Exploring Collagen Parameters in Pure Special Types of Invasive Breast Cancer

One of the promising tools to evaluate collagen in the extracellular matrix is the second-harmonic generation microscopy (SHG). This approach may shed light on the biological behavior of cancers and their taxonomy, but has not yet been applied to characterize collagen fibers in cases diagnosed as invasive breast carcinoma (BC) of histological special types (IBC-ST). Tissue sections from 99 patients with IBC-ST and 21 of invasive breast carcinoma of no special type (IBC-NST) were submitted to evaluation of collagen parameters by SHG. Tissue microarray was performed to evaluate immunohistochemical-based molecular subtype. In intratumoral areas, fSHG and bSHG (forward-SHG and backward-SHG) collagen parameters achieved their lowest values in mucinous, papillary and medullary carcinomas, whereas the highest values were found in classic invasive lobular and tubular carcinomas. Unsupervised hierarchical cluster analysis and minimal spanning tree using intratumoral collagen parameters allowed the identification of three main groups of breast cancer: group A (classic invasive lobular and tubular carcinomas); group B (IBC-NST, metaplastic, invasive apocrine and micropapillary carcinomas); and group C (medullary, mucinous and papillary carcinomas). Our findings provide further characterization of the tumor microenvironment of IBC-ST. This understanding may add information to build more consistent tumor categorization and to refine prognostication.

IBC-ST refers to the architectural growth of the tumors, defining special morphological and cytological patterns, which have been consistently associated with distinctive clinical presentation and/or outcomes 3 . However, although histological grade identifies prognostic subgroups in IBC-NST, among the IBC-ST some subtypes show high grade but still bear favorable prognosis (e.g. medullary, mucinous and invasive apocrine breast carcinomas) 3 . Interestingly, genetic and transcriptomic features of breast cancers has been correlated with histological grade, and microarray-based genomic signatures for histological grades have been devised [14][15][16][17] .
The use of new investigation tools in IBC-ST has been limited partly due to their relative low prevalence with consequent lower interobserver reproducibility, skewing their systematic investigation in class discovery [10][11][12][13] and class prediction 16,[18][19][20][21] . This is the case of tumor microenvironment, especially the extracellular matrix (ECM), which has been intensively studied in tumor progression and outcome 22 . Among the various components of ECM, collagen fibers are the most abundant in breast cancer stroma 23,24 . The restructuring of collagen pattern may be either induced by tumor cells, or occur as a result of an integrated modification between neoplastic cell and stroma, suggesting that progression of a tumor relies not only on changes of cell function, but also on ECM alterations.
One of the promising tools to evaluate collagen in the ECM during cancer progression is the second-harmonic generation microscopy (SHG) [25][26][27] . This approach may shed light on the biological behavior of cancers and their taxonomy 28,29 . It allows for the identification of collagen maturation process by distinguishing collagen diameters. Thus, backward SHG images exhibit punctate distribution attributable to small-diameter, segmental collagen, which permits recognition of fibrillogenesis in immature tissue by directly imaging backward-propagating SHG (bSHG). By contrast, images from the mature fibrils are identical in the forward (fSHG) and backward directions 30 . Although it has not been applied to IBC-ST, several studies have shown that this is an efficient means to describe common breast diseases, differentiate benign from malignant breast tissue and establish prognosis in invasive ductal carcinoma [31][32][33][34][35] . It should be emphasized that studies also indicate that certain aspects of collagen deposition may lead to a more aggressive behavior of the tumor, like greater amount of collagen, presence of perpendicular fibers to the tumor bounder and thicker collagen fibers 34,[36][37][38] . Thus, the goals of the present study are: (1) to investigate peri-and intratumoral collagen parameters in various histological subtypes of pure IBC-ST, using IBC-NST as standard samples, and (2) correlate these parameters with pathological and clinical features. . Among all 120 cases, most carcinomas were of histological grade III (n = 48; 40.0%), followed by grade I (n = 43; 35.8%) and grade II (n = 29; 24.2%) carcinomas. The majority of the patients did not present lymph node metastases (n = 54; 45.0%); these were found in 25 patients (20.8%). According to the status of immunohistochemical expression of hormone receptors (HR) and HER2/neu protein used as surrogate for molecular types, cases were classified as follows: HR+HER2− was the most frequent subtype (n = 55; 45.8%), followed by HR−HER2− (n = 26; 21.7%), HR+HER2+ (n = 4; 3.3%), and HR−HER2 + (n = 4; 3.3%). In 31 cases (25.8%), patients had no further material for immunohistochemical staining, thus immunohistochemical-based molecular subtyping was not possible for these patients. www.nature.com/scientificreports www.nature.com/scientificreports/ Collagen features in breast cancer. The correlation between collagen parameters (quantity, uniformity and organization) in peritumoral areas was poor, contrasting to high correlation coefficients found in intratumoral areas (Supplementary Table 1 and Fig. 1).

