2.1.

Mouse Models and Sample Preparation

Mice transgenic for all three human globin genes were utilized [B6;129-Hbatm1(HBA)Tow Hbbtm2(HBG1,HBB*)Tow/Hbbtm3(HBG1,HBB)Tow/J, Jackson Laboratory #013071].²⁹ SCD mice exclusively expressed human sickle beta globin, alpha globin, and gamma globin. Littermate control mice or wild type (WT) exclusively expressed normal human beta globin, alpha globin, and gamma globin. Mice examined here ranged in age from 30 to 40 weeks. Younger mice (approximately 8 weeks) samples were analyzed, but found to already present splenic disruption which is not fully characterized here, but indicates that a proper disease progression study must be carried out at even earlier stages of development. Under IACUC protocol 100677-0516, animals were euthanized by ketamine-xylazine overdose and exsanguinated after becoming unresponsive to toe pinch. Additionally, it should be stated that no effort was made to distinguish between signals associated with SCD and those associated with hypoxia since hypoxia itself is associated with SCD in these models and in humans, and therefore, information gained from a study of hypoxia-induced biomarkers is important to understanding SCD. Four spleens from each tissue type, i.e., SCD and WT, were examined via MPM and two of the spleen samples from each sample type were histologically processed after MPM imaging. For MPM imaging, samples were placed, unstained and unprocessed, on coverslips in a pool of phosphate buffered saline to prevent drying while imaging and placed on the stage of the inverted microscope.

Following MPM imaging, two spleen samples from each tissue type were prepared by standard histological methods (i.e., mechanical sectioning and staining) for white light microscopy. Tissues were fixed with 10% neutral buffered formalin and processed for paraffin wax embedding. Sections were cut in $4\text{-}\mu \mathrm{m}$ thick sections and stored at room temperature overnight for drying. The slides were then stained with hematoxylin and eosin (H&E). H&E slides are included in this study as a “gold standard” tissue analysis method for comparison with MPM.

Additionally, histology slides were prepared to label iron deposits in tissue with Prussian Blue (PB) stain (Sigma Aldrich). Slides were deparaffinized and hydrated to distilled water. Immediately following hydration, the slides were incubated in working PB staining solution for 20 min at room temperature. The slides were then rinsed in several changes of distilled water and counterstained with nuclear fast red, cleared, and finally cover slipped using resinous mounting medium. PB slides were included in this study as a direct iron-complex detection method to compare with MPM objects.

2.2.

Image and Spectra Acquisition

MPM imaging was performed on a Nikon A1R-MP Confocal system using a $40\times$, 1.15 NA, water-immersion, or $20\times$. 75 NA MIMM objective detecting 3 channels (Blue: $482/35$, Green: $525/50$, and Red: $575/50$) by Episcopic GaAsP PMTs. Excitation was performed at 800 nm at modest laser power levels (10–50 mW) generated by femtosecond pulsed Mai Tai HP DeepSee laser. An IR stop filter (680 nm short pass, OD 8+) was used to pass TPEF photons and reject reflected/scattered excitation light. Samples were excited at $\lambda =800\mathrm{nm}$, corresponding to a maximum in the two-photon excited florescence rate as well as readily available pulsed laser. Three-channel imaging was performed with the following set of collection parameters: frame rate:$1/4\mathrm{fps}$, line averaging: $8\times$, $1536\times 1536\mathrm{px}$ ($1986\times 1986\mu {\mathrm{m}}^2$) frame size, equal detector gain and offset settings for each channel which are comparable to other intrinsic fluorescence MPM studies.^18,19 To ensure that these parameters did not result in visible photo damage or photobleaching, single frames were imaged for several minutes (much longer than the typical acquisition time) and qualitatively assessed for decrease in brightness or tissue damage with time. Spectrum analysis was performed using 32-channel PMT collecting from 400 to 650 nm at a 10-nm wavelength resolution. Bright field imaging was performed on a Nikon 90i Upright with a $20\times$, 0.75 NA objective to acquire $2560\times 1920\mathrm{px}$ ($870\times 652\mu {\mathrm{m}}^2$) images. Postprocessing, e.g., file type conversion, object segmentation and analysis, pseudocoloring, cropping, z-stacking, intensity mapping, and spectrum smoothing, was performed on images and data after acquisition via MATLAB® (Mathworks, Natick, Massachusetts), ImageJ (NIMH, Bethesda, Maryland) and Imaris (Zurich, Switzerland).

