1. Introduction

About 66% of breast cancer patients are diagnosed with localized disease [1]. While adjuvant chemotherapy reduces the risk of metastatic recurrence in high-risk breast cancer patients, it can significantly impact quality of life and may not be as beneficial in low-risk breast cancer patients. However, the morbidity associated with breast cancer leads to a majority of breast cancer patients with localized disease receiving adjuvant chemotherapy and enduring its side effects. Gene expression signatures, such as Oncotype DX, can evaluate the benefits versus risks of chemotherapy in estrogen receptor positive (ER+), HER2-negative breast cancer patients; however, such tests are expensive and not easily accessible at all clinics [2,3]. More recently, the detection of circulating tumor DNA in the plasma of patients with breast cancer was shown to predict metastatic recurrence in breast cancer with high sensitivity and specificity [4]; however, such methods require a sufficiently large tumor that can shed biomolecules into the blood stream. There is a need for approaches that can noninvasively evaluate the primary tumor, irrespective of size, and determine the risk of metastatic recurrence so that patients with localized disease that will not metastasize do not receive adjuvant therapy.

Diffuse reflectance spectroscopy (DRS) is a low-cost and non-invasive tool that can provide quantitative measures of bulk tissue structure and function. DRS uses optical fibers to illuminate tissue with non-ionizing visible to near-infrared wavelengths of light and collect the diffusely reflected light from the tissue bulk. The light incident on tissue is subject to scattering and absorption from tissue constituents that play a key role in disease progression. Specifically, the diffusely reflected light contains information regarding scattering from cells, cell organelles, and collagen as well as absorption by hemoglobin and other tissue-specific absorbers. Several studies have used DRS to determine the accuracy of distinguishing between benign and malignant breast tissue [5–10], and identify surgical margins [11–13] in human breast tumor samples.

Here, we used DRS in a controlled animal study to determine whether there are measurable optical differences in tumors with differences in eventual metastatic outcome. We utilized a panel of isogenic breast cancer cell lines of varying metastatic potential – 4T1, 4T07, 168FARN, and 67NR – to grow tumors in mice and perform DRS measurements. Each cell line in this panel can complete only specific steps in the invasion-metastasis cascade (primary tumor formation, invasion, intravasation, extravasation, and metastatic node formation), with only the 4T1 capable of robust metastases. Previous work from our lab has shown that there are significant metabolic differences between these cell lines in vitro [14]. Our results demonstrate specific trends and significant differences in tissue scattering, vascular oxygenation and hemoglobin content of these cell lines. In addition to acquiring optical spectra from the four primary tumors, we also acquired spectra from tumors grown from clones of the highly metastatic 4T1 cells that had metastasis-promoting genes knocked down. These preliminary results in murine tumors demonstrate that DRS can potentially identify microenvironmental differences between indolent and aggressive primary tumors.

2. Materials and methods

2.1 Cell culture and tumor xenografts

The cell lines used in this study - 4T1, 4T07, 67NR, and 168FARN – were originally derived from a spontaneous breast tumor growing in a Balb/c mouse and were provided by Dr. Fred Miller (Karmanos Cancer Institute) [15,16]. The TWIST gene plays a key role, especially in this panel of cell lines, in promoting metastasis [17]. CRISPR/Cas-9 was used to generate an indolent version of 4T1 by knocking out the TWIST gene (4T1-TWIST KO). In addition to TWIST, we also investigated the effects of silencing two other genes identified in the same study – FOXC2 and CXCR3. A clonal population of 4T1 cells with FOXC2 knocked down (hereafter referred to as FOXC2-KD) and corresponding vector control (FOXC2-VC) were a kind gift from Dr. Sendurai A. Mani (MD Anderson Cancer Center, Houston, TX) [18]. Similarly, the 4T1-CXCR3-KD shRNA knockdown clones and their vector control (CXCR3-VC) were obtained from Dr. Li Yang (National Cancer Institute, Bethesda, MD) [19]. Cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) with the addition of 10% (v/v) fetal bovine serum (FBS), 2 mM L-glutamine, 1% (v/v) nonessential amino acids, and 1% (v/v) penicillin-streptomycin. The cells were cultured in a humidified incubator set to 5% CO₂ and 37°C and passaged when they reached approximately 80% confluence. Cells were used within the first 10 passages for all experiments. Each of the 9 cell lines (approx. 150-250,000 cells suspended in 100 µl of saline; 4 million for the 168FARN) were injected into the flanks of 10 Balb/c mice to grow tumors (n = 90; 10/cell line). All experiments were approved by the Institutional Animal Care and Use Committee at the University of Arkansas (IACUC protocol# 20026).

