Diffusion tensor imaging of articular cartilage at 3T correlates with histology and biomechanics in a mechanical injury model

Abstract

Purpose:

We establish a mechanical injury model for articular cartilage to assess the sensitivity of diffusion tensor imaging (DTI) in detecting cartilage damage early in time. Mechanical injury provides a more realistic model of cartilage degradation compared with commonly-used enzymatic degradation.

Methods:

Nine cartilage-on-bone samples were obtained from patients undergoing knee replacement. 3T DTI (0.18×0.18×1 mm³) was performed before, one week and two weeks after (control, mild and severe) injury, with a clinical Radial Spin-Echo DTI (RAISED) sequence used in our hospital. We performed stress-relaxation tests and used a quasilinear-viscoelastic (QLV) model to characterize cartilage mechanical properties. Serial histology sections were dyed with Safranin-O and given an OARSI grade. We then correlated the changes in DTI parameters with the changes in QLV-parameters and OARSI grades.

Results:

After severe injury the mean diffusivity increased after one and two weeks, whereas the fractional anisotropy decreased after two weeks (p<0.05). The QLV-parameters and OARSI grades of the severe injury group differed from the baseline with statistical significance. The changes in mean diffusivity across all the samples correlated with the changes in the OARSI grade (r=0.72) and QLV-parameters (r=−0.75).

Conclusion:

DTI is sensitive in tracking early changes after mechanical injury, and its changes correlate with changes in biomechanics and histology.

Introduction

Post-traumatic osteoarthritis (PTOA) arises as a long-term consequence of a joint injury, such as anterior cruciate ligament (ACL) rupture. During non-contact ACL rupture, high compression and shear forces act on the articular cartilage with devastating consequences. Hours after ACL rupture, high concentrations of proteoglycan (PG) and type II collagen fragments are found in the synovial fluid^1,2,3,4. Although the concentration of PG and collagen fragments decreases with time after injury, for years afterwards it remains significantly higher than in asymptomatic controls^1,2. The synovial fluid of ACL rupture patients has also shown evidence of increased PG synthesis⁵. This leads to an early transient increase in the total PG content⁴, which has also been observed in longitudinal MRI studies^6,7,8. The degradation of type II collagen in humans, however, has shown a persistent increase in collagenase cleavage and denaturation over a period of years^3,4. There is histological evidence that the degeneration of type II collagen occurs during a transient increase in the total PG content⁴. Thus, the assessment of the injured cartilage requires an assessment of both components of the cartilage matrix, namely, the PG content and the integrity of the collagen network.

Diffusion tensor imaging (DTI) has emerged as an imaging biomarker of cartilage integrity that can assess both PG content and collagen architecture, thus providing a new way of tracing cartilage degradation following knee injury. Since the first DTI study showed the potential of DTI to track collagen architecture of articular cartilage⁹ many studies have focused on understanding the meaning of diffusion measurements in cartilage^{10,11,12,13,14,15}. The current interpretation of DTI metrics is that the changes in the level of PG modulates the mean diffusivity (MD) while the collagen structure modulates the fractional anisotropy (FA)^{10,11,12,13,14,15,16}. Ex vivo experiments have shown that DTI detects the earliest signs of OA observed in histology. One recent study classified 43 samples with 95% accuracy¹², while one of the first clinical studies has similarly shown DTI to provide high accuracy (92%) in the early diagnosis of OA^17,18; this is more accurate than T₂ relaxation time (60%)¹⁷ and Na imaging (78%)¹⁹.

Our aim is to investigate whether DTI biomarkers are sensitive to changes occurring in articular cartilage as a consequence of knee injury. Here we implement an injury model that mimics the injury mechanism during ACL rupture, in which controlled mechanical overloading is applied to the cartilage samples in order to produce either a mild or a severe level of injury, as required ^20,21. Thereafter, cartilage degradation occurs as a consequence of the mechanical damage and the biological response triggered by the injury. To test the ability of DTI to monitor these early changes we acquired MRI, histology and biomechanics measurements before and two weeks after injury. MRI was performed on a clinical MRI scanner using an optimized knee DTI sequence that is normally used in vivo (see subsection Methods:MRI).

Our objectives were:

• to show the feasibility of the experimental setup which includes MRI in whole-body 3T scanners, biomechanics and injury of cartilage;

• to characterize the changes in DTI, histology and biomechanics before and after injury;

• to quantify the correlation between the changes in biomechanics and histology with the changes in DTI indices (MD and FA).

