2.1.

Methods

The experiments were carried using a home-built microscope, whose essential features are illustrated in Fig. 2 . A reflection mode configuration was chosen, as this enabled the en-face examination of cartilage samples still attached to the subchondral bone, minimizing the disruption of the collagen architecture.

Fig. 2

The optical elements of the nonlinear microscope (not to scale).

Excitation was at $800\mathrm{nm}$ using a tuneable $100\text{-}\mathrm{fs}$ pulsed Ti:saphire laser with a repetition rate of $82\mathrm{MHz}$ (Spectra Physics, Mountain View, California, USA) The selection of the excitation wavelength was based on a number of considerations. Previous studies have produced conflicting results as to which wavelength most efficiently excites SHG in collagen,^{7, 25} but we found that in our system, the signal of both SHG and TPF peaked between 780 and $820\mathrm{nm}$ , and the use of longer wavelengths increases tissue penetration and minimizes cellular damage.⁶ A 1.0-NA water dipping objective with a $2\text{-}\mathrm{mm}$ working distance was employed (Fluor $60\times$ ) and data acquisition was via a photon-counting photomultiplier tube and a PC-based photon counting card (Hamamatsu-Hamamatsu Japan). The laser fundamental was removed by a short pass dichroic beamsplitter (CVI laser, Albuquerque, New Mexico, USA) and a bandpass filter (CVI laser, Albuquerque, New Mexico, USA). Blocking of the fundamental was greater than 9 OD and the transmission of the TPF and SHG light about 30%. To change between SHG and TPF imaging, the filter directly below the PMT was changed, with a $450\text{to}520\text{-}\mathrm{nm}$ bandpass filter being used for TPF detection and a $410\text{to}390\text{-}\mathrm{nm}$ filter for SHG detection (CVI laser, Albuquerque, New Mexico, USA). For polarization measurements, two $800\text{-}\mathrm{nm}$ half wave plates were added to the system. The first wave plate (CVI laser, Albuquerque, New Mexico, USA) was placed before the dichroic beamsplitter, orientated to ensure that completely p-polarized entered the beamsplitter. The second (Meadow lark, Frederick, Colorado, USA) was placed directly behind the back aperture of the microscope objective to allow the polarization of the light incident on the sample to be rotated. This arrangement was necessary because our dichroic beamsplitter introduced differential phase-shifts between s and p-polarized states when used in a reflection mode, which would convert the incident linear polarization state to an elliptical state if the polarization was rotated before the dichroic. The photon-counting PMT used for the detection of the TPF and SHG signal was polarization insensitive, and therefore the effect the wave-plate had on the TPF light passing back through the objective would not affect the experimental results.

Axial imaging was performed by raster scanning the beam using two scanning galvanometer mirrors (Cambridge Technologies Incorporated, Lexington, Massachusetts, USA). An axial scan was generated by moving the sample upward on a motor stage with an accuracy of about $1\mu \mathrm{m}$ (Physik Instrumente, Palmbach, Germany). The field of view of the microscope objective allowed images up to $125\times 125\mu \mathrm{m}$ to be acquired, the size being controlled by the voltage applied to the galvanometer mirrors. The images reported in this study were $100\times 100\mu \mathrm{m}$ and composed of $500\times 500\text{pixels}$ . The pixel dwell time was $10\mu \mathrm{s}$ .

The laser power used for imaging was approximately $40\mathrm{mW}$ exiting the microscope objective. For the polarization sensitivity measurements, a power of $30\mathrm{mW}$ was used for acquisition of surface images, but power was increased with scanning depth so that at $200\text{-}\mu \mathrm{m}$ depth, the power exiting the objective was $140\mathrm{mW}$ .

In polarization sensitivity experiments, the polarization of the light incident on the sample was rotated in $10\text{-}\mathrm{deg}$ steps using the half wave plate behind the microscope objective. The total intensity from the area scan was measured, and the intensities for each angle were combined to create the polarization sensitivity curve. This process was repeated at depth steps into the tissue of $10\mu \mathrm{m}$ in tendon and healthy cartilage and $15\mu \mathrm{m}$ in cartilage lesions. The depth steps were chosen as a compromise between collecting a very detailed dataset and minimizing the total scan time.