Results
In general, collagen quantity, uniformity and organization were higher in peritumoral regions compared to intratumoral areas (Supplementary Table 2). In classic invasive lobular, tubular, micropapillary and metaplastic carcinomas, there was no statistical difference between fSHG peri-and intratumoral collagen organization. The same was observed for bSHG peri-and intratumoral collagen organization in classic invasive lobular and micropapillary carcinomas.
Peritumoral collagen parameters for both fSHG and bSHG fibers did not differ significantly across histological subtypes ( Table 2 and Supplementary Fig. 2). In contrast, in intratumoral areas, fSHG and bSHG collagen quantity, uniformity and organization achieved the lowest values in mucinous, papillary and medullary carcinomas, whereas the highest values for these parameters were found in classic invasive lobular and tubular carcinomas ( Peritumoral and intratumoral bSHG collagen quantity (p = 0.008 and p = 0.003, respectively) and uniformity (p = 0.009 and p = 0.005, respectively) were higher in histological grade I tumors compared to grade III tumors. However, there was no difference between peri-and intratumoral collagen quantity and uniformity, considering histological grade II tumors and grade I or histological grade II and grade III tumors. Peri-and intratumoral collagen organization was similar across histological grades (Supplementary Table 4). Peritumoral collagen quantity, uniformity and organization did not differ in relation to lymph node status and immunohistochemical-based subtype. However, intratumoral collagen bSHG fiber uniformity and organization were higher in HR−HER2− tumors compared to HR−HER2+ tumors (Supplementary Table 5).
The multivariate recursive partitioning method, using intratumoral collagen parameters, showed that fSHG and bSHG quantity, uniformity and organization were lower in mucinous, papillary and medullary tumors (p < 0.001). On the other hand, the three parameters were significantly higher in classic invasive lobular and tubular carcinomas (p = 0.005) compared to the other histological subtype. IBC-NST, micropapillary, metaplastic and invasive apocrine carcinomas presented an intermediary fSHG and bSHG intratumoral collagen parameters ( Supplementary Fig. 3).
Unsupervised hierarchical cluster analysis using intratumoral collagen ( Fig. 2 and Supplementary Figs 4 and 5) allowed the identification of three main groups of breast cancer. Group A: high intratumoral collagen quantity, uniformity and organization (classic invasive lobular and tubular carcinomas); group B: intermediary intratumoral collagen quantity, uniformity and organization (invasive ductal, metaplastic, invasive apocrine and micropapillary carcinomas); and group C: low intratumoral collagen quantity, uniformity and organization (medullary, mucinous and papillary carcinomas). Supervised analysis using estrogen receptor (ER) expression or molecular subtype presented the same result. Figure 3 visually summarizes the distinct groups of breast carcinomas according to intratumoral collagen parameters.  www.nature.com/scientificreports www.nature.com/scientificreports/ Mucinous, medullary and metaplastic breast carcinoma subclassification. Collagen parameters did not differ between mucinous A and B breast carcinoma. In addition, peritumoral collagen parameters in typical and atypical medullary were similar. Intratumoral fSHG collagen quantity (p = 0.017) and uniformity (p = 0.044) were higher in atypical medullary breast cancer; these findings may be extended to bSHG fiber quantity (p = 0.054) and uniformity (p = 0.054); fSHG and bSHG fiber organization was similar in both tumor subtypes (p = 0.198 and p = 0.064, respectively).
