2.1.

Tissue Collection

Animal procedures were performed in accordance with a protocol approved by the University of Wisconsin–Madison Institutional Animal Care and Use Committee. Mice were housed and handled in AAALAC-accredited facilities in accordance with guidelines from the National Institutes of Health (NIH) Guide for Care and Use of Laboratory Animals. All female mice were purchased from Jackson Laboratories (C57Bl/6J, 00064). In this study, twenty 8-week-old female C57Bl/6 mice were fed either a control (normal) diet (ND, 5% kcal from fat; Teklad Global #2018) ($n=10$) or HFD (60% kcal from fat, Test Diet #58126) for 16 weeks to induce obesity ($n=10$). Purified diets contained equal amounts of vitamins and micronutrients. All mice were given water and the same chow ad libitum. To generate tumors, $1\times {10}^6$ EO771 cells, which were derived from a spontaneous adenocarcinoma in a C57Bl/6 mouse,³⁴ were suspended in matrigel and injected into the inguinal mammary glands of HFD ($n=5$, HT) or ND-fed ($n=5$, NT) mice.

The healthy tissue samples comprised the entire fourth mammary gland and were $\sim 500{\mathrm{mm}}^3$ in size. When tumors reached 1.5 cm in diameter, mice were euthanized and a quarter of the tumor, $\sim 130{\mathrm{mm}}^3$ in size, including both the edge as well as the interior of the tumor, was sectioned. Samples were snap frozen immediately upon excision in optimal cutting temperature (OCT) medium in a bath of methanol and dry ice. Once frozen, the blocks were stored at $-80°\mathrm{C}$ and sent on dry ice to Tufts, where they continued to be stored in a $-80°\mathrm{C}$ freezer until imaging. An identical protocol was followed for tissue freezing and all samples were shipped to Tufts at the same time. All samples were imaged within a period of a month from each other.

2.2.

Imaging

An identical protocol was followed for tissue thawing and preparation prior to imaging. Specifically, each tissue was cut in half immediately after being taken out of the $-80°\mathrm{C}$ freezer, and half of the sample together with OCT were placed in a petri dish, allowed to thaw for at least 5 min at room temperature, then washed in phosphate-buffered saline (PBS) twice to remove the OCT. The other half was placed back in the $-80°\mathrm{C}$ freezer. The thawed tissue was placed on a glass-bottom dish with the incision side placed against the coverslip for imaging. To keep the samples hydrated, one drop of PBS was added on top of each sample.

Images ($512\times 512\text{pixels}$; $387.5\times 387.5\mu \mathrm{m}$) of the excised breast tissues were obtained using a Leica TCS SP8 confocal microscope equipped with a tunable (680 to 1300 nm) femtosecond laser (InSight Deep See; Spectra-Physics; Mountain View, California) and a water-immersion 40× objective (NA 1.1). The collagen SHG signals were measured using 920-nm excitation and collected by an external HyD detector with a $460\pm 20\text{-}\mathrm{nm}$ emission band-pass filter. A $1162.5\text{-}\mu \mathrm{m}\times 1162.5\text{-}\mu \mathrm{m}$ composite image stack, which is composed of nine adjacent stacks, was acquired from each sample over a depth of $80\mu \mathrm{m}$ with a step of $2\mu \mathrm{m}$. Spectral images were collected using 740- and 860-nm excitation wavelengths, over emission wavelengths in the 390- to 750-nm range, with a detection bandwidth of 10 nm, using an internal HyD detector. Spectra from 12 fields were collected for each of the NN and HN groups, while 18 spectral images were acquired from each of the NT and HTgroups. Spectra and spectral images were acquired from five independent tissue blocks from each of the HN, NT, and HT groups, and four independent tissue blocks from the NN group. One of the NN tissue blocks was used to optimize image acquisition parameters. To minimize bias introduced in our calculations based on the specific regions that we imaged, we selected our imaging regions following a 2-D tile scan that covered at least a $5\text{-}\mathrm{mm}\times 5\text{-}\mathrm{mm}$ region of the tissue and selected an area that represented the main tissue features (e.g., fiber rich versus fatty rich regions).

