Variations in Three-Dimensional Cancellous Bone Architecture of the Proximal Femur in Female Hip Fractures and in Controls

Authors


  • Presented at the 44th annual meeting of the Orthopaedic Research Society, New Orleans, Louisiana, 1998.

Abstract

Cubes of cancellous bone were obtained from proximal femora of women with hip fractures (n = 26) and from female cadaveric controls (n = 32) to compare architecture and mechanics between groups. Specimens were scanned on a microcomputed tomography system. Stereologic algorithms and model-based estimates were applied to the data to characterize the three-dimensional cancellous microstructure. Cubes were mechanically tested to failure to obtain mechanical properties. Specimens from control subjects had significantly higher bone volume fraction, trabecular number, and connectivity than specimens from patients with hip fractures; no difference in trabecular thickness was observed between groups. Both maximum modulus and ultimate stress were significantly higher in the control than in the fracture group, consistent with the higher bone volume found in the control group. No statistical differences in any of these architectural or mechanical variables were found when groups were matched for bone volume. Specimens from both patients with hip fractures and controls demonstrated strong relationships between trabecular number and bone volume fraction that were statistically equivalent, suggesting that for a given bone mass, both groups have the same overall number of trabeculae. However, there was an architectural difference between fracture and control groups in terms of the three-dimensional spatial arrangement of trabeculae. Fracture specimens had a significantly more anisotropic (oriented) structure than control specimens, with proportionately fewer trabecular elements transverse to the primary load axis, even when matched for bone volume. Relationships between mechanical and architectural parameters were significantly different between groups, suggesting that fracture and control groups have different structure-mechanics relationships, which we hypothesize may be a consequence of the altered three-dimensional structure between groups.

INTRODUCTION

Osteoporotic hip fracture continues to be one of the most serious public health problems facing society today. More than a quarter million hip fractures occur every year,(1) resulting in significant morbidity and mortality. Twelve to twenty percent of all cases ultimately lead to death, and long-term nursing home care is required for half of those who survive.(2) Financial costs of osteoporosis in the United States alone are estimated at $5–10 billion per year, and are expected to increase further with the increasing size and longevity of the population.(1)

Numerous studies have been performed to examine the potential causes and risk factors for fractures of the proximal femur. Despite significant associations between hip fracture and both low bone mass and falling, especially falls to the side,(3–8) neither bone mass nor falling are sufficient to definitively categorize who will sustain a fracture. Previous work has shown that fracture patients as a whole may have lower bone mass than controls, but there is still substantial overlap in measurements between groups.(3,5,6,8–11) Other studies demonstrated no differences in bone mineral content of the femoral neck(4) or femoral head(12) between patients with femoral neck fracture and healthy age-matched controls. In addition, it is estimated that 90% of hip fractures result from a fall, but only 5% of falls result in a hip fracture.(8) Thus it seems evident that factors other than bone mass and falls are important to hip fracture risk.

One such potential factor whose role in hip fracture has not been systematically explored is cancellous architecture. Architecture influences cancellous bone mechanical properties, and therefore may have implications for hip fractures. Trabecular orientation and connectivity are both related to cancellous bone strength and stiffness.(13–15) Trabecular bone that has no preferential axis of orientation (isotropic) will resist loads equally in any direction. In contrast, anisotropic specimens (contain axis of preferred orientation) will be stronger and stiffer when loaded along the direction of dominant trabecular orientation and weaker when loaded along an axis where the trabeculae are less oriented. Few studies have compared microstructures from fracture and control groups to assess whether or not they are similar. Furthermore, these studies were limited to examination of two-dimensional histological sections of bone from the iliac crest.(9–11,16–20) While measures of trabecular number and thickness can be quantified from histological sections, three-dimensional connectivity cannot be determined. The ilium is commonly used in bone histomorphometry studies since this site lends itself most conveniently for biopsy and samples can therefore be obtained from a greater number of subjects. However, the use of an iliac model to study osteoporotic fracture, particularly with respect to the proximal femur, has been questioned.(6,21,22) The ilium does not experience the same functional loads as the proximal femur, nor does it have a similar microstructure.(23,24) The ilium has a relatively isotropic curved plate morphology, whereas the proximal femur has a rod-plate or rod-rod morphology with preferred axes of orientation (anisotropic). Moreover, the ilium is not associated with fracture, and thus may not serve as a meaningful comparison for the proximal femur. To determine if architecture has a role in the fracture of the proximal femur, systematic, regionally specific three-dimensional architectural measurements made at the fracture site are needed. This may be particularly important for femoral neck fractures, since this region has a relatively thin cortical shell surrounding a larger volume of trabecular bone that likely is responsible for carrying the majority of the loads transmitted across the hip.

