The Presence of Cyclooxygenase 2, Tumor-Associated Macrophages, and Collagen Alignment as Prognostic Markers for Invasive Breast Carcinoma Patients

Abstract

Inflammation, and the organization of collagen in the breast tumor microenvironment, is an important mediator of breast tumor progression. However, a direct link between markers of inflammation, collagen organization, and patient outcome has yet to be established. A tumor microarray of 371 invasive breast carcinoma biopsy specimens was analyzed for expression of inflammatory markers, including cyclooxygenase 2 (COX-2), macrophages, and several collagen features in the tumor nest (TN) or the tumor-associated stroma (TS). The tumor microarray cohort included females, aged 18 to 80 years, with a median follow-up of 8.4 years. High expression of COX-2 (TN), CD68 (TS), and CD163 (TN and TS) predicted worse patient overall survival (OS). This notion was strengthened by the finding from the multivariate analysis that high numbers of CD163⁺ macrophages in the TS is an independent prognostic factor. Overall collagen deposition was associated with high stromal expression of COX-2 and CD163; however, total collagen deposition was not a predictor for OS. Conversely, local collagen density, alignment and perpendicular alignment to the tumor boundary (tumor-associated collagen signature-3) were predictors of OS. These results suggest that in invasive carcinoma, the localization of inflammatory cells and aligned collagen orientation predict poor patient survival. Additional clinical studies may help validate whether therapy with selective COX-2 inhibitors alters expression of CD68 and CD163 inflammatory markers.

Yearly, more than 1.7 million women are diagnosed as having breast cancer worldwide. Despite the improvement in early detection and treatment, 31% of women diagnosed as having breast cancer will succumb to this disease.¹ Tumor-associated macrophages (TAMs) play a dynamic and multifaceted role in breast cancer development. They produce different outcomes, depending on their protumor or antitumor behavior. TAMs have served a dual role in breast cancer. They can produce an antitumorigenic effect by activation of ILs and interferon, while also promoting a tumorigenic environment by secreting diverse cytokines, growth factors, and proteases.² Protumoral TAMs aid in the processes of angiogenesis, proliferation, immunosuppression, degradation of the extracellular matrix (ECM), promotion of breast tumor epithelial cell migration, and metastasis.3, 4 CD68 and CD163 are glycoproteins that are expressed in human monocytes and tissue macrophages.5, 6 CD163 is a scavenger receptor that is overexpressed by macrophages in an anti-inflammatory environment,⁷ and it is considered a highly specific monocyte/macrophage marker for polarized protumoral macrophages.6, 8, 9 On the other hand, CD68 is a pan-macrophage marker that recognizes both protumoral and antitumoral macrophages.⁵ There is evidence of macrophages in the tumor microenvironment expressing high cyclooxygenase 2 (COX-2) levels, and prostaglandin E2 (PGE2) production downstream of COX-2 is one of the key molecules that facilitates the protumoral capabilities of TAMs.10, 11 Through COX-2 enzymatic production of PGE2, macrophages can be stimulated to produce cytokines and growth factors that will promote more proinflammatory cell recruitment and the development of colitis-associated tumorigenesis.¹¹ Chemoprevention by celecoxib in mammary carcinoma was first demonstrated in a 7,12-dimethylbenz(a)anthracene rat model, with significant reduction in incidence, number, and size of tumors.¹² Elevated expression levels of COX-2 and PGE2 are key elements in the inflammatory response of mouse mammary tumors that arise in collagen-dense tissue, which leads to increased recruitment of TAMs, elevated levels of several cytokines, and enhanced proliferation; they ultimately promote tumor development.¹³ Consequently, it was found that COX-2 inhibition with celecoxib caused the decrease in the presence of TAMs in conjunction with diminished cytokine levels, smaller and less proliferative tumors specifically in the collagen-dense tumors. In addition, COX-2 inhibition decreased the amount of collagen in the stroma, suggesting that COX-2, through enzymatic production of PGE2, has an important role in collagen deposition and macrophage recruitment to the tumor microenvironment and is essential in the growth and spread of mammary tumors.¹³

There is evidence that the immune response and ECM play a critical role toward the development and progression of breast cancer. There are several reports demonstrating that expression levels of COX-2 are elevated in breast, colorectal, and other carcinomas in comparison to normal tissue.14, 15 Unlike COX-1, which is a constitutively expressed enzyme,¹⁶ COX-2 is an inducible enzyme that is activated at sites of injury as part of the inflammatory response.¹⁷ Cyclooxygenases are responsible for the biosynthesis of prostaglandins, including PGE2, which has been associated as a major contributor to many cancers.¹⁸ COX-2 expression can be modulated by cytokines, ILs, hormones, growth factors, genetic mutations, and PGE2 itself, promoting its own biosynthesis.19, 20, 21, 22 COX-2 overexpression is observed in 40% to 75% of invasive breast carcinoma cases and correlates with more aggressive types of tumors and poor patient prognosis.23, 24 Concordantly, several epidemiologic studies demonstrated that COX-2 inhibition by nonsteroidal anti-inflammatory drugs is associated with decreased breast cancer recurrence and increased survival.15, 25, 26, 27 In addition, changes in collagen fiber structure, known as tumor-associated collagen signatures (TACSs),²⁸ are associated with tumor progression. These structures are characterized by the deposition of bundled straight collagen (TACS-2) that becomes oriented perpendicular to the tumor-stromal boundary (TACS-3); these structures are thought to provide an avenue for cell egression and dissemination.²⁸ These collagen structures are observed in human breast cancer, and the presence of TACS-3 collagen is an independent predictor of poor patient outcome.²⁹ Recent findings suggest that high COX-2 expression and increased levels of aligned collagen are the driving force for the development of ductal carcinoma in situ in a postpartum mammary gland involution mouse model.³⁰ Thus, it is important to know whether there is an association between cancer inflammation and collagen structure in invasive breast cancer.

Furthermore, the tumor microenvironment is complex and heterogeneous. It includes not only tumor epithelial cells, which are often found as foci of cells termed a tumor nest (TN), but also tumor-associated stroma (TS) components, such as stromal cells, immune cells, ECM, and blood vessels, which each may serve as a potential treatment target depending on the patient's tumor molecular expression characteristics.³¹ On the basis of the tumor-promoting factors secreted by cells expressing COX-2 and by TAMs, identifying the site of infiltration in the tumor microenvironment may reveal unknown associations with patient treatment response or survival. For instance, in melanoma patients, high tumor stroma presence of TAMs is associated with poor overall survival (OS), whereas in colorectal cancer, infiltration of TAMs in the tumor nest is indicative of poor patient prognosis.32, 33 Thus, it is imperative to study whether these markers of cancer-associated inflammation are localized to a specific compartment in the tumor microenvironment, which may serve as a more relevant prognostic indicator and perhaps guide treatment in invasive breast cancer.

