M. Malcangio, A. Fox, J. Rediske, T. Pellas, B. Kidd, S. Bevan, and J. Winter. Pain related behaviour in two models of osteoarthritis in the rat knee. Pain 112:83–93, 2004.

39. Fu, S. C., Y. C. Cheuk, L. K. Hung, and K. M. Chan. Limb Idleness Index (LII): A novel measurement of pain in a rat model of osteoarthritis. Osteoarthr. Cartil. 20:1409–1416, 2012.

40. Fuller, D., M. Sandhu, N. Doperalski, M. Lane, T. White, M. Bishop, and P. Reier. Graded unilateral cervical spinal cord injury and respiratory motor recovery. Respir. Physiol. Neurobiol. 18:1199–1216, 2013.

41. Gabriel, A. F., M. A. E. Marcus, W. M. M. Honig, G. H. I. M. Walenkamp, and E. A. J. Joosten. The CatWalk method: a detailed analysis of behavioral changes after acute inflammatory pain in the rat. J. Neurosci. Methods 163:9–16, 2007.

42. Geiger, B., H. Steenbock, and H. Parsons. Lathyrism in the rat. J. Nutr. 6:427– 442, 1932.

43. Gerwin, N., A. M. Bendele, S. Glasson, and C. S. Carlson. The OARSI histopathology initiative - recommendations for histological assessments of osteoarthritis in the rat. Osteoarthritis Cartilage 18 Suppl 3:S24–34, 2010.

44. Glasson, S., M. Chambers, W. Van Den Berg, and C. Little. The OARSI histopathology initiative - recommendations for histological assessments of osteoarthritis in the mouse. Osteoarthr. Cartil. 18:S17–S23, 2010.

45. Glasson, S. S., T. J. Blanchet, and E. A. Morris. The surgical destabilization of the medial meniscus (DMM) model of osteoarthritis in the 129/SvEv mouse. Osteoarthritis Cartilage 15:1061–9, 2007.

46. Goldring, M. B., and S. R. Goldring. Articular cartilage and subchondral bone in the pathogenesis of osteoarthritis. Ann. N. Y. Acad. Sci. 1192:230–237, 2010.

47. Goldring, M. B., and K. B. Marcu. Cartilage homeostasis in health and rheumatic diseases. Arthritis Res. Ther. 11:224, 2009.

48. Gonzalez-Rothi, E., G. Armstrong, A. Cerreta, G. Fitzpatric, P. Reier, M. Lane, A. Judge, and D. Fuller. Forelimb muscle plasticity following unilateral cervical spinal cord injury. Muscle Nerve 53:475–478, 2016.

49. Gravel, P., M. Tremblay, H. Leblond, S. Rossignol, and J. A. de Guise. A semi-

147

automated software tool to study treadmill locomotion in the rat: from experiment videos to statistical gait analysis. J. Neurosci. Methods 190:279–88, 2010.

50. Grynpas, M., B. Alpert, I. Katz, I. Lieberman, and K. Pritzker. Subchondral bone in osteoarthritis. Calcif. Tissue Int. 49:20–26, 1991.

51. Gu, X. I., P. E. Palacio-Mancheno, D. J. Leong, Y. A. Borisov, E. Williams, N. Maldonado, D. Laudier, R. J. Majeska, M. B. Schaffler, H. B. Sun, and L. Cardoso. High resolution micro arthrography of hard and soft tissues in a murine model. Osteoarthr. Cartil. 20:1011–1019, 2012.

52. Guilak, F. Biomechanical factors in osteoarthritis. Best Pract. Res. Clin. Rheumatol. 25:815–823, 2011.

53. Hamers, F. P., A. J. Lankhorst, T. J. van Laar, W. B. Veldhuis, and W. H. Gispen. Automated quantitative gait analysis during overground locomotion in the rat: its application to spinal cord contusion and transection injuries. J. Neurotrauma 18:187–201, 2001.