Thus, the objectives for this study were threefold. The first was to characterize the three-dimensional cancellous microstructure and mechanical properties of the proximal femur in women who had had femoral neck fractures and in nonfracture controls. The second set of objectives were designed to test how microstructure and mechanical properties were related to bone mass in the fracture and control groups. That is, we hypothesized that for a given bone mass, the two groups will arrange the cancellous microstructure or behave mechanically in a way that is fundamentally different. Finally, we sought to make a more detailed examination of specimens from both groups whose bone mass measures overlap to determine if any architectural or mechanical distinctions were present.

MATERIALS AND METHODS

Specimen acquisition

Excised femoral heads were recovered from women at St. Joseph Mercy Hospital (Ypsilanti, MI, U.S.A.) who had undergone arthroplasty for neck fractures of the proximal femur. Specimens were wrapped in a saline-soaked paper towel and stored frozen until future use. Patient age and date of surgery were recorded. Specimens from fractures occurring due to extreme trauma (motor vehicle accidents, falls from ladders, etc.) were not included in the study. All activities were approved by institutional review boards and administrative units. Specimen retrieval was coordinated with the surgical nursing staff at the hospital.

Control specimens were obtained from the Anatomical Donations Program at the University of Michigan (Ann Arbor, MI, U.S.A.). Femurs were harvested from female cadavers, and a cut was made at the base of the femoral neck and perpendicular to its axis using a hand saw. The distal fragment was discarded, and the proximal segment was cleaned of any remaining muscle tissue and stored frozen until future use. The age and cause of death of each specimen were recorded. Persons who had cancerous bone lesions were not included in the study.

Specimen preparation

Eight-millimeter-thick frontal slices, centered about the insertion of the ligament teres, were cut from fracture and control specimens using a diamond saw. Slices were contact-radiographed. All films were screened by a radiologist to identify specimens showing pathology so they could be excluded from the study. Reasons for exclusion included a lytic lesion (n = 1) and a previously undiagnosed fracture in a control specimen (n = 1). Osteoarthritis was not considered a reason for exclusion.

X-ray films were used as templates to fabricate 8-mm cubes from the region immediately inferior to the epiphyseal line containing primary compressive trabeculae and oriented along their axis (Fig. 1). This axis will be subsequently referred to as the inferosuperior (IS) axis of each cube. This particular site was chosen for study because it was the location closest to the fracture site from which a cube could be consistently obtained. Cubes could not be prepared from three fracture specimens because of damage caused to the cancellous bone by surgical instrumentation. For one fracture specimen, the cube was cut ∼3 mm to the right of the standard location in order to avoid damaged regions and obtain a cube of intact bone. On the basis of gross morphological examination, no differences in bone volume or structure between the two regions were apparent.

Figure FIG. 1.

Location within proximal femur from which 8-mm-cube specimens were prepared. All frontal slices were radiographed, and the X-ray film served as a template to remove a cube immediately inferior to the epiphyseal line containing primary compressive trabeculae and oriented along their axis. This axis is referred to as the inferosuperior IS axis throughout the study.

After all exclusions, the data set consisted of 32 control specimens (mean age 75.1 years) and 26 fracture specimens (mean age 80.8 years) (Table 1). Cubes were stored frozen until future use.

Table Table 1. Number, Mean Bone Volume Fraction, and Mean Age of Specimens in Fracture and Nonfracture Control Groups for the Complete Data Set and for the EBMS
original image

Architectural characterization

Bone cubes were removed from the freezer and immediately scanned with a microcomputed tomography system. Data were reconstructed on a 50 μm mesh to characterize three-dimensional cancellous architecture as previously described.(15,23,25) Briefly, three independent parameters (Pp, PL, and Euler-Poincaré number) were measured for each cube. Pp represents the number of bone voxels per total number of voxels in an analysis region. PL is defined as the number of intersections between bone and nonbone components per length of test line. The Euler-Poincaré number represents the extent of connectedness between trabecular elements, where (1 − Euler Number) represents the number of branches in the structure that can be removed without breaking the structure into two parts. This measure of connectivity is described in detail by Feldkamp et al.(23)

