Original Article

Tumor Biology
April 2018: 1–12
Ó The Author(s) 2018
Collagen analysis by second-harmonic Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
generation microscopy predicts DOI: 10.1177/1010428318770953
journals.sagepub.com/home/tub

outcome of luminal breast cancer

Rodrigo A Natal1 , José Vassallo1,2, Geisilene R Paiva3, Vitor B Pelegati4,5,

Guilherme O Barbosa6, Guilherme R Mendonc xa1, Caroline Bondarik1,
Sophie F Derchain7, Hernandes F Carvalho5,6, Carmen S Lima8,
Carlos L Cesar4,5 and Luı́s Otávio Sarian7

Abstract
Second-harmonic generation microscopy represents an important tool to evaluate extracellular matrix collagen struc-
ture, which undergoes changes during cancer progression. Thus, it is potentially relevant to assess breast cancer devel-
opment. We propose the use of second-harmonic generation images of tumor stroma selected on hematoxylin and
eosin–stained slides to evaluate the prognostic value of collagen fibers analyses in peri and intratumoral areas in patients
diagnosed with invasive ductal breast carcinoma. Quantitative analyses of collagen parameters were performed using
ImageJ software. These parameters presented significantly higher values in peri than in intratumoral areas. Higher intra-
tumoral collagen uniformity was associated with high pathological stages and with the presence of axillary lymph node
metastasis. In patients with immunohistochemistry-based luminal subtype, higher intratumoral collagen uniformity and
quantity were independently associated with poorer relapse-free and overall survival, respectively. A multivariate
response recursive partitioning model determined 12.857 and 11.894 as the best cut-offs for intratumoral collagen quan-
tity and uniformity, respectively. These values have shown high sensitivity and specificity to differentiate distinct out-
comes. Values of intratumoral collagen quantity and uniformity exceeding the cut-offs were strongly associated with
poorer relapse-free and overall survival. Our findings support a promising prognostic value of quantitative evaluation of
intratumoral collagen by second-harmonic generation imaging mainly in the luminal subtype breast cancer.

Keywords
Breast cancer, second-harmonic generation, collagen fibers, tumor microenvironment, prognosis

Date received: 25 December 2017; accepted: 19 March 2018

1
Laboratory of Investigative and Molecular Pathology, Center for
7
Investigation in Pediatrics (CIPED), Faculty of Medical Sciences, State Department of Obstetrics and Gynecology, Faculty of Medical Sciences
University of Campinas, Campinas, Brazil and CAISM—Women’s Hospital, State University of Campinas,
2
Department of Anatomic Pathology, A.C. Camargo Cancer Center, São Campinas, São Paulo, Brazil
8
Paulo, Brazil Oncology Section, Department of Internal Medicine, Faculty of Medical
3
Laboratory of Experimental Pathology (LAPE), CAISM—Women’s Sciences, State University of Campinas, Campinas, São Paulo, Brazil
Hospital, State University of Campinas, Campinas, Brazil
4
Department of Quantum Eletronics, ‘‘Gleb Wataghin’’ Institute of Corresponding author:
Physics, State University of Campinas, Campinas, Brazil Luı́s Otávio Sarian, Department of Obstetrics and Gynecology, Faculty of
5
INFABIC—National Institute of Science and Technology on Photonics Medical Sciences and CAISM—Women’s Hospital, State University of
Applied to Cell Biology, Campinas, Brazil Campinas, Rua Tessália Vieira de Camargo, 126, Campinas 13083-970,
6 São Paulo, Brazil.
Department of Structural and Functional Biology, Institute of Biology,
State University of Campinas, Campinas, Brazil Email: [email protected]

Creative Commons Non Commercial CC BY-NC: This article is distributed under the terms of the Creative Commons
Attribution-NonCommercial 4.0 License (http://www.creativecommons.org/licenses/by-nc/4.0/) which permits non-commercial
use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and
Open Access pages (https://us.sagepub.com/en-us/nam/open-access-at-sage).
2 Tumor Biology