2.2 CRISPR/Cas9 for the deletion of TWIST gene

The sgRNA guide tool provided by the Zhang laboratory (MIT, Cambridge, MA) was used to identify sgRNA to aim at the TWIST gene. The 20-base pair (5’-TTGCTCAGGCTGTCGTCGGC-3’) sgRNA was cloned into pCasGuide-EF1a-GFP plasmids through the services of OriGene (Rockville, MD). E. Coli bacteria were used to expand the plasmid, which was isolated utilizing the QIAGEN Plasmid Maxi Kit. Plasmid transfection was achieved by seeding 4T1 cells in a 6-well plate at a concentration of one million cells/well for an incubation time of 24 hours. 10 µg of plasmid were added to Lipofectamine 3000 to be added to the 4T1 cells. After 24 to 48 hours, a Nikon TiE fluorescence microscope workstation with a CoolSnap HQ2 camera was used to detect the signals expressing green fluorescent proteins (GFP), indicating cell transfection. A PBS solution was used to suspend the transfected 4T1 cells at a concentration of two million cells per mL, which were subsequently filtered into a FACS tube with a 50 µm filter. The cells were sorted on a FACS Aria III System (BD Biosciences, San Jose, CA). The 5% percent of transfected cells with the highest GFP expression were classified as cells with the greatest CRISPR / Cas9 plasmid concentration. The classified cells were incubated for 7 to 14 days, and the formed cell colonies were separated into 13 clones to produce a cell population, having the least TWIST expression. Western blots were used to confirm TWIST expression in these clones.

2.3 Diffuse reflectance spectroscopy

The DRS system used here has been described in previous studies [20,21]. Briefly, the system consists of a tungsten-halogen lamp (HL-2000, Ocean Optics; Dunedin, Florida) for illumination, a portable spectrometer for spectral acquisition (Flame, Ocean Optics; Dunedin, FL), and a fiber-optic probe (fiber dia. = 200 µm, NA = 0.22; FiberTech Optica, Ontario, Canada) with source-detector separation of 2.25 mm. The diffusely reflected light (475-600 nm) was collected by five peripheral fibers, which surround four central source fibers arranged in a circle (Fig. 2(B) in [21]). Optical spectra from each tumor were recorded when tumors reached a volume of 200 mm³ while animals were under anesthesia (1.5% v/v isoflurane mixed with 100% oxygen). Previous work has shown that this combination of isoflurane and oxygen leads to minimal changes in vascular oxygenation and is comparable to measurements made with no anesthesia [22]. For the four primary cell lines, about 3-5 spectra were obtained from each tumor. The integration time for each spectrum is 150 milliseconds. After completion of studies on the four primary tumor groups, we modified our study protocol to ensure at least 10 spectra were acquired from each tumor to ensure better coverage of the tumor. About 10-20 optical spectra were obtained from each of the vector control and knockdown tumors. Data analysis was performed in MATLAB (MathWorks, Natick, MA). Tumors were excised from animals following optical measurements, placed in optimal cutting temperature (O.C.T.) compound, and frozen down in liquid nitrogen.

 

Fig. 1. A. Representative diffuse reflectance spectra (black line) and the corresponding LUT model-based fit (red line) from tumors drawn from different groups. B. Absorption coefficient and C. Reduced scattering coefficient as a function of wavelength for the different tumor groups.

Download Full Size | PDF

 

Fig. 2. Comparison of functional and structural optical properties across isogenic cell lines of varying metastatic potential. A. Tissue scattering. B. Total hemoglobin content. C. Vascular oxygenation. D. Oxygenated hemoglobin. E. Deoxygenated hemoglobin. Error bars represent standard error of the mean (SEM). * denotes statistically significant differences at p < 0 .05, ** - p < 0.01, *** - p < 0.001, and **** - p < 0.0001.