Stereologic algorithms(15) based on point and intercept counting and model based estimates(17) assuming a plate structure were applied to the three-dimensional data sets to determine bone volume fraction, trabecular number, connectivity, and trabecular thickness in terms of these threeindependent parameters according to the following equations:

equation image(1)
equation image(2)
equation image(3)
equation image(4)

In addition, an anisotropy algorithm based on measurements of mean intercept length in several directions was utilized to determine orthogonal directions of primary trabecular orientation and the structural parameter PL in each of the three directions (PLi, i = 1,2,3). The mean intercept length method was described by Whitehouse.(26) The value is based on calculating PL for test lines that are rotated at 5° increments. Using the mean intersection counts per unit test line length for all angles and plotting them in polar coordinates allows one to define an ellipsoid whose axes provide a measure of orientation relative to reference point. For this study, the direction of greatest orientation is indicated by i = 1, and the direction of least orientation by i = 3, with PL1 < PL2 < PL3. The degree of anisotropy, a relative measure of orientation within cubes, was calculated as the ratio between PL3 and PL1, and represents the ratio of the number of trabecular elements aligned in the most- to least-oriented direction in the cube(15):

equation image(5)

Mechanical testing

To evaluate mechanical properties, cubes were wetted with saline and mechanically tested in uniaxial compression at room temperature. All tests were performed on an MTS machine (Model 810, MTS Systems Corp., Eden Prairie, MN, U.S.A.) at constant displacement rate of 1%/sec, monitoring load with the system load cell and displacement with an LVDT mounted between unlubricated loading platens. Data-acquisition software (LabVIEW, National Instruments Corp., Austin, TX, U.S.A.) and a microcomputer (Macintosh IIci, Apple Computers Inc., Cupertino, CA, U.S.A.) were used to collect load and displacement data throughout the test. Specimens were loaded preyield to 0.4% strain in the anteroposterior and mediolateral directions, and then to failure to 15% strain in the IS direction with no end restraint. Each test was preceded by ten preconditioning cycles, where the specimen was loaded to 0.4% strain. Specimens were kept moist with saline during testing. Data were analyzed to determine two mechanical parameters: maximum modulus (Emax, in megapascals), defined as the maximum slope within the linear region of the stress-strain curve of the failure test, and the ultimate stress (σult, in megapascals) (Fig. 2).

Figure FIG. 2.

Typical stress-strain curve for destructive mechanical test performed in the inferosuperior direction. Data were analyzed for maximum modulus (Emax), defined as the maximum slope within the linear region of the stress-strain curve, and for ultimate stress (σult).

Data sets

Two sets of data were examined in this study (Table 1). The first was the complete original data set, which included all subjects. The second set is what will subsequently be referred to as the “equal bone mass subset” (EBMS). The EBMS was analyzed to determine if there were any architectural parameters other than bone mass that would distinguish the fracture group from the control group. This subset included only those individuals with bone volume fraction in the range of the mean bone volume fraction of the fracture group ±1 SD. It was intended to represent individuals from both groups whose bone volume measures overlap. For the complete data set, the control group was significantly younger in age and had significantly higher bone volume fraction than the fracture group. No statistical differences in age or bone volume fraction existed between groups within the equal bone mass subset (Table 1).

Data analysis

Student's t-test was used to determine differences in architectural and mechanical variables between fracture and control groups for the complete data set and the EBMS. Linear regression analysis was used to examine relationships between architectural and mechanical parameters, and analysis of covariance was used to compare slopes and intercepts of those regressions between fracture and control groups. Commercial software (SYSTAT, SYSTAT, Inc., Evanston, IL, U.S.A.) was used to perform all analyses, and p < 0.05 was considered significant.

RESULTS

Architecture

For the complete data set, controls had significantly higher bone volume fraction, trabecular number, and connectivity than fracture cases (Fig. 3). There was no difference in trabecular thickness between groups (Fig. 3). In contrast, no statistical differences in any of these architectural variables were found between fracture and control groups when the equal bone mass subset was analyzed (Fig. 3).

Figure FIG. 3.