In this report, we tested the hypothesis that localization of high COX-2 expression, high infiltration of TAMs, and increased aligned stromal collagen orientation will be associated with a worse patient outcome. The goal of this study was to assess whether localizations of COX-2 and the macrophage markers CD68 and CD163 were associated with collagen features and examine their independent prognostic value in a large US cohort study of invasive carcinoma patients. Herein, we report that high collagen content is associated with high expression levels of COX-2 and CD163 macrophages. Localization of COX-2 in the TN compartment, in particular, is associated with a worse patient prognosis. Moreover, high infiltration of CD68⁺ and CD163⁺ macrophages in the TN and both TN and TS, respectively, predicts worse OS. Finally, CD163 in the TS is an independent prognostic marker for poor patient OS. Taken together, these data suggest that tumor localization of inflammatory markers as well as collagen features are associated with patient survival in invasive breast cancer. Most important, total collagen deposition is associated with specific markers of inflammation. Further clinical studies may help validate whether therapy with selective COX-2 inhibitors, such as celecoxib, alters expression of CD68 and CD163 inflammatory markers in invasive breast carcinoma patients.

Materials and Methods

Invasive Breast Carcinoma Patient Samples

The invasive breast carcinoma tissue microarray (TMA) used in this study was obtained from the University of Wisconsin Biobank. The University of Wisconsin Comprehensive Cancer Center Health Sciences Institutional Review Board approved the study protocol (OS10111), to build and use deidentified patient tissue and select clinical patient information with a waiver of consent. Thus, all research using this TMA was Institutional Review Board exempt (Institutional Review Board approval 2010-0405). This protocol retrospectively collected deidentified data and archived tissue. The use of the TMA was the most efficient way to test hundreds of precious breast cancer tumor specimens without exhaustion of clinical specimens. Because hundreds of tumor samples fit in the same slide, all clinical cases in this report were subjected to the same experimental staining process at the same time, reducing experimental error, cost, and time. This invasive breast carcinoma TMA consisted of 371 female cases diagnosed from 1999 to 2009, staged I to III at diagnosis, and 15 normal cases divided into three slides. For this study, normal cases were excluded. The TMA sections were 4 μm thick, and each case contained triplicate punch biopsies of 0.6 mm in diameter, which were placed adjacent to each other. Deidentified patients' demographics and clinical information corresponding to each case were entered into a database. Receptor status and Ki-67 data were obtained from manual medical record review, registry database, and immunohistochemistry staining testing.

Antibodies

The following primary antibodies were used for immunofluorescence experiments: COX-2 (160126; Cayman, Ann Arbor, MI), pan-cytokeratin (ab9377; Abcam, San Francisco, CA), CD163 (NB110-59935; Novus Biologicals, Littleton, CO), and CD68 (ab955; Abcam). Horseradish peroxidase–conjugated anti-rabbit (ab7090; Abcam) or anti-mouse (ab97023; Abcam) was used as secondary antibody.

Histology and Multiplex Immunofluorescence

For histology, tissues were formalin fixed and prepared by standard methods for paraffin embedding and sectioning, as previously described.³⁴ One TMA set was stained with hematoxylin and eosin to aid in visualization of tumor and stroma composition and architecture. For immunofluorescence, the protocol was followed as previously described.¹³ Briefly, formalin-fixed, paraffin-embedded tissues were subject to standard deparaffinization, dehydration, and antigen retrieval with Citra Plus (HK080-5K; Biogenex, Fremont, CA) for 20 minutes, and blocking with BLOXALL (SP-6000; Vector Laboratories, Burlingame, CA), avidin/biotin (SP-2001; Vector Laboratories), and normal serum. Then, the TMA sections were subjected to the TSA Plus kit for tissue labeling, following manufacturer's protocols (fluorescein NEL741E001KT, Cy 3.5 NEL744E001KT, and Cy 5 NEL745E001KT; Perkin Elmer, Waltham, MA). Briefly, primary antibodies were incubated as follows: pan-cytokeratin (1:1000, 1 hour), COX-2 (1:1000, overnight), CD163 (1:1000, 1 hour), and CD68 (1:1000, 1 hour). Horseradish peroxidase–conjugated anti-rabbit or anti-mouse (1:500) was added for 10 minutes, following 10-minute incubation with TSA Plus kit working solution, including wanted fluorophore. Tissues underwent the antigen retrieval step for 20 minutes if the same tissue would be subjected to multiple labelings before counterstaining with DAPI for 2 minutes at 1:10,000 (D21490; Thermo Fisher Scientific, Waltham, MA). TMAs were mounted with ProLong diamond (P36961; Thermo Fisher Scientific).

Masson Trichrome Staining

To assess collagen deposition in the TMA samples, Masson trichrome staining kit (SS1026-MAB-250; Cancer Diagnostics Inc., Durham, NC) was used on formalin-fixed, paraffin-embedded sections, and a standard staining protocol provided by the manufacturer was applied. Briefly, formalin-fixed, paraffin-embedded TMAs were deparaffinized and rehydrated. Tissues were fixed in Bouin solution for 1 hour at 56°C to improve staining quality and then immersed in Weigert iron hematoxylin working solution and then Biebrich scarlet-acid fuchsin for 10 minutes each. Next, tissues were differentiated in phosphomolybdic-phosphotungstic acid solution for 10 minutes and transferred to aniline blue solution for 5 minutes. Staining of tissue was further differentiated by 1% acetic acid solution for 2 minutes; then, tissues were dehydrated through a series of ethanol immersions, cleared in xylene, and mounted. Samples were subjected to washes with distilled water between staining and differentiation steps. Collagen fibers stained blue, cell nuclei stained black, and cell cytoplasm, muscle tissue, and keratin stained red.

Nuance and InForm Software

The Vectra system was used to acquire TMA core images with a 20× air objective and Vectra software version 2.0.8 (Perkin Elmer), with analysis performed as previously described.13, 34 A spectral library was generated using image cubes to delineate distinctive spectral curves for each of the fluorophores and a DAPI counterstain to correct for background effects and, subsequently, to accurately identify positive staining and quantitation of biomarkers using InForm version 2.1 software (Perkin Elmer). With the system's automated slide handling and pattern recognition–based image analysis, samples were segmented, analyzed, and quantified in an objective manner. Machine learning–based algorithms for tissue and subcellular compartment separation were used and were >95% for precision (Supplemental Figure S1). Algorithms using the guidance of epithelial cell marker, pan-cytokeratin, were generated for separating tissue compartments into tumor nest (epithelium) and tumor-associated stroma. This process allowed us to identify nuclei, as stained with DAPI, to accurately assign associations for positive staining of a marker to a specific compartment in the biopsy sample. Of the image data set, 10% was used to generate algorithms for each experimental analysis. For COX-2 expression, a semiquantitative scoring method (H-score) was used, which reflects both percentage of positive cells and intensity of the staining. To calculate H-score, the following formula was used: [1 × (% cells with 1⁺ intensity) + 2 × (% cells with 2⁺ intensity) + 3 × (% cells with 3⁺ intensity)], with scoring ranging from 0 to 300. For CD68 and CD163, markers, the percentage positivity analysis was applied, which looked into the number of cells that are positive for that particular marker normalized to the total number of cells present in the tissue sample. For COX-2/CD68 and COX-2/CD163 expression levels, the percentage double positivity analysis was used, which looked into the number of cells that were double positive for COX-2 and a macrophage marker at the same time and normalized to total number of cell present in the tissue sample. For Masson trichrome, images were analyzed by the positive number of blue pixels (collagen)/the total amount of pixels in the core area and multiplied by 100 to obtain percentage area of collagen deposition. For all of the analyses, there were a maximum of three cores per case. The average of the cores was used as the final score for each case.