Introduction that this approach was predictive of local invasion and

that the visualization of radially distributed COL fibers
Second-harmonic generation (SHG) has become an
(TACS-3) correlated with increased local tumor inva-
important tool in evaluating collagen (COL) structure
sion and poorer long-term survival in patients with
in the extracellular matrix (ECM) to assess cancer
breast cancer.16–18 These reports speculated that the
development.1,2 This technique may represent a more
dispersal of tumor cells into the surrounding tissue and
objective automatic and quantitative method for cancer
vasculature was facilitated by modified COL in the
prognostication. Current strategies to determine the stroma, as also demonstrated in vitro.19 In addition, a
prognosis and treatment of breast cancer are ruled by recent study has demonstrated that higher COL density
tumor size, lymph node status, histological grade, was linked to positive axillary lymph nodes.20 It was
expression of steroid hormone receptor, and human also shown that treatment with trastuzumab could
epidermal growth factor receptor 2 (HER-2) status. determine alterations in COL organization.21
However, these factors insufficiently explain the biolo- Molecular categorization of breast cancer has been
gical and clinical behavior of breast cancer because widely used to define optimal therapeutic modalities.22
tumors that have similar traditional predictive and In brief, breast cancer could be classified into five
prognostic factors can achieve distinct outcomes.3,4 immunohistochemistry (IHC)-based subtypes, surro-
It has been proposed that the composition and orga- gate for molecular subtypes, as follows: luminal A
nization of the ECM in breast cancer might have prog- (expression of steroid hormone receptors, HER-2 nega-
nostic value because changes in the ECM appear to be tive and low histological grade); luminal B (expression
associated with the release of extra and intracellular of steroid hormone receptors, HER-2 negative and
signals and the synthesis of proteases.3,5 These factors, high histological grade); luminal-HER-2 (expression of
in turn, influence tumor behavior, such as its growth, steroid hormone receptors, HER-2 positive); HER-2-
invasion, and metastasis.3,5 Of the elements of the overexpression (no expression of steroid hormone
ECM, COL fibers are the most abundant component receptors, HER-2 positive); and triple-negative (no
of breast cancer stroma.6,7 expression of steroid hormone receptors, HER-2 nega-
In the past decade, SHG has been applied to analyze tive).23 However, when genomic status together with
noncentrosymmetric molecules, such as COL, at high the cell proliferation activity was compared between
resolution without fluorescent staining, photodamage, IHC-based subtypes, it was found that the luminal A
or photobleaching.8,9 As indicated by its name, SHG and luminal B subtypes should be further divided into
shows a strong signal precisely at twice the frequency more groups,24 reinforcing heterogeneity within the
and half the wavelength of the excitation beam,10 with luminal subtype breast cancer. Moreover, in 2012, the
high specificity. Previous work showed that SHG is genomic landscape of breast cancer was extensively
quite independent of the excitation wavelength for the characterized using next-generation sequencing
typical Ti:Sapphire laser tuning range (680– (NGS).25 These studies have provided new opportuni-
1000 nm),11,12 so if one has to choose the fundamental ties to advance treatment of luminal tumors but
wavelength to match the filters, they have to detect the revealed remarkable intertumor heterogeneity, which
SHG without any detrimental effect to the signal. It is highlighted new challenges.26,27
a coherent signal, meaning that the electromagnetic Our purpose in this study is to evaluate the prognos-
fields generated by aligned permanent dipoles add up, tic value of COL changes in ECM of breast cancer,
and the intensity would be proportional to the square mainly addressing invasive ductal carcinoma of IHC-
of the number of dipoles. Therefore, SHG of aligned based luminal subtype, which is more frequent and
COL fibers, highly crystalline, is more intense than may present heterogeneous prognosis. Thus, COL
other forms. Literature shows that type I COL with its alterations were quantitatively assessed by SHG and
three hydrogen bond a-chains and self-assembled into correlated with clinicopathological features and tumor
fibrils is the most intense SHG emitter in most human progression (relapse-free (RFS) and overall survival
tissues.8 SHG provides image analyses in great detail, (OS)). We also intend to determine the reproducibility
with qualitative and quantitative measurements.13–15 of this technique. This quantitative analysis of COL
In a pivotal study in an animal model, Provenzano and its correlation with a variety of clinicopathological
et al.16 defined three tumor-associated COL signatures features represents a distinctive approach for breast
(TACSs). TACS-1 was established at an early stage of cancer and supports the clinical value of TACS. We
tumor progression and was characterized by a localized observed that COL quantity, uniformity, and organiza-
increase in the deposition of COL near the tumor. tion (see section ‘‘Materials and methods’’ for more
TACS-2 was distinguished by the straightening of COL detail) parameters were significantly different in peri
fibers around the tumor boundaries, and in TACS-3, and intratumoral areas. We detected an association
COL was aligned perpendicularly to the tumor bound- between intratumoral COL uniformity with higher val-
aries.16 Qualitative studies using TACSs demonstrated ues of body mass index (BMI), advanced pathological
Natal et al. 3

