Characterizing Trastuzumab-Induced Alterations in Intratumoral Heterogeneity with Quantitative Imaging and Immunohistochemistry in HER2+ Breast Cancer

Abstract

The purpose of this study is to investigate imaging and histology-based measurements of intratumoral heterogeneity to evaluate early treatment response to targeted therapy in a murine model of HER2+ breast cancer. BT474 tumor-bearing mice (N = 30) were treated with trastuzumab or saline and imaged longitudinally with either dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) or ¹⁸F-fluoromisonidazole (FMISO) positron emission tomography (PET). At the imaging study end point (day 4 for MRI or 7 for PET), each tumor was excised for immunohistochemistry analysis. Voxel-based histogram analysis was performed on imaging-derived parametric maps (i.e., K^trans and v_e from DCE-MRI, SUV from ¹⁸F-FMISO-PET) of the tumor region of interest to measure heterogeneity. Image processing and histogram analysis of whole tumor slice immunohistochemistry data were performed to validate the in vivo imaging findings. Trastuzumab-treated tumors had increased heterogeneity in quantitative imaging measures of cellularity (v_e), with a mean Kolmogorov-Smirnov (K-S) distance of 0.32 (P = .05) between baseline and end point distributions. Trastuzumab-treated tumors had increased vascular heterogeneity (K^trans) and decreased hypoxic heterogeneity (SUV), with a mean K-S distance of 0.42 (P < .01) and 0.46 (P = .047), respectively, between baseline and study end points. These observations were validated by whole-slice immunohistochemistry analysis with mean interquartile range of CD31 distributions of 1.72 for treated and 0.95 for control groups (P = .02). Quantitative longitudinal changes in tumor cellular and vascular heterogeneity in response to therapy may provide evidence for early prediction of response and guide therapy for patients with HER2+ breast cancer.

Introduction

Intratumoral heterogeneity refers to the cellular diversity and regional differences in the microenvironment within a tumor. These variations can be phenotypic or genotypic in nature and are defined as a diverse collection of subpopulations within a single tumor [1], [2]. Heterogeneity describes the nonuniform distribution of these subpopulations and the resulting alterations to the tumor microenvironment. For example, diverse proliferation rates and metabolic activity of cancer cells, a characteristic of intratumoral heterogeneity, lead to varied regions of cellular density causing gradients in nutrient concentration and pH levels across the tumor microenvironment [3], [4], [5]. Furthermore, variations in quality and distribution of vasculature within a tumor, another characteristic of intratumoral heterogeneity, lead to the formation of hypoxic niches and result in nonuniform drug delivery, contributing to the rise of more resistant and aggressive disease [3], [5], [6]. Increased intratumoral heterogeneity is associated with variable treatment response and poorer patient prognosis [5], [7] and provides a significant challenge to maximizing the efficacy of treatment as it adversely affects tumor response to therapies and increases the manifestation of resistant disease [1], [8], [9].

Approximately 20% of breast cancer cases in the United States are diagnosed as human epidermal growth factor receptor 2 positive (HER2+), a subtype of breast cancer associated with decreased pathological response rates to treatment, a heightened level of metastatic potential, and more aggressive disease compared to other breast cancer subtypes [10], [11], [12], [13]. While HER2+ patients are identified by HER2 overexpression, heterogeneous distributions of intratumoral HER2 expression are often observed [14], [15], contributing to varied manifestation of disease across patients. Trastuzumab, a monoclonal anti-HER2 antibody, is a targeted therapy used in the treatment of HER2+ breast cancer and has led to reduced recurrence of disease and dramatic improvements in patient survival [12], [16], [17]. In addition to its cytotoxic mechanisms, trastuzumab has been shown to alter tumor vasculature and oxygenation as a secondary mechanism of action [18], [19], [20], [21]. Although trastuzumab has improved the treatment of locally advanced HER2+ breast cancer, only 30% of patients respond effectively to trastuzumab therapy in combination with other standard-of-care treatments [12], [16], [22], [23]. For those patients who do not respond to therapy, earlier predictive measures of treatment response have the potential to optimally guide interventions, thereby improving outcomes while reducing exposure to ineffective treatments.