Second Harmonic Generation Microscopy

To visualize collagen fibers in the TMA samples, a custom multiphoton microscope was used to obtain second harmonic generation images. This system was described previously by Bredfeldt et al.³⁵ Excitation (780 nm) was used with a Nikon Plan Apo 20×/0.75 numerical aperture (Nikon, Melville, NY) air objective lens. Second harmonic generation light was collected in the forward direction with a 0.54 numerical aperture condenser lens and a 22-nm narrow band emission filter centered at 390 nm (Semrock, Rochester, NY). Images were obtained as a Z-stack of three images spaced 3 μm apart, then z-projected to improve field flatness. Individual images of 1024 × 1024 pixels were captured using an electronic zoom of 3, resulting in an image size of 360 μm².

Computer-Based Quantitation of Collagen Features

Collagen fiber quantitation and analysis from second harmonic generation images was performed with the open source software packages CT-FIRE version 1.3 Beta2³⁶ and CurveAlign version 3.0,³⁷ which are available for download (http://loci.wisc.edu/software/home, last accessed March 7, 2017). CT-FIRE software was used to calculate the overall length, width, and straightness of each individual collagen fiber in the second harmonic generation image, and it provided summary statistics for the TMA. CurveAlign software was used to gather additional data with respect to local collagen features around the TN. A boundary separating the TS from epithelial cells in the TN was drawn, and the area measured had a radius of 50 pixels from the TN boundary. TACS-3 includes aligned collagen fibers that are >60 degrees perpendicular to the TN boundary. TACS-2 includes aligned collagen fibers that are >45 to 60 degrees perpendicular to the TN boundary. CurveAlign software measured the angle of fibers with respect to the boundary, density, and alignment (fiber alignment in relation to its neighboring fibers) of the segmented collagen fibers with respect to that boundary. The fiber features for analysis were chosen because they are altered in various types of cancer when compared with normal tissue or are predictors of poor patient prognosis.29, 38

Statistical Analysis

All statistical analyses and graph preparation were based on SAS version 9.4 (SAS Institute, Cary, NC) and R software version 3.2.2 (R Core Team, Vienna, Austria). All statistical tests were two sided, and P < 0.05 was considered statistically significant, unless adjustment for multiple testing was deemed more appropriate. In this case, the P values were adjusted using the Bonferroni correction. The correlations between expression of COX-2, CD68, or CD163 and amount of tumor or stroma content were calculated using Spearman correlation test. A Spearman ρ correlation value of 0.4 was considered an indication of weak-to-moderate correlation, and a value >0.7 was considered a strong correlation. The data were divided into quartiles to reduce the range of data points and CIs. High COX-2 expression and high infiltration of TAMs included the cases that were in the upper 75th percentile of the data for the specific marker. The associations between biomarker expression and clinicopathologic parameters were assessed by a Wald test, where simultaneously all of the clinical parameters were entered into a logistic model and compared for their effect contribution (predictors) to high COX-2, CD68, or CD163 expression per compartment. The correlation between a biomarker and a continuous covariate was tested by an analysis of variance model to test whether the biomarker was significantly associated with the continuous covariate on the basis of the F-statistic. The χ² statistic was used to measure association between categorical variables. For univariate analysis, survival curves for OS or progression-free survival (PFS) were calculated according to the Cox proportional hazards model. PFS was calculated from patient age at diagnosis to the occurrence of metastases or death from breast cancer, whichever occurred first. Different survival curves were compared with the log-rank method. Multivariate survival analysis was performed using the Cox proportional hazards regression model. In determining the final Cox model, stepwise selection procedure was used and the criterion for adding or removing a covariate during the selection was P = 0.1, which is the default choice in SAS version 9.4. All significant variables in the univariate analysis and COX-2 TS and CD68 TN were included as candidate variables. Age at diagnosis and tumor stage were directly included in the final model. P < 0.05 was used as a significance level to interpret the selected multivariate survival analysis model. For all analyses, COX-2, CD68, CD163, COX-2/CD68, and COX-2/CD163 were converted into categorical variables, from 0 (no expression) to 3 (high expression), on the basis of their corresponding quartiles. For the time-to-event analyses, H-scores for COX-2 were converted into binary data and high COX-2 H-score included all data falling into the upper 75th quartile.

Results

Characterization of the Invasive Carcinoma Cohort

A total of 371 female confirmed invasive carcinoma cases were included in this study. Patients in this cohort were diagnosed from 1999 to 2009. At diagnosis, their age ranged from 26 to 94 years, with a median age of 54 years (mean, 56). Most of the cohort were racially white (96%), and 1.9% were African American. Of the patients, 13% had a family history of breast cancer and 6% had a clinical history of a prior breast cancer. Moreover, tumors were of ductal invasive carcinoma type (83%) and stage I or II (41% or 47%, respectively). Positive receptor status for tumors was as follows: 80% for estrogen receptor (ER), 72% for progesterone receptor (PR), and 17% for human epidermal growth factor receptor 2 (HER2). Following the St. Gallen consensus 2013,³⁹ 39% of the tumors had a Ki-67 proliferation score of ≥14%; for molecular subtypes, 44% of the tumors were luminal A, 27% were luminal B, and 12% were triple negative. For more detailed patient demographics and clinicopathologic features, see Table 1. Furthermore, median duration of patient follow-up was 8.4 years, during which 22% of patients experienced cancer recurrence within a period of 2.3 years and 26% were deceased (55% of those because of breast cancer) in a median time frame of 3.5 years. For more detailed patient follow-up data, see Supplemental Table S1.

Table 1.

