1 INTRODUCTION

Anterior cruciate ligament (ACL) injury is the most frequent intra-articular injury of the knee and occurs at an incidence of 68.6 per 100,000.1, 2 Young individuals participating in cutting and pivoting sports are at highest risk of ACL injury, and injury rates are on the rise.2 The incidence of ACL injury increased by 74% in those under 25 years during a recent 15-year period.3 An unfortunate but common consequence of ACL injury is the development of posttraumatic knee osteoarthritis (PTOA) at an early age. ACL injury increases the risk of future osteoarthritis (OA) by over eightfold within 11 years of injury.4 Thus, the risk for total knee arthroplasty after ACL injury is 20 times greater during the third decade of life and 7.5 times greater during the fourth decade of life compared to the overall population.5 The rapidly increasing incidence of ACL injuries in young populations will likely result in a greater burden of PTOA in young adults in the years ahead. Mechanisms underlying the early development of PTOA after ACL injury are not well understood. Thus, interventions to prevent or delay cartilage breakdown after ACL injury do not exist.

Articular cartilage is avascular, and chondrocytes and the extracellular matrix of articular cartilage are thus dependent on repetitive, cyclic loading to promote tissue health. Dynamic joint loading results in stronger collagen and increased proteoglycan concentrations in cartilage.6 Meanwhile, inadequate joint loading results in cartilage that is thinner, softer and more susceptible to breakdown.7 T2 relaxation time is a quantitative magnetic resonance imaging (qMRI) marker that provides an early trajectory of joint health before macroscopic MRI or radiographic changes occur after ACL injury. T2 relaxation times can detect changes in the water content and morphologic changes in collagen organization of the articular cartilage.8, 9 Thus, increased T2 relaxation times in the injured limb may indicate the potential for asymmetric chondral degradation over time. Other techniques, such as Th1rho and dGMERIC, have been established for quantitative evaluation of cartilage; however, these are based primarily on assessing proteoglycan content. Abnormal walking patterns have been linked to increasing T2 relaxation times after ACL injury.10-15 Although preoperative changes in T2 relaxation time have been reported,16 limited evidence exists regarding the relationship between walking patterns and T2 relaxation time before ACL reconstruction (ACLR). Due to the precedent for T2 relaxation in relation to physical activity (PA), this was selected over other techniques.

Although individuals spend less time in PA roughly 2–3 years (mean: 27.8 ± 17.5 months) after ACLR compared to uninjured matched controls,17 our understanding of PA levels before ACLR is limited. It is also unknown if daily magnitudes of PA immediately after ACL injury are associated with change in T2 relaxation times in the cartilage of the injured knee. Because articular cartilage structure is dependent on repetitive and dynamic joint loading, an integrated approach using PA levels and walking biomechanics may provide an enhanced understanding of total daily knee joint loading. An understanding of the association between cumulative measures of joint loading (i.e., step counts and knee moments during gait) and measures of cartilage structure is needed to inform modifiable strategies to limit cartilage degeneration after ACL injury.

The purpose of this study was to determine if measures of knee joint loading (i.e., knee joint biomechanics during gait, PA levels) are associated with T2 relaxation times in the articular cartilage of the knee within 1 month of ACL injury. Because alterations in quantitative MRI markers of cartilage structure occur rapidly after ACL injury,16 the 1-month time point was chosen to identify immediate changes in T2 relaxation time within a clinically reasonable period to recruit and enroll participants to inform the timing of future preventative interventions. We hypothesized that lower measures of knee joint loading would be associated with longer (worse) T2 relaxation times. A secondary aim was to determine if interlimb differences in gait biomechanics and cartilage T2 relaxation times are present immediately after ACL injury. We hypothesized that in the injured knee lower joint angles and moments but longer T2 relaxation times would be present.