2.2.

Materials

Tissue samples were obtained fresh from the abattoir, cartilage from the metacarpophalangeal joints and tendon from the deep flexor tendon of horses. The age of the samples was estimated by the abattoir staff and ranged from $5\text{to}20+\text{years}$ . Some samples were frozen prior to imaging, but this had no discernable effect on the images. At the laboratory, the joint was opened and a full depth plug of cartilage and several millimeters of subchondral bone was excised from the apex of the joint using a jeweller’s saw. During measurement, samples were maintained at room temperature and kept moist by immersion in $0.15\text{-}\mathrm{M}$ saline.

3.1.

Surface Measurements on Cartilage and Tendon

Figure 3 shows the angular variations in intensity of SHG and TPF at the surface of cartilage and tendon. The images were acquired at a depth of $10\mu \mathrm{m}$ from the surface zone of fibers aligned parallel to the surface (Fig. 1). The polarization angle is defined as the angle between the collagen fiber and the polarization of the incident radiation, and in both tissues TPF varied only weakly with angle. The visibilities $[(I_{\text{peak}}-I_{\text{trough}})∕(I_{\text{peak}}+I_{\text{trough}})]$ for tendon and cartilage were $(0.045\pm 0.005)$ and $(0.04\pm 0.02)$ , respectively, ( $\text{mean}\pm \mathrm{SD}$ , measurements from five samples). Variations in the intensity with polarization angle were comparable in magnitude with the measurement noise. It therefore appears that TPF generation in both tissues does not have any significant intrinsic polarization sensitivity, or the generation is polarization sensitive but the sources are randomly orientated. As found by previous researchers,^{15, 16, 17, 18, 19} the SHG signal from tendon showed a strong polarization dependence, with the greatest intensity when the laser excitation was parallel with the collagen fibrils. The peak in the SHG signal corresponded with the long axis of the collagen fibers visible in the SHG images to within $\pm 7\mathrm{deg}$ . The SHG signal from the cartilage surface also depended on polarization and had a period of $180\mathrm{deg}$ , suggesting a preferred orientation of the collagen fibers in the superficial zone of the cartilage. Though this was not observed in previous studies on SHG from cartilage,²¹ it is consistent with well-established histological evidence.^{1, 26}

Fig. 3

Polarization sensitivity curves for both TPF and SHG taken at the surface in (a) cartilage and (b) tendon, with the $x$ axis showing the angle between the collagen fiber axis and the polarization of the laser fundamental. The intensities are normalized with respect to the mean intensity for each curve. Data were acquired at a depth of $10\mu \mathrm{m}$ beneath the articular surface.

The visibility of the intensity-polarization angle plots was calculated as a measure of the degree of collagen fiber organization in the tissue. For tendon, the visibility was $0.29\pm 0.04$ , and for cartilage the visibility was $0.11\pm 0.05$ ( $\text{mean}\pm \mathrm{SD}$ , measurements from six samples). These differences are consistent with the perceived differences in the level of collagen fiber organization in the two tissues, but different levels of intrafibrillar organization or optical properties between type-I and -II collagen may also contribute.

To investigate whether there were differences in collagen fiber organization within the $100\text{-}\mu {\mathrm{m}}^2$ scanning area, we compared SHG images with the polarizer set to provide maximum signal with those with the polarizer rotated through $90\mathrm{deg}$ . Representative images and the resulting ratio image are shown in Fig. 4 . Histological studies have described the pericellular matrix as basket-like and containing finer collagen fibrils.^{1, 26, 27} Therefore, we were expecting less polarization sensitivity and consequently a ratio close to 1 in these regions. However, these differences were not apparent in the ratio image, and although the ratio varied between 0.5 and 2 on a length scale of a few microns, the variations showed no obvious correlation with position. On the basis of these observations, it was felt that averaging over a $100\times 100\text{-}\mu \mathrm{m}$ image area was suitable for subsequent polarization sensitivity investigations. A smaller area will provide information on the collagen fiber arrangement on lower hierarchical levels instead of the zonal architecture described in Fig 1. Averaging over longer length scales is not required, as the polarization sensitivity results averaged over the $100\times 100\text{-}\mu \mathrm{m}$ image area are repeatable, and therefore a larger imaging area will produce the same results.