Comparison of architectural parameters between fracture and nonfracture control groups for the complete data set and for the EBMS: (A) bone volume fraction, (B) trabecular number, (C) connectivity, (D) trabecular thickness. For the complete data set, controls had significantly higher bone volume fraction, trabecular number, and connectivity than fracture cases, but there was no difference in trabecular thickness. No statistical differences in any of these architectural parameters were found between groups when analyzing the EBMS. Bars indicate SD (p < 0.05, Student's t-test).

For all cubes, the direction of greatest orientation, indicated by the lowest PL value, corresponded to the primary compressive trabeculae (IS) axis. The anatomic anteroposterior direction had the next greatest degree of orientation, and the corresponding mediolateral direction had the least orientation. For the complete data set, PLi was significantly higher for controls than for fracture cases in all three principal directions (Fig. 4). However, there were no statistical differences in individual PLiwhen analyzing the EBMS (Fig. 4).

Figure FIG. 4.

Comparison of PL1, PL2, and PL3 between fracture and nonfracture control groups for the complete data set and for the EBMS. For the complete data set, PLi was significantly higher for controls than for fracture cases in all three principal directions. There were no statistical differences in individual PLi when analyzing the EBMS. Bars indicate SD (p < 0.05, Student's t-test).

There was a significant difference in relative orientation within cubes between groups, irrespective of bone mass. As indicated by the degree of anisotropy, the fracture specimens had a significantly more anisotropic structure than the controls, even when analyzing the EBMS (Fig. 5).

Figure FIG. 5.

Comparison of degree of anisotropy between fracture and nonfracture control groups for the complete data set and for the EBMS. Fracture cases had a significantly more anisotropic structure, even when matched for bone volume. Bars indicate SD (p < 0.05, Student's t-test).

Mechanical properties

Mechanical results were consistent with the architectural findings. Both maximum modulus and ultimate stress were significantly higher in the control than in the fracture group, consistent with the higher bone volume found in the control group (Fig. 6). When matched for bone volume, there were no differences between groups in either of these mechanical parameters (Fig. 6).

Figure FIG. 6.

Comparison of maximum modulus and ultimate stress along the IS direction between fracture and nonfracture control groups for the complete data set and for the EBMS. For the complete data set, controls had significantly higher maximum modulus and ultimate stress than fracture cases. No statistical differences in these mechanical properties were found between groups when analyzing the EBMS. Bars indicate SD (p < 0.05, Student's t-test ).

Regression analyses

Both fracture and control groups demonstrated strong linear relationships between trabecular number and bone volume fraction. Regression equations were statistically equivalent, indicating that for a given bone mass, specimens from both groups have the same trabecular number (Fig. 7). Connectivity and bone volume fraction were not related for either group.

Figure FIG. 7.

Linear regression plots of trabecular number (Tb.N) versus bone volume fraction for fracture and nonfracture control groups. Regressions were significant (p < 0.05) for both groups. Slopes and intercepts of the regressions were statistically equivalent between groups (ANCOVA, p < 0.05).

Both fracture and control groups demonstrated significant relationships between architectural and mechanical variables. However, the regression lines differed significantly between groups, suggesting that specimens from fracture and control groups have fundamentally different relationships between structure and mechanics. Positive linear relationships between maximum modulus and bone volume fraction and between maximum modulus and trabecular number were observed for specimens from both groups (Figs. 8, 9). In both cases, the regression for the control group had a steeper slope and a lower intercept than that for fracture cases.

Figure FIG. 8.

Linear regression plots of Emax vs. bone volume fraction for fracture and nonfracture control groups. Regressions were significant (p < 0.05) for both groups. Slopes and intercepts of the regressions were significantly different between groups (ANCOVA, p < 0.05).

Figure FIG. 9.

Linear regressions of Emax versus trabecular number (Tb.N) for the fracture and nonfracture control groups. Regressions were significant (p < 0.05) for both groups. Slopes and intercepts of the regressions were significantly different between groups (ANCOVA, p < 0.05).

Ultimate stress also had a significant positive relationship with bone volume fraction and trabecular number, but slopes and intercepts of the regressions for fracture and control groups were not significantly different.