2.3.

Image Processing and Analysis

2.4.

Tissue Staining

Hematoxylin and eosin (H&E) and anti-$\alpha$-smooth muscle actin (SMA) stainings were performed on both tumor and nontumor tissues by UWCCC Experimental Pathology Laboratory. H&E staining was used to evaluate the tissue structures while α-SMA is a marker for cancer-associated fibroblasts, which could be one source of collagen remodeling within the tumor microenvironment. Tumors were fixed in 10% neutral buffered formalin for 24 to 48 h, embedded in paraffin and sectioned at $5\mu \mathrm{m}$. Paraffin-embedded sections were rehydrated in graded alcohol and incubated with 0.5% ${\mathrm{H}}_2{\mathrm{O}}_2$ for 20 min. Sections underwent antigen retrieval in 0.01 M citrate buffer ($\mathrm{pH}=6.0$) at 95°C for 20 min and stained using the M.O.M. Kit (Vector BMK-2202) according to package instructions. Tissue sections were incubated with anti-α-SMA antibodies (Sigma Aldrich, A2547) followed by Alexa Fluor 546 goat anti-rabbit IgG (Life Technologies, A11010), then counterstained with $250\mathrm{ng}/\mathrm{ml}$ 4′,6-diamidine-2′-phenylindole dihydrochloride (DAPI) for 5 min. Slides were mounted with the Slow-Fade mounting kit (Life Technologies, S2828). Sections were imaged using a Nikon Eclipse 80 t microscope with NIS Elements BR 3.2 software.

2.5.

Statistics

All image processing was performed in MATLAB_R2018a software. Statistical analysis was performed in JMP Pro 14 (SAS Institute, Cary, North Carolina). Significance in the difference of the values of the different metrics among different groups was calculated based on one-way ANOVA and post-hoc Tukey tests. Comparisons were performed to assess differences either as a result of the presence of tumor or in response to diet (i.e., NN was compared to HN and NT groups, HN was compared to NN and HT, NT was compared to NN and HT). Significant differences at $\alpha =0.05$ or lower are reported. Five parameters (minimum, lower quantile, median, higher quantile, and maximum) were used to generate the box plots, and mean ± standard error of the mean were used for the bar graphs.

3.1.

Collagen Fibers Are More Aligned in Tumor Tissues than in Normal Tissues, while Obesity Further Enhances the Alignment Level Preferentially in Tumor Tissues

Representative SHG stacks ($1162.5\mu \mathrm{m}\times 1162.5\mu \mathrm{m}\times 80\mu \mathrm{m}$) and SHG optical sections ($1162.5\mu \mathrm{m}\times 1162.5\mu \mathrm{m}$) from tumor- and nontumor-bearing breast tissues from mice fed either a normal or HFD are shown in Fig. 1. The presence of more highly aligned and less dense fibers in tumor bearing tissues is easily perceived from these images. The corresponding 3-D variance coded optical sections enhance visualization of these features, with fibers represented by bluer hues (lower variance/better alignment) and occupying a smaller fraction of the field in the tumor tissues. The dotted region in the upper left corner of the NT (normal diet fed, tumor) images corresponds to cross-sections of fibers oriented primarily along the transverse direction, as indicated more clearly in Fig. S1 in the Supplementary Material.

Fig. 1

Representative volume renderings (pseudocolored green) (left column), 2-D optical sections (middle column) and corresponding 3-D variance color-coded collagen fibers (right column) from the four different tissue types. (a) NN, (b) HN, (c) NT, and (d) HT. Scale bar and color bar are the same for the panels in columns 2 and 3. $\text{Scale bar}=200\mu \mathrm{m}$.