Fig. 4

SHG images acquired with the polarizer at the position of (a) maximum intensity and (b) rotated through $90\mathrm{deg}$ , with the arrows indicating the polarization angle. (c) shows the ratio of the two images. The dark regions in (a) and (b) represent the locations of the chondrocytes. In the ratio image, these areas are circled. (Field $100\times 100\mu \mathrm{m}$ , arrows indicate the polarization of the laser light.) The dark region in the top right quarter of the image represents an area where the polarization sensitivity varies from that found in the rest of the scan area.

3.2.

Depth Variations in Tendon

To determine how the effects of biattenuance and birefringence in the overlying tissues affected measurements of changes in polarization sensitivity with depth, we undertook an initial study on tendon, whose structure and fiber orientations are constant with depth and can be observed directly in the multiphoton images. The angular dependence of SHG and TPF was measured at depths up to $200\mu \mathrm{m}$ and values averaged over a $100\text{-}\mu {\mathrm{m}}^2$ field in three samples are summarized in Fig. 5 . Although the orientation of the collagen fibers remained constant with depth, the SHG polarization sensitivity pattern changed dramatically with depth, as is shown in Fig. 5a. The TPF polarization sensitivity in Fig. 5b shows an increase with depth, with strong peaks in intensity appearing perpendicular to the collagen fiber orientations. This indicates that the optical properties of the overlying tissue are contributing to the measured patterns. Figures 5c and 5d show the natural log of the unnormalized data for the SHG and TPF polarization, respectively. These plots have been included because they show the attenuation of the light through the tissue.

Fig. 5

Polarization dependence of (a) SHG and (b)TPF in tendon at different depths [collagen fibers orientated at $90\mathrm{deg}$ to the original incident beam polarization $\left(0\mathrm{deg}\right)$ , intensities normalized by the mean intensity of the image field at each depth]. (c) and (d) show the natural log of the unnormalized polarization sensitivity. (e) shows the ratio of maximum intensity of the TPF/ the minimum intensity of the TPF as a function of depth in the tissue.

From the TPF data, it was possible to estimate an upper bound for the biattenuance of the tissue at the excitation wavelength. Surface imaging demonstrated that the TPF excitation efficiency has no intrinsic polarization sensitivity, which in turn suggests an isotropic orientation of fluorescence excitation dipoles within the focal volume. The variation in TPF emission at depth can thus be presumed to arise from “biattenuance” (a term introduced by Kemp ²³ to describe differential absorption or scattering of the excitation and/or emitted light for different incident polarization states). Let us assume that the superficial layer contains a well-defined transmission axis such that field strengths of light linearly polarized parallel and perpendicular to this axis experience a differential attenuation coefficient given by $2\pi \Delta \chi ∕{\lambda }_0$ , where $\Delta \chi$ is the biattenuance of the medium. Further, let us assume that these axes are coincident with the birefringence fast and slow axes of the superficial layer, i.e., we effectively assume that the superficial tissue can be described by a complex birefringence $\Delta n+i\Delta \chi$ . Incident linearly polarized light aligned with either of these directions will thus remain polarized along these directions with increasing depth, and the differential attenuation of the excitation light between these directions will attain the maximum value. Linearly polarized light incident at other angles will undergo conversion to elliptical polarization states with increasing depth, and will thus experience an intermediate degree of attenuation. We thus identify the angles at which maximum and minimum TPF intensities occur as defining the directions of the fast and slow axes. Recalling that our detection system is polarization insensitive, if the emitted TPF were itself unpolarized, then the emitted light would propagate to the detector with equal efficiency, regardless of the polarization direction of the excitation light. We could then use the difference in the maximum and minimum TPF intensities to infer $\Delta \chi$ at the excitation wavelength a follows.