2 METHODS 2.1 Participants

Participants between 15 and 35 years of age were enrolled within 1 month of ACL injury before ACLR for this ongoing, prospective cohort study (Level 2 evidence). Older individuals were excluded due to higher risk of baseline cartilage degeneration. Exclusion criteria included a previous injury or surgery to either knee, concomitant Grade III tear to other knee ligaments, meniscus tear with anticipated meniscectomy by the treating orthopaedic surgeon, acute chondral lesions or degenerative cartilage changes identified on postinjury MRI, or open growth plates requiring altered ACLR technique (i.e., physeal-sparing). Additional exclusion criteria included history of inflammatory disease, immune compromise, chronic use of nonsteroidal antiinflammatory drugs, history of cortisone injection during the prior 3 months, current pregnancy, or contraindications to MRI. This study was approved by the Institutional Review Board at the University of Nebraska Medical Center. All participants provided written informed consent.

2.2 Self-reported participant characteristics

Participants reported age, sex, race, and preinjury cutting and pivoting activity level (Level 1: soccer, basketball, etc.; Level 2: tennis, baseball, etc.18, 19) in surveys within the REDCap electronic data capture tools hosted at the University of Nebraska Medical Center.20 Height was measured using a portable stadiometer with shoes off.

2.3 MRI acquisition and T2 relaxation time

Participants sat for 30 min before MRI acquisition to unload knee cartilage due to acute effects of loading on T2 relaxation time.21 Each MRI scan began between 4:15–5:45 p.m. to control for the effect of daily activity on qMRI markers.22 Bilateral MR image data (injured knee first) were acquired on a 3-Tesla Phillips Ingenia MRI scanner using a 16 channel transmit/receive knee coil (Phillips North America Corporation) in slight knee flexion and neutral rotation. For T2 mapping, a spin echo (SE) sequence with multiple echoes (MSE) was acquired with these parameters: TR = 2700 msec; 10 echoes with the echo times TEi = i × 10 msec (i = 1, …, 10); FOV: 120 × 120 mm; acquisition matrix = 252 × 250; slice thickness = 3.0 mm; slice gap = 0.5 mm; range of slices = 23–31; pixel size = 0.3125 × 0.3125 mm; echo train length = 10; number of averages = 1. In addition to MSE, fat suppressed proton density weighted SE sequence in axial, coronal, and sagittal orientations and a sagittal T1 weighted SE were also included in the MRI protocol.

Multi-echo MRI data at each pixel were fit to the signal equation to generate T2 maps using Levenberg-Marquardt nonlinear least squares algorithm (Si = the signal at echo time TEi, and S0 = signal at TE = 0) within Interactive Data Language (Harris Geospatial Solutions Inc.). First echo data were not used in the fitting to minimize the errors due to stimulated echoes.23

Manual cartilage segmentation was completed in ITK-SNAP software24 on reference images corresponding to TE = 40 msec from MSE data used for generating T2 maps for which we have demonstrated reliability (intrarater intraclass correlation coefficients [ICC] [n = 12]: femoral: 0.759; tibial: 0.775; interrater ICC [n = 12]: femoral: 0.949; tibial: 0.930). Before uninjured segmentation, a manual and affine registration technique was used to register the injured reference images to the uninjured knee using 3-D Slicer software.25 The goal of this registration procedure was to provide an initial segmentation mask for the uninjured knee cartilage to reduce processing time. The combined registration was applied to the injured segmentation mask, overlaid on the uninjured reference images, and manually adjusted to anatomically match uninjured knee cartilage.