Methods

Sample Preparation and Handling

Knee cartilage samples were drawn from cartilage obtained from two patients (55 and 42 years-old) who underwent knee replacement surgery at the New York University Hospital for Joint Diseases. The institutional review board of NYU School of Medicine approved this study and informed consent was obtained. In the cartilage we identified macroscopically intact areas that had a smooth, shiny and white surface. We drilled nine 7-mm-diameter cylindrical cartilage-on-bone samples perpendicularly to the articular surface. The cartilage layer was trimmed using a specifically made Plexiglas hole puncher of 4-mm inner diameter, and a ceramic scalpel so as to avoid any metal debris causing MRI artifacts. The trimming ensures that no cartilage injury was caused by the sharp bone edges. The remainder of the trimmed cartilage was used for baseline histology analysis (see subsection Methods: Histology). During the 2-week period, the samples were incubated at 37 °C on a medium with antibiotic, Dubbelcco's Modified Eagle Medium, DMEM (Gibco, 11965), to prevent bacterial infection.

The Experimental Design

In this experiment we use a model for traumatic injury where macroscopically intact cartilage samples are mechanically overloaded. We performed DTI measurements at three time-points, as shown in Fig.1: prior to injury, and then one and two weeks after injury; we refer to these as ‘week 0’, ‘week 1’ and ‘week 2’. Having histology and mechanical tests performed at week 0 and 2 (but not at week 1, to minimize the risk of contamination during biomechanical tests), we compared the DTI measurements with the histology and the mechanical properties.

Figure 1:

After harvesting, the cartilage samples were incubated at 37 °C for at least two days to allow the drilling trauma to settle. Week 0 measurements included histology (from the shaved cartilage preparation), MRI and, within six hours of MRI, mechanical testing. This step is followed by the mechanical injury. The nine samples were split into three equally-sized groups, undergoing no/mild/severe injury. After injury the samples were incubated for exactly one week, after which a second MRI measurement was taken. The last measurements at week 2 involved a third MRI measurement, a second mechanical test, and histology.

After harvesting, the cartilage samples were incubated at 37 °C for at least two days to allow the drilling trauma to settle. Week 0 measurements included histology (from the shaved cartilage preparation; see subsection Methods: Sample Preparation and Handling), MRI and, within six hours of MRI, mechanical testing. This step is followed by the mechanical injury. The nine samples were split into three equally sized groups, undergoing zero, mild, or severe injury. After injury the samples were incubated for exactly one week, after which a second MRI measurement was taken. The last measurements at week 2 involved a third MRI measurement, a second mechanical test, and histology.

MRI

All samples were imaged with a 3T scanner (Magnetom Skyra, Siemens Healthcare, Erlangen, Germany), with a maximum gradient strength of 45 mT/m, using a specially built 4 cm butterfly coil²². We acquired a high resolution T₂-weighted image for cartilage thickness measurements (repetition time/echo time = 7680/70 ms, in-plane resolution = 0.18×0.18 mm², slice thickness = 1.3 mm). DT images were acquired using an optimized knee DTI protocol, the radial spin-echo diffusion tensor imaging (RAISED) sequence²³, which includes a 2D phase navigator for motion correction, with TR/TE = 1500/49 ms, in-plane resolution = 0.18×0.18 mm², slice thickness = 1.2 mm, bandwidth = 300 Hz/pixel, 360 radial views per diffusion-weighted image (spokes). This b = 300 s/mm² sequence had Δ = 19.0 ms and δ = 14.5 ms. The MRI protocol consisted of a set, repeated once more, of one unweighted (b=0) and six diffusion-weighted measurements applied along six spatially optimized gradient directions calculated with the downhill simplex minimization (DSM) method²⁴, thus giving 14 measurements in total. The expected isotropic attenuation factor is e^−bD~0.63 for b=300 s/mm² and D=1.5×10⁻³ mm²/s. The total acquisition time was 2 h. We used external marks on the samples to guarantee a reproducible positioning across the scanning sessions. Outside the incubator, and during the scan, the cartilage samples were immersed in DMEM (see also Methods: Sample Preparation). To avoid the formation of air bubbles the samples were kept continuously immersed in DMEM fluid, therefore minimizing contact with air. The cartilage plugs holder, which is placed inside the DMEM-filled container, has a hole drilled in the middle of its top, so as to let any trapped air to escape. To reach thermal stability, 2-3 hours before scanning the samples were placed in the scanner room.

Image Processing

The diffusion tensor was calculated using an in-house (but relatively simple and standard) Matlab routine (MathWorks Inc, Natick, MA). Instead of the conventional 3D diffusion tensor, in each voxel we fit a cylindrically symmetric tensor²⁵, characterized by two eigenvalues (λ₁ and λ₂ = λ₃) and their corresponding eigenvectors. These eigenvalues produce two important indices: the mean diffusivity and the fractional anisotropy²⁶, expressed respectively as:

$$\begin{gathered} \text{MD}=({\lambda }_1+2{\lambda }_2)∕3 \\ FA=(3∕2)[({\lambda }_1-MD{)}^2+2({\lambda }_2-MD{)}^2]∕[({\lambda }_1^2+2{\lambda }_2^2)] \end{gathered}$$ (1)

We segmented the cartilage on the b=0 images using in-house software²⁷, and automatically split the cartilage into two layers, often referred to as “superficial” and “deep”, each occupying one-half of the thickness. We calculated summary statistics, the mean and the standard deviation, of the MD and FA for the bulk cartilage, for each cartilage layer and for the DMEM fluid outside the cartilage. The confidence intervals in the estimated indices were derived from bootstrap sampling²⁸.