Methods to characterize tumor heterogeneity from noninvasive imaging data have recently been investigated to elucidate responses to treatment [7]. Clinically, qualitative descriptions often describe aspects of tumor heterogeneity from images, such as radiotracer “hot-spots” [24] or morphological descriptors regarding lesion spiculation [25], [26]. However, these descriptions are subjective to observer evaluation of the tumor [7], [24]. Signal intensity histogram analysis [27], [28], [29] and image texture analysis [30], [31], [32], [33] are more quantitative methods used to characterize intratumoral heterogeneity in anatomical magnetic resonance imaging (MRI) data and computed tomography (CT) data. However, anticancer treatment interventions alter cellular and microenvironment heterogeneity, and these techniques often do not directly describe the biological heterogeneity present within the tumor. Additionally, initial changes in tumor microenvironment heterogeneity after treatment could serve as an early metric for treatment response and be used to guide therapy. Quantitative imaging strategies, such as dynamic contrast-enhanced (DCE) MRI [34] and ¹⁸F-fluoromisonidazole (¹⁸F-FMISO) positron emission tomography (PET) [35], [36], can noninvasively and longitudinally measure biological and molecular tumor microenvironment alterations throughout anticancer treatment strategies [37], [38], [39]. DCE-MRI can be used to quantitatively characterize tissue vascular features related to blood flow, vessel permeability, and tissue volume fractions, parameters which have demonstrated the ability to predict treatment response in both preclinical [40], [41], [42] and clinical [43], [44], [45] studies of breast cancer. ¹⁸F-FMISO-PET can noninvasively measure hypoxia through evaluation of the radiotracer uptake within the tumor and has been studied as a metric for early treatment response in preclinical models of breast cancer [18], [46]. Parametric maps extracted from DCE-MRI and ¹⁸F-FMISO-PET imaging provide voxel-level quantitative data of vasculature, cellularity, and oxygenation, thereby characterizing variability across the 3D tumor volume. Currently, analyses of quantitative imaging data commonly focus on summary statistics of whole tumor regions of interest (ROIs), such as the average or median tumor value of an individual imaging metric. By summarizing the tumor ROI with a single summary statistic, potentially valuable information regarding the heterogeneity of the tumor is discarded [7], [47].

In this contribution, we utilize quantitative DCE-MRI and ¹⁸F-FMISO-PET imaging of a murine model of HER2+ breast cancer to evaluate longitudinal alterations of intratumoral cellular and vascular in response to trastuzumab therapy. Quantitative metrics derived from these imaging strategies directly relate to the biology of the imaged tissue, allowing for the longitudinal evaluation of physiological tumor heterogeneity. While trastuzumab has been shown to alter vasculature and improve therapeutic delivery [18], [19], [20], [48], longitudinal alterations in whole tumor heterogeneity have not been characterized. We employ histogram analysis of physiological parameters derived from the quantitative imaging data to characterize heterogeneity of the tumor microenvironment and correlate these findings with the gold standard of immunohistochemistry analysis. We hypothesize that quantitatively characterizing the entire tumor (instead of summarizing the cellular and vascular status with a single number) will provide increased biological information enabling a more precise separation of treatment groups.

Materials and Methods

Cell Culture and Animal Model

BT474 cells, a HER2+ human breast cancer cell line (ATCC, Manassas, VA) with an established positive response to trastuzumab [48], [49], were cultured in improved minimal essential medium (Invitrogen, Carlsbad, CA) with 10% fetal bovine serum and 1% insulin at 37°C with 5% CO₂. Data from two independent cohorts of mice as part of previous studies were used in this study: one cohort from previous MRI studies [19], [40] and one from a previous PET imaging study [18]. Female nude athymic mice (Charles River Laboratories, Raleigh, NC) were implanted subcutaneously with a 0.72-mg, 60-day release, 17β-estradiol pellet (Innovative Research of America, Sarasota, FL). Approximately 24 hours after 17β-estradiol implant, 10⁷ BT474 breast cancer cells in serum-free media and 20% growth factor–reduced Matrigel were injected subcutaneously into the flank of the mouse. Tumors were allowed to grow until they reached ~225 mm³ (4-6 weeks after inoculation). Mice were randomly assigned to trastuzumab-treated (10 mg/kg) or saline control groups. Intraperitoneal injections of trastuzumab or saline were given on days 0 (baseline) and 3.

Magnetic Resonance Imaging

While details are provided in previous studies [18], [19], [40], we briefly summarize the MRI procedure. One cohort of mice (N = 20) was imaged using a 7-T small-animal MRI scanner [Agilent Technologies (formally Varian), Palo Alto, CA] to obtain DCE-MRI on days 0, 1, and 4 (Figure 1). Two days prior to baseline imaging, mice were implanted with a jugular catheter for exogenous delivery of the MRI contrast agent. DCE-MRI acquisition parameters were as follows: transaxial slices, 64 × 64 matrix, 1 mm slice thickness, 28 × 28 × 15 mm³ field of view. A multislice inversion recovery snapshot fast low angle shot gradient echo sequence was used to obtain precontrast T₁ maps using seven inversion times ranging from 250 to 10,000 milliseconds. Dynamic T₁-weighted images were acquired using a spoiled gradient echo sequence with a temporal resolution of 12.8 seconds for 20 minutes. Precontrast images were collected for 2 minutes prior to the administration of 0.05 mmol/kg gadolinium-diethylenetriaminepentaacetic acid (Magnevist, Bayer, Whippany, NJ) through an automated syringe pump. For each mouse at each imaging time point, a high-resolution T₂-weighted anatomical image was utilized to manually draw ROIs around the tumor boundary for all slices containing tumor (Supplementary Figure 1).