Following Kemp, ²³ we note that $\Delta \chi$ defines the ratio of the field strengths polarized along the two axes at a depth $\Delta z$ , such that

These estimates represent only an upper bound for the biattenuance, however, because even with an isotropic distribution of fluorophores, the TPF may still be partially polarized. This is because the incident polarization vector will most efficiently excite parallel-aligned dipoles, resulting in the TPF being partially polarized in the direction of the incident laser polarization (unless the fluorophores rotate on a time scale fast compared with the fluorescence lifetime, which is unlikely in collagen).²⁹ This means that the emitted light will be transmitted to the detector with different efficiencies for the two orthogonal polarization states, so that the intensity ratio will then depend on the value of $\Delta \chi$ at both the excitation and emission wavelengths. The precise relationship between our value and the true values at 400 and $800\mathrm{nm}$ will depend on the degree of fluorescence anisotropy. This highlights the need for further research into not only identifying but also characterizing the properties of endogenous two-photon fluorophores.

For the SHG, the pattern of polarization with depth in tendon represents the combined effects of the biattenuance, birefringence, and the intrinsic polarization sensitivity of the collagen SHG. Tendon is birefringent, with a higher refractive index for light polarized parallel to the collagen fibers.¹⁹ The birefringence $\left(\Delta n\right)$ is approximately 0.0045 for equine flexor tendon,³ indicating that a $180\text{-}\mu \mathrm{m}$ thickness of tendon would act as a whole wave plate at $800\mathrm{nm}$ . It should be possible to use this information to recreate our polarization sensitivity patterns, using either a Jones or a Muller matrix (if scattering is found to be a significant effect) to describe the effects of the overlying tissue, and the Freund model¹⁶ to describe the polarization sensitivity of collagen fibers. However, the effects of beam geometry, such as varying interaction lengths and ray-to-fiber inclination angles across the beam profile, are likely to be significant for the large numerical aperture objective we employed, and it would be a nontrivial task to include these in the calculations. Such studies will form the basis of further work to be reported in the future.

3.3.

Depth Variations in Cartilage

TPF and SHG polarization sensitivity data were taken on the same field in samples of healthy cartilage, and representative plots are shown in Fig. 6 . It proved possible to acquire data at depths up to $200\mu \mathrm{m}$ . This region is marked on the histological section taken of the sample after imaging [Fig. 6e]. From this we can see it corresponds to the superficial and transitional zones shown in Fig. 1. The TPF polarization measurements were affected by photobleaching, and although exposure time was kept to a practical minimum, it was impossible entirely to eliminate the effect. The photobleaching in cartilage was a more significant effect than in tendon. This may indicate that a different fluorophore is responsible for the fluorescence in the two tissues, or that the differences in the extracellular matrix environment effect the photobleaching rates. In contrast to tendon, the TPF images did not show strong polarization sensitivity. At $150\text{to}200\mu \mathrm{m}$ , four peaks were evident but their visibility was very small $(<0.05)$ compared to that in tendon $(\sim 0.6)$ at a similar depth. It is therefore apparent that biattenuance is not as significant in cartilage as it is in tendon, and the structural bases of this difference require further investigation. If this is interpreted as a null result for detecting biattenuance in cartilage, we can place an upper bound for the possible biatteunance of $6\times {10}^{-5}$ by assuming that periodic variations should be at least two times greater than the variations due to noise or photobleaching.

Fig. 6

Polarization sensitivity with depth in cartilage, (a) SHG and (b) TPF (intensities normalized by the mean field intensity at each depth). (c) and (d) show the natural log of the unnormalized polarization sensitivity for SHG and TPF. (e) shows a histological section taken from the sample after the polarization sensitive measurements, viewed through crossed polarizers to reveal collagen organization.

The variation in SHG polarization with depth showed a clear repeatable pattern in all the normal samples of cartilage investigated. At the surface of the tissue, there were two intensity maxima, but at greater depths two additional peaks occurred between the original peaks. There was also an increase in visibility of the peaks with depth into the tissue. The depth at which the additional peaks appeared varied between 60 and $90\mu \mathrm{m}$ in different samples.