Biomechanics

Incremental non-confined stress-relaxation testing of the articular cartilage was carried out on fully hydrated and equilibrated samples, using a BOSE Electroforce 3200 instrument (BOSE, Minneapolis, MN, USA). In the test, using a displacement-controlled ramp function, the cartilage tissue was compressed by advancing the actuator, at a loading rate of γ=5 μm/s, at 5% of the specimen's thickness in each of the four steps, with a relaxation period of 150 s between the steps, and thus achieving a total cartilage compression of 20% of cartilage thickness. Throughout the experiment the specimens were surrounded by gauze that was soaked in antibiotic medium DMEM (see subsection Methods: Sample Preparation and Handling).

The cartilage overloading was performed at three levels: three samples were given no injury (0 N compression), three were given a mild injury (120 N) and three were given a severe injury (190 N) using a strain rate of 0.07 s⁻¹ (gel diffusion rate)²⁹, imparting the injury load in 2 s.

We fitted a quasi-linear viscoelastic (QSV) model to the biomechanical injury data measurements. The QSV model accounts for the nonlinear poro-viscoelastic behavior of articular cartilage, using an exponential stress-strain model to represent the constitutive equations^30,31

$${\sigma }_{\mathrm{e}}=\mathrm{A}\left({\mathrm{e}}^{\mathrm{B}\epsilon (\mathrm{t})}-1\right),$$ [1]

where σ_e is the elastic response, A is the elastic stress constant, B is the elastic power constant and ε(t) is the applied strain. By definition σ_e is the tensile stress instantaneously generated when a step function of ε(t) is imposed on the tissue. Due to the poro-viscoelastic properties of cartilage, the stress measured, σ, is not exactly σ_e, but σ_e modulated by the transient response of the cartilage. In the QSV approximation this modulation is modeled as the convolution of dσ_e/dt with a propagator G(t) called the relaxation function,

$$\mathrm{G}(\mathrm{t}\mid \mathrm{C},{\tau }_1,{\tau }_2)=\frac{1+\mathrm{C}\left[{\mathrm{E}}_1\left(\frac{\mathrm{t}}{{\tau }_2}\right)-{\mathrm{E}}_1\left(\frac{\mathrm{t}}{{\tau }_1}\right)\right]}{1+\mathrm{C}\text{ln}\left(\frac{{\tau }_2}{{\tau }_1}\right)},$$ [2]

where C is a dimensionless constant of stress relaxation, τ₁ and τ₂ are relaxation times and E₁ indicates the exponential integral function. In our case the strain function was ε(t)=ε₀+γt, for t between 0 and the ramp up time t_r, with γ=5 μm/s and ε= ε₀+ γt_r for t>tr. The strain in each step was Δ ε= γt_r. The stress measured during t>t_r is then,

$$\begin{gathered} \sigma (\mathrm{t})={\int }_0^{{\mathrm{t}}_{\mathrm{r}}}\text{dz}\mathrm{G}(\mathrm{t}-\mathrm{z}\mid \mathrm{C},{\tau }_1,{\tau }_2)\frac{\mathrm{d}{\sigma }_{\mathrm{e}}(\epsilon )}{\mathrm{d}\epsilon }\frac{\mathrm{d}\epsilon }{\text{dz}}=\frac{\mathrm{A}}{1+\mathrm{C}\text{ln}\left(\frac{{\tau }_2}{{\tau }_1}\right)}\left({\mathrm{e}}^{\mathrm{B}\Delta \epsilon }-1+\text{CQ}(\mathrm{t})\right) \\ \mathrm{Q}(\mathrm{t})=\mathrm{B}\gamma {\int }_0^{{\mathrm{t}}_{\mathrm{r}}}{\mathrm{e}}^{\mathrm{B}\epsilon (\mathrm{t}-\mathrm{z})}\left[{\mathrm{E}}_1\left(\frac{\mathrm{z}}{{\tau }_2}\right)-{\mathrm{E}}_1\left(\frac{\mathrm{z}}{{\tau }_1}\right)\right]. \end{gathered}$$ [3]

For the fitting of the QSV model we first used a grid search based on initial parameter estimates as found in literature (A=17, B=5, C=4, τ₁=35, τ₂=63)³¹. The local minimum in this grid was then used as an initial guess for the optimization that uses the simplex minimization algorithm implemented in MATLAB’s “fminsearch” function. This fitting provided stable parameters; to further test the stability in the parameter estimation, we repeated the fitting across bootstrap datasets²⁸ (these datasets are constructed by randomly subsampling on the original dataset).