The changes in alignment are quantified first using the 3-D stacks and the corresponding 3-D directional variance metrics (Fig. 2), calculated using a $5.3\times 5.3\times 6\text{-}\mu {\mathrm{m}}^3$ localized window, which yielded significant levels of differences among different tissue types (Fig. S2 in the Supplementary Material). The decrease in sensitivity of the 3-D variance to report on differences between some of the groups as the window size increases reflects the presence of heterogeneities in alignment that are characterized more accurately with smaller window sizes. The averaged PDFs of the localized 3-D variance values of each sample within a group are shown in Fig. 2(a), clearly demonstrating significant shifts in the shape of these distributions toward smaller values within tumor tissues (NT and HT) compared to normal ones (NN and HN). While the distributions for normal breast tissues appear fairly similar independent of the diet to which the mice were exposed to (NN and HN), the shape of the 3-D variance PDF corresponding to tumor tissue from mice fed a HFD (HT) is even further weighted toward smaller values when compared to the NT one. The mean 3-D variance, the PDF’s peak 3-D variance value, its skewness (defined as the ratio of the widths to the right and the left of the location where a perpendicular drawn from the peak of the distribution crosses the FWHM line), and FWHM range report these trends [Figs. 2(b)–2(e)]. 3-D variance PDF shapes vary considerably among optical sections from a given tissue group and all the PDFs are shown in Fig. S3 in the Supplementary Material. Within the normal tissues, we observe that fields that include a duct or a significant number of adipocytes tend to have broader and/or low-value shifted PDFs. While it is harder to observe variations in the PDF shapes of the tumor tissues that depend on a somewhat consistent fashion on the level of cellularity or adipocity, it is clear that these PDFs are shifted to lower 3-D variance values when compared to the normal tissues, and we see such shifts more consistently for the HFD-fed tumor-bearing mice. The KLD values reported in Fig. 2(f) illustrate that the 3-D variance PDF differences are highest for the tumor versus normal comparisons (NN versus NT and HN versus HT) with the HFD leading to significant enhancement in those differences.

Fig. 2

(a) Averaged PDFs of the voxel-wise 3-D collagen fiber variance after KDE. (b) The mean value, (c) the peak of the distribution, (d) its skewness, and (e) the FWHM range are shown. * and ** denote significance at $a=0.05$ and $a=0.01$, respectively. The color with which ** is indicated denotes the comparison group. (f) KLD values for corresponding group comparisons. Each box plot conveys the median, the first and third quantiles, and the minimum and maximum values.

3.2.

Tumor Tissues Have Lower Collagen Fiber Density and Less Correlated SHG Texture Features

While the HFD-fed mice in our studies gained significantly more weight than those fed the ND [Fig. S4(A) in the Supplementary Material], and the tumor weights in HFD-fed mice were significantly larger than those from ND-fed mice [Fig. S4(B) in the Supplementary Material], the tumors appeared similar for the different diet-treatment groups using H&E staining [Figs. S4(C)–S4(D) in the Supplementary Material], as well as the expression of positive cancer-associated fibroblasts via immunohistochemical $\alpha$-SMA staining [Figs. S4(E) and S4(F) in the Supplementary Material]. Thus, these traditional 2-D staining procedures did not suggest differences in extracellular matrix remodeling. To further characterize 2-D SHG image texture, we used a GLCM-based approach. Representative SHG optical sections, along with corresponding cloned SHG images are shown in Fig. 3. From the SHG segmented images [Figs. 3(a2)–3(d2)], it is easy to perceive a significant decrease in fiber density in the tumor tissues of the breast cancer animal model we employed for this study. This decrease appears to be enhanced with obesity (i.e., HFD), as shown quantitatively in Fig. 3(h), where fiber density is reported as the ratio of SHG-positive pixels over the number of all the pixels within each SHG stack acquired from each tissue. A significant decrease in the $D_{50}$ value extracted from correlation analysis of the SHG image texture features is also highlighted in Figs. 3(a4)–3(d4), resulting probably from a combination of changes in the density and size of collagen fibers. Significant differences in the $D_{50}$ value PDFs for the different tissue types from all the images we acquired are shown in Fig. 3(e) and quantified by corresponding differences in the mean values of these distributions [Fig. 3(f)]. There are significant differences between tumor- and nontumor-bearing groups (NT versus NN and HT versus HN); however, an HFD does not lead to significant differences between the HT and NT groups, unlike the 3-D variance metrics [Fig. 3(g)]. Nevertheless, the KLD comparisons indicate that differences are indeed higher between the HN and HT groups than the NN and NT groups.