Download Full Size | PDF

2.4 Quantification of tissue optical properties and statistical analysis

We determined the scattering, total hemoglobin concentration (cHb), and vascular oxygen saturation (sO₂) using an empirical lookup table-based inverse model [23,24] that we have previously employed in animal studies [25–27]. Light scattering in tissue was assumed to have a negative power-law dependence on wavelength: ${\mathrm{\mu}_\textrm{s}}^\mathrm{^{\prime}}(\mathrm{\lambda } )= {\mathrm{\mu}_\textrm{s}}^\mathrm{^{\prime}}({{\mathrm{\lambda }_0}} ).{({{\raise0.7ex\hbox{$\mathrm{\lambda }$} \!\mathord{/ {\vphantom {\mathrm{\lambda } {{\mathrm{\lambda }_0}}}} }\!\lower0.7ex\hbox{${{\mathrm{\lambda }_0}}$}}} )^{ - \textrm{B}}}$, where λ₀ is a reference wavelength at which light absorption is minimum and is set to 600 nm and ${\mathrm{\mu}_\textrm{s}}^\mathrm{^{\prime}}(\mathrm{\lambda } )$ is the wavelength-dependent reduced scattering coefficient. The absorption coefficient is calculated as the linear sum of absorption coefficients of individual absorbers, namely HbO₂ and dHb, and animal skin: ${\mathrm{\mu}_\textrm{a}}(\mathrm{\lambda } )= [{\textrm{Hb}} ][\mathrm{\alpha }{\mathrm{\sigma }_{\textrm{Hb}{\textrm{O}_2}}}(\mathrm{\lambda } )+ ({1 - \mathrm{\;\ \alpha }){\mathrm{\sigma }_{\textrm{dHb}}}(\mathrm{\lambda } )} ]+ [{\textrm{Ml}} ]\textrm{mel}(\mathrm{\lambda } )$ where [Hb] and [Ml] respectively are total hemoglobin concentration and skin absorption. α is vascular oxygen saturation representing the ratio of oxygenated (HbO₂) to total hemoglobin concentration [Hb]. The extinction coefficients of these absorbers have previously been established [28].

An ordinary one-way analysis of variance (ANOVA) followed by post-hoc Tukey HSD testing was implemented to determine statistically significant differences in optical properties between tumors derived from the different parental and modified cell lines. The optical parameters were calculated for each spectrum acquired and the average optical parameters are determined for each mouse for the purposes of plotting optical properties.

We employed a random forest approach to determine if DRS-based optical properties could accurately differentiate between the different tumor groups and discriminate between tumors that could only accomplish specific steps of the invasion-metastasis cascade. While all four tumor groups are capable of primary tumor formation, only 168FARN, 4T07, and 4T1 are capable of intravasation, only 4T07 and 4T1 are capable of extravasation, and only 4T1 can form viable metastatic nodes at distant sites [15,16]. We implemented the TreeBagger class in MATLAB with 120 trees based on Breiman’s original algorithm [29]. A leave-one-mouse-out training strategy was implemented to prevent representation of test mice in the training dataset. The random forests classifiers are trained on optical parameters determined from all optical spectra from each mouse. The label predicted by a majority of the optical parameters for each mouse determined the final predicted label for the test mouse. We used a four-class classifier for differentiating between tumor groups (67NR, 168FARN, 4T07, or 4T1) and a binary classification (positive or negative) for intravasation, extravasation, and metastasis. The ground truth for each stage is based on previous published literature.

2.5 Quantification of lung metastases

To assay for metastatic growth, the 4T1, 4T1-TWIST-KO, 4T1-CXCR3-VC, 4T1-CXCR3-KD, 4T1-FOXC2-VC, and 4T1-FOXC2-KD were injected into the tail vein of Balb/c mice (150,000 cells in 100 µl of saline; 5/group). Mice were euthanized approximately 30 days following injection and lungs were extracted, placed in OCT, and snap-frozen in liquid nitrogen for evaluation of metastatic nodes. To evaluate metastatic nodes, the OCT from the snap-frozen lungs was removed by thawing the tissue in 20 mL of PBS for 15 min at room temperature. Each lung was then placed in a centrifuge tube in 5 mL of Bouin’s solution at room temperature and incubated for 3 days. Metastatic nodes were counted using calipers. Nodes larger than 2 mm were classified as macrometastases and modules smaller than 2 mm were classified as micrometastases.