Figure 1.

Schedule of tumor implantation, imaging, and treatment schedule for separate MRI and PET study cohorts. During the MRI study, tumors are imaged at day 0, 1, and 4, while during the PET study, tumors are imaged at day 0, 1, 3, 4, and 7.

Supplementary Figure 1.

Anatomical landmarks on MRI and PET/CT data. Each panel shows a transaxial image of the mouse cross-section with various anatomical landmarks indicated with white arrows to orient the reader. A high-resolution T₂-weighted image is shown in panel A, along with the outline of the manually drawn tumor ROI. Panel B shows a CT image and panel C shows an ¹⁸F-FMISO-PET/CT overlay image. Additionally, panel C shows an example outline for the manually drawn tumor ROI.

Using a population-derived [50] arterial input function [50], [51], DCE-MRI data within the tumor ROI were fit (MATLAB, Natick, MA) to the standard Kety-Tofts model [52], [53] [Equation (1)] to extract parametric maps of the volume transfer coefficient (K^trans) and the extravascular extracellular volume fraction (v_e):

$$C_t\left(t\right)=K^{\text{trans}}\underset{0}{\overset{t}{\int }}{C}_{\rho }\left(u\right)\cdot exp\left(-\left(K^{\text{trans}}/v_{\mathrm{e}}\right)\left(t-u\right)\right)\mathit{du}.$$ (1)

Voxels which yielded nonphysiological values (e.g., K^trans < 0 min⁻¹, K^trans > 5 min⁻¹, v_e< 0, v_e> 1) from the DCE-MRI fitting routine were removed from subsequent analysis. The average (and 95% confidence interval) percent of tumor voxels removed was 16.5% (2.5%).

Positron Emission Tomography

While details are provided in a previous study [18], we briefly summarize the PET procedure. One cohort of mice (N = 10) was imaged with ¹⁸F-FMISO-PET on days 0, 1, 3, 4, and 7 (Figure 1). At each imaging time point, mice were injected with approximately 350 μCi of ¹⁸F-FMISO via retro-orbital injection prior to PET/CT imaging. After 70 minutes, a high-resolution CT scan was acquired for anatomical localization. To quantify volumetric uptake of ¹⁸F-FMISO, list-mode data were collected at 90 minutes postinjection. Images were reconstructed into transaxial slices (128 × 128 × 95) with voxel sizes of 0.95 × 0.95 × 0.8 mm³. For each voxel, the standardized uptake value (SUV) was calculated (MATLAB) using Equation (2):

$$\mathrm{SUV}=\frac{C_{\mathrm{img}}}{\mathrm{ID}/\mathrm{BW}},$$ (2)

where c_img is the average activity concentration in the voxel, ID is the injected dose, and BW is the animal body weight in grams. For each mouse at each imaging time point, PET/CT overlays were utilized to manually draw ROIs around the tumor boundary for all slices containing tumor (Supplementary Figure 1).

Histogram and Statistical Analysis

Longitudinal alterations in heterogeneity were assessed using voxel-based frequency histogram analysis of imaging parameters in R (https://www.r-project.org/, RStudio). Group frequency histograms for K^trans, v_e, and SUV were generated by averaging individual mouse distributions across each treatment group at each imaging time point. Differences between parameter distributions were quantified using the two-sample Kolmogorov-Smirnov (K-S) statistic, a distance measure between each sample pair's empirical distribution functions. The K-S distance is a measure of distribution similarity ranging from 0 to 1, with a value closer to 0 representing more similar distributions and a value closer to 1 representing more different distributions; however, it is important to note that the K-S distance between distributions does not indicate the direction of change between distributions. The K-S statistic was calculated between the parameter distributions obtained at baseline and subsequent imaging time points for each individual mouse to quantify longitudinal alterations in intratumoral heterogeneity. The distance between parameter distributions were compared between treatment groups using the nonparametric Wilcoxon rank sum test. A P value of less than .05 was considered significant.

Immunohistochemistry

On the final imaging time point (i.e., day 4 for MRI and day 7 for PET), each animal was sacrificed and the tumor excised for histology. For the ¹⁸F-FMISO-PET cohort of mice, 1 hour prior to sacrifice, animals were intravenously injected with pimonidazole (Hypoxyprobe, Burlington, MA). After tumor excision, tumors were sectioned at the largest cross section corresponding to the in vivo imaging plane and fixed in 10% formalin for 48 hours. The tissue was then stored in 70% ethanol for 72 hours, processed, and embedded in paraffin. Slices were sectioned at 5-μm thickness and floated onto charged slides and dried overnight. For each tumor, tissue sections were stained with hemotoxylin and eosin (H&E), anti-CD31 (DAKO, Carpinteria, CA), anti-Ki67 (DAKO, Carpinteria, CA), or antipimonidazole (Hypoxyprobe, Burlington, MA). Slides were digitally scanned in high-resolution (20×) brightfield with a Leica SCN400 (Leica Biosystems, Nussloch, Germany) slide scanner. Images were processed using Bio-Formats [54] (http://openmicroscopy.org) and MATLAB software.