DISCUSSION

Osteoporosis and bone fragility have principally been studied with regard to their association with low bone mass.(1–3) However, it is becoming increasingly evident that additional factors are likely important to fully characterize the etiology of osteoporotic fracture. One such factor is cancellous microstructure. The thickness and number of individual trabecular elements and the manner in which they are connected to each other will affect overall mechanical behavior of the cancellous continuum.(13–15) Microstructures with greater connectivity and alignment of trabeculae along the load axis demonstrate higher stiffness and strength. Altered iliac structure has been demonstrated in vertebral fracture patients and controls of equal bone mass, with fracture patients having fewer number of slightly thicker trabeculae.(11,18) It is believed that the consequential loss of connectivity in the structure of the patients contributed to their fracture risk.

In previous studies which compared iliac bone from patients with hip fractures and controls, the only structural index examined was bone volume.(9,10,16) These studies reported mixed results, with one demonstrating relatively lower bone volume in fracture patients (consistent with the present study)(9) and others showing no difference between groups.(16,10) Hordon and Peacock(20) measured trabecular number and thickness of the iliac crest but reported no differences compared to age- and sex-matched controls. Consistent with Hordon and Peacock's findings, in the present study we found no differences in these structural parameters (or in connectivity) when specimens of fracture patients and controls were matched for bone volume.

Results from the present study are not directly comparable to previous work because we characterized the microstructure of the proximal femur rather than the iliac crest. To our knowledge, this is the only study which has examined the microstructure of the proximal femur in patients with hip fractures and controls.

One objective of this work was to examine whether or not patients who sustain a fracture have a geometric arrangement of bone that differs from controls of comparable bone mass. No significant differences in connectivity, trabecular number, or trabecular thickness were found between groups when matched for bone volume. Furthermore, regressions of trabecular number on bone volume fraction for fracture cases and controls were statistically equivalent, suggesting that for a given bone mass, both groups have the same overall number of trabeculae. However, the architecture of the fracture and control groups differ in terms of the three-dimensional spatial arrangement of trabeculae. Specimens from fracture patients have a significantly more anisotropic structure than that of specimens from controls, with proportionately fewer trabecular elements transverse to the primary load axis. This architectural distinction was maintained even when specimens were matched for bone volume.

It should be noted that although the difference in the degree of anisotropy (the ratio between PL3 and PL1) between specimens from fracture patients and control subjects within the EBMS was significant (Fig. 3), no differences were observed between groups for either PL1 or PL3 (Fig. 4). In order for the ratio to be higher, either PL3 must be higher, PL1 must be lower, or both. In this study, PL3 was slightly higher and PL1 was slightly lower in fracture cases than controls, although these differences were not significant (Fig. 4). This indicates that fracture cases have proportionately more trabeculae aligned along the primary load axis (and thus proportionately fewer transverse trabeculae) than controls within a given cube, but it does not suggest in terms of absolute numbers that fracture cases as a group definitively have either more trabeculae along the primary axis or fewer along the transverse axis than controls.

The finding that the ratio of PL3/PL1 is distinct between groups while neither PL1 nor PL3 is not may seem contradictory, but can be explained by a more careful examination of the data within the EBMS. Note that although fracture cases have slightly higher bone volume than controls, the overall number of trabeculae is slightly reduced, with the higher bone volume being achieved by having somewhat thicker trabeculae (Fig. 3). The difference in mean PL3 values between groups was 2.5 times that of the difference in mean PL1 values (Fig. 4), suggesting that the slight reduction in trabecular number in the fracture group is a result of disproportionate losses of transverse trabeculae. Because the difference in PL3 values exerts a greater influence on the difference between the ratio of PL3/PL1between groups, we hypothesize that the relatively fewer transverse trabeculae leads to the difference in anisotropy between groups. Similar losses of transverse elements leading to an increased anisotropic structure occur in human vertebral bone with age.(27)

Altered three-dimensional structure would be expected to affect overall mechanical properties. The relatively fewer transverse trabeculae in the fracture group may lead to reduced cross-bracing and a greater propensity for buckling of trabeculae oriented along the load axis, as well as a reduced resistance to transverse loads. Regressions of modulus on both bone volume and trabecular number are significantly different between groups (Figs. 8, 9), suggesting that for a given overall measure of structure derived from analysis of the entire cube, the two groups will have a different modulus. While these different structure-mechanics relationships between fracture and control groups may be related to the difference in anisotropy between the two groups, there may be other factors which adversely affect the material properties in the fracture group, such as increased microdamage levels or differences in trabecular hard tissue properties. However, recent work by Mori et al.(28) showed no differences in microdamage levels in the femoral head between femoral neck fracture patients and age-matched controls. While several experimental approaches have been utilized to measure tissue properties and to examine aging effects,(29–32) we are not aware of any studies comparing tissue properties in fracture patients and controls.