stage, and presence of axillary lymph node metastasis, Fluorescent in situ hybridization (FISH), used to detect
all features related to unfavorable prognosis. More HER-2 amplification, was performed using the HER-2/
importantly, intratumoral COL quantity and unifor- neu probe (VYSIS 36-161060, Des Plaines, IL, USA).
mity in breast cancer were related to distinct outcomes Immunohistochemical typing was performed as fol-
within the luminal IHC-based subtype. lows: luminal A (ER and or PR+/HER-2–/histological
grades I or II), luminal B (ER and or PR+/HER-2–/
histological grade III), luminal-HER-2 (ER and or
Materials and methods PR+/HER-2+), HER-2-overexpression (ER–/PR–/
Patients and human tissues HER-2+), and triple-negative (ER–/PR–/HER-2–). As
the immunohistochemical expression of Ki67 is still
This cohort study used retrospective archival material subject of standardization (due to preanalytical issues,
from the Department of Pathology and patients’ data technical and assessment variations), histological grade
compilation, at the Center for Integral Attention to was used, as suggested elsewhere.22 For statistical pur-
Women’s Health (CAISM), University of Campinas poses, we combined all cases with hormone receptor
(Unicamp, São Paulo, Brazil). Invasive ductal breast expression (luminal A, luminal B, and luminal-HER-2)
cancer tissue specimens were obtained from 60 female as luminal subtype; the HER-2-overexpression and the
patients who were undergoing primary mastectomy or triple-negative were maintained as such.
quadrantectomy from January 2008 to July 2009 to
perform at least 5-year survival analyses. Exclusion
criteria comprised previous chemotherapy or radiother- SHG imaging
apy, history of other cancer types, and medical follow- The peri and intratumoral areas selected for COL
up shorter than 5 years. All subjects granted informed, assessment on H&E-stained breast cancer tissue sec-
written consent before being enrolled. The study is fully tions were examined at the National Institute of
compliant with the Declaration of Helsinki (approved Sciences and Technology on Photonics Applied to Cell
by the ‘‘Comitê de Ética em Pesquisa da Unicamp,’’ Biology (INFABiC, Unicamp). SHG microscopy was
approval number 087/2008). performed on a Zeiss LSM 780-NLO confocal system
Formalin-fixed, paraffin-embedded, and hematoxy- (Carl Zeiss AG, Göttingen, Germany) using a 40 3 /1.3
lin and eosin (H&E)-stained tissue sections obtained oil immersion EC Plan-Neofluar objective. The high
from the surgical specimens were reviewed to examine numerical aperture (1.3) was necessary to provide the
histopathological parameters as described28 and to spatial resolution to observe the fibrils. An excitation
select areas to be analyzed by SHG microscopy by two wavelength of 780 nm, with an approximately 100-fs
pathologists (J.V. and G.R.P.). Both pathologists, who width pulse at an 80 MHz repetition rate, was provided
did not perform the SHG analyses, independently by a Tsunami Ti:Sapphire laser (Spectra-Physics,
made selection of three intra and peritumoral areas. Irvine, CA, USA). Laser power in the objective lens,
Intratumoral areas corresponded to fibrosis within neo- incident on the sample, was around 80 mW, with circu-
plasia, admixed with groups of neoplastic cells; peritu- lar polarization. Acquisition time of each image was
moral areas corresponded to fibrosis at the border of around 60 s.
neoplasia, adjacent to cancer cells, at the interface with Only the SHG forward signal at 390 nm with the
non-neoplastic, morphologically normal tissue. Areas 0.55 NA—WD 26 mm condenser lens was collected. A
with inflammatory infiltrate, vessels, and elastosis were short-pass SP690 (Omega Filters, Brattleboro, VT,
avoided (Figure 1(a) and (b)). Analyses of COL para- USA) was used to filter the 780 nm excitation29 and
meters in the areas selected by each pathologist were was followed by a filter cube with a long-pass LP490 at
made separately and compared in order to circumvent 45° to avoid two photons excited fluorescence (TPEF)
selection bias. and an SP430 at 90° to filter only the SHG signal. Two
In order to avoid endogenous autofluorescence of detectors were used to capture the TPEF signals
tissue, 3% hydrogen peroxide washes were performed (photomultiplier (PMT) detector) and SHG signals
(four incubations, 30 s each). (non-descanned (NDD) detector), and only the latter
was considered for this study (Figure 2). A spectral
analysis of the emission between 350–430 nm did show
Hormonal receptor and HER-2 status that only a narrow peak at 390 nm consistent with
IHC for estrogen receptor (ER) and progesterone SHG signal appeared. TPEF eosin staining fluores-
receptor (PR) was performed using clones 1D5 and cence signal was also detected, but in the 530 nm range.
PgR 636, respectively (Dako, Carpenteria, CA, USA). We used the same procedure described by Burke
For HER-2, the polyclonal antibody A0458 (Dako) et al.1 to take into account day-to-day variations in
was used. Antigen–antibody reactions were detected optical alignments, comprising the acquisition of a nor-
using the ADVANCEä HRPSystem (Dako).28 malization factor, provided by the SHG image of a
4 Tumor Biology