Lateral and medial compartments were defined using the center of the intercondylar notch for both the femur (LFC and MFC, respectively) and the tibia (LTC and MTC, respectively). Femoral and tibial cartilage in each compartment (LFC, MFC, LTC, MTC) were further divided into anterior, weightbearing, and posterior regions according to the location of the meniscus horns in the sagittal plane (Figure 1) using axial MR images to verify meniscus horn location. Anterior and posterior tibial cartilage was not analyzed due to few pixels in these regions. The patellar cartilage comprised a single region. A board-certified, fellowship-trained musculoskeletal radiologist confirmed accuracy of segmentation masks and region of interest (ROI) boundaries. Cartilage masks were overlaid on T2 maps to extract mean T2 relaxation time within each ROI (six femoral, two tibial, and one patellar). Pixels with T2 relaxation times less than 10 ms or more than 90 ms were excluded to remove outliers due to fitting errors.26 A T2 relaxation time interlimb ratio (ILR) was calculated in each ROI (T2 ILR = injured limb/uninjured limb).13 Thus, an ILR more than 1.00 indicates longer T2 relaxation time in the injured compared to uninjured knee.

The femoral and tibial cartilage segmentation masks in each compartment (lateral compartment pictured above) were divided into anterior, weightbearing, and posterior regions as defined by the location of the meniscus horns in the sagittal plane. The anterior and posterior (not pictured) tibial cartilage was not used in analysis. Thus, analyses included three femoral regions (anterior, weightbearing and posterior) in each compartment (lateral and tibial) and a single weightbearing region in each compartment (lateral and tibial). The patellar cartilage comprised a single region. LFC, lateral femoral condyle; LTC, lateral tibial condyle [Color figure can be viewed at wileyonlinelibrary.com] 2.4 Gait biomechanics

Three-dimensional motion capture data were collected using an 8-camera system (Qualysis AB) sampled at 120 Hz and two embedded force plates (Bertec Corporation) sampled at 1080 Hz. Passive, 14-mm retroreflective markers were placed on skeletal landmarks of the trunk, pelvis and lower extremities (Figure 2). Rigid shells each with four markers were placed at the lateral shanks and thighs.

Individual markers (represented by green circles) were placed on bony landmarks of the trunk and lower extremities with rigid shells of markers placed at the thighs and shanks. Anterior view is on the left. Posterior view is on the right. Images generated in Visual 3D software [Color figure can be viewed at wileyonlinelibrary.com]

Participants stood in anatomical position for a 1-s static trial. Markers at the first metatarsal heads, malleoli, femoral epicondyles, and anterior superior iliac spines (ASIS) were removed before gait trials. Participants completed five gait trials with valid kinematic and kinetic data on each limb at a self-selected, comfortable walking speed. Average gait speed was calculated along a 5.4 m walkway during the first three gait trials and maintained within 5% for all remaining trials.

Labeled marker data were exported to Visual 3D software (C-Motion, Inc.) for custom data postprocessing. Target and ground reaction force (GRF) data were low-pass filtered using a fourth-order bidirectional Butterworth filter with a cutoff frequency of 6 Hz. A cutoff frequency of 6 Hz was chosen after completing residual analysis of kinetic data as described by Winter.27 Briefly, residuals were calculated for cutoff frequencies from 0.1 to 49.9 Hz at increments of 0.1 Hz using gait trials from the first ten participants in this study. Residuals were normalized to the maximum residual value. The linear section of high frequency residuals was defined as the collection of points where the values of the residuals' second discrete time derivative were below 0.0001. Using linear regression, the cutoff frequency was equal to the y-intercept of the linear regression line. The average cutoff frequency across both limbs was 4.5 ± 0.5 Hz. To account for two SDs of variance, a cutoff frequency of 6 Hz was chosen.

A subject-specific model was created using height (stadiometer) and mass (static trial) to determine segment lengths and joint centers. Virtual markers at bony landmarks were offset 9 mm toward the bone to account for half of the 14-mm marker and the 2-mm base.28 The ankle and knee joint centers were defined as the mid-point of the virtual medial and lateral malleoli and virtual medial and lateral femoral epicondyles, respectively. A Visual 3D composite pelvis was built from virtual ASIS and PSIS landmarks. The hip joint center was defined using estimates described by Bell and colleagues.29, 30 Knee joint moments were calculated using an inverse dynamics approach.27 The beginning and end of stance phase was determined using a 10 N threshold of the GRF. Variables of interest included the knee flexion angle (KFA) at initial contact, peak knee flexion angle (pKFA) and moment (pKFM) during loading response, and peak knee adduction moment (pKAM) during the first 50% of stance phase. Joint moments are reported as external moments. Knee excursion (kEXC) during loading response was defined by the difference in KFA from initial contact to pKFA. The impulse of the external KFM and KAM over the entire stance phase were calculated using the trapezoidal rule. External knee joint moments were normalized to mass (kilograms) and height (meters). Positive joint angles represent knee flexion. Positive joint moments represent knee flexion and adduction, respectively.

