2.1.

Breast Cancer Tissue Samples

The samples used throughout this study are primarily tissue microarray (TMA) slides, which are composed of 0.6- to 1.0-mm-diameter samples from cylindrical cores of paraffin embedded tissue specimens sectioned into 5-μm thick slices. One advantage of these samples is that due to their thickness the effect of subsequent scattering or absorption of SHG light is negligible [the absorption coefficient is ${\mu }_a\sim 20{\mathrm{cm}}^{-1}$, while the reduced scattering coefficient is ${\mu }_s^{\prime }\sim 10{\mathrm{cm}}^{-1}$.^30,31 Each 5-μm thick, $\sim 1.0\text{-}\mathrm{mm}\text{-}\text{diameter}$ specimen corresponds with a different patient/tumor and is mounted on a slide, with several tens of samples on each slide. The tumor specimens are staged by a certified pathologist, and every tenth section of the TMA block is used to create an H&E stained slide, which is analyzed by the pathologist to ensure that each sample is within the tumor and demonstrates properties of the diagnosis. Hence in the direction perpendicular to the plane of the sample, each tissue section is at least 50 μm from the tumor surface. In the plane of the sample, sections are chosen that are entirely within the tumor and do not contain significant healthy tissue, based upon inspection of the adjacent H&E stained slides. However, the distance from the outer edge of the core to the tumor surface is not known. Consequently any significant variation of tumor SHG properties with distance from the tumor edge may increase the variance in our measurements.

The breast cancer tissue microarray samples used throughout this study came from two sources. Samples of healthy breast tissue, as well as various grades and stages of IDC and ILC, were purchased from Biomax (Rockville Maryland, slides BR1921, BR20830, BR805, BR961, and BR954). Pure DCIS and IDC tissue microarray slides were received from Dr. Ping Tang and created in the Department of Pathology and Laboratory Medicine at the University of Rochester Medical Center.

In addition to the tissue microarray samples, several LCIS and ILC samples were generated from paraffin embedded tissue blocks, cut to 5 μm thickness. In these samples we utilized an adjacent H&E stained slide to identify the region of tumor, then generated a single SHG field of view from the center of tumor in the unstained slide. Table 1 summarizes which samples were used to create each data set.

Table 1

Table of the tissue microarray slides that were used, along with how many samples from each slide were used to create each set of results. The number given (i.e., “BR1921”) is the part number of the Biomax microarray slide, while “URMC” refers to the full sections or tissue microarray (TMA) slide manufactured by the University of Rochester Department of Pathology. Note that samples are “double counted” in that a sample that is IDC T1 and N0 would count in both the IDC T1 column as well as the IDC N0 column.

2.2.

Sample Grading and Staging

Invasive breast tumors are graded on a three-state classification system dependent upon tubule formation, nuclear polymorphism, and mitotic count, where Grade 1 has the most differentiated tissue with the best prognosis and Grade 3 is the least differentiated with the worst prognosis. The TNM staging system is a method of tumor classification determined by the size of the primary tumor ($T$), regional lymph node involvement ($N$), and presence of distant metastases ($M$). Staging of the primary tumor ($T$) as well as the metastatic events in the lymph node ($N$) incorporates multiple degrees of staging as opposed to metastatic staging ($M$), which is a binary system determined simply by the presence or absence of a distant metastases. The primary tumor staging is split into categories based on tumor diameters of between 1 to 20 mm ($T1$), between 20 to 50 mm ($T2$), and greater than 50 mm ($T3$). Tumors that have penetrated the chest wall or skin, independent of the size of the tumor are staged as $T4$ tumors. The lymph node (LN) staging scale begins at $N0$ indicating there are no metastatic events present in the LN that are greater than 0.2 mm or 200 cells. The two middle stages, $N1$ and $N2$, describe increasing involvement in the axillary LNs, or metastasis to the internal mammary LNs without spread to the axillary LNs. Specifically, $N1$ tumors have either micrometastases, one to three axillary LN metastases, or sentinel LN metastases detected through biopsy. $N2$ tumors are characterized by four to nine axillary LN metastases, or clinically detected sentinel LN metastases. $N3$ stage is the most far reaching metastatic LN events, including either greater than 10 axillary events with at least one event greater than 2 mm, infraclavicular LN metastases, or clinically detected internal mammary and axillary LN metastases.³² Note that grading information was not available for ILC samples, which are uniformly low grade, nor was M stage information available because distant metastases for this tumor type, if they occur, usually appear many years after initial diagnosis and hence are not known during preparation of the TMA. All samples were classified based upon the aforementioned grading and staging scheme by a certified pathologist, and all parties were blinded to the classification of the samples during image acquisition and analysis.