Fig. 3

Textural analyses and representative figures. Representative detected SHG intensity images from each tissue group (a1-d1), along with corresponding SHG-positive masks (a2-d2), cloned SHG images (a3-d3), and correlation intensities as a function of pixel distance (a4-d4). (a1-a4) Normal diet, no tumor; (b1-b4) HFD, no tumor; (c1-c4) normal diet, tumor; (d1-d4) HFD, tumor. ($1\text{pixel}\approx 0.76\mu \mathrm{m}$). Asterisks and blue lines signify for each plot the distance where correlation value declines to 50% of its initial value (D50) for the corresponding image. (e) PDF of D50 values from all optical sections assessed for each treatment group. (f) Box plots of the mean D50 values of each group. (g) KLD values of each D50 PDF group pairs. (h) Fiber density within the image stacks assessed for each group. $\text{Scale bar}=50\mu \mathrm{m}$; colorbar and scalebar are the same for columns (1) to (3), panels (a)-(d). * and ** denote significance at $a=0.05$ and $a=0.01$, respectively. The color with which ** is indicated denotes the comparison group. The intensities of images listed in columns (1) and (3), panels (a)-(d) were rescaled and separated into 64 different intensity levels, which were assigned with different pseudocolors to range from blue to red.

3.3.

Collagen Associated Fluorescence is Significantly Higher within Breast Tumor Tissues of Obese Mice

Average fluorescence emission spectra from each group of breast tissues we considered are shown from the SHG-positive pixels for 740- and 860-nm excitation in Fig. 4. The SHG peak is present at 430-nm emission within the 860-nm excitation spectra, as expected. All spectra were normalized to their corresponding 430-nm emission peaks so that the fluorescence signal intensity is effectively normalized to the collagen content, representing essentially a measure of fluorescent collagen-cross-link density. Especially, for nontumor-bearing mice, there is a strong peak in the 600 nm range of the spectrum, which likely emanates from porphyrins in the chow used to feed the mice and has been reported previously.^40,41 While all mice were fed the same chow, the absence of strong porphyrin contributions within the tumor tissues is likely associated with changes in heme biosynthesis, which have been reported for a number of related enzymes for several cancers, including human breast cancers.⁴² To remove contributions from this peak, we performed spectral deconvolution to extract the shapes of the fluorescence emission that describe the spectra, which are shown in Figs. 4(b) and 4(e). The porphyrin spectral shape contributions are consistent for the two excitation wavelengths; however, the remaining fluorescence emission contributions have distinct shapes for the two excitation wavelengths, with the spectrum acquired at 740-nm excitation having broader features. This likely indicates the presence of more than one fluorophore in the tissue. However, the overall fluorescence intensity of this nonporphyrin-like emission follows very similar trends for the different tissue types for both wavelengths, as shown in Figs. 4(c) and 4(f). These panels indicate that obesity increases significantly the fluorescence that is present within tumor tissues selectively, and this increase leads to the detection of statistically different levels of fluorescence between tumor and nontumor tissues in the case of the HFD-fed mice, as well as between the tumor tissues from the mice fed with HFD or normal diet.

Fig. 4

Graphs of CAF spectra at (a) 740 nm and (d) 860 nm excitation acquired with emission in the 390- to 790-nm region (steps: 10 nm) for NN, HN, NT, and HT groups. Spectral deconvolution results of the emission spectra for (b) 740 nm and (e) 860 nm excitation. Values for the weights of peak 1 in each group are shown in (c) and (f) as mean ± standard error of the mean (SEM). (*$a=0.05$, **$a=0.01$). The color with which * is indicated denotes the comparison group.