54. Hamers, F. P. T., A. J. Lankhorst, T. J. A. N. V. A. N. Laar, W. B. Veldhuis, and W. H. Gispen. Automated Quantitative Gait Analysis During Overground Locomotion in the Rat : Its Application to Spinal Cord Contusion and Transection Injuries. J. Neurotrauma 18:, 2001.

55. Hannan, M. T., D. T. Felson, and T. Pincus. Analysis of the discordance between radiographic changes and knee pain in osteoarthritis of the knee. J. Rheumatol. 27:1513–1517, 2000.

56. Hayami, T., M. Pickarski, G. A. Wesolowski, J. McLane, A. Bone, J. Destefano, G. A. Rodan, and L. T. Duong. The Role of Subchondral Bone Remodeling in Osteoarthritis: Reduction of Cartilage Degeneration and Prevention of Osteophyte Formation by Alendronate in the Rat Anterior Cruciate Ligament Transection Model. Arthritis Rheum. 50:1193–1206, 2004.

57. Hayami, T., M. Pickarski, Y. Zhuo, G. A. Wesolowski, G. A. Rodan, and L. T. Duong. Characterization of articular cartilage and subchondral bone changes in the rat anterior cruciate ligament transection and meniscectomized models of osteoarthritis. Bone 38:234–243, 2006.

58. Hayashi, D., A. Guermazi, and D. J. Hunter. Osteoarthritis year 2010 in review: Imaging. Osteoarthr. Cartil. 19:354–360, 2011.

59. Hedrick, T. L. Software techniques for two- and three-dimensional kinematic measurements of biological and biomimetic systems. Bioinspir. Biomim. 3:034001, 2008.

60. Hildebrand, M. Symmetrical gaits of horses. Science (80-. ). 150:701–708, 1965.

148

61. Hildebrand, M. Analysis of Asymmetrical Gaits. J. Mammology 58:131–156, 1977.

62. Hildebrand, M. The Quadrupedal Gaits of Vertebrates. Bioscience 39:766–774, 1989.

63. Hill, D., P. Batchelor, M. Holden, and D. Hawkes. Medical image registration. Phys. Med. Biol. 46:R1–R45, 2001.

64. Hruska, R., S. Kennedy, and E. Silbergeld. Quantitative Aspects of Normal Locomotion in Rats. Life Sci. 25:171–180, 1979.

65. Hull, C. L. The rat’s speed-of-locomotion gradient in the approach to food. J. Comp. Psychol. 17:393–422, 1934.

66. Hunter, D. J., a. Guermazi, G. H. Lo, a. J. Grainger, P. G. Conaghan, R. M. Boudreau, and F. W. Roemer. Evolution of semi-quantitative whole joint assessment of knee OA: MOAKS (MRI Osteoarthritis Knee Score). Osteoarthr. Cartil. 19:990–1002, 2011.

67. Hunter, D. J., J. J. McDougall, and F. J. Keefe. The symptoms of osteoarthritis and the genesis of pain. Med. Clin. North Am. 93:83–100, xi, 2009.

68. Hwang, P. Y., K. D. Allen, M. F. Shamji, L. Jing, B. A. Mata, M. A. Gabr, J. L. Huebner, V. B. Kraus, W. J. Richardson, and L. A. Setton. Changes in Midbrain Pain Receptor Expression, Gait and Behavioral Sensitivity in a Rat Model of Radiculopathy. Open Orthop. J. 6:383–391, 2012.

69. Im, H., J. Kim, X. Li, N. Kotwal, D. R. Sumner, J. Andre, V. Wijnen, F. J. Davis, D. Yan, B. Levine, J. L. Henry, and J. S. Kroin. Alteration of Sensory Neurons and Spinal Response to An Experimental Osteoarthritis Pain Model. Arthritis Rheum. 62:2995–3005, 2011.

70. Intema, F., T. Thomas, D. Anderson, J. Elkins, T. Brown, A. Amendola, F. Lafeber, and C. Saltzman. Subchondral bone remodeling is related to clinical improvement after joing distraction in the treatment of ankle osteoarthritis. Osteoarthr. Cartil. 72:181–204, 2011.