2.6 Quantification of collagen

Masson’s trichrome is used to stain collagen fibers, fibrin, muscles, and erythrocytes. The stain produces red keratin and muscle fibers, blue collagen, pink cytoplasm, and dark brown cell nuclei. Tumors frozen down following optical measurements were sectioned (10 µm sections), mounted on glass slides and stained with Masson’s trichrome. Tumor sections were imaged with a Nikon TE fluorescence microscope. To calculate the percentage of collagen-positive pixels, RGB images of Masson’s trichrome-stained tumor sections were processed to preserve blue pixels while setting all other pixels to zero. A binary mask of blue pixels was created, and the percentage of collagen-positive pixels was calculated based on the ratio of blue pixels to total pixels within the image. Masking was performed to remove any artifacts due to staining or sectioning, especially at boundaries.

3. Results

3.1 Diffuse reflectance spectroscopy reveals differences in total hemoglobin content and tissue scattering in isogenic breast tumors of varying metastatic outcome

Figure 1(A) illustrates representative DRS spectra at a tumor volume of 200 mm³ from each breast cancer cell line. Please note that the 4T1 vector controls corresponding to CXCR3 and FOXC2 knockdown are not presented here. The solid black line indicates the in vivo measurements and the solid red lines the LUT model fit. Based on the model fits to the DRS spectra, we determined the wavelength-dependent absorption and scattering coefficients for each tumor. Representative absorption spectra from each tumor group are presented in Fig. 1(B). The absorption peaks of hemoglobin are clearly visible in the absorption spectra between 540 and 580 nm. While there do not appear to be large differences in the magnitude of the absorption coefficient across the different tumor groups, some tumor groups, such as the 168FARN demonstrate the prominent dual peaks at 542 and 576 nm associated with oxygenated hemoglobin and hence a more oxygenated tumor. The wavelength-dependent reduced scattering coefficients illustrate differences in both magnitude and slope between the different tumor groups (Fig. 1(C)). Specifically, the 4T1, 4T1-TWIST KO, and 67NR have lower scattering intensities compared with the other tumor groups while the 4T1 has a greater scatter slope compared to all the other tumor groups. The magnitude and slope of the reduced scattering coefficient are typically associated with scatter density and size, respectively [30].

We quantified the tumor optical properties from the absorption and scattering coefficients. The data in each subplot of Fig. 2 are arranged in order of increasing metastatic potential. Tissue scattering presented in Fig. 2(A) is the average reduced scattering coefficient across the entire wavelength range (475-600 nm). The 67NR tumors had the lowest tissue scattering of the four tumor groups and were significantly different from the 168FARN (p < 0.0001), 4T07 (p < 0.0001), and 4T1 (p = 0.0003) tumors. In addition, tissue scattering from the 4T1 tumors was significantly lower than the 4T07 tumors (p = 0.003). Total hemoglobin content (Fig. 2(B)) was lowest in the 4T1 tumors and was significantly different between the 67NR and 4T1 tumors (p = 0.04). The 67NR tumors had the lowest vascular oxygenation of the four tumor groups (Fig. 2(C)). Decomposing the total hemoglobin content into its constituents based on oxygenation revealed that the 67NR tumors had the lowest levels of oxygenated hemoglobin (Fig. 2(D)) and the highest levels of deoxygenated hemoglobin (Fig. 2(E)), which was significantly higher than that of the 4T1 tumors (p = 0.01).