2.3.

Imaging

The excitation light is a Spectra Physics MaiTai Ti:Sapphire laser at 810 nm, with 100 fs pulses at 80 MHz. It is directed to the sample through an Olympus BX61WI upright microscope, with beam scanning and image acquisition controlled by an Olympus Fluoview FV300 scanning system. Before entering the scan box, the laser passes through a Berek compensator (Model 5540, New Focus) adjusted such that the excitation light reaching the objective lens is circularly polarized (verified as $<2\%$ variation in transmitted power versus angle of an analyzer set after the dichroic and before the objective lens). An Olympus UMPLFL20XW water-immersion lens ($20\times$, 0.5 NA) is used to focus excitation light and capture backward propagating SHG signal. After passing through the objective, the signal is separated from the excitation beam using a 670-nm short-pass dichroic mirror, filtered using a 405-nm filter (HQ405/30m-2P, Chroma, Rockingham, Vermont), and collected by a photomultiplier tube (Hamamatsu HC125-02). In the forward direction an Olympus 0.9 NA optical condenser was used to collect the signal, reflected by a 475-nm long-pass dichroic mirror (475 DCSX, Chroma, Rockingham, Vermont) in order to remove excess excitation light, filtered by a 405-nm filter (HQ405/30m-2P, Chroma, Rockingham, VT) and captured by photomultiplier tube (Hamamatsu HC125-02).

2.4.

F/B Analysis

Forward and backward images were simultaneously collected as a stack of 11 images spaced 3 μm apart, with a 660-μm field of view from unstained samples of IDC, DCIS, ILC, LCIS, and healthy breast tissue. One stack was made from the geometric center of each TMA sample. For the non-TMA samples, the image field in an unstained section was chosen based upon imaging of an adjacent H&E stained section to identify the center of the visible tumor structure. Image analysis was conducted in with ImageJ Software.³³ Each stack was maximum intensity projected, serving as an “autofocus” for the effectively single layer of collagen that exists in these 5-μm sections and producing a single-image pair for each sample. Projected images were background subtracted using a maximum intensity projection of an 11 image scan taken with a closed shutter. Day to day variations in optical alignments were accounted for by imaging one tissue sample (not included in the data pool) each day as a standard SHG sample and determining a normalization factor for each detector pathway that rendered the signal from that standard sample constant over time. For each image, a common threshold was applied to all images taken in that imaging session and chosen by a blinded observer to distinguish collagen pixels from background pixels. The threshold of the backward-collected image was then used to create a mask, in which all of the pixels above threshold were set to 1, and all of the pixels below the threshold were zero. That mask was then used to exclude background (i.e., non-fiber) pixels from consideration and the average pixel value of fiber (i.e., non-background) pixels was calculated from an F/B ratio image. Due to the small sample thickness (5 μm), we assume that the measured F/B ratio is due solely to the original ratio of forward-emitted and backward-emitted SHG and is not significantly affected by subsequent backscattering of either component.^30,31 Note that unless otherwise specified (see Sec. 2.5) all quantification of SHG F/B is performed in unstained specimens.

2.5.

Analysis of Spatial Heterogeneity of F/B

H&E stained sections of DCIS, LCIS, and healthy breast were imaged with a CCD camera (Olympus, SP-350) through the microscope eyepiece and an SHG F/B image of the same approximate field of view was generated as described above. Based upon the tissue H&E staining patterns imaged by the CCD camera, in the SHG images regions of interest within 50 μm of ductal or lobular structures were outlined with ImageJ, producing at the same time regions of interest greater than 50 μm from ductal or lobular structures. The average F/B ratio within those regions of interest (per-ductal, peri-lobular, or distant from either feature) was then calculated as described in the previous section.