71. Jacobs, B. Y., H. E. Kloefkorn, and K. D. Allen. Gait analysis methods for rodent models of osteoarthritis. Curr. Pain Headache Rep. 18:456, 2014.

72. Janusz, M. Induction of osteoarthritis in the rat by surgical tear of the meniscus: Inhibition of joint damage by a matrix metalloproteinase inhibitor. Osteoarthr. Cartil. 10:785–791, 2002.

73. Johnson, V., and D. Hunter. The epidemiology of osteoarthritis. Best Pract. Res. Clin. Rheumatol. 27:743–755, 2014.

74. Johnstone, B., M. Alini, M. Cucchiarini, G. R. Dodge, D. Eglin, F. Guilak, H.

149

Madry, A. Mata, R. L. Mauck, C. E. Semino, and M. J. Stoddart. Tissue engineering for articular cartilage repair--the state of the art. Eur. Cell. Mater. 25:248–67, 2013.

75. Keefe, F. J., J. C. Lefebvre, J. R. Egert, G. Affleck, M. J. Sullivan, and D. S. Caldwell. The relationship of gender to pain, pain behavior, and disability in osteoarthritis patients: the role of catastrophizing. Pain 87:325–334, 2000.

76. Kloefkorn, H. E., B. Y. Jacobs, A. M. Loye, and K. D. Allen. Spatiotemporal gait compensations following medial collateral ligament and medial meniscus injury in the rat: correlating gait patterns to joint damage. Arthritis Res. Ther. 17:287, 2015.

77. Kraus, V. B., J. L. Huebner, J. DeGroot, and A. Bendele. The OARSI histopathology initiative - recommendations for histological assessments of osteoarthritis in the guinea pig. Osteoarthr. Cartil. 18:1–38, 2010.

78. Krizsan-agbas, D., M. K. Winter, L. S. Eggimann, J. Meriwether, N. E. Berman, P. G. Smith, and K. E. Mccarson. Gait Analysis at Multiple Speeds Reveals Differential Functional and Structural Outcomes in Response to Graded Spinal Cord Injury. J. Neurotroma 31:846–856, 2014.

79. Kuyinu, E. L., G. Narayanan, L. S. Nair, and C. T. Laurencin. Animal models of osteoarthritis: classification, update, and measurement of outcomes. J. Orthop. Surg. Res. 11:19, 2016.

80. Lakes, E. H., and K. D. Allen. Gait Analysis Methods for Rodent Models of Arthritic Disorders: Reviews and Recommendations. Osteoarthr. Cartil. , 2016.doi:10.1016/j.joca.2016.03.008.This

81. Lane, M. A., D. D. Fuller, T. E. White, and P. J. Reier. Respiratory neuroplasticity and cervical spinal cord injury: translational perspectives. Trends Neuroscie 31:538–547, 2009.

82. de Lange-Brokaar, B. J. E., A. Ioan-Facsinay, G. J. V. M. van Osch, a-M. Zuurmond, J. Schoones, R. E. M. Toes, T. W. J. Huizinga, and M. Kloppenburg. Synovial inflammation, immune cells and their cytokines in osteoarthritis: a review. Osteoarthritis Cartilage 20:1484–99, 2012.

83. Laurin, J., E. Dousset, S. Mesure, and P. Decherchi. Neuromuscular recovery pattern after medial collateral ligament disruption in rats. J. Appl. Physiol. 107:98– 104, 2009.

84. Lawrence, R. C., D. T. Felson, C. G. Helmick, L. M. Arnold, H. Choi, R. A. Deyo, S. Gabriel, R. Hirsch, M. C. Hochberg, G. G. Hunder, J. M. Jordan, J. N. Katz, H. M. Kremers, and F. Wolfe. Estimates of the prevalence of arthritis and other rheumatic conditions in the United States. Part II. Arthritis Rheum. 58:26–35, 2008.