2.5 Physical activity

PA was measured using a 3-axis accelerometer (wGT3X-BT; Actigraph Corporation) sampled at 100 Hz. This accelerometer reliably measures step counts across varying gait speeds.31 Participants wore the accelerometer on the right iliac crest for 7 days beginning the day after MRI and biomechanics testing during all waking hours except when in water. Data were processed within Actilife 6 software (Actigraph Corporation).32 Activity counts, which represent the weighted sum of the number of accelerations, were calculated for each 1-min interval to identify wear periods and calculate PA levels. A valid week of data required 4 days with at least 10 h of wear to provide a reliable estimate of PA behavior.32-34 Nonwear periods were defined as intervals of at least 90 min with activity counts equal to zero with no more than 2 min of activity counts between 1 and 99.33 The variable of interest was mean steps per day.

2.6 Statistical analysis

Nominal data were described using counts and proportions. Continuous data were described using means, SDs, and 95% confidence intervals. Paired t tests were used to determine if gait biomechanics and cartilage T2 relaxation times differed between the injured and uninjured knee.

Hierarchical multiple regression models were used to determine the association between knee joint loading predictors (daily step counts, kEXC, and KAM impulse) with the outcome of T2 relaxation time ILR in each ROI. kEXC and KAM impulse parameters were defined as interlimb differences (injured minus uninjured). Daily step counts and kEXC were chosen because they represent global joint loading measures. KFM variables were not included because it demonstrated high collinearity with kEXC (peak KFM: r = .711; KFM impulse: r = .590). KAM impulse was chosen because it represents the relative joint loading balance in the frontal plane (i.e., relative loading between the medial versus lateral tibiofemoral compartment) throughout stance phase.35 Age, sex (female = 0; male = 1), and the presence of meniscus injury (medial meniscus for medial compartment, lateral meniscus for lateral compartment, and any meniscus injury for patellar cartilage analyses) (n = 0; yes = 1) were entered as covariates in the first block, followed by the knee joint loading predictors in the second block. Age and concomitant meniscus injury were included as covariates because they increase the odds for developing knee OA earlier after ACL injury.36, 37 Multiple regression assumptions of independent observations, individual predictor linearity, collective predictor linearity and homoscedasticity of residuals, and residual normality were tested using the Durbin-Watson statistic, partial regression plots, scatterplots of unstandardized predicted values versus studentized residuals, and histograms and P-P plots, respectively. No outliers (standardized residual >3 SDs) were identified. A p value of less than .05 was set a priori.

3 RESULTS

Descriptive statistics for age, race, mass, height, and sex are presented in Table 1. All but two participants participated in Level 1 cutting and pivoting activities (e.g., soccer, basketball) before ACL injury.18, 19 Participants were enrolled and completed MRI and biomechanical testing at an average of 25 days after injury (Table 1). Approximately one-quarter of participants had a concomitant medial meniscus tear and nearly one-half had lateral meniscus injury (Table 1). Accelerometer wear, daily steps counts, and gait speed during biomechanics testing are presented in Table 2.

