2.1.

Samples

The LCO staining solution, containing a mix of qFTAA and hFTAA, was prepared by using $1\mathrm{mg}{\mathrm{ml}}^{-1}$ stock solutions prepared in 2 mM NaOH and stored at 4°C while awaiting use. Synthesizing of qFTAA and hFTAA was done as described by Klingstedt et al.⁹ Before staining, fresh solutions were prepared by diluting the stock to $1:500$ in phosphate buffered saline (PBS). Then qFTAA and hFTAA were mixed in a ratio of $2:1$, resulting in a final concentration of 2.4 μM qFTAA and 0.7 μM hFTAA, in assay solutions. Brain tissues from 19 mice aged 6 to 23 months were employed in the study to facilitate the investigation of the age dependence of spectral variations in the plaques. About 10 μm cryo sections were prepared and kept at −20°C until needed. The LCO staining solution containing a mix of qFTAA and hFTAA was applied to the cryosections. The sections were washed $3\times 5\min$ with PBS and mounted with Dako fluorescence mounting medium. The slides were left in room temperature in the dark overnight and sealed with nail polish. Structures for qFTAA and hFTAA can be seen in Fig. 1. See Nyström et al. for further details.¹⁴

Fig. 1

Chemical structures for the LCOs qFTAA and hFTAA.

2.2.

Hyperspectral Imaging

After preparation, the samples were imaged using a Leica DM6000 B fluorescence microscope mounted with a SpectraCube module from applied spectral imaging (ASI). An excitation wavelength filter with peak transmission at 405 nm and a full width at half maximum of $\approx 20\mathrm{nm}$ and a long-pass filter allowing emission above 450 nm were used to block back-propagating light from the excitation.

2.3.

Correlation Analysis

The correlation analysis, see Ellingsen et al. for the details,⁸ is based on calculating the cross-correlation coefficient $\rho$ between the measured spectrum in a pixel ($M$) and a reference spectra ($R$). This is repeated for every pixel in the hyperspectral image above a certain background threshold, yielding a new image with the correlation coefficient. If there are additional reference spectra, the process is repeated, yielding one correlation image for each reference spectrum. For the remainder of this work, the correlation coefficient will be referred to as the correlation.

3.1.

General Approach for Collecting Data

Hyperspectral images were recorded using the fluorescent microscope (Leica) equipped with an ASI Sagnac interferometric hyperspectral imager, see Sec. 2.2 for details. The initial hyperspectral images, each containing several plaques, were thresholded to remove background (signal to noise of more than 10 for the intensity image, with the same level for individual spectral channels) and then the data of individual plaques were extracted using operator selected region of interest (ROI). Figure 2 shows an example of the thresholded image and Fig. 3 shows a resulting cropped image.

Fig. 2

Intensity image (sum over spectrum) with regions of interest showing the seven cropped areas used in the rest of the analysis. Color scale is in counts. The threshold of 3500 counts is shown as dark blue. The mouse age was 686 days.

Fig. 3

Intensity image of crop number 7 in Fig. 2 colored in the natural colors of the eye using the algorithm described in Ref. 16. (See Video 1 MPEG, 0.8 MB) [URL: http://dx.doi.org/10.1117/1.JBO.18.10.101313.1] for a wavelength scan over the image.

Totally, 19 different AD mice were examined and data were collected from between 5 and 34 plaques (with majority of them above 20) for each age (some mice showed more plaques than others). Correlation images were then generated from this set of raw data using reference spectra from plaques stained with pure qFTAA and hFTAA, respectively. The two reference spectra are shown in Fig. 4. For more details on the acquisition of these spectra, see Nyström et al.¹⁴ As can be seen in Fig. 4, the references are well defined, with distinct spectral differences. When mixed and used in double staining the resulting spectra will be harder to separate due to the occurrence of both stains and a background that can vary depending on the location within the sample and the sample itself.

Fig. 4

The reference spectra for qFTAA ans hFTAA recorded in single-stain experiments.