150

85. Li, G., J. Yin, J. Gao, T. S. Cheng, N. J. Pavlos, C. Zhang, and M. H. Zheng. Subchondral bone in osteoarthritis : insight into risk factors and microstructural changes. Arthritis Res. Ther. 15:, 2013.

86. Little, C., M. Smith, M. Cake, R. Read, M. Murphy, and F. Barry. The OARSI histopathology initiative - recommendations for histological assessments of osteoarthritis in sheep and goats. , 2010.

87. Liu-Bryan, R. Synovium and the Innate Inflammatory Network in Osteoarthritis Progression. Curr. Rheumatol. Rep. 15:1–12, 2014.

88. Lohmander, L., P. Englund, L. Dahl, and E. Roos. The Long-term Consequence of Anterior Cruciate Ligament and Meniscus Injuries Osteoarthritis. Am. J. Sports Med. 35:1756–1769, 2007.

89. Lories, R. J., and F. P. Luyten. the Bone–Cartilage Unit in Osteoarthritis. Nat. Publ. Gr. 7:43–49, 2010.

90. Lotz, M. K., and V. B. Kraus. New developments in osteoarthritis. Posttraumatic osteoarthritis: pathogenesis and pharmacological treatment options. Arthritis Res. Ther. 12:211, 2010.

91. Mabey, T., and S. Honsawek. Cytokines as biochemical markers for knee osteoarthritis. World J. Orthop. 6:95–105, 2015.

92. Machado, A. S., D. M. Darmohray, J. Fayad, H. G. Marques, and M. R. Carey. A quantitative framework for whole-body coordination reveals specific deficits in freely walking ataxic mice. Elife 4:1–22, 2015.

93. Mahjoub, M., F. Berenbaum, and X. Houard. Why subchondral bone in osteoarthritis? The importance of the cartilage bone interface in osteoarthritis. Osteoporos. Int. 23 Suppl 8:S841–6, 2012.

94. Malfait, A. M., C. B. Little, and J. J. McDougall. A commentary on modelling osteoarthritis pain in small animals. Osteoarthritis Cartilage 21:1316–26, 2013.

95. Mankin, H. J., H. Dorfman, L. Lippiello, and A. Zarins. Biochemical and Metabolic Abnormalities in Articular Cartilage from Osteo-Arthritic Human Hips. J. Bone Jt. Surg. 53-A:523–537, 1971.

96. Martinez, M., M. Delcour, M. Russier, Y. Zennou-Azogui, C. Xerri, J.-O. Coq, and J.-M. Brezun. Differential tactile and motor recovery and cortical map alteration after C4-C5 spinal hemisection. Exp. Neurol. 221:186–97, 2010.

97. McDonough, A. L., M. Batavia, F. C. Chen, S. Kwon, and J. Ziai. The validity and reliability of the GAITRite system’s measurements: A preliminary evaluation. Arch. Phys. Med. Rehabil. 82:419–25, 2001.

151

98. McDougall, J. J., R. C. Bray, and K. A. Sharkey. Morphological and immunohistochemical examination of nerves in normal and injured collateral ligaments of rat, rabbit, and human knee joints. Anat. Rec. 248:29–39, 1997.

99. McErlain, D. D., C. T. G. Appleton, R. B. Litchfield, V. Pitelka, J. L. Henry, S. M. Bernier, F. Beier, and D. W. Holdsworth. Study of subchondral bone adaptations in a rodent surgical model of OA using in vivo micro-computed tomography. Osteoarthr. Cartil. 16:458–469, 2008.

100. McErlain, D. D., V. Ulici, M. Darling, J. S. Gati, V. Pitelka, F. Beier, and D. W. Holdsworth. An in vivo investigation of the initiation and progression of subchondral cysts in a rodent model of secondary osteoarthritis. Arthritis Res. Ther. 14:R26, 2012.

101. McGraw, K. O., and S. P. Wong. Forming inferences about some intraclass correlations coefficients. Psychol. Methods 1:390–390, 1996.