2.6.

Effects of H&E Staining on SHG F/B

In DCIS, LCIS, and healthy breast samples, pairs of adjacent 5-μm sections were cut and mounted on glass slides, and one of the pair was H&E stained. SHG F/B of each adjacent section were analyzed as described in F/B Analysis above and compared to determine the effects of H&E staining on SHG F/B. In addition to quantifying the effects of staining on SHG F/B averaged over the entire 600-μm field of view, peri-ductal regions, peri-lobular regions, or regions distant from either feature were independently analyzed as described in Analysis of Spatial Heterogeneity of F/B above.

2.7.

Statistics

Statistical analysis was performed using Prism 5 software (GraphPad, La Jolla, California). Statistically significant data was defined as a $p$-value less than 0.05. Comparisons of two groups were analyzed using unpaired student’s t-tests. If the F-test for this data returned a value $<0.05$, a Mann-Whitney nonparametric t-test was used. For grouped analysis a one-way ANOVA was used with a Newman-Keuls multiple comparison post-hoc test to determine the difference between groups. If the Barrett’s test for equal variance results in a $p$-value less than 0.05, a non-parametric Kruskal-Wallis test with a Dunn’s post-hoc test was performed.

4.1.

Biology of Ductal Carcinoma

The changes in SHG F/B in Fig. 2(a) demonstrate that there is no significant alteration in fibrillar collagen microstructural properties between healthy breast tissue and DCIS, but that there is an alteration in them as the tumor progresses from an in situ to an invasive carcinoma. This is consistent with the fact that the majority of ${\mathrm{SHG}}^{+}$ collagen fibers in healthy tissue are surrounding the ducts, so as the tumor cells exclusively fill in the ducts in DCIS, the surrounding collagen microstructural properties remain relatively unaffected. It is only when the tumor cells begin to invade the breast tissue outside of the duct that they induce changes in microstructural properties of the surrounding fibrillar collagen, as observed in the statistically significant difference between SHG F/B of IDC and healthy breast tissue [Fig. 2(a)].

Taking a closer look at this phenomenon, we examined whether invasion through the basement membrane into the surrounding tissue (i.e., the transition from DCIS to IDC) causes an immediate change in the F/B ratio by comparing SHG F/B to the grade of the IDC tumor [Fig. 2(b)]. Our results showed that in Grade 1 IDC the F/B ratio is still not significantly different from the DCIS tissue nor the healthy tissue, but as IDC progresses to a Grade 2 or 3 tumor, the SHG F/B ratio becomes significantly different. This suggests that the overall fibrillar collagen microstructural properties as determined by the F/B ratio are constantly evolving away from the healthy state as the IDC tumor cells become less differentiated.

Regarding the changes in primary tumor F/B seen throughout IDC tumor staging, there is interestingly no difference between the IDC samples that are at different primary tumor sizes ($T$ stage), although they are all different than the healthy tissue cohort. Presumably, the transition from healthy to tumor SHG F/B occurs when tumors are small enough to not be included in our cohorts of tumor samples, which means that it occurs when the tumor diameter is less than 1 mm. Once tumors are greater than 1 mm and hence large enough to be included in the $T1$-4 grading scheme, the size of the primary tumor does not influence collagen microstructural properties as quantified by SHG F/B. Combined with the fact that F/B of Grade 1 IDC does not differ significantly from healthy tissue, this suggests that during early tumor growth and progression, there is a complex interplay between tumor size and grade, whereby small high-grade tumors may produce a more significantly altered matrix than larger low-grade ones.