For quantifying heterogeneity from immunohistochemistry data, a square tessellation technique was used derived from the work by Maley et al. [55]. Automated segmentation of histological sections stained with H&E or anti-Ki67 generated masks for individual nuclei across the entire tumor slice (Supplementary Figure 2). Automated segmentation of H&E data involved converting images to CIELAB color space and then using a k-means clustering algorithm to segment out nuclei. Automated segmentation of anti-Ki67 data involved converting images to HSV color space and then using Otsu thresholding to segment Ki-67+ nuclei. Automated segmentation of anti-CD31 and anti-pimonidazole data (Supplementary Figure 2) involved converting images to HSV color space and then using Otsu thresholding to segment stained areas across the entire tumor slice.

Supplementary Figure 2.

Example histology sections and binary masks. Panel A shows an H&E-stained histology section as well as a binary mask of individual nuclei across the section. Panel B shows a CD31-stained section with its associated mask for CD31+ vessels. Panel C shows a Ki67-stained section alongside a binary mask of Ki67+ nuclei. Panel D shows a pimonidazole-stained section, alongside its binary mask of pimonidazole stained area. The image segmentation algorithm was able to ignore false Ki67 and pimonidazole stained regions of necrosis.

To then analyze cellular density across the tissue slice, the segmented image was divided into blocks of 150 μm × 150 μm, and individual connected components were counted within each block to yield a heat map of nuclei counts (Supplementary Figure 3). To analyze heterogeneity of vascularity or hypoxia across the tissue, the anti-CD31 or antipimonidazole segmented image, respectively, was similarly divided into blocks, and percent stain-positive area was quantified within each block. The interquartile range (IQR) was calculated for each resulting distribution of heat map blocks, excluding nontissue areas from analysis. The nonparametric Wilcoxon rank sum test was used to assess differences between treatment groups, with a P value less than .05 considered significant.

Supplementary Figure 3.

Immunohistochemistry (IHC) data processing pipeline. The first panel shows an example histological section stained with H&E. The second panel shows the result of image segmentation, displaying a binary mask of individual nuclei across the histological section. The third panel displays a grid overlaid on the H&E-stained section, indicating the individual image blocks, within which the segmented nuclei would be counted (or, in the case of CD31 or pimonidazole staining, percent stained area would be quantified). The final quantification within each block is displayed in the fourth panel, as a heat map, which is ultimately converted into a distribution, shown in the final panel. For the H&E and Ki67 stains, the distribution of nuclei counts was quantified. For the CD31 and pimonidazole stains, a vessel and pimonidazole percent area was quantified, respectively.

Correlations Between Quantitative Imaging and Histology Heterogeneity

To biologically validate quantitative imaging measures of intratumoral heterogeneity, individual mouse parameter distributions were correlated to corresponding histology data. Parameter distributions of the tumor ROI within the central slice at the study end point (day 4 for MRI, day 7 for PET) imaging data were used to calculate IQR values for each individual mouse parameter distribution. IQR values for distributions of heat map blocks from histology data were correlated with IQR values for parameter distributions of the matching central slice from quantitative imaging data. Comparisons of heterogeneity were made for the following: CD31 vessel area and K^trans, H&E nuclei count and v_e, and pimonidazole percent stained area and SUV. Correlations were tested using Pearson's product-moment correlation, and a P value less than .05 was considered significant.

Results

Histogram Analysis of Quantitative DCE-MRI

Figure 2 shows representative control and trastuzumab-treated K^trans and v_e parameter maps derived from DCE-MRI data. Quantitative analysis of K^trans distributions showed a longitudinal increase in vascular heterogeneity upon trastuzumab treatment (Figure 3A). Figure 3, B-C shows box and whisker plots of K-S distance measures for each individual mouse between baseline and day 1 (Figure 3B), and baseline and day 4 (Figure 3C) K^trans distributions. On day 1, there were no significant differences in K^trans distributions between control and treated groups, with a mean K-S distance from baseline of 0.23 for both treatment groups (P = .53, Figure 3B), indicating no change in vascular heterogeneity. On day 4, individual control K^trans distributions showed little change in vascular heterogeneity from baseline, with a mean K-S distance of 0.21 (Figure 3C). Trastuzumab-treated mice, however, revealed a longitudinal increase in vascular heterogeneity on day 4, as indicated by the increase in the tails of the grouped distribution (Figure 3A). Furthermore, K^trans distributions in the trastuzumab-treated cohort showed an increased mean K-S distance of 0.42 on day 4 (P < .01, Figure 3C), demonstrating increased vascular heterogeneity.

Figure 2.