Histology

Sample workup for histology involved standard preparation procedures, including formalin fixation (4%), decalcification in 7% ethylenediaminetetraacetic acid (EDTA), dehydration in ascending ethanol series, paraffin embedding, and 5-μm sectioning (Schlittenmikrotom, Jung AG, Heidelberg, Germany). Tissue samples were serial sectioned and stained with safranin-O. Eight non-consecutive serial sections from each sample were given an OARSI grade by two independent readers, with any discrepancies solved by mutual agreement. To test the intra-reader variability, the two readers performed a second OARSI scoring in a subset of 24 slides containing 105 histological cuts. For each sample the average OARSI grade over the 8 scored sections was used.

Comparative Analysis

We analyzed the relationships between the DTI measurements, the histology (cartilage composition and integrity), and the cartilage mechanical properties. In particular, we examined the correlations of the two-week changes in DTI indices (MD and FA) with the biomechanical parameters and the histology grade; we also analyzed the correlation between the histology and the biomechanical parameters. In this study these correlations between differences in parameters across two weeks are simply referred to as "correlations"; our occasional use of the "pooled correlation" refers to the measurements for week 0 and week 2 as being treated independently, and then grouped together.

We chose to report DTI measurements for two cartilage layers, rather than one or three. Due to the challenges of harvesting the cartilage from human knees, the thin samples varied greatly in their thickness and shape. We judged that producing a single measurement for one layer gives too blunt an index for the cartilage. On the other hand, due to the limited number of samples in this experiment and the limited thickness, three layers would introduce high variance, making less clear the message of this study.

Statistical Methods

With 9 samples in total, and 3 samples per group, we can show an effect size of 2.8 sigma and have 80% power to detect correlations that are 0.67 or higher. The correlations of the DTI indices with the histology and biomechanical parameters were calculated using the Pearson’s correlation coefficient r. The partial correlations were used for the pooled correlations. The significance of the correlations was assessed via Student’s t-test. Inter- and intra-reader agreement for OARSI scoring was assessed with the weighted Cohen’s kappa statistic. The differences between groups were assessed using the paired t-test if the differences were investigated on the same samples at different time points, and the unpaired t-test for differences between the groups of samples. We used the t-test after ensuring that the data was normally distributed using the Kolmogorov-Smirnov test. A comparison of the histology grades was done using Wilcoxon rank tests, comparing grades at week 0 and week 2. One-way ANOVA and Kruskal-Wallis tests were used to test for multiple group comparisons with Bonferroni correction. We assume a p-value below 0.05 as indicating statistical significance.

Results

Figure 2 shows DTI maps at week 0 and week 2. The values at baseline and two weeks of MD and FA for each group are given in Table 1.

Figure 2:

DTI indices, mean diffusivity (MD, on the left) and fractional anisotropy (FA, on the right) for four representative samples. The MD is in μm²/ms units, and FA varies from 0 to 1. (In each cartilage the surface points towards the centre.)

Table 1:

Statistics for MD, FA (bulk, superficial and deep layer), OARSI grade, and the QLV-parameters across all cartilage injury classes. Boldface indicates 2 weeks’ values being statistically significantly different from baseline (week 0, p<0.05). MD has units of μm²/s, A has MPa, τ1 and τ2 have units of s. The numbers represent the mean and, in brackets, the standard deviation.

	Week 0			Week 2
	no injury	mild injury	severe injury	no injury	mild injury	severe injury
MD bulk	1.33 (0.02)	1.28 (0.05)	1.36 (0.04)	1.35 (0.09)	1.36 (0.04)	1.52 (0.08)
MD superficial	1.36 (0.02)	1.33 (0.07)	1.41 (0.03)	1.42 (0.05)	1.42 (0.07)	1.57 (0.07)
MD deep	1.30 (0.05)	1.23 (0.06)	1.31 (0.08)	1.29 (0.15)	1.30 (0.04)	1.46 (0.14)
FA bulk	0.25 (0.03)	0.29 (0.03)	0.25 (0.02)	0.30 (0.04)	0.27 (0.09)	0.19 (0.01)
FA superficial	0.24 (0.05)	0.29 (0.03)	0.25 (0.03)	0.30 (0.06)	0.26 (0.09)	0.18 (0.04)
FA deep	0.26 (0.01)	0.29 (0.03)	0.24 (0.04)	0.31 (0.06)	0.28 (0.09)	0.20 (0.04)
A	0.78 (0.26)	1.78 (0.80)	1.40 (0.82)	0.60 (0.18)	0.94 (0.41)	0.43 (0.11)
B	15.15 (1.26)	12.59 (1.11)	16.02 (2.32)	15.49 (0.84)	12.62 (0.91)	18.46 (3.49)
C	87.9 (5.9)	76.3 (50.0)	72.8 (40.0)	78.7 (9.9)	84.3 (52.1)	163.6 (68.4)
τ1	35.2 (1.9)	34.5 (1.7)	35.0 (1.0)	34.4 (2.5)	34.0 (2.9)	32.3 (1.3)
τ2	57.7 (11.9)	54.2 (10.9)	56.2 (4.8)	55.1 (12.7)	54.0 (14.0)	42.2 (2.9)
OARSI	0.61 (0.30)	0.88 (0.41)	0.77 (0.16)	0.98 (0.25)	1.50 (0.62)	3.40 (1.30)