Table 1. Participant characteristics and concomitant meniscus injury is presented for all 27 participants Variable Mean (SD) or count (%) 95% CI Age (years) 19.8 (5.0) 17.8–21.8 Race Asian 2 (7.4) Black or African American 2 (7.4) Hispanic, Latino or Spanish 3 (11.1) White 20 (74.1) Mass (kg) 74.4 (15.5) 68.3– 80.5 Height (m) 1.70 (0.09) 1.67–1.74 Sex (female) 15 (55.6) Preinjury activity level (Level 1)18, 19 25 (92.6) Time From ACL injury (days) 24.6 (4.7) 22.8–26.5 Medial meniscus tear (Yes) 7 (25.9) Lateral meniscus tear (Yes) 12 (44.4) Medial or lateral meniscus tear (Yes) 14 (51.9) Note: Mean (SD) and 95% confidence intervals provided for continuous data. Counts (percentage) provided for categorical data. % = Percentage. Abbreviations: CI, confidence interval; kg, kilograms; m, meters. Table 2. Accelerometer wear, daily steps counts, and gait speed during biomechanics testing is presented for all 27 participants Variable Mean (SD) 95% CI Accelerometer wear (days) 5.6 (1.3) 5.1–6.2 Daily Accelerometer wear (min) 878.7 (145.7) 821.0–936.3 Daily step count 6274.6 (2500.7) 5285.4–7263.9 Gait Speed (m/s) 1.38 (0.21) 1.29–1.46 Note: Mean (SD) and 95% confidence intervals are provided. Abbreviations: CI, confidence interval; m, meter; min, minutes; sec, second. 3.1 Gait biomechanics and T2 relaxation time

Compared to the uninjured limb, the injured limb demonstrated approximately 3° greater pKFA during loading response of gait (Table 3). However, this was accompanied by an average of 6° more knee flexion at initial contact resulting in less kEXC (Figure 3). A lower pKAM and KAM impulse was observed in the injured limb (Table 3 and Figure 4). However, no interlimb differences were present for sagittal plane knee moments. There were no significant interlimb differences in T2 relaxation times within any cartilage ROI (Table 4).

Table 3. Knee flexion angle and sagittal and frontal plane joint moments during gait is presented for the injured and uninjured limb for all 27 participants Injured Uninjured Difference 95% CI p KFA at IC (°) 6.7 (5.8) 0.2 (4.2) 6.4 4.4– 8.5 <.001 pKFA (°) 21.0 (5.9) 18.4 (6.5) 2.6 0.5–4.7 .019 kEXC (°) 14.3 (3.7) 18.2 (4.4) −3.9 −5.5–−2.2 <.001 pKFM (N·m/kg·m) 0.36 (0.14) 0.40 (0.18) −0.04 −0.10–0.02 .224 KFM impulse (N·m·s/kg·m) 0.056 (0.022) 0.057 (0.028) −0.001 −0.010–0.007 .769 pKAM (N·m/kg·m) 0.21 (0.09) 0.28 (0.10) −0.07 −0.12–−0.03 .003 KAM impulse (N·m·s/kg·m) 0.071 (0.043) 0.097 (0.044) −0.026 −0.046–−0.006 .012 Note: Values in parentheses are SDs. The interlimb difference is presented with its 95% confidence interval. Boldface numbers indicate statistical significance (p values of <.05). ° = degrees. Abbreviations: CI, confidence interval; IC, initial contact; KAM, knee adduction moment; kEXC, knee flexion angle excursion; KFA, knee flexion angle; kg, kilogram; KFM, knee flexion moment; m, meter; N, newton; p, p value; pKAM, peak knee adduction moment; pKFA, peak knee flexion angle; s, second. Mean knee flexion angle for all 27 participants during stance phase of gait. Participants walked with a statistically significant (asterisks) greater knee flexion angle at initial contact and at the end of loading response (~25% of stance phase) but less knee excursion in the injured limb compared to uninjured limb. The shaded regions represent ±1 SD of mean knee flexion angle at each percentage of stance phase. °, degrees; %, percentage [Color figure can be viewed at wileyonlinelibrary.com] Mean external knee adduction moment for all 27 participants during stance phase of gait. Participants walked with a statistically significant (asterisks) smaller peak knee adduction moment during the first 50% of stance phase and a smaller knee adduction moment impulse over all of stance phase compared to uninjured limb. The shaded regions represent ±1 SD of the knee adduction moment at each percentage of stance phase. N, newton; m, meter; kg, kilogram; %, percentage [Color figure can be viewed at wileyonlinelibrary.com] Table 4. Mean T2 relaxation time (milliseconds) in the cartilage of each region of interest is presented for all 27 participants Injured Uninjured ILR 95% CI p LFC—anterior 46.8 (2.9) 47.1 (2.7) 0.99 0.97–1.02 .624 LFC—weightbearing 46.9 (3.3) 47.5 (4.3) 0.99 0.96–1.03 .484 LFC—posterior 42.9 (4.4) 42.7 (5.2) 1.01 0.96–1.06 .897 LTC—weightbearing 41.7 (3.7) 42.6 (4.6) 0.99 0.94–1.04 .446 MFC—anterior 46.9 (3.7) 46.4 (3.6) 1.02 0.97–1.06 .608 MFC—weightbearing 46.7 (4.0) 46.9 (3.6) 1.00 0.96–1.04 .826 MFC—posterior 42.3 (5.6) 40.8 (4.5) 1.05 0.98–1.12 .300 MTC—weightbearing 43.0 (4.1) 42.3 (3.7) 1.02 0.97– 1.08 .611 Patella 39.0 (2.6) 39.4 (3.1) 0.99 0.97–1.02 .468 Note: Values in parentheses are SDs. The interlimb ratio (ILR) is presented with its 95% confidence interval. Abbreviations: CI, confidence interval; LFC, lateral femoral condyle; LTC, lateral tibial condyle; MFC, medial femoral condyle; MTC, medial tibial condyle; p, p value. 3.2 Lateral tibiofemoral cartilage