102. McIlwraith, C., D. Frisbie, C. Kawcak, C. Fuller, M. Hurtig, and A. Cruz. The OARSI histopathology initiative - recommendations for histological assessments of osteoarthritis in the horse. Osteoarthr. Cartil. 18:S93–S105, 2010.

103. Mendes, C. S., I. Bartos, Z. Márka, T. Akay, S. Márka, and R. S. Mann. Quantification of gait parameters in freely walking rodents. BMC Biol. 13:50, 2015.

104. Mendes, C. S., I. Bartos, Z. Márka, T. Akay, S. Márka, and R. S. Mann. Quantification of gait parameters in freely walking rodents. BMC Biol. 13:50, 2015.

105. Mogil, J. S. Animal models of pain: progress and challenges. Nat. Rev. Neurosci. 10:283–94, 2009.

106. Mohan, G., E. Perilli, J. S. Kuliwaba, J. M. Humphries, I. H. Parkinson, and N. L. Fazzalari. Application of in vivo micro-computed tomography in the temporal characterisation of subchondral bone architecture in a rat model of low-dose monosodium iodoacetate-induced osteoarthritis. Arthritis Res. Ther. 13:R210, 2011.

107. Mohan, G., E. Perilli, I. H. Parkinson, J. M. Humphries, N. L. Fazzalari, and J. S. Kuliwaba. Pre-emptive, early, and delayed alendronate treatment in a rat model of knee osteoarthritis: Effect on subchondral trabecular bone microarchitecture and cartilage degradation of the tibia, bone/cartilage turnover, and joint discomfort. Osteoarthr. Cartil. 21:1595–1604, 2013.

108. Mullenix, P., S. Norton, and B. Culver. Locomotor damage in rats after X- irradiation in utero. Exp. Neurol. 48:310–324, 1975.

109. Negus, S. S., T. W. Vanderah, M. R. Brandt, E. J. Bilsky, L. Becerra, and D. Borsook. Preclinical assessment of candidate analgesic drugs: recent advances and future challenges. J. Pharmacol. Exp. Ther. 319:507–514, 2006.

152

110. Neo, H., J. I. Ishimaru, K. Kurita, and A. N. Goss. The effect of hyaluronic acid on experimental temporomandibular joint osteoarthrosis in the sheep. J. Oral Maxillofac. Surg. 55:1114–1119, 1997.

111. Neu, C. P. Functional imaging in OA: Role of imaging in the evaluation of tissue biomechanics. Osteoarthr. Cartil. 22:1349–1359, 2014.

112. Orita, S., T. Ishikawa, M. Miyagi, N. Ochiai, G. Inoue, Y. Eguchi, H. Kamoda, G. Arai, T. Toyone, Y. Aoki, T. Kubo, K. Takahashi, and S. Ohtori. Pain-related sensory innervation in monoiodoacetate-induced osteoarthritis in rat knees that gradually develops neuronal injury in addition to inflammatory pain. BMC Musculoskelet. Disord. 12:134, 2011.

113. Parfitt, A. M., M. K. Drezner, F. H. Glorieux, J. A. Kanis, H. Malluche, P. J. Meunier, S. M. Ott, and R. R. Recker. Bone histomorphometry: standardization of nomenclature, symbols, and units. Report of the ASBMR Histomorphometry Nomenclature Committee. , 1987.

114. Pearson, R. G., T. Kurien, K. S. S. Shu, and B. E. Scammell. Histopathology grading systems for characterisation of human knee osteoarthritis - reproducibility, variability, reliability, correlation, and validity. Osteoarthr. Cartil. 19:324–331, 2011.

115. Pham, D. L., C. Xu, and J. L. Prince. Current Methods in Medical Image Segmentation. Annu. Rev. Biomed. Eng. 2:315–337, 2000.

116. Piel, M. J., J. S. Kroin, A. J. Van Wijnen, R. Kc, and H. J. Im. Pain assessment in animal models of osteoarthritis. Gene 537:184–188, 2014.