3.2 Supervised classification based on tumor optical properties accurately classifies tumors at each step of the metastatic cascade

Having observed differences in the optical properties of indolent and aggressive breast tumors, we sought to determine if these differences could be leveraged to accurately classify tumors based on their propensity to accomplish each stage of the metastatic cascade. (Fig. 3). Using a leave-one-mouse-out random forest classifier on the tumor optical properties, we first determined if the tumors within each group could be accurately identified. We found that 100% of 67NR tumors (7/7), 90% of 4T1 tumors (9/10), 70% of 4T07 tumors (5/7), and 50% (3/6) of 168FARN tumors were correctly classified into their respective groups. All intravasation-negative tumors (67NR) were correctly classified as such while the model correctly identified 20/23 intravasation-positive tumors across the 168FARN, 4T07, and 4T1 groups. All 67NR tumors (7/7) were classified as extravasation-negative and all 4T07 tumors (7/7) as extravasation-positive while 3/10 4T1 tumors and 3/6 168FARN were incorrectly classified as extravasation-negative and extravasation-positive, respectively. Finally, all 168FARN (6/6) and 4T07 tumors (7/7) were correctly predicted to be metastasis-negative while 5/7 67NR tumors and 9/10 4T1 tumors were identified correctly as being metastasis-negative and metastasis-positive, respectively.

 

Fig. 3. Classification of primary tumors and metastatic stage-specific phenotypes. The results of leave-one-mouse-out random forest classification are presented for each stage of the metastatic cascade – Primary tumor formation, intravasation, extravasation, and metastasis. A 4-class classifier was used for primary tumor classification and binary classification was used for all other stages. For each stage, the ground truth for each tumor group is indicated above the doughnut holes (negative or positive).

Download Full Size | PDF

3.3 Inhibition of metastasis-promoting genes in metastasis-capable tumors leads to changes in tumor optical properties in the direction of non-metastatic tumors

Next, we determined if knocking out genes responsible for promoting tumor metastasis would cause measurable differences in the tumors derived from these modified cell lines. We used CRISPR/Cas-9 to delete TWIST from the 4T1 cell line and obtained shRNA knockdown versions of the 4T1 cell line for CXCR3 and FOXC2. We found that deletion of TWIST led to a small but statistically insignificant increase in the tissue scattering compared with the parental 4T1 cell line (Fig. 4(A)). We also found an increase in total hemoglobin content (Fig. 4(B)) and a decrease in vascular oxygenation (Fig. 4(C)) in the TWIST KO cell line compared with the parental 4T1 line. The FOXC2 KD tumor group demonstrated a decrease in tissue scattering and vascular oxygenation and an increase in total hemoglobin. None of these changes were statistically significant. Finally, the CXCR3 KD tumor group showed no differences in any optical properties in comparison to the corresponding vector control. We then investigated how a supervised classifier trained on the four original tumor groups (67NR, 168FARN, 4T07, and 4T1) would classify tumors derived from these modified cell lines. Because the vector control (VC) for the TWIST KO cell line was the 4T1 that was part of the original group, data from that cell line is not shown here. Most of the TWIST KO tumors were classified as 4T1 tumors (7/8) with one tumor being classified as the non-metastatic 67NR (Fig. 4(D)). The FOXC2 KD tumors were classified as either 4T1 (5/10), 4T07 (4/10), or 67NR (1/10) and the CXCR3 KD tumors were classified as 4T1 (4/10), 4T07 (3/10), or 168FARN (3/10). The FOXC2 VC tumors were classified as either 4T07 (5/8) or 168FARN (3/8) while the CXCR3 VC tumors were classified into all 4 primary tumor groups – 168FARN (4/9), 4T07 (3/9), 67NR (1/9), and 4T1 (1/9).

 

Fig. 4. A. Comparison of functional parameters between vector\control and the corresponding knockdown\knockout cell lines. A. Tissue scattering. B. Total hemoglobin content. C. Vascular oxygenation. Error bars represent standard error of the mean (SEM). D. Results of random forest classification of tumors from the three 4T1-variant cell lines (knockdown/knockout and control). We used a four-class classifier trained on data from the original tumor panel - 67NR, 168FARN, 4T07, and 4T1.

Download Full Size | PDF

3.4 Tissue scattering measured with DRS in the primary tumor groups – 4T1, 4T07, 168FARN, and 67NR – is correlated with Masson’s trichrome staining of collagen

Due to the statistically significant differences observed in tissue scattering among the four primary tumor groups, we stained the excised tumor sections with Masson’s trichrome to compare the levels of collagen. Within the four primary tumors, the percentage of blue or collagen-positive pixels followed the same trend as tissue scattering, with the 67NR tumors showing the lowest percentage of collagen and 4T07 showing the highest level of collagen (Fig. 5). In addition, the tissue scattering from the primary tumor groups (4T1, 4T07, and 67NR) was highly correlated with the levels of collagen (data not shown). However, the same trends are not seen in the modified cell line group, where we observed significant differences in collagen levels between the vector control and knockdown cell lines despite the absence of significant differences in tissue scattering between these tumor groups.