After controlling for age, sex, and concomitant lateral meniscus injury in the full regression models, daily step counts, kEXC, and KAM impulse accounted for an additional 35.8% of the variability in T2 relaxation time ILR's in the weightbearing LFC cartilage, 44.8% in the posterior LFC cartilage, and 64.5% in the weightbearing LTC cartilage (Table 5). KAM impulse was the only significant factor of T2 relaxation times in the weightbearing cartilage of the LFC (β = .713; p = .001). KAM impulse also most strongly associated with T2 relaxation times in the posterior cartilage of the LFC (β = .799; p < .001) and the weightbearing cartilage of the LTC (β = .956; p < .001). KAM impulse always positively associated with T2 relaxation times in the lateral compartment, indicating that asymmetrically lower KAM impulse in the injured knee was associated with asymmetrically shorter T2 relaxation time in the injured knee (ILR <1.00). Higher step counts (posterior LFC: β = .478; p = .035; weightbearing LTC: β = .371; p = .025) and less kEXC (posterior LFC: β = −.481; p = .026; weightbearing LTC: β = −.403; p = .012) also associated with longer T2 relaxation in the posterior LFC and weightbearing LTC of the injured knee.

Table 5. Results of linear regression models with daily physical activity (step counts), interlimb difference in knee flexion angle excursion, and interlimb difference in knee adduction moment impulse as independent variables and interlimb ratios in T2 relaxation times in the lateral tibiofemoral regions of interest as the outcome of interest, after adjusting for age, sex, and concomitant meniscus injury Cartilage region R2 R2 change p Factor Unstandardized B β p LFC—anterior 0.265 0.246 .116 Age <0.001 .033 .877 Sex −0.008 −.063 .754 Meniscus injury −0.029 −.239 .255 Daily PA <0.001 .159 .550 kEXC −0.001 −.045 .859 KAM impulse 0.670 .554 .026 LFC—WB 0.507 0.358 .011 Age −0.003 −.156 .375 Sex −0.018 −.101 .540 Meniscus injury 0.032 .183 .286