Representative longitudinal (progressive) control and treated animal parametric maps and voxel frequency distributions derived from DCE-MRI (K^trans and v_e) over 4 days. One representative animal from each treatment group is displayed in each column, while each row represents a separate time point showing a single slice of the MR image along with the associated whole tumor ROI parameter voxel distributions for that time point. The trastuzumab (TRA)-treated mouse shows a longitudinal increase in vascular heterogeneity, as measured by the K^trans distribution (A). The TRA-treated mouse additionally shows increased heterogeneity in cellularity, as measured by the v_e distribution (B).

Figure 3.

Alterations in K^trans distributions for treatment groups over time. Panel A shows average K^trans distributions for each treatment group over time, with dashed lines representing the standard deviation. Panels B and C display box and whisker plots of K-S distances between baseline and day 1 (B) and baseline and day 4 (C) for the K^trans distributions for each individual mouse from the control (CTL) and trastuzumab (TRA) -treated groups. Individual mouse K-S distances are shown as points overlaid on each box and whisker plot (B and C). The K^trans distributions for TRA-treated tumors show a longitudinal increase in vascular heterogeneity with increased frequency in the distribution tail (A), as well as an increased mean K-S distance of 0.42 compared to CTL mean K-S distance of 0.21 (**P < .01, C).

Quantitative analysis of v_e distributions showed a longitudinal increase in cellularity heterogeneity upon trastuzumab treatment (Figure 4A). Figure 4, B and C show box and whisker plots of K-S distance measures for each individual mouse between baseline and day 1 (Figure 4B), and baseline and day 4 (Figure 4C) v_e distributions. On day 1, there were no significant differences in v_e distributions between control and treated groups, with a mean K-S distance from baseline of 0.18 and 0.22, respectively (P = .44, Figure 4B), indicating no change in cellularity heterogeneity. On day 4, individual control v_e distributions showed little change in cellularity heterogeneity from baseline, with a mean K-S distance of 0.22 (Figure 4C). Trastuzumab-treated mice, however, showed a longitudinal increase in cellularity heterogeneity on day 4, as observed by the increase in the tails of the grouped distribution (Figure 4A). Furthermore, v_e distributions for the trastuzumab-treated cohort showed an increased mean K-S distance of 0.32 on day 4 (P = .05), demonstrating increased cellularity heterogeneity.

Figure 4.

Alterations in v_e frequency distributions for treatment groups over time. Panel A shows average frequency v_e distributions for each treatment group over time, with dashed lines representing the standard deviation. Panels B and C display box and whisker plots of K-S distances between baseline and day 1 (B), and baseline and day 4 (C) for the v_e distributions for each individual mouse from the CTL and TRA-treated groups. Individual mouse K-S distances are shown as points overlaid on each box and whisker plot (B-C). The v_e distributions for the TRA-treated tumors show a longitudinal increase in cellular heterogeneity with increased frequency in the distribution tail (A), as well as an increased mean K-S distance of 0.32 compared to a CTL mean K-S distance of 0.22 (P = .05, not significant, C).

Histogram Analysis of Quantitative ¹⁸F-FMISO-PET

Representative control and trastuzumab-treated SUV images are shown in Figure 5. Quantitative analysis of SUV distributions showed longitudinal decrease in heterogeneity of oxygen distribution upon trastuzumab treatment (Figure 6A). Figure 6, B-E shows box and whisker plots of K-S distance measures for each individual mouse between baseline and day 1 (Figure 6B), day 3 (Figure 6C), day 4 (Figure 6D), and day 7 (Figure 6E) SUV distributions. On day 1, there were no significant differences in SUV distributions between control and treated groups, with a mean K-S distance from baseline of 0.18 and 0.12, respectively (P = .79, Figure 6B), indicating no change in hypoxia heterogeneity. On day 3, individual control SUV distributions showed little change in heterogeneity of oxygenation from baseline with a mean K-S distance of 0.12 (Figure 6C). Trastuzumab-treated mice, however, revealed a longitudinal decrease in hypoxia heterogeneity on day 3, with a narrowing of the treated SUV distribution (Figure 6A). Additionally, individual treated mouse SUV distributions showed an increased mean K-S distance of 0.42 on day 3 (P = .05), demonstrating decreased heterogeneity of oxygenation across the tumor. By day 7, control mice demonstrated a longitudinal increase in hypoxic heterogeneity, with a widening of SUV distributions (Figure 6A) and an increased mean K-S distance of 0.28 from the baseline distribution (Figure 6E). Conversely, on day 7, trastuzumab-treated mice showed a longitudinal decrease in hypoxic heterogeneity, with a narrowing of SUV distributions (Figure 6A) and an increased mean K-S distance of 0.46 from the baseline distribution (P = .047, Figure 6E).

Figure 5.