 

Fig. 5. Evaluation of collagen-positive pixels in Masson’s Trichrome stained images. A. Representative images of Masson’s trichrome stained tumor sections from each of the four primary tumor groups. The scale bar represents 150 mm. Percentage of collagen-positive pixels in the B. Primary tumors and C. Modified tumors. * indicates statistically significant differences at p < 0.05 and ** indicates p < 0.01.

Download Full Size | PDF

We evaluated the metastatic outcome of the 4T1 and other 4T1-modified cell lines. Deletion of TWIST led to a near complete elimination of macrometastatic nodules in the mouse lungs (Fig. 6(A)). In our hands, we did not observe significant differences in macrometastatic nodules between the vector control and knockdown versions of FOXC2 and CXCR3 inhibited 4T1 cell lines (Fig. 6(B)). There were significantly more micrometastatic nodules in the lungs of mice injected with knockdown clones compared with their vector controls.

 

Fig. 6. Mean metastatic nodule count of the micrometastases and macrometastases of the tail vein injected mice. Mean metastatic count vector\control animals against knockout\knockdown cell lines for TWIST, FOXC2, and CXCR3 genes. Error bars represent error of the mean (SEM). * denotes statistically significant differences at p < 0.05.

Download Full Size | PDF

4. Discussion

Metastasis is frequently described as an inefficient process with early studies demonstrating that there are very few cells within the primary tumor that are capable of metastasis [31,32]. Other studies have determined that the propensity for metastasis is encoded in the bulk of the primary tumor [33], and identified gene expression signatures that predict risk of metastatic recurrence across multiple organs [34,35]. Despite widespread genetic heterogeneity in cancers [36], these genetic changes invariably result in tissue-level changes to structure and function that can be detected using methods that are sensitive to such changes. In this study, we used diffuse reflectance spectroscopy to interrogate tumor biology in a controlled model of metastatic cancer progression and investigated whether the knockdown of metastasis-promoting genes results in detectable tissue-level changes. The murine breast cell lines used in this study are sibling lines derived from a single spontaneous murine breast tumor and have similar growth rates but different metastatic outcomes. Their similar genetic background provides an excellent model for comparison of tumor properties.

We observed statistically significant differences in tissue scattering between the four primary tumor groups – 67NR, 168FARN, 4T07, and 4T1. The major sources of scattering from tissue are cells and cell organelles, such as nuclei and mitochondria, and collagen. The 67NR tumors had the lowest levels of scattering and 4T07 tumors had the highest tissue scattering (Fig. 2(A)). Analysis of Masson’s Trichrome staining on these tumors found a very similar trend in collagen staining (Fig. 5(B)) that was consistent with our optical measures of tissue scattering. Increased accumulation of collagen and remodeling of the collagen-containing extracellular matrix is associated with cancer progression and metastasis [37,38]. These observations contrast with other reports investigating structural differences between the 4T1 and 67NR tumors. A previous report from our group used the same tumors under investigation in this study to acquire ex vivo Raman scattering spectra (following in vivo DRS measurements presented here and tumor excision) and found decreasing levels of collagen-like species with increasing metastatic potential (67NR > 168FARN > 4T07) [39]. Bendau et al. found no differences in collagen morphology between 67NR and 4T1 tumors based on H&E images but found differences in curvature and alignment based on second harmonic generation (SHG) imaging of collagen [40]. When we analyzed the tissue scattering in the tumor groups that had knockdown of metastasis-promoting genes, we did not observe significant differences in tissue scattering. On the other hand, there were statistically significant differences in Masson’s trichrome staining of collagen between the vector control and knockdown cell lines. These opposite trends require further investigation to determine whether the lower tissue scattering and minimal collagen found in 67NR tumors is more of a cell line-specific observation or a feature of non-metastatic tumors.