While it cannot be discerned from the present study whether or not the altered structure of the proximal femur was indeed causative of fracture, one of the most intriguing questions raised is why these distinct structures exist. Assuming that bone adapts to its mechanical environment, we could hypothesize that either fracture patients have a different mechanical environment with fewer stimuli transverse to the primary load direction; or that the loading environment is the same but the mechanotransduction and/or bone adaptation response mechanisms are different (and perhaps flawed).

An altered mechanical environment in fracture patients implies altered loading about the hip, which could result from individual differences in anatomy or neuromuscular activity. Variations in the femoral neck angle, the location and angle of attachment of muscle insertions, muscle firing patterns, or muscular strength could affect the magnitude and direction of loads and hence bone's adapted structure.

A more plausible hypothesis is that the altered structure in patients with hip fractures results from differences in bone adaptation mechanisms. The decision of how to best distribute bone throughout the structure, particularly at critically low levels of bone mass, seems to differ between groups. In comparing regressions of maximum modulus on both bone volume fraction and trabecular number, for lower values of bone volume fraction and trabecular number, the regression lines indicate that the fracture cases will have a higher modulus than the controls (Figs. 8, 9). As bone mass approaches reduced levels, the fracture patients seem to preferentially preserve trabecular elements parallel to the primary load axis while allowing those transverse to the load axis to be resorbed. This adaptation results in mechanical properties being greater than those of controls in the direction of the primary load axis while compromising properties in the transverse directions. We hypothesize that this jeopardizes the overall structural integrity of the proximal femur and makes it more susceptible to fracture during a fall.

The techniques used in this study imposed some limitations which require further discussion. Since complete medical histories were not available for either fracture or control patients, both groups could include persons who may have had medical conditions and lifestyle habits that could affect bone structure, such as prolonged steroid use, smoking, or limited physical activity. While these factors may result in reduced bone mass compared to individuals of similar age, for this study, age was not the primary variable of interest. Because the focus of the study was to compare the bone of fracture cases and controls with comparable bone mass in an effort to structurally distinguish those who fracture from those who do not, we do not feel that these factors would have a significant effect on the interpretation of the data.

Another limitation concerns the location within the femoral head from which samples were taken. Most fracture specimens in this study came from patients who had a subcapital or femoral neck fracture. Previous studies suggest that trochanteric fractures represent a more severe form of osteoporosis and may have an etiology distinct from that of subcapital or femoral neck fractures. Patients with trochanteric fractures have greater bone loss and may experience more trabecular thinning as opposed to losses in trabecular number than those with other types of hip fracture.(10,19) This issue could not be addressed in the present study because specimens could not be obtained from patients who experienced intertrochanteric fractures. Intertrochanteric fractures and those which occur closer to the base of the femoral neck are typically treated with fracture fixation devices and are unlikely to result in the use of a prosthesis. Hence, no specimens from these patients would be available. Although the sample location chosen was more within the femoral head than the femoral neck, this location was the one closest to the fracture site from which a bone sample could be consistently obtained.

In summary, there were differences in overall cancellous bone architecture and mechanical properties of the proximal femur between women with hip fractures and controls. Even when matched for bone volume, fracture cases had a significantly more anisotropic three-dimensional structure than controls, with proportionately fewer trabecular elements transverse to the primary load axis. This altered structure may discriminate between patients who have a high risk of fracture and those with equal bone mass that are at less risk. Fracture and control groups demonstrated different relationships between mechanics and structure. Further studies are necessary to more rigorously examine how and to what extent the difference in structural anisotropy explains the different structure-mechanics relationships between groups, as well as other aspects of mechanical behavior.

Acknowledgements

We thank the St. Joseph Hospital Orthopaedic Surgery Department (Ypsilanti, MI) for their assistance in obtaining hip fracture specimens, as well as Dr. Larry Matthews, Dr. Robert Goulet, Dr. A. Bouffard, Dr. Kathe Derwin, Dr. Steve Elder, Barbara McCreadie, and Tiffany Powell for their contributions to this work. Supported by National Institutes of Health grants AR34399, and 2T32 AG00114.

Ancillary