2.3.

Image Processing and Object Segmentation

Objects which could be spectrally distinguished were segmented, and associated morphological properties were extracted by the following method. The algorithm employed for this task is summarized in Fig. 1. Raw image files (.ND2) were pseudo colored and scaled with proper pixel-per-micron scaling factors for analysis in ImageJ. The scaling factor for images acquired by MPM was $1.3\mu \mathrm{m}/\text{pixel}$ and for images acquired by standard optical microscopy was $0.34\mu \mathrm{m}/\text{pixel}$. Thirty images from each of the three imaging modalities employed to image the two sample types were used for a total of 180 images processed by this procedure. Each $\mathrm{N}\times \mathrm{M}\times 3$ RBG colored image was then transformed to LAB color space by the makecform and applycform functions where $L$ is the lightness of the pixel, $A$ is a number associated with a color on the magenta-green color scale, and $B$ is a similar number assigned to colors based on its position on the blue–yellow color scale. This allows segmentation by color while ignoring modest variations in intensity which may result from slight changes in acquisition parameters. The different colors present in the image are then clustered via a $k$-means minimization algorithm with respect to square Euclidean distance. MPM and PB images were clustered well using only 5 colors and 6 repetitions, whereas H&E segmentation required 10 colors and 13 repetitions to achieve satisfactory color segmentation as summarized in Table 2. The distance between colors in LAB space between objects of interest in the H&E samples is smaller than that between colors for PB (which specifically stains iron blue for contrast) and MPM. Colors are segmented into individual images and viewable as RBG images. The images containing the color cluster of objects of interest are chosen for further processing based on the mean red (R), green (G), and blue (B) values of the cluster compared to the predetermined values of objects of interest. For example, clusters from an MPM dataset were automatically selected based on the highest R and lowest B of the mean cluster values. Selected cluster images are converted into black and white binary files where noise removal and other binary functions can be performed. Isolated objects of less than $1.7\mu \mathrm{m}$ (scaled according to image type) in diameter were excluded as noise. The morphology of each connected component is extracted by the regionprops function and analyzed for statistical significance. Success in segmenting objects of interest was facilitated and visualized by highlighting connected components. Figure 2 graphically summarizes the image processing algorithm.

Table 2

Table summarizing color segmentation and cluster selection variables.

Fig. 1

Process flow of color segmentation algorithm demonstrated on a single frame of sickle cell disease (SCD) splenic tissue. The original image is segmented using a $k$-means clustering method in which the RGB images are converted into LAB color space and the square Euclidean distance between A and B points is minimized according to a given number of colors. The cluster containing the object of interest is selected using a previously determined set of color criterion and converted into a binary image. The connected components are analyzed for morphology of each object and labeled in the original image with highlighting to facilitate analysis of the accuracy of the segmentation process. Object densities are computed using appropriate scaling factors. Images are then zoomed and crop for convenient viewing.

Fig. 2

Emission spectra of SCD tissue and common intrinsic fluorophores. NADH, FAD, and bilirubin are shown in solid lines as reported in Refs. 12 and 24 and experimentally measured intrinsic emission spectra of bulk tissue relevant to sickle cell disease are shown in dashed lines. While NADH and FAD are thought to be the dominant contributors of autofluorescence in tissue, two-photon excited fluorescence (TPEF) of spleen tissue excited by 800 nm (green dashed) light is a combination of multiple fluorophores partially explained by FAD, small regions of long-wave emission (maroon dashed) and intrinsic emission of splenic pulp (blue dashed). The emission spectra of extracted and solid hemin (purple dashed) also present with a unique fluorescence spectra similar in trend to the spectra of long-wave emission nodules.