Figure 3 shows how DTI detects progressive cartilage damage within two weeks in the severe injury group (further statistics on all week 0 and week 2 groups are given in Table 1). The SNR averaged over all the samples at the three time points was 33.3±7.1 (minimum 22.3, maximum 60.5) in b=0 images and 22.3±4.4 (range: 14.8 to 37.9) for diffusion weighted measurements, which is comparable to the SNR range we measured in vivo. For the severely injured samples, the MD increases in time, whereas FA decreases. The MD achieves a group difference that is statistically significant for both weeks 1 and 2 (relative to week 0), but FA achieves significant changes only for week 2. On average, the severe injury samples group indicates a two-week change in bulk MD of 0.18±0.08 μm²/ms and in FA of −0.06±0.03. In the case of mild injury, two-week changes were not significant either in MD (0.08±0.08 μm²/ms) or in FA (−0.02±0.11). For the controls, no significant change was observed in MD (0.02±0.09 μm²/ms) or in FA (0.05±0.06). There were no significant differences in baseline (week 0) MD or FA among the control, mild injury and severe injury groups.

Figure 3:

Bulk MD (left panel) and FA (right panel) values in the sever injury group at week 0, 1, and 2. In this box-and- whiskers plot the lower and upper bounds of the box indicate the 25th and 75th percentile, respectively, the red line identifies the population mean and the whiskers include all data. The star indicates any statistically significant difference with respect to the population at week 0.

Table 1 gives the measured QSV-model of all five model parameters at week 0 and week 2 for each group (Fig. 4 illustrates the stress relaxation curves obtained before and after injury for a representative cartilage sample). At week 2 all five parameters were significantly different to baseline in the severe injury group, but not in the no injury or mild injury groups. After two weeks the stiffness parameter A dropped by 70% (p<0.05) in the severe injury group, by 48% (p=0.62) in the mild injury group, and by 14% (p=0.62) in the control group. There was no significant difference in any of five QSV parameters among the three injury groups at baseline week 0. The control group showed lower stress constant A. Referring to Eq.3, A is a factor that modulates stress σ(t); this can be seen in Fig.4 where, at baseline, the no-injury sample exhibits lower stress peaks compared with that in the severe sample.

Figure 4:

Illustrative incremental stress-relaxation curves for no injury (top) and severe injury (bottom). Blue dots and red dots indicate the curves before injury and two weeks after injury, respectively. Notice the decrease in the peak loads after injury but not in the control group. Post-injury curves (in red) relax faster than pre-injury curves. It is both, the amplitude of the strain rate as well as the rate of relaxation, that affect the biomechanics model fitting (see Methods: Biomechanics).

Figure 5 shows examples of week 0 and week 2 safranin-O stained slides for the control and severe injury groups. The OARSI grade for native week 0 cartilages was low, at an average of 0.75±0.36 (Table 1). There was no difference in the OARSI grades for week 0 between the groups with severe (0.77±0.16), mild (0.88±0.41) and no injury (0.61±0.30). The OARSI grade at week 2 for the severe injury samples was 3.40±1.30 (+2.63 difference, p=0.009, Wilcoxon), for the mild injury it was 1.50±0.62 (+0.62 difference, p=0.036), and for the no injury it was 0.98±0.25 (+0.37 difference, p=0.082). Cohen’s κ for inter-reader variability was 0.66 (p<0.001, with a confidence interval of [0.46, 0.87]), showing substantial agreement between the readers. The mean Cohen’s κ for intra-reader was 0.85 (p<0.001, with a confidence interval of [0.76 0.95]), showing perfect agreement between the two readings.

Figure 5:

Illustrative safranin-O histology sections for a no-injury and a severe injury sample. Samples were positioned in paraffin so that histology sections were in the same plane as the MRI. For consistency the week 0 and week 2 cartilages from each sample was stained together. Both samples had similar OARSI grade at baseline (0.5 and 0.7). Two weeks after injury the no injury sample showed no change in histology grade (0.7) while the severe injury sample showed deep cracks with matrix loss compatible with a OARSI grade of 4.3.

Table 2 presents the correlations between the DTI indices (MD and FA) and the five parameters of the QLV biomechanics model, for the full thickness (bulk) and for each cartilage layer. Only the change in elastic stress constant A shows statistically significant negative correlation with MD values, for the bulk cartilage (r=−0.683, p=0.043) as well as the deep layer (r=−0.753, p=0.019). The pooled correlations of DTI indices in weeks 0 and 2 (results not shown) with pooled QLV parameter A showed no statistically significant correlation.