117. Pritzker, K. P. H., S. Gay, S. a. Jimenez, K. Ostergaard, J. P. Pelletier, K. Revell, D. Salter, and W. B. van den Berg. Osteoarthritis cartilage histopathology: Grading and staging. Osteoarthr. Cartil. 14:13–29, 2006.

118. Roemer, F. W., F. Eckstein, D. Hayashi, and A. Guermazi. The role of imaging in osteoarthritis. Best Pract. Res. Clin. Rheumatol. 28:31–60, 2014.

119. Roubille, C., J.-P. Raynauld, F. Abram, P. Paiement, M. Dorais, P. Delorme, L. Bessette, A. D. Beaulieu, J. Martel-Pelletier, and J.-P. Pelletier. The presence of meniscal lesions is a strong predictor of neuropathic pain in symptomatic knee osteoarthritis: a cross-sectional pilot study. Arthritis Res. Ther. 16:1–7, 2014.

120. Rushton, R., H. Steinberg, and C. Tinson. Effects of a single experience on subsequent reactions to drugs. Br. J. Pharmacol. Chemother. 20:99–105, 1963.

121. Saarakkala, S., P. Waris, V. Waris, I. Tarkiainen, E. Karvanen, J. Aarnio, and J. M. Koski. Diagnostic performance of knee ultrasonography for detecting degenerative changes of articular cartilage. Osteoarthritis Cartilage 20:376–81, 2012.

153

122. Saïed, A., E. Chérin, H. Gaucher, P. Laugier, P. Gillet, J. Floquet, P. Netter, and G. Berger. Assessment of articular cartilage and subchondral bone: subtle and progressive changes in experimental osteoarthritis using 50 MHz echography in vitro. J. bone Miner. Res. 12:1378–86, 1997.

123. Sandhu, M. S., B. J. Dougherty, M. A. Lane, D. C. Bolser, P. A. Kirkwood, P. J. Reier, and D. D. Fuller. Respiratory recovery following high cervical hemisection. Respir. Physiol. Neurobiol. 169:94–101, 2009.

124. Schaible, H.-G. Mechanisms of Chronic Pain in Osteoarthritis. Curr. Rheumatol. Rep. 14:549–556, 2012.

125. Scheff, S. W., A. G. Rabchevsky, I. Fugaccia, J. A. Main, and J. E. Lumpp. Experimental modeling of spinal cord injury: characterization of a force-defined injury device. J. Neurotrauma 20:179–193, 2003.

126. Schwartz, M. H., J. P. Trost, and R. A. Wervey. Measurement and management of errors in quantitative gait data. Gait Posture 20:196–203, 2004.

127. Shrout, P. E., and J. L. Fleiss. Intraclass correlations: Uses in assessing rater reliability. Psychol. Bull. 86:420–428, 1979.

128. van der Sluijs, J. a, R. G. Geesink, A. J. van der Linden, S. K. Bulstra, R. Kuyer, and J. Drukker. The reliability of the Mankin score for osteoarthritis. J. Orthop. Res. 10:58–61, 1992.

129. Smith, M. D. The normal synovium. Open Rheumatol. J. 5:100–6, 2011.

130. Somers, T. J., F. J. Keefe, J. J. Pells, K. E. Dixon, S. J. Waters, P. A. Riordan, J. A. Blumenthal, D. C. McKee, L. LaCaille, J. M. Tucker, D. Schmitt, D. S. Caldwell, V. B. Kraus, E. L. Sims, R. A. Shelby, and J. R. Rice. Pain Catastrophizing and Pain-Related Fear in Osteoarthritis Patients: Relationships to Pain and Disability. J. Pain Symptom Manage. 37:863–872, 2009.

131. Sophia Fox, A. J., A. Bedi, and S. A. Rodeo. The basic science of articular cartilage: structure, composition, and function. Sports Health 1:461–8, 2009.

132. Sterling, M., G. Jull, and A. Wright. The effect of musculoskeletal pain on motor activity and control. J. Pain 2:135–45, 2001.