There were no differences in collagen parameters comparing metaplastic breast carcinoma with squamous cells and with matrix-producing breast carcinomas, except for peritumoral fSHG collagen organization, that was higher in metaplastic breast carcinoma with squamous cells (p = 0.017).

Discussion
Our results allowed grouping breast carcinomas according to the intratumoral collagen parameters (Fig. 3). Group A, characterized by higher intratumoral collagen quantity, uniformity and organization, was enriched for classic invasive lobular and tubular carcinomas; group B, comprised by intermediary intratumoral collagen quantity, uniformity and organization, was enriched for IBC-NST, metaplastic, invasive apocrine and micropapillary carcinomas; and group C, covered lower intratumoral collagen quantity, uniformity and organization, was enriched for www.nature.com/scientificreports www.nature.com/scientificreports/ medullary, mucinous and papillary carcinomas. Furthermore, it would be worth highlighting that IBC-NST are not only heterogeneous at the morphologic and molecular level, but also at the clustering analysis performed herein; cases of IBC-NST could be found in any of the possible three groups identified, even if we used a supervised clustering analysis. www.nature.com/scientificreports www.nature.com/scientificreports/ The findings on intratumoral collagen corroborate the disparities between morphological classification and molecular biology. Heterogeneous tumor biology may be found within the same histological subtype, whereas distinct histological categories share common molecular features 39,40 . As such, morphologic subtyping may be challenging not only because of subjectivity, but also as a result of molecular similarities. Peritumoral collagen fibers presented similar parameters among all histological subtypes of breast cancer, suggesting a common endpoint, shared by them, in the mechanism of collagen deposition at tumor periphery. This acknowledgement may support the pivotal interplay of tumor microenvironment in cancer outcome.
Classic invasive lobular and tubular carcinomas, constituting group A in the present study, have been described as showing remarkably similar immunohistochemical and transcriptomic profiles 39 . Notwithstanding, invasive lobular carcinoma can be differentiated from tubular carcinoma based on the expression levels of E-cadherin, which is absent in the former 41,42 . Regarding collagen parameters, both histological subtypes presented similar intratumoral features, and shared the same hierarchical clustering, supporting the acquaintance between them. These findings provide structural evidence for the hypothesis that classic invasive lobular and tubular carcinomas, both frequently members of the low-grade breast carcinoma group, might originate from the same family of low-grade precursors 41 . Lopez-Garcia et al. support the existence of a "low-grade breast neoplasia family"; nevertheless, the transcriptomes of these lesions display small, yet important differences, which together with their distinct biological behavior, warrant their separation as discrete entities 43 .
In opposition to the subtypes above, a clear disparity in the clinical, morphological, molecular or immunohistochemical profiles was found within the other two groups of tumors with similar intratumoral collagen parameters, groups B and C 39,40,44 . This is evidence that, although the microenvironment may importantly contribute to progression of cancer in all phases, factors intrinsic to the neoplastic cells may respond for other important features, as the phenotype and genetic characteristics 39,40 .
Subtypes included in our group B have been described as presenting discrepant clinicopathological behavior. Metaplastic and micropapillary carcinomas present mostly poor outcome, in spite of ER status, respectively, negative and positive; apocrine carcinoma, tipically ER-negative, present variable outcome 45,46 . In a study of genomic profiling, carcinomas with apocrine differentiation, invasive micropapillary and metaplastic carcinomas showed the highest frequencies of chromosome aberrations (i.e. gene amplifications, gains and losses) 40 . Thus, although this group seems heterogeneous as for clinical outcome and for the variable expression of ER, the collagen parameters behaved similarly, and paralleled the high numbers of chromosome aberrations. It may be assumed that the collagen signature of group B is predictive of higher frequency of finding genetic alterations.