Patient Characteristics for the Study

Patient demographics	Value
Total cases included in the study∗	371
Date of diagnosis	1999–2009
Age in years at diagnosis, means ± SD	56 ± 14
Race
White	96 (356)
Other†	4 (15)
Family history of breast cancer	13 (47)
Prior cancer site
Breast	6 (24)
Multiple, including breast	0.5 (2)
Nonbreast	4 (15)

Clinicopathologic features	Value
Histology type
Ductal	83 (308)
Lobular	9 (34)
Mammary	7 (27)
Other‡	1 (2)
Tumor stage
I	41 (153)
II	47 (173)
III	12 (45)
Tumor grade
1	24 (87)
2	42 (152)
3	34 (125)
Tumor size in mm, mean (range)	26.5 (2–150)
Lymph node–positive status	40 (150)
Receptor status
ER+	80 (292)
PR+	72 (260)
HER2+	17 (44)
Ki-67 ≥ 14%§	39 (139)
Molecular subtypes§
Luminal A	44 (109)
HER2+ luminal B	11 (26)
HER2− luminal B	16 (40)
Triple negative	12 (30)

Data are expressed as % (n) unless otherwise indicated.

ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; PR, progesterone receptor.

^∗A total of 352 cases were included for cyclooxygenase 2 (COX-2) data analysis, and 313 were included for CD68, CD163, COX-2/CD68, and COX-2/CD163 data analyses.

^†Includes the following: Asian (n = 3), black (n = 7), Hispanic (n = 2), and other/unknown (n = 2).

^‡Includes the following: adenoid cystic (n = 1) and phyllodes (n = 1).

^§Follows the St. Gallen consensus 2013.³⁹

Characterization and Distribution of COX-2, CD68⁺, and CD163⁺ Macrophages in Invasive Carcinoma

From the 371 invasive breast carcinoma cases, TMA cores from 352 cases (95%) were suitable for measuring COX-2 expression, and 313 cases (84%) were included to measure the presence of all macrophages (CD68), the fraction of macrophages that are protumorigenic (CD163), and the coincident presence of the COX-2/CD68 and COX-2/CD163 combinations. Cases of exclusion were attributable to poor tissue quality, low tumor cell content, or loss of cores after TMA processing. From the included cases, COX-2, CD68⁺, and CD163⁺ cells were present in both TN and TS of invasive breast carcinoma cases (Figure 1). Overall COX-2 expression was more common and associated with TN (P < 0.0001), whereas macrophage presence, as labeled with CD68 or CD163, was associated with the TS (P < 0.0001 for both) (Supplemental Table S2). Many tumor samples had none or low levels of macrophage infiltration (CD68: TN = 33% and TS = 41%; and CD163: TN = 40% and TS = 44%), whereas most of the samples had some level of COX-2 expression (TN = 99% and TS = 99%). Supplemental Table S3 contains, in more detail, frequencies of cases with expression levels for COX-2, CD68, and CD163. In addition, it was observed that cells could be double positive for both CD68 and CD163 or single positive for either of these macrophage markers. This clearly suggests that there are more than two populations of tumor-associated macrophages and they can express COX-2.

Figure 1.

Immunofluorescence staining of inflammatory cells in the tumor nest (TN) or tumor-associated stroma (TS) of an invasive carcinoma tumor. A: Composite multiplex immunofluorescence image of an invasive carcinoma tumor depicting tumor and stromal cells. Cyclooxygenase 2 (COX-2)–positive cells and CD68⁺ and CD163⁺ macrophages are observed in both TN and TS. B: Magnified split-channel images of TN and TS areas demarked by the white window of composite. Nuclear labeling with DAPI (blue), epithelial cell labeling with pan-cytokeratin (PCK; green), COX-2 staining (orange), CD68 macrophage/monocyte staining (magenta), and CD163 protumoral macrophage/monocyte staining (red) are shown (20× objective). Scale bars: 15 μm (A); 7.5 μm (B).

Associations between COX-2, CD68⁺, and CD163⁺ Macrophages and Clinicopathologic Features in Invasive Carcinoma

To be able to assess the association of COX-2, CD68⁺, and CD163⁺ macrophages and to examine whether localization of these are specific to the TN or TS tissue compartments in the invasive carcinoma cohort, an image-based approach was used to analyze these biomarkers' association with several clinicopathologic features. Cases that had expression for COX-2 above the median of the data, and expression of CD68 and CD163 >75th percentile of the data, were considered as high expression or high infiltration in the tumor microenvironment for each specific marker.

Several of the classic clinicopathologic biomarkers were associated with the presence of COX-2 in the TN, including tumor size and Ki-67, ER, and PR status (Table 2). In addition to the previous factors listed, COX-2 expression in the TS was associated with tumor grade. Furthermore, several features were associated with macrophage infiltration in invasive carcinoma tumors. As expected, the presence of macrophages in either compartment was associated with several of the known biomarkers of poor survival, such as tumor size, grade, and Ki-67 expression (Table 2). Moreover, to see if COX-2 is produced by macrophages, CD68⁺ and CD163⁺ macrophages that were also COX-2⁺ were quantitated. CD68⁺/COX-2⁺ macrophages were associated with tumor size, grade, and Ki-67 expression. Tumors with CD163⁺/COX-2⁺ macrophages were significantly associated with tumor size and ER status (Table 2). Overall, these data indicate that COX-2, CD68, and CD163 have a role in tumorigenesis, in which their increased expression is associated with different clinicopathologic factors.

Table 2.

Association of Clinicopathologic Predictors with COX-2, CD68, CD163, and Macrophages Expressing COX-2 in Tumor Nest and Tumor Stroma in Invasive Breast Cancer

Variable	Clinicopathologic features	COX-2				CD68				CD163				COX-2/CD68				COX-2/CD163
		Tumor nest		Tumor stroma		Tumor nest		Tumor stroma		Tumor nest		Tumor stroma		Tumor nest		Tumor stroma		Tumor nest		Tumor stroma
		Statistic test	P value	Statistic test	P value	Statistic test	P value	Statistic test	P value	Statistic test	P value	Statistic test	P value	Statistic test	P value	Statistic test	P value	Statistic test	P value	Statistic test	P value
F-test statistic	Age	0.069	0.792	1.793	0.181	0.134	0.940	0.056	0.814	0.305	0.822	5.529	0.001	0.162	0.922	0.858	0.464	1.249	0.292	1.091	0.353
	Tumor size	7.574	0.006	7.789	0.006	8.984	0.000	17.980	0.000	5.283	0.001	7.013	0.000	4.427	0.005	11.190	0.000	2.210	0.087	4.204	0.006
	Ki-67	8.736	0.003	36.920	0.000	5.151	0.002	7.526	0.006	6.346	0.000	2.790	0.041	4.750	0.003	8.785	0.000	2.643	0.050	3.554	0.015
χ2 Test statistic	Tumor grade	4.965	0.084	17.477	0.000	20.961	0.002	5.838	0.054	16.682	0.011	16.684	0.011	17.395	0.008	21.546	0.001	12.711	0.048	7.918	0.244
	Tumor stage	8.137	0.017	6.484	0.039	20.442	0.002	11.262	0.004	10.307	0.112	13.325	0.038	10.411	0.108	15.428	0.017	10.288	0.113	9.080	0.169
	Nodal stage	0.775	0.379	2.672	0.102	9.100	0.028	3.858	0.050	8.686	0.034	4.973	0.174	2.354	0.502	5.911	0.116	5.122	0.163	3.338	0.342
	ER status	18.898	0.000	11.175	0.001	10.776	0.013	2.879	0.090	9.327	0.025	15.362	0.002	12.718	0.005	11.837	0.008	13.753	0.003	7.506	0.057
	PR status	9.044	0.003	6.972	0.008	8.210	0.042	1.172	0.279	11.965	0.008	14.829	0.002	9.215	0.027	9.967	0.019	4.775	0.189	5.939	0.115
	HER2 status	0.145	0.703	0.496	0.481	3.367	0.339	0.017	0.897	1.451	0.694	2.264	0.519	5.238	0.155	4.356	0.226	5.661	0.129	3.869	0.276