133. Stevenson, G. W., H. Mercer, J. Cormier, C. Dunbar, L. Benoit, C. Adams, J. Jezierski, A. Luginbuhl, and E. J. Bilsky. Monosodium iodoacetate-induced osteoarthritis produces pain-depressed wheel running in rats: Implications for preclinical behavioral assessment of chronic pain. Pharmacol. Biochem. Behav. 98:35–42, 2011.

134. Swagerty, D. L., and D. Hellinger. Radiographic assessment of osteoarthritis. Am. Fam. Physician 64:279–286, 2001.

154

135. Teeple, E., G. D. Jay, K. A. Elsaid, and B. C. Fleming. Animal models of osteoarthritis: challenges of model selection and analysis. AAPS J. 15:438–46, 2013.

136. Thakur, M., A. H. Dickenson, and R. Baron. Osteoarthritis pain: nociceptive or neuropathic? Nat. Rev. Rheumatol. 10:374–80, 2014.

137. Toia, F., T. Giesen, P. Giovanoli, and M. Calcagni. A systematic review of animal models for experimental neuroma. J. Plast. Reconstr. Aesthetic Surg. 68:1447– 1463, 2015.

138. Torello, M. W., J. Czekajewski, E. A. Potter, K. J. Kober, and Y. K. Fung. An automated method for measurement of circling behavior in the mouse. Pharmacol. Biochem. Behav. 19:13–17, 1983.

139. Tsushima, H., M. E. Morris, and J. McGInley. Test-Retest Reliability and Inter- Tester Reliability of Kinematic Data from a Three-Dimensional Gait Analysis System. J. Japanese Phys. Ther. Assoc. 6:9–17, 2003.

140. Turmezei, T. D., and K. E. S. Poole. Computed tomography of subchondral bone and osteophytes in hip osteoarthritis: The shape of things to come? Front. Endocrinol. (Lausanne). 2:1–9, 2011.

141. Wang, F., S. M. Ieee, E. Stone, W. Dai, M. Skubic, and J. Keller. Gait Analysis and Validation Using Voxel Data. , 2009.

142. Wei, L., B. C. Fleming, X. Sun, E. Teeple, W. Wu, G. D. Jay, K. A. Elsaid, J. Luo, J. T. Machan, and Q. Chen. Comparison of differential biomarkers of osteoarthritis with and without posttraumatic injury in the Hartley guinea pig model. J. Orthop. Res. 28:900–6, 2010.

143. Whishaw, I., and B. Kolb. The Behavior of the Laboratory Rat: A Handbook with Tests. Oxford: Oxford University Press, 2005, 499 pp.

144. Wilson, D. J., B. K. Smith, J. K. Gibson, K. Byung, B. C. Gaba, and J. T. Voelz. Accuracy of Digitization Using Automated and Manual Methods. Phys. Ther. J. 79:558–566, 2015.

145. Wong, P., V. I. Leo, M. Low, T. W. Mak, X. Zhang, and P. T. Reilly. Targeted ANP32E Mutant Mice Do Not Demonstrate Obvious Movement Defects. PLoS One 8:4–10, 2013.

146. Wooley, C. M., R. B. Sher, A. Kale, W. N. Frankel, G. A. Cox, and K. L. Seburn. Gait analysis detects early changes in transgenic SOD1(G93A) mice. Muscle Nerve 32:43–50, 2005.

147. Woolf, C. Central sensitization: Implications for the diagnosis and treatment of pain. Pain 152:S2–15, 2011.

155

BIOGRAPHICAL SKETCH

Heidi Elise Kloefkorn received a Bachelor of Science degree in biomedical engineering from the Georgia Institute of Technology before being accepted to the

University of Florida where she continued her education. Heidi received her Doctor of

Philosophy under the guidance of Kyle D. Allen from the University of Florida's J.

Crayton Pruitt Family department of biomedical engineering. Heidi published several papers in peer-reviewed scientific journals and mentored more than a dozen undergraduate projects during her time in the Orthopaedic Biomedical Engineering

Laboratory at the University of Florida.

156