On the other hand, group C, enriched for medullary, mucinous and papillary carcinomas have been related to a more indolent clinical behavior 45 . The three subtypes in group C have been described with similar gross growth pattern in image exams, presenting as a round or oval tumor with well-defined contours 47 . Genetic results demonstrated that papillary carcinomas are a homogenous special histological subtype of breast carcinoma, with good prognosis. They tend to present low rates of lymph node metastasis and low p53 expression, ER-positive, low frequency of gene copy number aberrations and high prevalence of PIK3CA mutations. The genomic profile of its three morphological variants (i.e. encapsulated, solid and invasive papillary) has shown remarkably similar 46 . Likewise, mucinous breast carcinoma display ER-positivity, low level of genetic instability and rare recurrent amplifications [48][49][50][51] . These genetic features reasonably approximate both tumors in hierarchic clusters 39 . In opposition, medullary carcinomas have been shown to share the basal-like immunophenotype (i.e. HR negative, HER-2 negative and P-cadherin positive), and high frequency of chromosome aberrations 39,40,[52][53][54][55] . In spite of this divergence in profiling, medullary carcinomas show, in common with the other two members of group C, www.nature.com/scientificreports www.nature.com/scientificreports/ the favorable prognosis and the lower intratumoral collagen quantity, uniformity and organization. As such, this group is homogeneous only in respect to clinical behavior for all three members 45 .
Metaplastic and medullary carcinomas, grouped herein as B and C, respectively, are consistently of triple-negative phenotype, and both were shown to display similar high frequency of gene copy number aberrations and comparable gene expression patterns 39,40,[52][53][54][55] . In relation to the concordance between the molecular subtyping and other prognostic gene signatures, patients with the basal-like type should be classified as aggressive tumor 9,56 . However, basal-like medullary carcinomas present a good outcome in opposition to metaplastic carcinoma 3 . Then, in contrast to molecular grouping, intrinsic intratumoral collagen parameters clustering proposed in this study assembled these subtypes of breast carcinomas more suitably.
The rarity of the entities analyzed herein represents a drawback. Additional studies of special subtypes of breast carcinoma will be required to validate these findings, and to determine the biological and clinical relevance of clustering these tumors according to the intratumoral collagen parameters. In special, studies on patients' outcome for each individual subtype will require larger cohorts. Also, the long collection period might affect the evaluation of collagen structure, although previous studies pointed out that tissue fixation time did not interfere with the availability of macromolecules 57,58 . Further, the present study evaluated images only with the 40×/1.3 oil immersion magnification, largely used in the literature. It would be interesting to specifically address the comparison of collagen parameters values between different magnifications, in order to evaluate the impact of microscopic resolution for these analyses.
Previous studies have already demonstrated the prognostic value of collagen parameters in breast cancer, showing that high disorganization of collagen fibers, i. e. TACS3, is indicative of poorer prognosis 34,37 . In a study on the evaluation of fibrils deposition in collagen fibers, it was also found that different patterns are related to the molecular subtypes 59 . However, most of them did not emphasize the usage of this technique on IBC-ST, with the exception of the study by Conklin et al., in which collagen parameters were evaluated in invasive lobular carcinoma 34 . In addition, these studies have not compared different collagen signatures in the histological subtypes. Further, the method applied in the present study was quantitative, in contrast to the qualitative or semiquantitative approach of the previous ones. Besides, the results presented herein could provide the basis for future automated analyses with specific mathematical algorithms for classification, as already described 60 . However, this application is still potential, as, to the best of our knowledge, our study represents the first approach to compare collagen parameters in IBC-ST.
In summary, our results do not allow a prognostic stratification, but provide information that could contribute to refine diagnosis. In that sense, an example would be the differential diagnosis between medullary carcinomas and triple-negative IBC-NST, since these entities belong to different groups, as proposed here. Our results provide a step forward to the characterization of the tumor microenvironment of IBC-ST. This understanding may add information to build more consistent tumor categorization, together with molecular findings. Grouping the different histological subtypes according to collagen parameters, as suggested herein, may represent a plausible explanation for the observation that tumors with different genetic profiles share similar clinical behavior. It potentially contributes to the refinement of the microenvironmental phylomorphology, and of the prognostication of breast cancer patients.