COX-2, cyclooxygenase 2; ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; PR, progesterone receptor.

The F-statistic was used to measure association between categorical and continuous variables or between continuous variables. The χ² statistic was used to measure association between categorical variables. P values are based on Bonferroni correction, where the significance level was divided by number of total tests per biomarker/compartment (n = 9), and significant values are shown in bold.

Furthermore, to study the inflammatory tumor microenvironment, the association of different collagen features was analyzed with expression of the inflammatory biomarkers COX-2, CD68, and CD163. Total collagen was assessed by Masson trichrome stain; however, properties of collagen fibers that are in local proximity to the tumor nest were also analyzed. Thus, a boundary was drawn between the tumor nest and tumor stroma and the angle of all of the collagen fibers (TACS-2 and TACS-3) measured with respect to that boundary within a 50-pixel radius. The density of fibers within this localized perimeter of fibers was also measured. Only stromal expression of COX-2 was associated with local collagen density (P = 0.001) (Table 3). However, total collagen deposition was associated with stromal COX-2 (P < 0.0001) and CD163 (P < 0.0001) (Table 3). In addition, the association of collagen features with clinicopathologic features was evaluated. Total collagen deposition was associated with tumor grade (P < 0.0001), ER status (P = 0.005), and Ki-67 expression (P < 0.0001) (Supplemental Table S4). Likewise, the tumor proliferation marker Ki-67 was associated with local collagen alignment (P < 0.0001) (Supplemental Table S4). As expected, collagen deposition increases in an inflammatory environment, and this effect is even present at the local level. It seems that the presence of dense collagenous stroma attracts macrophages and produces high levels of COX-2; however, these inflammatory components are not related to collagen alignment and orientation. In addition, many of the clinicopathologic biomarkers have a strong trend of association with collagen deposition and the angle of the fibers (TACS-2 and TACS-3), indicating a path for metastasis.

Table 3.

Association of Collagen Features with COX-2, CD68, CD163, and Macrophages Expressing COX-2 in Tumor Nest and Tumor Stroma in Invasive Breast Cancer

Collagen features	COX-2				CD68				CD163				COX-2/CD68				COX-2/CD163
	Tumor nest		Tumor stroma		Tumor nest		Tumor stroma		Tumor nest		Tumor stroma		Tumor nest		Tumor stroma		Tumor nest		Tumor stroma
	F statistic	P value	F statistic	P value	F statistic	P value	F statistic	P value	F statistic	P value	F statistic	P value	F statistic	P value	F statistic	P value	F statistic	P value	F statistic	P value
Collagen alignment (local)	0.114	0.736	0.670	0.414	0.277	0.842	0.030	0.862	1.099	0.350	0.671	0.570	0.574	0.633	1.585	0.193	0.747	0.525	1.817	0.144
Collagen deposition (total)	1.591	0.208	50.170	0.000	0.253	0.859	5.249	0.023	3.195	0.024	6.807	0.000	1.969	0.119	3.769	0.011	3.725	0.012	0.389	0.761
Collagen density (local)	3.261	0.072	12.350	0.001	0.154	0.927	0.171	0.679	0.256	0.857	1.156	0.327	2.916	0.035	1.362	0.255	0.465	0.707	0.395	0.757
TACS-3	0.139	0.710	1.698	0.194	2.035	0.109	0.085	0.771	0.412	0.745	0.109	0.955	2.265	0.081	0.471	0.703	3.290	0.021	1.785	0.150
TACS-2 and TACS-3	0.215	0.644	3.071	0.081	1.885	0.132	0.034	0.855	0.231	0.875	0.144	0.933	2.317	0.076	0.205	0.893	2.497	0.060	1.664	0.175

COX-2, cyclooxygenase 2; TACS, tumor-associated collagen signature.

The F statistic was used to measure association between categorical and continuous variables or between continuous variables. P values are based on Bonferroni correction, where the significance level was divided by number of total tests per biomarker/compartment (n = 5), and significant values are shown in bold.

Role of High COX-2 Expression and High Density of CD163⁺ and CD68⁺ Macrophage Infiltration in Patient Survival

The maximum duration of follow-up for the invasive carcinoma cohort was 13.6 years (median, 8.4 years). During this follow-up period, the cohort had 74% overall survival (95 deaths among 371 cases) and 22% disease recurrence (80 recurrences among 371 cases) (Supplemental Table S1). From the 80 cases that recurred, 76% (61 of 80 cases) passed away during follow-up. Several markers of poor patient survival, including high COX-2 expression in TN [P = 0.016, hazard ratio (HR) = 1.71], high infiltration of CD68 in TS (P = 0.007, HR = 2.12), and high infiltration of CD163 in both TN (P = 0.015, HR = 2.36) and TS (P < 0.0001, HR = 2.94) (Figure 2 and Table 4), were identified. In addition, the combined presence of high COX-2/CD68 expression in the TS (P = 0.005, HR = 2.15) and the combined presence of high COX-2/CD163 expression in both TN (P = 0.015, HR = 2.36) and TS (P < 0.0001, HR = 2.94) were associated with poor patient survival (Figure 2 and Table 4). Several clinicopathologic features, including age at diagnosis of invasive carcinoma, larger tumor size, higher grade and stage, lymph node–positive status, ER⁻ status, PR⁻ status, and higher Ki-67 index, were associated with a worse disease and worse progression-free survival outcome (Table 4). Taken all together, these data suggest that high COX-2 expression in the TN and high presence of CD68⁺ and CD163⁺ macrophages in both TN and TS are associated with poor patient overall survival. In addition, high infiltration of CD68⁺ and CD163⁺ macrophages, which also expressed high levels of COX-2, was associated with worse patient survival. Moreover, perpendicular collagen alignment to the tumor boundary TACS-3 (P = 0.049, HR = 1.65) and TACS-2/TACS-3 (P = 0.037, HR = 1.24) were predictors of poor patient overall survival. Interestingly, local collagen alignment (P = 0.007, HR = 0.06) and local collagen density (P = 0.047, HR = 0.96) were associated with better patient prognosis (Figure 3 and Table 4). In addition, collagen deposition was not a significant marker for OS (P = 0.561). None of the inflammatory biomarkers were associated with PFS. Altogether, several clinicopathologic, inflammatory, and stromal markers are associated with poor patient survival.