Representative longitudinal CTL and treated ¹⁸F-FMISO-PET images and SUV frequency distributions over 7 days. One representative animal of each treatment group is displayed in each column, while each row represents a separate time point showing a single slice of the PET image (tumor denoted by dashed circle) along with the associated whole tumor ROI SUV voxel distributions for that time point. The example TRA-treated mouse (right column) shows a decrease in hypoxic heterogeneity over time.

Figure 6.

Alterations in SUV frequency distributions for treatment groups over time. Panel A shows average frequency SUV distributions for each treatment group over time, with dashed lines representing the standard deviation. Box and whisker plots of K-S distances between baseline and day 1 (B), day 3 (C), day 4 (D), and day 7 (E) for the SUV distributions are shown for each individual mouse from the CTL CTL and TRA-treated groups. Individual mouse K-S distances are shown as points overlaid on each box and whisker plot (B-E). The SUV distributions from CTL tumors show a longitudinal increase in hypoxia heterogeneity with an increased voxel frequency in the distribution tail (A) and a K-S distance of 0.28 from baseline, at day 7 (E). Conversely, SUV distributions for the TRA-treated tumors show a longitudinal decrease in hypoxia heterogeneity with a narrowing of the distribution (A) and a K-S distance of 0.46 from baseline at day 7 (*P = .047, E).

Histogram Analysis of Immunohistochemistry Data

Figure 7 shows representative mouse histological sections acquired on day 4 (corresponding to the end point of the DCE-MRI study) stained with H&E, anti-CD31, or anti-Ki67 to measure cellular density, vessel area, and proliferating cellular density, respectively. Example mouse sections from each treatment group are shown in Figure 7, A-C, alongside associated heat maps and kernel density estimates of heat map distributions. Figure 7D displays box and whisker plots of IQR values for each stain's heat map distribution across treatment groups. Treatment cohort distributions are visualized in Supplementary Figure 4, A-B. On day 4, no statistically significant difference was observed between the trastuzumab-treated and control groups for H&E nuclei count distributions (mean treated IQR = 87.83, control IQR = 70.75, P = .57, Figure 7D), CD31 percent vessel area distributions (mean treated IQR = 1.59 control IQR = 1.42, P = .87, Figure 7D), or Ki67+ nuclei count distributions (mean treated IQR = 19.25, control IQR = 23.50, P = .57, Figure 7D).

Figure 7.

Representative histology from MRI study end point. From the MRI study end point (day 4), representative CTL and treated animal histology sections, associated heat maps, and distributions are shown for (A) H&E-stained nuclei, (B) anti-CD31 vessel area, and (C) anti-Ki67 positive nuclei. Each row in panels A-C shows a representative animal from each treatment group. The first column (IHC Stain) for panels A-C shows representative animal histology sections. The second and third columns for panels A-C show the corresponding representative heat maps and associated distributions, respectively, for (A) cellularity, (B) vessel percent area, and (C) Ki67+ proliferating cellularity. The representative CTL and TRA mice show similar cellularity and vascular heterogeneity (A-B). However, the TRA-treated mouse shows decreased levels of proliferating cellularity (C). These trends held similarly across the treatment cohorts, as seen in panel D, showing IQR values for heat map distributions of each mouse. No statistically significant differences in mean IQR values were observed between the treatment groups.

Supplementary Figure 4.

Cohort IHC stain distributions for MRI and PET study end points. Panel A shows cohort stain distributions for control and treated groups obtained at the MRI study end point (day 4). Panel C shows cohort stain distributions for control and treated groups obtained at the PET study end point (day 7). The distributions of H&E nuclei count are shown in the first row and the distributions of CD31 vessel percent area are shown in the second row of panels A and C. Ki67+ nuclei count distributions are shown in the third row of panel A, while pimonidazole percent area distributions are shown in the third row of panel C. To quantify the differences between cohort distributions, the K-S distance was calculated and shown in panels B and D.

Figure 8 shows representative mouse histological sections acquired on day 7 (corresponding to the end point of the ¹⁸F-FMISO-PET study) stained with H&E, anti-CD31, and antipimonidazole to measure cellular density, vessel area, and hypoxia, respectively. Example mouse sections from each treatment group are shown in Figure 8, A-C, alongside associated heat maps and kernel density estimates of heat map distributions. Figure 8D displays box and whisker plots of IQR values for each stain's heat map distribution across treatment groups. Treatment cohort distributions are visualized in Supplementary Figure 4, C-D. On day 7, no statistically significant difference was observed between the trastuzumab-treated and control groups for H&E nuclei count distributions, with a mean IQR of 84.5 and 114.0, respectively (P = .21, Figure 8D). Trastuzumab-treated tumors had increased vascular heterogeneity, with CD31 percent vessel area distributions showing a mean IQR of 1.72 compared to control mean IQR of .95 (P = .02, Figure 8D). Additionally, trastuzumab-treated tumors had decreased hypoxia heterogeneity, with pimonidazole percent area distributions showing a mean IQR of 0, compared to control mean IQR of 8.05 (P < .01, Figure 8D).