2.4.

Computations and Statistical Methods

All statistics reported here are computed using built in MATLAB functions. The densities reported in Sec. 3.3 were based on object counts per frame, converted into objects per area reported in square mm and averaged over several frames belonging to the same individual sample. To determine the statistical significance of resulting densities, $t$ test2 was used to compare $1\times 4$ the mean density vectors of each tissue type imaged by MPM without assuming equal variances. The results of the two- sample $t$ test determine acceptance or rejection of the null hypothesis, which states that two independent samples come from distributions with equal means. A $p<0.05$ here has been used to indicate significant difference. To identify these spectral features, spleens from four WT and four SCD mice were imaged at unique locations throughout the intact spleen by MPM. Two of the WT and two of the SCD spleens were then mechanically sectioned and stained with PB and H&E and imaged by standard microscopy. A total of 180 images, i.e., two sample types imaged by three imaging methods, were processed by the method described in Sec. 2.3 to generate the densities presented in Sec. 3.3.

3.1.

Intrinsic Emission in Splenic Tissue

For the study of SCD, splenic tissue was chosen as it known to be compromised by SCD and is important to immune system function, serving as a blood filtration organ that surveys circulating blood for evidence of pathogens.^6,30,31 Importantly, the spleen suffers severe damage in SCD, with accumulation of tissue damage and autoinfarction leading to functional asplenia in most adults with SCD. In Fig. 2, the normalized emission spectra of known intrinsic fluorophores and the measured spectra of splenic tissue are presented. While NADH and FAD are considered to be the dominant fluorophores responsible for autofluorescence in most tissues, the emission of bulk splenic tissue presents as a combination of fluorescent emissions only partially explained by the presence of FAD and NADH. Figure 3 shows the example images of SCD spleen tissue and solid hemin both imaged by MPM. In splenic tissue, of note are the broad emission peaking at 550 nm and the long-wave emission growing at 650 nm as well as the clear SHG peak at 400 nm caused by connective tissues largely composed of structured collagen of the spleen outer capsule.³² Additionally, when the spectra are measured over smaller regions of interest that appear to be homogenous, indicated by colored arrows and boxes in Fig. 3, emission from these regions can be isolated as spectrally distinct. The long-wave emission of nodules found in spleen tissue is measured and shown as a dashed maroon line in Fig. 3. The emission spectra of solid hemin, an iron-binding heme compound, were also measured and are indicated in Fig. 2 with a dashed purple line. When the normal iron-storage mechanisms of ferritin are exceeded, iron will deposit adjacent to the ferritin-iron complexes in the cell. Histologically, these amorphous iron deposits are referred to as hemosiderin and though are chemically not well characterized, are thought to be composed of ferritin, denatured ferritin, other iron-binding heme complexes with similar structures as hemin, and other poorly defined materials.²⁶

Fig. 3

Characteristic multiphoton microscopy (MPM) images of intrinsic fluorescence in tissue. (a) Bulk spleen tissue of SCD mouse excited at 800 nm and (b) solid hemin with salt crystals. Colored box and arrows indicate regions correlated with spectra measured in Fig. 2. Green dotted box shows a region of representative bulk spleen tissue, maroon arrows show long-wave nodules, cyan arrows indicate splenic pulp regions, and purple arrows indicate deposits of solid hemin. Scale bar indicates $200\mu \mathrm{m}$.

In order to further characterize the TPEF signals and the diagnostic relevance of MPM in the study of SCD, the samples were analyzed according to traditional biological methods to determine and identify specific observed optical signatures which can be used to study immune dysregulation in SCD.

3.2.