Figure 2.

Overall survival with respect to cyclooxygenase 2 (COX-2) expression and macrophage infiltration in invasive carcinoma tumors. A–F: Overall survival curves according to COX-2 expression and CD68⁺ and CD163⁺ macrophages in the tumor nest or tumor-associated stroma of invasive carcinoma cases. G–J: Overall survival curves according to COX-2 expression in both CD68⁺ and CD163⁺ macrophages in the tumor nest or tumor-associated stroma of invasive carcinoma cases. P < 0.05 was considered statistically significant.

Table 4.

Univariate Cox Regression Analysis for OS and PFS

Variables	Level	OS		PFS
		P value	HR	P value	HR
Age at diagnosis	NA	0.000	1.03	0.039	0.98
Tumor size	NA	0.000	1.02	0.000	1.02
Tumor grade	NA	0.000	1.82	0.000	2.27
Tumor stage	Overall	0.000	NA	0.000	NA
	1	Reference	1.00	Reference	1.00
	2	0.000	2.33	0.000	3.36
	3	0.000	4.76	0.000	5.94
Lymph node positive	NA	0.003	1.84	0.000	2.98
ER positive	NA	0.000	0.32	0.001	0.44
PR positive	NA	0.000	0.36	0.001	0.47
HER2 positive	NA	0.897	0.96	0.914	1.04
Ki-67 (proliferation)	NA	0.000	1.02	0.000	1.02
COX-2 TN	NA	0.016	1.71	0.249	1.33
COX-2 TS	NA	0.484	1.18	0.147	1.44
CD68 TN	Overall	0.150	NA	0.359	NA
	1	Reference	1.00	Reference	1.00
	2	0.804	1.12	0.081	0.28
	3	0.120	1.57	0.915	1.03
	4	0.030	1.85	0.965	1.01
CD68 TS	Overall	0.007	NA	0.791	NA
	1	Reference	1.00	Reference	1.00
	2	ND	ND	ND	ND
	3	0.007	2.12	0.791	1.07
	4	ND	ND	ND	ND
CD163 TN	Overall	0.001	NA	0.927	NA
	1	Reference	1.00	Reference	1.00
	2	0.031	2.54	0.930	0.95
	3	0.277	1.41	0.818	0.93
	4	0.000	2.86	0.620	1.16
CD163 TS	Overall	0.002	NA	0.712	NA
	1	Reference	1.00	Reference	1.00
	2	0.940	1.04	0.271	0.56
	3	0.002	2.52	0.920	1.03
	4	0.002	2.54	0.943	1.02
COX-2/CD68 TN	Overall	0.089	NA	0.755	NA
	1	Reference	1.00	Reference	1.00
	2	0.297	1.42	0.418	0.76
	3	0.904	1.04	0.356	0.73
	4	0.028	1.98	0.808	0.93
COX-2/CD68 TS	Overall	0.015	NA	0.143	NA
	1	Reference	1.00	Reference	1.00
	2	0.825	1.09	0.440	0.71
	3	0.986	0.99	0.193	0.64
	4	0.005	2.15	0.262	1.37
COX-2/CD163 TN	Overall	0.008	NA	0.535	NA
	1	Reference	1.00	Reference	1.00
	2	0.458	1.33	0.500	1.27
	3	0.003	2.76	0.183	1.57
	4	0.015	2.36	0.848	1.07
COX-2/CD163 TS	Overall	0.002	NA	0.203	NA
	1	Reference	1.00	Reference	1.00
	2	0.792	1.12	0.328	0.67
	3	0.036	1.97	0.547	0.82
	4	0.000	2.94	0.230	1.42
Collagen alignment (local)∗	NA	0.007	0.06	0.234	0.28
Collagen deposition (total)†	NA	0.561	0.68	0.119	0.32
Collagen density (local)∗	NA	0.047	0.96	0.026	0.95
TACS-3∗	NA	0.049	1.65	0.526	1.20
TACS-2 and TACS-3∗	NA	0.037	1.24	0.599	1.06

Significant P values are shown in bold.

ER, estrogen receptor; HER2, human epidermal growth factor receptor 2; HR, hazard ratio; NA, not applicable; ND, no data fell in that categorical level; OS, overall survival; PFS, progression-free survival; PR, progesterone receptor; TACS, tumor-associated collagen signature; TN, tumor nest; TS, tumor-associated stroma.

^∗Computational measurement of collagen organization from second harmonic generation images. Local collagen density measured from the tumor-associated stroma area that falls within a 50-pixel radius in length from the TN boundary. TACS-3 includes aligned collagen fibers that are >60 degrees perpendicular to the TN boundary. TACS-2 includes aligned collagen fibers that are >45 to 60 degrees perpendicular to the TN boundary.

^†Total collagen deposition measured with Masson trichrome staining as area of blue pixels.

Figure 3.

Collagen alignment perpendicular to the tumor nest (TN) is associated with poor overall survival. A and B: Representative hematoxylin and eosin images of a survivor and nonsurvivor breast cancer case. Yellow boxed areas correspond to yellow boxed areas in C and D. C and D: Second harmonic generation images depicting collagen structure in these patient samples. E and F: Enlarged second harmonic generation images from yellow boxed areas in C and D. E: Arrows depict collagen fiber alignment following the boundary of the tumor nest (green dashed line). F: Arrows depict collagen fibers aligned perpendicular to the TN (green dashed line), known as tumor-associated collagen signature 3. G and H: Curvelet analysis images of collagen fibers. TN boundaries are drawn in yellow. Fibers that fall within a 50-pixel radius of this boundary (green) were considered for the local collagen feature analysis. Fibers overlaid with red curvelets are not included in the analysis. I and J: Heat map of collagen fiber angles. The orientation of collagen fibers with respect to the TN boundary was pseudocolored according to the following scale: green indicates 10 to 45 degrees; yellow, >45 to 60 degrees; and red, >60 to 90 degrees (fibers <10 degrees were not assigned a color). Scale bars = 20 μm (A–J).