Table 2:

The correlation between the DTI indices, MD and FA, and the five biomechanics model parameters (see subsection Methods: Biomechanics), at both the start (week 0) and the end of the experiment (week 2). The biomechanics model parameters are continuous variables, so Pearson’s r is used. The numbers in bold font indicate statistical significance.

	A		B		C		τ1		τ2
	r	p	r	p	r	p	r	p	r	p
change in MD
bulk cartilage	−0.683	0.043	0.550	0.125	0.197	0.611	−0.583	0.100	−0.296	0.440
deep layer	−0.753	0.019	0.574	0.106	0.171	0.660	−0.567	0.111	−0.289	0.451
surface layer	−0.496	0.174	0.469	0.203	0.246	0.524	−0.528	0.144	−0.289	0.451
change in FA
bulk cartilage	0.473	0.198	−0.337	0.375	−0.331	0.384	0.299	0.434	0.212	0.584
deep layer	0.540	0.133	−0.339	0.372	−0.166	0.670	0.336	0.377	0.222	0.566
surface layer	0.307	0.422	−0.236	0.542	−0.394	0.294	0.205	0.597	0.136	0.728

Table 3 shows the correlations of the DTI indices with the histology grades. Only the surface layer correlation is significant; here, too, FA correlations were not significant. The pooled correlations were significant for MD and FA in the bulk and superficial layers.

Table 3:

The correlation between the DTI indices and measures for cartilage damage as evaluated by histology OARSI scoring (see subsection Methods: Biomechanics), at both week 0 and week 2. ‘Histology-pooled’ refers to correlations with weeks 1 and 2 estimates pooled together, as opposed to the change in estimates after the two weeks. The average over discrete OARSI grades produces continuous variables, so Pearson’s r is used. The numbers in bold font indicate statistical significance.

	Histology		Histology-pooled
	r	p-value	r	p-value
MD
bulk cartilage	0.62	0.075	0.68	0.003
deep layer	0.51	0.162	0.47	0.056
surface layer	0.72	0.029	0.77	0.0003
FA
bulk cartilage	−0.53	0.143	−0.51	0.038
deep layer	0.34	0.378	−0.34	0.177
surface layer	−0.61	0.081	−0.54	0.026

Figure 6 visualizes the most important results arising from Tables 2 and 3. The correlations between the histology OARSI grades and the biomechanics parameters (not shown) were significant for B, C, τ₂, τ₂ (r=0.73, 0.74, −0.79, −0.76 respectively) but, as shown in the first subplot in Fig. 6, not for A (r=−0.60, p=0.089). There was also statistical significance between the pooled correlations of the OARSI grades with the biomechanics parameters A (r=− 0.50, p=0.033), B (r=0.51, p=0.027), τ₁ (r=−0.57, p=0.012) and τ₂ (r=−0.51, p=0.032), but not with C (p=0.26).

Figure 6:

The correlations for the bulk cartilage, plotting the change (between week 0 and week 2) in biomechanics, DTI, and histology measurements across all samples. For biomechanics we show the percentage change of QLV model parameter A; for histology we show the difference (increase/decrease) in OARSI grades (whose raw grades vary from 0=healthy to 6=bone remodeling); for DTI we show the difference in bulk MD (in μm²/ms units) and FA (min=0, max=1). The last two subplots refer to DTI measurements in the surface; other subplots refer to bulk cartilage measurements. Markers in blue denote samples with severe injury, the green markers are for mild injury samples, and the red markers are for controls.

Discussion

To examine the feasibility of DTI to detect early in time the changes in cartilage after injury, we implemented a realistic model of injury, in which a direct mechanical impact on the cartilage triggers a biological response. Our results show that DTI can detect early degenerative changes, and that the changes in diffusion parameters correlate with the changes in mechanical properties and histology. Moreover, our results have translational value, as our experiment was performed in a clinical scanner using the same sequences and protocols that can be used in vivo. In the next four paragraphs we explain the design of our experiment and the methods used (the MRI, the biomechanics and the histology).

The design of this experiment differs from earlier work. Previous models of cartilage damage, which have been used for the validation of MRI parameters, use enzymatic degradation (mostly with trypsin³²) that degrades the cartilage matrix in an artificial way that does not corresponds with the early degenerative changes leading to PTOA. Contrary to enzymatic degradation, the mechanical injury of cartilage triggers a biological response with increased PG turnover and a disruption of the collagen matrix^1,5,20. This response contains most of the changes that a joint undergoes after injury, and thus represent the current most realistic model of cartilage degradation. However, components such as the inflammatory response are not captured by our experiment. We allowed two weeks to see the damage to the cartilage after the mechanical impact, as also used in previous work²⁹. However, previous studies that analyzed synovial fluid from ACL injured patients, found a significant increase in collagen Type 2 (predominantly found in cartilage) and PG fragments in SF even hours after the injury, indicating that catabolic damage starts soon after the traumatic event^1,2.