Materials and Methods
Case selection. Formalin-fixed, paraffin-embedded (FFPE) tissue specimens from 120 patients with pure invasive breast carcinoma were consecutively retrieved from four centers of Pathology (two in Brazil, one in Portugal and one in France). Tissues were obtained from primary lumpectomy or mastectomy; no patient had undergone primary chemo-or radiotherapy. Tumors were classified based on the WHO criteria 3 : 21 IBC-NST, 7 classic invasive lobular, 13 tubular, 23 mucinous, 6 invasive micropapillary, 9 invasive papillary, 15 medullary, 9 metaplastic and 17 apocrine carcinomas, from January 2000 to January 2015. Mucinous carcinoma were subdivided into hypocellular mucinous (mucinous A; n = 13) and hypercellular mucinous carcinoma (mucinous B; n = 10) based on the criteria of Capella et al. 61 . All cases were reviewed on hematoxylin and eosin-stained (H&E) sections by at least two experienced breast pathologists (GRP, CAA, RVF and FCS). The study is fully compliant with the Declaration of Helsinki (approved by the "Comitê de Ética em Pesquisa da Unicamp").
In order to confirm diagnosis and evaluate immunohistochemical-based molecular subtype 39,44 , two representative intratumoral areas of 2 mm in diameter from each case were selected to build a tissue microarray (tissue microarray (TMA) builder 20010.02, Histopathology, Pécs, Hungary) for the immunohistochemical analysis. Further, three representative regions from peri-and intratumoral areas were selected on the same H&E sections (4 µm each), and marked for SHG methodology ( Supplementary Information 1).
Immunohistochemical staining and evaluation. TMA sections were deparaffinized, rehydrated and submitted to antigen retrieval. The primary antibodies used, dilutions and detection methods are briefly described in Supplementary Table 6. Seven cases of medullary carcinoma kindly provided by Camille Franchet were analyzed in whole sections. The evaluation of immunohistochemical results was centrally performed by two experienced breast pathologists (CAA and FCS).
Staining for estrogen-receptor alpha (ER), progesterone receptor (PR) and androgen receptor (AR) were evaluated according to the previously described guidelines 62 . For p63, keratin 5 and P-cadherin, samples were scored as positive when ≥10% cells were immunoreactive 63 . Membrane expression for HER-2 and EGFR was evaluated as previously described 64 . Ki-67 was evaluated by counting 1000 neoplastic cells in areas with highest positivity; cases with <20% positive cells were considered to have low proliferation, and the others, high proliferation 65 Fig. 4 and Supplementary Fig. 7). Three peri-and three intratumoral areas selected for collagen assessment on H&E stained sections were examined in this condition, and digital images were collected for further evaluation. Peritumoral regions were defined as the fibrous tissue at the tumor borders, in the transition with non-neoplastic tissue. Intratumoral regions corresponded to fibrous bands within groups of neoplastic cells. In cases with multiple tumor sections, the most representative one was chosen.
Evaluation of collagen fibers. SHG images were stratified regarding collagen parameters using image pattern analysis methods. Collagen parameter refers to the pixels characterization in the SHG image, being evaluated through three features: quantity, uniformity and organization. Quantitative analysis of collagen parameters was performed in SHG images using ImageJ (http://imagej.nih.gov/ij/) and OrientationJ plug-in 66 . For this purpose, 16 representative areas to cover entire image (256 × 256 pixels) were performed in each image. The final value of each parameter represents the average of the values obtained in the three images; the value of each image denotes the average of the 16 representative areas of each image. (Supplementary Information 1). IBC-NST, being the most common subtype of breast cancer, was used as a comparison parameter for the other cases. Therefore, special cases may present collagen parameters equal, higher or lower than IBC-NST. Furthermore, a previous report demonstrated that the area of the cellular component does not represent a confounding variable in the collagen parameters evaluation 67 , so hypo-or hypercellular tumors were evaluated in the same way. www.nature.com/scientificreports www.nature.com/scientificreports/ statistical analyses. Collagen quantity and uniformity were log-transformed to base e for statistical purposes; statistical analyses were performed using R (https://cran.r-project.org/). Shapiro-Wilk test was performed to analyze the data distribution of collagen parameters. Comparisons were performed using t-tests or analysis of variance (ANOVA). Pearson's Correlation coefficients were calculated to analyze the correlation between collagen parameters. A multivariate recursive partitioning model was fit to determine groups of collagen between breast cancer subtypes. Hierarchical clustering analysis was performed, using Euclidean distance and Ward's clustering algorithm 49,51,68 and minimal spanning tree ( Supplementary Information 1). p values were corrected using Benjamini and Hochberg adjustment to minimize α errors, and adjusted p values less than 0.05 were considered significant.