To describe how clinicopathologic factors and experimental markers jointly affect patient survival and their prognostic significance, a multivariate analysis was performed. All covariates that had a P < 0.05 were considered statistically significant. After adjusting for patient age at diagnosis and tumor stage, multivariate analysis for overall survival demonstrated that high expression of stromal CD163 (P = 0.001, HR = 2.90) is an independent predictor of poor patient survival, as are tumor size (P = 0.035, HR = 1.05) and positive lymph node status (P = 0.005, HR = 2.57) (Table 5). TACS-3 was borderline significant as an independent prognostic factor of poor OS (P = 0.066, HR = 1.72). In contrast, localized collagen alignment (without consideration of angle) had a protective effect in patient mortality (P = 0.01, HR = 0.05) (Table 5). This result is also observed in the univariate analysis (Table 4). It seems that alignment is protective for disease progression because often most collagen fibers were oriented parallel to the TN boundary. This parallel arrangement of collagen fibers likely has the effect of containing the tumor epithelium, preventing tumor cell invasion to metastasis, reflected by a low hazard ratio (Table 5). When collagen fibers are reoriented perpendicular to the tumor boundary (TACS-3), patients have poor survival (Table 5 and Figure 3). Moreover, multivariate analysis for progression-free survival revealed that positive lymph node status and high proliferation (Ki-67) were the only features that had a negative effect on patient survival after breast cancer recurrence (P = 0.005, HR = 2.75; and P < 0.0001, HR = 1.02, respectively) (Supplemental Table S5). Interestingly, only tumors containing moderate, but not high, expression of COX-2 in CD68⁺ macrophages had a protective effect on patient PFS (P = 0.005, HR = 0.3) (Supplemental Table S5). This result was not observed in the univariate analysis because none of the biomarkers had an impact in PFS. Overall, the univariate analysis suggests that high expression of COX-2 in the TN and high presence of macrophages in the TS are poor predictors of OS. Most important, the CD163⁺ macrophage is an independent biomarker of breast cancer poor OS. In addition, TACS-3 seems to be the collagen feature that is conductive to poor OS, and not collagen alignment.

Table 5.

Multivariate Cox Regression Analysis for OS

Variables	OS
	HR	95% CI	P value
Age at diagnosis	1.05	1.03–1.07	0.000
Tumor stage
I	1	Reference
II	0.88	0.41–1.87	0.734
III	0.96	0.31–2.93	0.938
Tumor size	1.01	1.00–1.02	0.035
Lymph node status (positive)	2.57	1.33–4.97	0.005
ER positive	0.42	0.22–0.81	0.010
PR positive	0.47	0.22–0.88	0.018
CD163 TS
1	1	Reference
2	1.19	0.42–3.37	0.739
3	3.84	1.99–7.4	0.000
4	2.90	1.52–5.54	0.001
Collagen alignment (local)∗	0.05	0.00–0.48	0.010
TACS3†	1.72	0.97–3.07	0.066

Significant P values are shown in bold.

ER, estrogen receptor; HR, hazard ratio; OS, overall survival; PR, progesterone receptor; TACS, tumor-associated collagen signature; TS, tumor-associated stroma.

^∗Computational measurement of collagen organization from second harmonic generation images.

^†TACS-3 includes aligned collagen fibers that are >60 degrees perpendicular to the tumor nest boundary.

Discussion

COX-2 expression is often seen in 50% to 70% of breast cancers, whereas dense macrophage infiltration is observed in a minority of cases. Herein, we tested the hypothesis that the tissue localization of COX-2 overexpression and high infiltration of protumoral TAMs have an association with high collagen alignment and invasive carcinoma patient outcome. We find that collagen deposition is significantly associated with high stromal expression of COX-2 and CD163⁺ macrophages. In addition, tumors with high infiltration of protumoral macrophages in both the tumor and stromal compartments associate with tumor size, grade, stage, and proliferation, suggesting a role for macrophages in tumor progression. Interestingly, compartmentalization of high COX-2 expression in the TN is associated with decreased OS in invasive breast carcinoma patients. High infiltration of CD68⁺ macrophages in the TS and CD163⁺ macrophages in both TN and TS impairs OS. In addition, high expression of COX-2 in CD68⁺ macrophages in TS and CD163⁺ macrophages in TN and TS leads to poor patient OS. Finally, the multivariate analysis reveals that CD163 in the TS may be an independent prognostic marker for OS in invasive breast cancer. These data suggest that invasive breast carcinoma patients whose tumors express high COX-2 or have high infiltration of protumoral macrophages (CD163⁺) might benefit from a selective COX-2 inhibitor therapy, such as celecoxib. Additional clinical studies will be needed to validate the results from this study.

Many studies have demonstrated the implication of COX-2 in the development and progression of many cancers, including breast and colon cancers.¹⁴ Besides cytokines, ILs, hormones, and growth factors, COX-2 can also be up-regulated by specific mutations in Wnt, Ras, and HER2 oncogenes.19, 20, 21 The correlation between COX-2 and HER2 has been controversial and may be attributable to the type and number of cases included in the cohort, such as varying patient demographics, tumor characteristics, and disease stage, among others.⁴⁰ In this study, however, when all clinicopathologic variables were jointly compared for their effect in COX-2 expression, HER2 receptor expression did not correlate with COX-2 expression. However, there was an association between COX-2 and ER/PR status. In fact, other reports found that high COX-2 expression was associated with decreased OS in ER⁺ breast cancer tumors, confirming our results.23, 40, 41, 42 The proliferation marker Ki-67 has been previously associated with COX-2,23, 41 and an association was found between Ki-67 and all markers of tumor inflammation. A key mechanism for these correlations is that the up-regulation of COX-2/PGE2 promotes induction of aromatase cytochrome P450 (CYP19) and aromatase-catalyzed estrogen in a local paracrine manner, causing uncontrolled epithelial cell proliferation, and PGE2-induced cytochrome P450 1B1 (CYP1B1) in epithelial cells promotes paracrine estrogen to be converted into estrogen quinones, which has mutagenic effects.⁴³

To our knowledge, this is the first study to look at COX-2 and its expression in different tumor tissue compartments: tumor nest and tumor-associated stroma. The univariate analysis revealed that only high COX-2 expression in the TN resulted in reduced OS. Similarly, a study by Ristimäki et al,²³ which used the largest patient cohort of 1984 cases of different types of breast cancer, found that COX-2 expression was only negatively associated with OS. However, other studies containing patient cohorts from different regions of the world found associations for overall COX-2 overexpression (includes whole tissue) in breast cancer patients for OS, PFS, and relapse-free survival.40, 41, 42, 44 Clearly, this study demonstrates that the specific tissue compartment where COX-2 overexpression is localized is an important factor for predicting patient prognosis.