As tissues progress from healthy to $N0$ IDC (IDC with no LN involvement), the SHG F/B ratio statistically significantly decreases, and remains low as tumors progress to $N1$ (small metastatic events or $\le 3$ axillary events) and $N2$ (4–9 axillary events or clinically detected internal mammary LN metastases). Interestingly, as tumors progress to $N3$ stage ($>10$ events, further reaching LN metastases, or metastases to both internal mammary and axillary LN) SHG F/B becomes indistinguishable from healthy F/B [Fig. 3(c)] and is markedly different from the pooled group of other IDC tumors [Fig. 3(d)]. This evolution of SHG F/B suggests two intriguing possibilities. The first is based upon the view that $N$-stage progression is a chronological progression, with a single IDC advancing from having no LN metastases, through the subsequent $N$ stages, and ending as an IDC with distant LN metastases. From that perspective, this data demonstrates that during tumor progression fibrillar collagen microstructural properties evolve in ways that affect SHG F/B and raises the possibility that this evolution may influence efficient travel of metastatic tumor cells to distant LNs. After the initial evolution of collagen microstructural properties from normal to abnormal as the tissue progresses from healthy to an $N0$ tumor, there is a subsequent “normalization” back to the microstructural properties of healthy fibrillar collagen (as quantified by F/B) that may facilitate distant metastasis and lead to the $N3$ highly metastatic case. Alternatively, these matrix changes may not represent an evolution of individual tumors in time, but suggest instead that those tumors that do not undergo an initial decrease in SHG F/B as the tissue first progresses from healthy to cancerous and maintains a high F/B, may be more likely to produce extensive distant metastases. Although the F/B ratio reveals that there is a similarity in the collagen microstructural properties between the $N3$ tumor and healthy tissue, inspection of SHG images of healthy and $N3$ samples reveal that these similarities do not extend to the overall morphology of the fibers, as these are noticeably different between the two tissue states as shown in Fig. 4(d).

In traditional $M$ staging, IDC tumors are divided into $M0$ (no known metastases to distant non-lymph-node organs) and $M1$ (any number of metastases to distant non-lymph-node organs). For IDC tumors, there is no relationship between SHG F/B and $M$ status group [Fig. 3(b)]. This suggests either that there is no relationship between fibrillar collagen microstructural properties and the ability of the IDC to metastasize to different (non-LN) organs, or that the act of pooling all numbers of metastases to any non-LN organs into the single category of “$M1$” has hidden any subtle differences such as those that were evident in IDC when classified according to the more finely divided $N$ stage classification scheme.

4.2.

Biology of Lobular Carcinoma

Like IDC, ILC causes a change in the fibrillar collagen microstructural properties in breast tissue, as indicated by a decrease in the F/B ratio relative to healthy tissue [Fig. 4(a)]. However, LCIS F/B is significantly different from healthy tissue F/B whereas DCIS is not [Fig. 4(b)]. This difference in matrix alterations is consistent with the significant difference in behavior of the two diseases. Unlike DCIS, which is categorized as cancer and if left untreated is likely to progress to IDC,³⁴ LCIS is described as a development of abnormal cells, which is not believed to progress to ILC (although it does correlate with increased likelihood of the patient developing ILC or IDC later in life).³

In contrast to IDC, the stage of LN involvement [Fig. 4(c)] has no effect on the F/B ratio of ILC, and all $N$ stages are statistically different from healthy tissue. The stable F/B ratio throughout increasing LN involvement could result from the ILC’s characteristic lack of a strong inflammatory response after penetrating the basement membrane,³⁵ a response that would otherwise aid in the remodeling of the stromal collagen through recruitment of fibroblasts and macrophages.^13,14 In ILC there is a change in SHG F/B with tumor $T$ stage, specifically a loss of F/B difference relative to healthy tissue that occurs in the $T4$ stage of tumor progression [Fig. 4(c) and 4(d)], which is when the tumor interacts with tissues surrounding the breast, such as chest and skin. The marked difference between behavior of IDC and ILC at this stage contrasts with the close quantitative similarity of F/B values between the tumor types at earlier $T$ stages [compare Figs. 3(a) and 4(d)]. This suggests that the mechanisms by which ILC and IDC recruit stromal cells to produce the collagenous matrix is subtly different, a difference that only becomes evident once the source of stromal cells shifts from breast tissue to the chest wall and/or the skin.