We use DTI to track the changes after injury, since it is sensitive to PG content and collagen integrity^{10,11,12,13,14,15}. Previous ex vivo experiments have observed that diffusion anisotropy is correlated to collagen architecture^9,13,15. Specifically, the direction of higher diffusivity in cartilage shows a pattern that corresponds to the collagen architecture^9,11,15. Raya et al.¹¹ investigated further the relationship between collagen architecture and the first eigenvector (i.e. the direction of largest diffusion) with scanning electron microscopy (SEM), and found an excellent correlation of the zonal architecture measured from the first eigenvector (r²=0.87, P <0.01). A subsequent experiment¹⁰ used trypsin to degrade PG in human cartilage, revealing a modulation in the diffusion indices, with MD showing a high correlation with the change in PG content (the correlation of MD with PG loss was r²=0.86, P<0.007).

We fit a cylindrically symmetric tensor (with only one radial diffusivity, alongside the axial one) to the DTI data. From a recent study on ex vivo cartilage data this model provides a significantly less noisy fitting²⁵; additionally, we are not aware of any compelling (empirical or theoretical) reasons that justify two different radial cartilage diffusivities.

The mechanism of injury and the parameters used here are based on previous studies. The injury model is based on the work of Morel et al.²⁹. They monitored after mechanical injury the water content, chondrocyte activity/viability, proteoglycan level, as well as surface cracks, and found that the level of injury was dependent on the compressive strain rate: strain rates higher than an inherent tissue-specific (“gel diffusion”) mechanical relaxation rate of 0.07 s⁻¹ compressed the matrix fluid causing cell death and structural damage near the surface. In a subsequent study, Quinn et al.²¹ analyzed the effect of different maximum peak stress at the gel diffusion strain rate. An injury peak stress of 14 MPa (corresponding to 190N in our study) resulted in surface cracks in almost all cartilage plugs analyzed, while a 7.5 MPa injury (corresponding to 120N) lead to cracks in approximately one-quarter of the samples.

In histology, the safranin-O slides were evaluated using the OARSI scoring system. In the OARSI scoring two concepts are particularly important, the “grade” and the “stage”³³. By definition, “grade is defined as OA depth progression into cartilage”, whereas “stage is defined as the horizontal extent of cartilage involvement within one side of a joint compartment irrespective of the underlying grade”. We can only use the grade to assess the degree of pathology because of the small 7-mm-diameter osteochondral plugs. The baseline OARSI grades were low, though not 0, indicative that some incipient superficial changes were present. However, the variable relevant for our study is the change in the histology grade as a result of the mechanical insult.

The results in this experiment show that DTI has potential in tracking changes that follow mechanical injury. We see in Table 1 elevated diffusivity parameters after injury. In the mild injury group, across the two weeks, the MD showed a trend to increased values, but no change in FA, which corresponds with the slightly increased OARSI grade in this group. During this period, the severe injury samples showed a higher trend of increased MD values and decreased FA values. This indicates that catabolic pathways, activated as a consequence of injury, are responsible for the progressive breakdown of the matrix. This is consistent with previous work by Quinn et al.²⁰ which analyzed the PG and collagen turnover of mechanically injured samples. They found that samples injured at the gel rate showed increased rates of PG turnover in the pericellular matrix compared to controls and a collagen turnover comparable to the one measured in controls²⁰. The increase in MD, which we observe, could thus be attributed to the breakdown of PG dominating the PG turnover. Compared with FA, the faster changing MD may indicate that the damage to the collagen happens at a slower pace than the PG depletion rate, unless this effect is all because the MD is indeed more sensitive to the changes.

Ultimately, the clinical use of these results would be limited by the variability of MD and FA parameters in the healthy population, since a baseline measurement of subjects with ACL injury will not be available. However, based on current preliminary clinical data that we are currently collecting, we see that the intra-subject variability in healthy cartilage is small compared to the range of changes seen in cartilage degradation^12,17,18.

Figure 4 gave us an insight into the relation between the DTI and the biomechanics parameters. An increase in MD in the damaged cartilage indicates a higher mobility of water molecules, i.e. a higher permeability of the cartilage matrix. From a mechanical point of view, a higher permeability results in a decrease of the drag coefficient of the interstitial fluid, resulting in a reduced stiffness (a reduction in elastic stress constant A, and increase in C, providing lower σ(t) values) and a reduction of the poro-viscoelastic behavior of cartilage (exhibited as a reduction in relaxation times τ₁, and τ₂)³⁴. The plots illustrate these changes, as seen in the stress-relaxation curves after injury, showing a clear drop in the stress peaks in each stress-relaxation step, and a faster relaxation to the equilibrium, indicating some loss of the poro-viscoelastic properties of cartilage.