Figure 8.

Representative histology from PET study end point. Representative CTL and treated histology sections and the associated heat maps and distributions are shown for (A) H&E-stained nuclei, (B) anti-CD31 vessel area, and (C) antipimonidazole-stained area are shown from the PET study end point (day 7). Each row in panels A-C shows a representative animal from each treatment group. The first column for panels A-C shows representative histology sections (column labeled IHC stains). The second and third columns for panels A-C show corresponding representative heat maps and the associated distributions, respectively, for (A) cellularity, (B) vessel percent area, and (C) pimonidazole percent area. The representative CTL and TRA mice show similar cellularity heterogeneity (A). However, the TRA-treated mouse shows increased vascular heterogeneity (B) and decreased levels of hypoxia (C). These trends held similarly across the treatment cohorts, as seen in panel D, showing IQR values for heat map distributions for each mouse. The TRA-treated cohort shows increased vascular heterogeneity with vessel percent area distributions having a mean IQR of 1.72 compared to the CTL IQR of 0.95 (*P = .02, D). Furthermore, the TRA-treated cohort shows decreased hypoxic heterogeneity with pimonidazole percent area distributions having a mean IQR of 0 compared to the CTL mean IQR of 8.05 (**P < .01, D).

Correlations Between Imaging and Histology Heterogeneity

Figure 9B shows a positive linear correlation between CD31 percent stained area heterogeneity (vessel percent area IQR) and quantitative imaging measures of vascular heterogeneity (K^trans IQR) with an r² = 0.33 (P = .05). No correlation was observed between v_e and H&E nuclei counts (Figure 9C). Figure 9D shows a positive linear correlation between pimonidazole percent stained area heterogeneity and quantitative imaging measures of hypoxia heterogeneity (SUV IQR) with an r² = 0.69 (P < .01).

Figure 9.

Correlation of quantitative imaging distributions to histological distributions at the study end point. Central imaging slices were selected from study end point MRI and PET data (A). IQR values were calculated for quantitative imaging parameter distributions for each mouse tumor ROI. IQR values were also calculated for heat map block distributions from individual mouse histology data. The correlation was calculated between IQR values of the following distributions: vessel percent area and K^trans (B), nuclei count and v_e (C), and pimonidazole percent stained area and SUV (D). Quantitative imaging of vascular permeability and perfusion (K^trans) showed a positive linear correlation with vessel percent area (B, r² = 0.33, P = .05). No correlation was observed between v_e and nuclei counts (C). Quantitative imaging of hypoxia (SUV) showed a positive linear correlation with pimonidazole percent stained area (D, r² = 0.69, P < .01).

Discussion

In this work, we demonstrated that measures of tumor microenvironment heterogeneity, derived from quantitative imaging, and their longitudinal alterations can be used to distinguish treatment response in a preclinical model of HER2+ breast cancer. Histogram analysis of DCE-MRI derived parametric maps revealed increased heterogeneity in cellularity upon trastuzumab treatment, with a widening of the v_e distribution for treated mice longitudinally, and compared to control tumors. The treated group showed increased spread of the v_e distribution, and this longitudinal increase in cellularity heterogeneity across trastuzumab tumors is expected with increased cell death. Histogram analysis of K^trans parametric maps showed increased vascular heterogeneity with trastuzumab treatment with a spreading of treated mice K^trans distributions over time and compared to control mice. Although a positive correlation is observed between CD31-stained vessel density and K^trans heterogeneity, immunohistochemistry analysis of vascular heterogeneity on day 4 showed no difference in CD31-stained vessel density between treatment and control. This is not surprising since K^trans is a mixed measure of vessel permeability and perfusion; while vessel density may be similar between treatment and control groups, vascular delivery is not solely a function of vessel density. Trastuzumab therapy revealed improved oxygenation across the tumor with a narrowing of ¹⁸F-FMISO-PET SUV distributions over time and compared to control mice. Pimonidazole staining of excised tissue from day 7 validated the observed decrease in hypoxic heterogeneity of the treated mice; in particular, there was a decrease in the mean IQR of distributions of pimonidazole percent area for treated mice compared to the control mice. These results are consistent with ¹⁸F-FMISO-PET results of decreased hypoxic heterogeneity observed in vivo. CD31 staining on day 7 was also in agreement with ¹⁸F-FMISO-PET results, having shown an increase in vessel density across the tumor, corresponding with increased vascular heterogeneity upon trastuzumab treatment. Our study showed quantitative imaging-derived measures of tumor microenvironment heterogeneity were consistent with immunohistochemical measures of microenvironment heterogeneity and able to distinguish treatment response in a preclinical model of HER2+ breast cancer.