Qualitative observation of plaques generally showed a diffuse structure surrounding a region of high fluorescence, defining a center of high fluorescence intensity. The plaque shapes were not distinct but essentially circular with small elliptical distortions, or with patches of smaller regions with high intensities arranged around an essentially circular main plaque. Two ways of identifying the plaque center was tested: (1) from the average of some pixels of highest intensity, or (2) by calculating the center of mass for the whole intensity distribution, see e.g., Adams.¹⁷ The latter approach was abandoned as it generated a much larger spread of data when calculating radial distributions, especially with plaques that were far from spherical. Therefore, approach (1) was adopted by defining a threshold for the background intensity (sum over the spectrum) image, and then finding the maximum after passing it through a $3\times 3$ median filter. The motivation for adopting this approach was that the plaques are actually sections of three-dimensional structures. A consequence would then be that the center would be the area where the plaque is “thickest” and therefore the fluorescence intensity should be the highest. Using a median filter ensures that a singular pixel with a high intensity, e.g., from noise or a measurement error, was not chosen as the center. An example of such a defined center is shown in Fig. 5 for a plaque that is very irregular in shape. The choice of method for finding the center for any of the methods will give some results where the center is ill-defined; however, owing to the nature of the statistical analysis, as long as the majority of the centers are found well enough, the overall results from the statistics should make the most probable distribution most significant.

Fig. 5

Image showing the sum over the spectrum for crop number 7 in in Fig. 2. The white plus in the figure marks the center used in the radial analysis. Color scale is in counts. The threshold of 3500 counts is shown as dark blue.

After the correlation images for all the plaques were generated, (see representative correlation images for the two stains in Fig. 6), the data set was further analyzed in terms of radial distribution and mouse age.

Fig. 6

Image showing the correlation between the qFTAA (left) or hFTAA (right) reference and the measured spectrum. The crop is the same as Fig. 5. Notice the big difference in the center of the plaque between the correlation value. The threshold of 3500 counts is shown as dark blue.

3.2.

Spectral Correlations Versus Radial Distribution

Having calculated the correlation for every pixel in an image, we now want to map its distribution, e.g., the radial distribution within a plaque. The distance or radius from the plaque center $r$ is given by

Fig. 7

The mean correlation and its standard deviation as a function of normalized distance from center (radius) for four mice of different ages. The mean and standard deviation are calculated over all analyzed plaques of each mouse. The age is indicated in each plot.

The findings so far are in agreement with the simple analysis in Nyström et al.¹⁴ based on a simple comparison of intensities of two spectral emission peaks in the center of a plaque, where the peak at 500 nm corresponds to the binding of qFTAA and the peak at 540 nm reflects the binding of hFTAA. The difference between the two references becomes even more pronounced the older the mice are, with the correlation eventually crossing each other somewhere around 400 days (see Fig. 7). From this, it can be deducted that there is a clear change over time with respect to the structure of the plaque, clearly inhibiting the binding of hFTAA in the center. Since the LCOs are sensitive to the conformation of the proteins within the $\mathrm{A}\beta$ plaque, this yields a very promising result for the study of structural changes within the plaque for different ages. Further details of the spectral correlation was obtained from a statistical analysis of all plaques, to be discussed below (Sec. 3.3).

3.3.

Statistical Analysis of Spectral Distribution

In addition to the four mouse ages shown in Fig. 7, data of 15 more AD mice were recorded, correlated with the references and the corresponding radial characteristics was calculated for each age. In order to plot the wealth of analyzed data, the relative abundance of the hFTAA and qFTAA spectral components are plotted as a function of mouse age at different relative radii as in Fig. 8. Hereby, we obtain an overview of the spectral characteristics for the essential regions of the plaques. This shows that the correlation for qFTAA and hFTAA in the plaque center varies significantly as a function of age, apparently crossing each other for mice at approximately 340–490 days of age. In the outer regions of the plaques (Fig. 8: lower panels), the relative abundance is relatively constant up to approximately 500 days of age. Here, hFTAA has a much larger propensity to bind with a correlation value close to 1. These results and conclusions are in agreement with the simpler analysis in Nyström et al.¹⁴ Interestingly, the correlation analysis shows a pronounced “second wave” as qFTAA increases its correlation value in the outer regions above 500 days. Also, a second crossing of the correlation values is prominent at plaque center (Fig. 8: upper left panel), indicating that there are different structural properties of plaques in the younger mice ($<500$ days) compared with the older ones, and that hFTAA staining starts to dominate the plaque centers again at $>600$ days. These results and conclusions are in agreement with the simpler analysis in Nyström et al.¹⁴ Variation in plaque morphologies in both man¹⁸ and mouse¹² has been documented earlier. However, our described method allows for both qualitative and quantitative analysis of images from biological specimens due to the ability to analyze the entire image for a series of images in a data set. Together with the ability to analyze radial distributions and smaller spectral differences, it provides a more detailed and accurate approach, with statistically verified results.

Fig. 8

Mean correlation as a function of age in days and normalized distance from the center. Notice that the $x$-axis (age) is uniformly spaced with respect to each case, and that there are three identical ages (245, 392, 531 days) which are from different mice and therefore are shown as different points. The outer red/blue borders denote the standard deviation.