Figure 1. Continued

a narrower interval (between 290° and 230°), in contrast with

intratumoral areas, which have lower number of fibers, that
demonstrate a linear increase throughout the entire interval of
angles. A Gaussian gradient and Gaussian window (s) of 20
pixels were used to represent the distribution of collagen fibers
in the diagram.

Figure 2. Schemata of laser workstation setup. Laser titanium:

sapphire-emitting pulsing ray of 80 MHz is deflected by a mirror
placed at 45°; ray crosses tissue section at specific regions of
interest (ROIs) chosen by the pathologists; a small amount of
laser (L) plus the SHG (second-harmonic generation) and two
photons excited fluorescence (TPEF) are directed to a filter,
short pass 690 nm (SP690), then eliminating L signal.
TPEF+SHG signals cross a second filter, long pass 490 nm
(LP490), to separate TPEF to a photomultiply tube detector
(PMD); a third filter, short pass 430 nm (SP430), will purify SHG
signal, which is detected in a non-descanned detector (NDD).
HWP: half-wave plate; PBS: polarizing beam splitter; L1 and L2:
convergent lens; QWP: quarter-wave plate; G1 and G2: Gauss
lens; OBJ: objective lens.

standard sample (human aorta tissue) at the beginning

of each experimental section. The field of view of this
objective was of 212 3 212 mm, but to avoid the edges
of the picture, where the intensity of SHG signal is
weaker, we used a zoom to choose an area of
177 3 177 mm, which appeared homogeneous. To elim-
Figure 1. Images (a) and (b) represent histological aspects of inate background signals, a blind observer physicist
peri and intratumoral collagen, respectively (H&E stain). SHG without any medical training applied a common thresh-
images represent the same regions shown in (a) and (b), ((c) and
old to all images comparing all images with one single
(d), respectively), demonstrating that collagen quantity and
control sample from human aorta tissue. After this
organization is higher in the former. Graph (e) indicates the
frequency of distribution, by orientation, of collagen fibers threshold procedure, all background signals, not
(continuous line for peritumoral and dashed line for related to the tissue COL, were blanked. Of 720 images,
intratumoral areas). Here, it is clear that peritumoral fibers are 666 were suitable for analysis (counting images from
more homogeneously orientated (between 290° and 230°) areas selected by both pathologists, 333 images from
than intratumoral fibers (peaks present in a broader interval of each). The 54 images excluded correspond to 18 sam-
angles). Graph (f) indicates the sum of collagen fibers on each ples with no peritumoral areas available, because only
angle. It is clear that increase of collagen fiber numbers occur in tumor was present in the histological specimens.
Natal et al. 5

Evaluation of COL fibers

SHG signals were stratified with regard to COL organi-
zation and structure in stroma using image pattern
analysis methods. Quantitative analysis of COL para-
meters was performed in SHG images using ImageJ
(http://imagej.nih.gov/ij/) and OrientationJ plug-in
Rezakhaniha et al.30 Quantity was determined as the
total sum of SHG signals intensity of the 1024 3 1024
pixels, which was directly associated with the amount
of COL molecules. These values were obtained using
the integrated density of the ImageJ software. Using
the OrientationJ plug-in, according to the methods
described by Rezakhaniha et al.,30 we calculated the
uniformity (energy) and organization (coherence).31
Uniformity (energy) reflects the density of COL fibers,
that is, how the fibers are distant (low values) or near
(high values) to each other. Organization (coherence)
reflects the direction of the COL fibers in space; thus,
values vary from 0 (completely isotropic areas, zero
directionality) to 1 (highly oriented structures). From
the mathematical definition of these parameters one
should expect a high correlation between them.
The OrientationJ first calculates  the structure tensor
 