Our study utilized voxel-based histogram analysis of quantitative imaging data and showed that trastuzumab treatment improved vascular delivery and alleviated hypoxia, resulting in more homogenous oxygenation across the tumor ROI. As it is widely known that tumors are spatially heterogeneous, summarizing an entire tumor with a single mean parameter value potentially eliminates a great deal of critical information. Histogram analysis of imaging data enables the quantitative characterization of that heterogeneity across the tumor volume that would be lost when only summary statistics are employed. While characterization of intratumoral heterogeneity using histogram analysis is not uncommon, histogram analysis of quantitative imaging data typically involves a bin-wise comparison of the test group's average histograms compared to that of the control [29]. Crokart et al. utilized this approach to histogram analysis of DCE-MRI parameter distributions to assess differences in perfusion characteristics between control and NSAID-treated tumors [56]. Additionally, Li et al. investigated histogram analysis of DCE-MRI derived parametric maps to evaluate heterogeneity of response to antiangiogenic therapy [57]. Comparatively, our work utilizes the K-S distance to measure differences of each mouse's parameter distribution posttreatment from its distribution at baseline, thereby personalizing the approach by comparing changes in heterogeneity at the individual tumor level.

Our observations of increased vascular heterogeneity upon trastuzumab treatment are further supported by other studies which have shown alterations in tumor vasculature and oxygenation in preclinical models of HER2+ cancer in response to trastuzumab therapy [20], [58], [59], [60], [61]. In a HER2+ murine model of brain cancer, Izumi et al. utilized optical imaging to show significant decrease in vascular permeability in response to trastuzumab therapy along with increased rates of survival [20]. Using hyperspectral imaging, with CD31 IHC staining for validation, McCormack et al. observed increased vessel density in HER2+ BT474 tumors upon trastuzumab treatment [58]. Furthermore, Hardee et al. utilized a transfected MCF7 cell line, constitutively expressing HER2, to demonstrate decreased hypoxia, as quantified by pimonidazole staining, in trastuzumab-treated MCF7^Her2 tumors compared to untreated MCF7^WT controls [60].

In an effort to combat treatment challenges posed by intratumoral heterogeneity, recent work has investigated methods to alter the tumor microenvironment and reduce vascular heterogeneity in the hopes of yielding more treatment sensitive tumors [62], [63], [64]. Here we demonstrated that trastuzumab therapy increased cellular and vascular heterogeneity and reduced heterogeneity in tumor oxygenation across the tumor. In previous work with the same murine model of HER2+ breast cancer, we have shown that sequential delivery of trastuzumab prior to half the standard dose of the cytotoxic therapy, doxorubicin, was as effective at reducing tumor volume as simultaneous delivery of trastuzumab and a full dose of doxorubicin [19]. These results taken together demonstrate the ability of trastuzumab to improve vascular delivery of cytotoxic therapy across the tumor. Furthermore, these results highlight the importance of longitudinal analysis of tumor microenvironment heterogeneity because with improved therapeutic delivery, cytotoxic drug dose and associated harmful side effects can be reduced.

One limitation of our study is the lack of biological validation at each imaging time point. Tumors were only excised at the end point of the quantitative imaging study; however, this was required for the longitudinal measurement of quantitative imaging parameters and analysis of tumor microenvironment heterogeneity over time. Additionally, biological validation was only performed on one to two cross-sectional slices of the excised tumor for each immunohistochemistry stain. While the entire tumor slice was quantitatively evaluated, the central slice is only representative of a small portion of the tumor. Future work will involve biological validation using the whole tumor and comparing biological heterogeneity measures to those observed with quantitative imaging. Another limitation of our study is the one-dimensional assessment of intratumoral heterogeneity through histogram analysis of quantitative imaging parameter maps. Most analyses of quantitative imaging data involve averaging across the tumor ROI, forgoing much of the information about physiological heterogeneity across the tumor. While we address this shortcoming utilizing voxel-based histogram analysis of quantitative imaging data, we do lose information regarding spatial heterogeneity. Future work will involve developing methodologies to analyze tumor microenvironment heterogeneity in 3D across the imaged tumor. Additionally, alternative techniques involving MR oximetry (e.g., oxygen-enhanced MRI or tissue oxygen level–dependent MRI) may provide the opportunity to collect similar oxygen data as ¹⁸F-FMISO-PET without the necessity of an exogenous, radioactive tracer [65], which is important for future clinical translation. Furthermore, this could allow for both oxygen and vascular (DCE-MRI) data from one imaging session, along with combined analysis of these parameters within a spatially aligned data set.

Conclusion

We have demonstrated the value in quantifying whole tumor, voxel-wise alterations in the tumor microenvironment heterogeneity over time and in response to therapy. In particular, DCE-MRI and ¹⁸F-FMISO-PET reveal quantitative longitudinal changes in the three-dimensional heterogeneity of tumor cellular, and vascular features could serve as early indicators for treatment response to help guide therapy for patients with HER2+ breast cancer.

The following are the supplementary data related to this article.