The ECM plays an important role in tumor development. In general, the stromal expression of all of the inflammatory markers evaluated in this study was associated with total collagen deposition. However, when the association of collagen features with clinicopathologic features was evaluated, total collagen deposition was found to be associated with tumor grade, ER status, and proliferation (Supplemental Table S4). Interestingly, collagen fiber orientation perpendicular to the tumor boundary (TACS-3), but not deposition, was associated with poor overall survival, whereas local collagen density was a protective marker for PFS (Figure 3 and Table 4). The multivariate regression analysis reveals that local collagen alignment has a protective effect for breast carcinoma patients; nevertheless, perpendicular alignment to the tumor boundary (TACS-3) may be an independent prognostic marker (borderline statistically significant) of poor OS (Figure 3 and Table 5). Overall, these data indicate that high expression levels of COX-2, CD68, and CD163 have a role in tumorigenesis, where their increased expression is associated with various properties of the ECM, including collagen deposition, alignment, and TACS-3. Most important, TACS-3 predicts invasive breast cancer patient overall mortality, whereas local collagen alignment without regard to the orientation of the fibers predicts a higher chance of survival to breast cancer patients.

The tumor microenvironment also contains a variety of stromal cells, such as tumor-associated macrophages. Recent evidence demonstrated that protumoral TAMs are involved in several tumor-promoting processes, including proliferation, promotion of breast tumor epithelial cell migration, and metastasis.3, 4 CD68 and CD163 are glycoproteins that are expressed in human monocytes and tissue macrophages and play a role in the immune response.5, 6 Although CD163 is considered a highly specific monocyte/macrophage marker for polarized protumoral macrophages,6, 8, 9 CD68 is a pan-macrophage marker that recognizes both protumoral and antitumoral macrophages.⁵ CD68 and CD163 are associated with larger tumor size, grade, and proliferation in both TN and TS, and CD163 expression is additionally associated with patient age and ER/PR status. Furthermore, the univariate analysis revealed that high expression of stromal CD68 and high expression of CD163 in both TN and TS are negatively associated with OS. These results differ from those of Medrek et al,⁴⁵ because they were able to find clinicopathologic and survival associations with both markers of macrophage infiltration exclusively in the tumor-associated stroma. Consequently, multivariate analysis revealed that high infiltration of stromal CD163⁺ macrophages is an independent negative prognostic factor for overall survival. In contrast, Medrek et al⁴⁵ found that CD68 in the TS is an independent prognostic marker for breast cancer–specific survival. Possible differences in results may be attributable to distinct patient cohorts: number of samples analyzed, ethnicity, variation in age, distribution of receptor status and molecular types, and patient treatment.

Moreover, there is evidence of macrophages expressing high COX-2 levels in the tumor microenvironment. It has been suggested that COX-2–derived PGE2 can enhance the protumor capabilities of TAMs.10, 11, 46 For instance, higher expression levels of COX-2 in TAMs has been suggested as a prognostic factor for melanoma progression, and dense infiltration of COX-2–expressing TAMs is observed after tumor irradiation, leading to early prostate cancer growth in mice.47, 48 In addition, COX-2 inhibition with etodolac caused loss of protumoral macrophage characteristics and diminished breast cancer metastasis.⁴⁹ Overall, the univariate survival analysis showed that high infiltration of COX-2/CD68⁺ macrophages in the TS and COX-2/CD163⁺ macrophages in both TN and TS led to reduced patient OS. Remarkably, a study by Li et al,⁴⁶ in a Chinese cohort of patients, found that COX-2⁺ TAMs, as measured with CD163 expression in the whole tissue, were associated with poor patient survival and may serve as an independent prognostic biomarker. Interestingly, in the multivariate analysis for PFS, after adjusting for age and tumor stage, cells that expressed moderate, but not high, levels of COX-2 and CD68 in the TS had a protective effect. This may mean that most of these macrophages have antitumoral effects protecting the patient from disease progression. However, this effect is not observed at the highest expression levels of CD68. Overall, these data suggest that high infiltration of protumoral TAMs, as labeled with CD163, is an independent prognostic factor of poor patient survival and may play an important role in breast cancer tumor development.

Furthermore, it would be important to evaluate the role of high COX-2 expression and its association with tumor-promoting TAMs in larger studies, which include more demographic data, such as menopausal status, body mass index, and mammographic density. Such data are important to take into consideration, because high body mass index in post-menopausal women and mammographic density have been previously reported as breast cancer risks.50, 51 Besides, it would be insightful to further evaluate the role of collagen matrix deposition and other related factors, such as mammographic density, to study plausible mechanisms in which the matrix has been remodeled in the tumor microenvironment because of both COX-2 overexpression and its inhibition. Previously, we reported the association between high expression of COX-2 or PGE2 and increased collagen deposition in collagen-dense tumors.¹³ In addition, other reports demonstrated that increased COX-2 and collagen I gene expression was correlated with reduced survival and shorter time to metastasis,³⁰ and collagen alignment is negatively associated with patient survival outcome.29, 35 Larger-scale studies would be needed to conclude whether COX-2, protumoral TAMs, and perpendicular collagen alignment to the tumor boundary (TACS-3) may be independent prognostic markers or whether they are features involved in initiating and promoting breast cancer. COX-2 inhibitors are relatively inexpensive when compared with standard cancer treatments. For instance, celecoxib, a selective COX-2 inhibitor, has tolerable adverse effects and makes tumor cells more susceptible to radiotherapy.⁵² Taken together, these data suggest that high infiltration of protumoral COX-2–expressing TAMs, coupled with aligned collagen orientation, predicts poor patient survival and may play key roles in breast cancer tumor development. On the basis of these data, we propose that breast cancer patients with increased COX-2 expression and COX-2–expressing CD163⁺ macrophages would benefit from COX-2 inhibitory therapy.

Supplemental Data

Supplemental Figure S1.

Tissue segmentation and biomarker analysis. Images for the process of tissue segmentation. Algorithms for tissue segmentation (ie, tumor epithelium versus tumor stroma), were generated by machine learning (Materials and Methods). A: Sample immunofluorescence image cube of mouse tumor stained with pan-cytokeratin (green), cyclooxygenase 2 (COX-2; orange), CD68 (magenta), and CD163 (red), and counterstained with DAPI (turquoise) (20× objective). B: Tissue segmentation map after training the software. Red indicates epithelium; green, stroma; and blue, other (empty space and debris/artifacts). C: Overlay of total object cell count map. Inset: A region to depict cell surface markers (colorful cell outlines); cells were further segmented into nuclei and membrane. D and E: Each object (cell) circled in green in C was associated to its respective tissue compartment: tumor nest (D) or tumor-associated stroma (E). Blue indicates negative cells for a marker; and red, positive cells for a marker. Debris was associated with the other category and not included in the data analysis. F: Cell count for COX-2 positivity. Blue indicates 0+; yellow, 1+; orange, 2+; and purple, 3+. The sum of the percentage of cells falling into each of the 0+ to 3+ categories was converted into an H-score. Scale bar = 100 μm (A–F). Original magnification, ×1200 (inset).