Though we cannot wholly disregard the influence that the bone profile may have on the stress-relaxation curves, there are some studies which indicate that for relatively small strain rates of 20% the stress response is mostly due to the cartilage. Studies that have analyzed the compression of cartilage under load with scanning electron microscopy³⁵ and with polarized light microscopy using a Hoffman clamp to compress cartilage^36,37,38 have shown that the largest adaptive changes in cartilage occur in the superior 50-60%. Nevertheless, the high stiffness of the subchondral bone, in which the deep radial zone anchors, can play an important role in the behavior of the deeper cartilage.

Table 2 gave us the correlations of the DTI indices with the mechanical parameters, as reflected in the QLV-model parameters. It showed a reduction of the elastic stress constant (A) and the viscoelastic relaxation times τ₁ and τ₂. As expected, we found a correlation between the change in MD and the change in A for the bulk and the deep layer. Surprisingly, the change in MD values on the deep layer was a better predictor of the change in A. Though this would need to be confirmed in experiments with larger sample size, our interpretation is that the integrity of the deep layer plays an important role in the overall stiffness of the cartilage matrix, even for strains of 20%. The absence of correlations with the FA is due to the fact that with an incremental stress relaxation test we are testing the PG component of cartilage, which is responsible for the poro-viscoelastic response of cartilage^39,40. Thus, given the interpretation that PG modulates MD and collagen modulates FA, we expect no correlation with FA. The pooled results are somewhat surprising. However, as table 1 shows there are large variations in the biomechanical parameters across the samples, likely reflecting anatomical differences. Intersample variability probably dominates in the pooled correlations, while intra-sample changes dominates in the relative correlations.

These results (from Table 2) are in agreement with previous results obtained with diffusion-weighted imaging^34,35. Using an elastic model, Juráš et al.⁴¹ correlated the diffusion constant with the mechanical properties measured with indentation (instantaneous and equilibrium modulus, and relaxation time). They found a negative correlation between the diffusion coefficient and the equilibrium modulus (r=−0.52) and the relaxation time (r=−0.73). However, the elastic model is a poor model of cartilage since it cannot provide an adequate quantitative explanation of the time-dependent transient response of cartilage constant strain. More recently, Aoki et al.⁴² investigated the relationship between diffusion and mechanical indentation parameters. In their study they performed a creep experiment (constant force) and used the Voigt model to model the viscoelastic response. They reported a statistically significant correlation between the diffusion coefficient in articular cartilage and the viscosity coefficient (r²=0.69; p<0.01) and mechanical relaxation time (the ratio between the elastic modulus and the viscosity constant, r²=0.75; p<0.01) as measured with an indentation experiment.

In Table 3 we observed a high correlation between the DTI indices with the OARSI histology grades. The change in histology grades after injury correlated better with the changes in MD in the superficial layer than in the deep layer. This is consistent with the pattern of cartilage degradation observed in safranin-O stained sections, where the surface is the first to degrade, followed by a deterioration propagating down the cartilage. One factor that may explain the better correlation of the MD, compared with FA, is that the histology grades take into account the PG composition (via the safranin-O staining) and the cell death and, to a much lesser extent, the collagen damage³³. The pooled correlations are in line with a previous study on human samples with early cartilage damage (n=43) that reported a significant correlations of bulk MD ρ=0.55 (r=0.68 here, corresponding to ρ=0.69), MD in the superficial layer ρ=0.63 (r=0.77 here, corresponding to ρ=0.64), and bulk FA ρ=−0.26 (r=−0.51 here, corresponding to ρ=−0.34)¹⁷.

Our study had several limitations. First, we possessed a relatively small number of samples (three for each injury group), and the cartilage plugs were drawn from two subjects, therefore providing limited biological variability. However, despite the small sample size we were able to establish important connections between DTI and cartilage integrity and function, and show feasibility of this experimental set-up. Second, the cartilage-on-bone samples were extracted from different anatomical regions of the knee, since we worked with cartilage obtained from knee replacement surgery. This can potentially explain some variability present in the data. Third, we did not measure the matrix turnover, and so we did not have a standard of reference for the cartilage composition.

In conclusion, we have demonstrated the feasibility of a realistic in vitro experimental model of cartilage injury to help test the sensitivity and specificity of imaging biomarkers of cartilage integrity. Our experimental setup is performed in clinical scanners with the same sequences used in vivo, thus providing an interpretative framework for DTI clinical studies. In particular, we have shown that DTI is sensitive to track early changes in collagen and PG after injury and correlates with changes in biomechanics and histology. DTI biomarkers provide an advantage as surrogates of histology and biomechanics, since they promise real-time indicators in the diagnosis and the monitoring of the disease, with potential to accelerate clinical trials of new therapeutic interventions.