hf x , f x iw f ,f
2 3 2 matrix J =    x y w , where fj are the
fx , fy w fy , fy w
derivatives along the principal ÐÐ axes, and the inner prod-
uct is defined as hg, hiw = R2 wðx, yÞg ðx, yÞhðx, yÞdxdy,
where wðx, yÞ is a Gaussian weight function. The trace
of this tensor is defined   as the Energy
½E = trace(J ) = hfx , fx iw + fy , fy w , and Coherence is Figure 3. Representative images of collagen parameters
defined as the ratio between the difference and the sum evaluated in this study: (a) and (b): quantity; (c) and (d):
of the two high and low eigenvalues of the tensor uniformity; (e) and (f): organization. High values are found in
images (a), (c), and (e) and low values are found in (b), (d), and (f).
C = l+  l =l+ + l . There is a mathematical link
between the two parameters, the denominator of this
ratio is the Energy itself, and one should expect a large
pathologist; the same was made for the peritumoral
correlation between the two.
For this purpose, nine representative areas areas. The mean COL parameters were compared to
(150 3 150 pixels) were randomly chosen in each recognized clinical and pathological breast cancer risk
image, and COL quantity, organization, and unifor- factors, such as patients’ age (\50 years and 50 years
mity were performed. The ratio between peri and intra- groups), BMI (\30 Kg/m2 and 30 Kg/m2 groups),
tumoral COL quantity was additionally calculated tumor pathological stage, immunohistochemical sub-
(Figure 3). type, and histological grade. For these comparisons, t-
tests or analysis of variance (ANOVA) were used.
Parameters were included in an ANOVA multifactor
Statistical analyses test to identify the independence of each. Next, peri
COL quantity and uniformity were log-transformed to and intratumoral COL parameters were compared per-
base e for statistical purposes, and the statistical analy- forming the same test, in order to evaluate any associa-
ses were performed using R. Shapiro–Wilk test was tions between both topographies. Furthermore,
performed to analyze the data distribution of COL Pearson correlation or Spearman correlation tests were
parameters. The concordance of the findings between applied to evaluate the correlation between peri and
the two pathologists was investigated using intraclass intratumoral COLs’ parameters. Cox proportional
correlation coefficients (icc).32 For this analysis, the hazard models were fit to determine the associations of
mean values from the three intratumoral areas selected COL parameters with RFS and OS. RFS and OS were
by a pathologist were correlated to the three corre- measured from the date of primary breast cancer diag-
sponding intratumoral areas selected by the second nosis to relapse or death, caused by breast cancer. To
6 Tumor Biology

determine the cut-off values for the COL parameters Table 1. Clinicopathological features of the 60 patients in this
for the survival analyses, a multivariate response recur- study.
sive partitioning algorithm was applied.31 Finally,
Features
Kaplan–Meier curves for RFS and OS were generated
for the significant variables, as determined by the recur- Age
sive partitioning model. Survival curves were compared Median (CR; years) 60 (33–88)
by the log-rank test. Significance was set at p \ 0.05. BMI
Median (CR; kg/m2) 28 (20–37)
Menopausal status
Yes (%) 40 (66.7)
Results No (%) 20 (33.3)
Pathological stage
Clinicopathological features I (%) 6 (10.0)
The median age of the 60 patients was 60 years (95% II (%) 28 (47.7)
III (%) 26 (43.3)
central range (CR) = 33–88 years); 40 (66.7%) patients Lymph node status
were postmenopausal, and the median BMI was Positive (%) 45 (75.0)
28 Kg/m2 (95% CR = 20–37 Kg/m2). The tumor was Negative (%) 15 (25.0)
pathological stage I in 6 (10.0%), II in 28 (47.7%), and Histological grade
III in 26 (43.3%) patients. Clinically positive axillary I (%) 3 (5.0)
II (%) 54 (55.0)
lymph nodes were found in 45 (75.0%) patients; the III (%) 3 (5.0)
remaining 15 (25.0%) patients were negative. Vascular invasion
Histological grading according to the Nottingham sys- Positive (%) 27 (45.0)
tem was I in 3 (5.0%), II in 54 (90.0%), and III in 3 Negative (%) 33 (55.0)
Lymph vessel invasion
(5.0%) patients. Lymph vessel invasion was present in
Positive (%) 18 (30.0)
18 (30.0%) patients and absent in 42 (70.0%), while Negative (%) 42 (60.0)
vascular invasion was present in 27 (45.0%) patients Perineural invasion
and absent in 33 (55.0%). Perineural invasion was pres- Positive (%) 14 (23.3)
ent in 14 (23.3%) and absent in 46 (76.7%) patients. By Negative (%) 33 (76.7)
Immunohistochemical-based subtype
IHC-based subtype (surrogate for molecular typing), Luminal (%) 43 (72.0)
43 (72.0%) patients had luminal subtype tumors (27 HER-2-overexpressing (%) 7 (11.0)
luminal A, 1 luminal B, and 15 luminal-HER-2), 7 Triple-negative (%) 10 (17.0)
(11.0%) had HER-2-overexpressing tumors, and 10 Relapse
(17.0%) had triple-negative tumors. Nine (15.0%) Yes (%) 9 (15.0)
No (%) 51 (85.0)
patients presented tumor relapse, and 7 (11.7%) died Death from disease
from disease. In one case of luminal A breast cancer, Yes (%) 7 (11.7)
follow-up data were not available (Table 1). No (%) 53 (88.3)