Quantification of the tumor optical properties revealed significantly higher total hemoglobin content in the non-metastatic 67NR compared with the metastatic 4T1 tumors (Fig. 2(B)) even though the mean tumor volumes for both groups were approximately the same (∼200 mm³). These observations are consistent with the report by Gerwing et al. who used magnetic resonance imaging to demonstrate homogeneous 67NR tumors with thick endothelial layers and intact blood vessels compared with the heterogeneous 4T1 tumors that showed extensive hemorrhage and compromised blood vessels [41]. While total hemoglobin content was higher, the vascular oxygenation (Fig. 2(C)) was lower within the 67NR tumors (mean SO₂ = 18.01%) compared to the 4T1 tumors (mean SO₂ = 30.34%). Taken together with the significantly high deoxygenated Hb (Fig. 2(E)) in the 67NR tumors, these data indicate that the lower SO₂ in the 67NR tumors can be attributed to higher oxygen consumption by the 67NR cancer cells and not due to challenges with oxygen supply. While these tumors were measured at a mean tumor volume of 200 mm³, the SO₂ for the 4T1 and 4T07 tumors in this study was higher compared with other spectroscopy studies that have investigated the same cell lines but at smaller tumor volume [42,43]. This could be attributed to the difference in oxygen levels used in anesthesia in these studies (100% in the current study versus 21% in previous work) or due to differences in cell passage available in different labs. What we found interesting was that the knockdown of metastasis-promoting genes, such as TWIST and FOXC2 led to an increase, though not statistically significant, in total Hb (Fig. 4(B)) and a decrease in vascular oxygenation (Fig. 4(C)) in those tumors, similar to what we observe in the non-metastatic 67NR tumors.

Using a random forest approach, we were able to classify each primary tumor into its respective tumor group with reasonable accuracy. In addition, these tumors were also accurately classified as positive or negative for each step of the metastatic cascade based on their established propensity. Of course, a true measure of whether the combination of optical properties here can be used as a general classifier for non-metastatic versus metastatic tumors is accurate classification of an independent dataset. When tested on cell lines that had metastasis-promoting genes knocked down either using siRNA, shRNA, or CRISPR, we observed mixed results. Both the FOXC2 KD and CXCR3 KD tumors were classified as a mix of 4T1, 4T07, and 168FARN. Interestingly, when we injected these cell lines into mice to assay for metastases, we found that both cell lines resulted in several micrometastases to the lungs (Fig. 6), consistent with its classification as some of the more metastatic cell lines. However, the TWIST KO tumors, which had demonstrated changes in optical properties in a direction towards the non-metastatic 67NR tumors (vascular oxygenation and total hemoglobin) were almost overwhelmingly classified as 4T1; in addition, there were no visible metastases in the lungs. When these same tumors were examined ex vivo using Raman spectroscopy in a previous study from our lab, the FOXC2 KD and CXCR3 KD tumors were similarly classified as a mix of 4T1, 4T07, and 168FARN while the majority of TWIST KO tumors were classified as 67NR [39]. Further studies are necessary to determine why the knockout of TWIST, sufficient to eliminate lung metastases, results in discernible molecular changes in the primary tumor that were identified by Raman spectroscopy but did not result in any significant functional changes as determined by diffuse reflectance spectroscopy.

This study does have some drawbacks. All tumors in this study were injected subcutaneously in the flanks of mice. Ideally, such studies should employ orthotopic tumors where cancer cells are injected in the mammary fat pad. The study involved multiple cell lines obtained from different labs; while these cell lines are well-established through peer-reviewed publications, there can still be differences due to differences in cell line handling and protocols. These results could also be specific to this group of cell lines and further studies are required in other breast cancer models of metastasis. Nevertheless, these results provide promising preliminary evidence that diffuse reflectance spectroscopy can identify measurable differences in the microenvironment of breast tumors that differ in their propensity for metastasis.

5. Conclusion

In conclusion, we have demonstrated that there are measurable differences in tissue scattering, total hemoglobin, and vascular oxygenation between indolent and aggressive tumors. The ability to identify likely metastatic outcome based on the primary tumor could greatly decrease the overtreatment of indolent tumors and allow us to develop personalized treatment plans for indolent and aggressive tumors that may be anatomically similar but functionally different.