Relating the changes in polarization sensitivity with depth to fiber orientation is a more complex problem in cartilage than in tendon. In contrast to tendon, the birefringence and biattenuance properties of the tissue are likely to change with depth, and this level of complexity cannot simply be described by a single Jones or Muller matrix. In addition, the collagen fibers form a 3-D network, so the fibers in the deeper zones no longer lie in the imaging plane, and the larger the angle between the collagen fiber and the imaging plane, the weaker the polarization dependence and the weaker the SHG signal.¹⁹ It seems therefore that it will be necessary to develop a numerical model, perhaps based on measurements of biattenuance and birefringence, in a series of thin slices of cartilage at increasing depth into the tissue.

3.4.

Cartilage Lesions

Spontaneous lesions were found on the apex of a large number of the metacarpophalangeal joints examined. The histology of these lesions has been discussed in detail elsewhere.³⁰ We report here on the detailed examination of three such lesions.

Figure 7 shows paired TPF and SHG images taken from the surface of the cartilage in the histologically normal tissue close to the lesion, at the periphery of the lesion, and in the center of the lesion. The former images are very similar to those described earlier and reported by previous authors.^{21, 31} At the periphery of the lesion, where the histology showed an intact surface but a loss of zonal structure, the TPF images appeared normal, but the SHG images showed abnormally high signal intensities from the pericellular matrix. In the core of the lesion, there were far fewer cells and the matrix also appeared abnormal with a rippling effect clearly visible. The rippling appears more strongly in the SHG than the TPF and so can be confidently ascribed to a collagen abnormality. Ratio images emphasized the difference in the pericellular matrix between the normal tissue and the tissue peripheral to the lesion, with the ratio of SHG/TPF much greater in the pericellular matrix at the periphery of the lesion.

Fig. 7

Cartilage lesion. (a) SHG and TPF images taken from a region $3\mathrm{mm}$ away from the center of the lesion, which appeared normal in histological sections. (b) Images from the periphery of the lesion, showing abnormalities in the SHG from the pericellular matrix, while the TPF image remains normal. (c) Images from the center of the lesion, showing loss of cells and rippling effect in the SHG image. (Field $100\times 100\mu \mathrm{m}$ ). The ratio images shown in the right-hand column highlight the changes in (b) the pericellular matrix and that (c) the rippling effect is stronger in the SHG image.

Figure 8 shows the measurements of both SHG and TPF polarization as a function of depth, in the core of the lesion and at the periphery, in macroscopically normal tissue. At the latter sites, the tissue showed the weak biattenuance in the TPF signal reported earlier, but at the core of the lesion, no polarization sensitivity was observed. Again we interpret this as a null result and place an upper bound of $6\times {10}^{-5}$ on the biattenuance, as for the TPF from healthy cartilage.

Fig. 8

The polarization dependence of TPF and SHG in the vicinity of a lesion, with site 1 being macroscopically normal-looking tissue close to the lesion, site 2 being approximately central to the lesion and site 3 being tissue in the periphery of the lesion.

The SHG depth profiles showed larger changes than the TPF in the vicinity of the lesion, and so Fig. 9 shows a set of nine depth scans taken at $1\text{-}\mathrm{mm}$ intervals across the lesion. The first three scans covered regions that were histologically normal, and the scans were indistinguishable from those obtained in normal tissue at remote sites. However, at the site of the fourth scan, the histological structure was disrupted with the cartilage zones no longer clearly identifiable, and the SHG pattern in the four peak pattern beginning to be lost. The remaining scans were grossly abnormal in showing only two intensity maxima. In addition, the angle at which the peaks in intensity occurred was not constant with depth. This indicates a change in the angle of the fibers, as observed in x-ray diffraction studies on similar lesions.⁴ The intensity of the polarization sensitivity was greatest in the center of the lesion. This is consistent with a previous study that was able only to measure polarization sensitivity in diseased and not in healthy cartilage.²¹

Fig. 9

Polarization dependence of SHG with depth for a series of ten sites at $1\text{-}\mathrm{mm}$ lateral positions through the lesion. Position 1 is anterior to the lesion and position 9 is posterior. At position 6, at the center of the lesion, the structure was too broken to detect any pattern of polarization sensitivity with depth. [All scales as in (a)].