Splenic Tissue in Sickle Cell Disease

The spleen is involved in filtering blood and red blood cells and represents an important immune organ.³² It has been suggested elsewhere^6,7 that the spleen plays a significant role in the high incidence of chronic infection in SCD, and a deeper study of this organ may lead to answers concerning immune dysregulation. Macroscopically, even in young mice, the SCD spleen is greatly enlarged compared to WT spleen. However, the mechanisms by which this organ becomes less functional are unknown. MPM investigations could provide in vivo measurements of this system in order to begin identifying some of the mechanisms at the cellular and subcellular levels for splenic dysfunction. Figure 4 is an example of spleen tissue imaged first by MPM and then compared to PB and H&amp;E stained slices of tissue in similar regions. These examples are representative of the dataset. SCD and WT images are shown for comparison. Though several potential optical biomarkers were observed, the distinction of disease-state tissue chosen to be analyzed further in this study is the notable increase of nodules of long-wave emission (Fig. 4 white arrowheads). The additional biomarkers include: higher fluorescence intensity in the SCD tissue, indicating higher cellular density; morphological features such as voids (Fig. 4 arrows); the demonstration of a higher degree of disruption of cellular organization; and higher hemosiderin deposition in SCD tissue. In order to analyze some of the features pointed out in this section, an object counting algorithm as described above was developed to segment objects by color for direct and repeatable object counting and morphological analysis.

Fig. 4

Comparison of (top) typical wild type (WT) and (bottom) SCD depth resolved splenic tissue (a, e) imaged by MPM and (b, f) single slices, (c, g) stained with Prussian blue (PB) and (d, h) stained with hematoxylin and eosin. Voids (white arrows) were observed in the SCD tissue. Nodules of long-wave emission (white arrowheads) were observed in both samples types, but at higher density in SCD samples compared to WT. In the histologically prepared slices, brown hemosiderin deposits (white arrowheads) and blue stained iron deposits (yellow arrowheads) were also observed and were suspected to be the same object identified as nodules of long-wave emission in MPM. Scale bar indicates $200\mu \mathrm{m}$. Rotating z-stacks of spleen tissue, (still shown on the left) are available as multimedia files. Video 1, (MPEG, 6568 KB) [URL: http://dx.doi.org/10.1117/1.JBO.20.6.066001.1] shows the WT, and Video 2 (MPEG, 4798 KB) [URL: http://dx.doi.org/10.1117/1.JBO.20.6.066001.2] shows the SCD.

3.3.

Object Analysis

The iron deposits, first spectrally distinguished, were analyzed and compared to iron deposits found via mechanical sectioning and optical microscopy methods. Objects identified as potential optical signatures associated with disease state were compared by basic morphological aspects. Figure 5 shows the representative results of the object segmentation algorithm described in Sec. 2.3. The cross-sectional area of the nodules of long-wave emission in MPM images of SCD tissue and the blue-stained iron deposits of PB images of SCD tissue were measured. The cross-sectional area of MPM long-wave nodules was determined to be $22.4\pm 7.3\mu {\mathrm{m}}^2$ and the area of PB stained iron deposits was found to be $25.6\pm 8.0\mu {\mathrm{m}}^2$. Further, object densities were computed for each frame and individual sample densities were computed by averaging over several frames. Consistently, the mean object density in this exploratory population for SCD tissue by any imaging method was higher than densities computed for WT samples. The mean object densities for each mouse found for each imaging method are shown in Table 3.

Fig. 5

Examples of typical object segmentation results for each imaging method and tissue type. (b–l) Iron deposits are highlighted as such: (a–d) pink for MPM, (e–h) yellow for PB and (i–l) green for H&E. It can be seen from (b–l) the highlighted regions that the SCD (bottom of each block) image contains higher numbers of iron-complex deposits when compared with the WT (top of each block). Scale bars are $200\mu \mathrm{m}$.

Table 3

Object densities computed for each sample according to imaging method. Densities are computed as the mean object density of several frames from each sample and reported as thousand objects per square mm along with standard error in parenthesis.

The significance of these results, including the spectral emission, size of objects and density of objects, is discussed further in the next section.