CR: 95% central range; BMI: body mass index.

SHG evaluation in areas selected by the two

pathologists: concordance analysis (icc) Comparison between peri and intratumoral COL
Values obtained for the areas selected by the two parameters
pathologists had satisfactory inter-rater agreement (icc
Peritumoral areas presented higher COL quantity
. 0.70). Concordance was maximal for the average of
(p \ 0.001), uniformity (p \ 0.001), and organization
the three images taken for each region, selected by each
(p = 0.006). In addition, there was strong correlation
pathologist. For peri and intratumoral COL quantity,
between peri and intratumoral COL quantity and uni-
icc was 0.912 (p \ 0.001; CI: 0.850–0.949) and 0.936
formity (p \ 0.001, r = 0.614 for both parameters), but
(p \ 0.001; CI: 0.895–0.961), respectively. For peri and
there was no correlation between peri and intratumoral
intratumoral COL uniformity, icc was 0.985 (p \ 0.001;
COL organization (p = 0.479, r = 0.101; Table 2). Peri
95% confidence interval (CI): 0.974–0.992) and 0.948
and intratumoral COL SHG image and distribution
(p \ 0.001; 95% CI: 0.914–0.969), respectively. For peri
are shown in Figure 1(c)–(f), with their related H&E-
and intratumoral COL organization icc was 0.834
stained areas. In intratumoral areas, there were a lower
(p \ 0.001; CI: 0.727–0.902) and 0.886 (p \ 0.001; CI:
frequency of COL in the same orientation and greater
0.816–0.930), respectively.
variations in orientation of fibers.
Natal et al. 7

Table 2. Comparison of peri and intratumoral collagen fibers parameters.

Parameters Peritumoral Intratumoral pcom Correlation (r) pcor

Quantity 13.07 (0.60) 12.06 (0.79) \0.001 0.614 \0.001

Uniformity 11.78 (0.79) 10.68 (1.16) \0.001 0.614 \0.001
Organization 0.23 (0.08) 0.19 (0.08) 0.006 0.101 0.479

pcom: the level of significance comparing the parameters between peri and intratumoral collagen fibers: all parameters present significantly higher
values in peritumoral collagen. pcor: the level of significance when peri and intratumoral parameters were correlated: while quantity and uniformity
of collagen fibers increase or decrease with similar behavior in both locations, fiber organization did not show the same correlation. The data were
expressed as mean (standard deviation).

Table 3. Cox proportional hazards analysis of intratumoral COL parameters.

Luminal subtype breast cancer

Relapse-free survival Overall survival
Factor Intratumoral Intratumoral
HR p 95% CI padj HR p 95% CI padj
Lower Upper Lower Upper

Quantity 4.127 0.031 1.135 15.010 4.127 7.838 0.017 1.446 42.480 0.031
Uniformity 3.198 0.009 1.325 7.716 3.198 1.787 0.270 0.644 4.961 0.256
Organization 4.488 0.765 EL 0.529 0.009 0.440 EL 0.497

HR: hazard ratio; 95% CI: 95% confidence interval.

p demonstrates univariate analysis, while padj demonstrates a multivariate analysis (axillary lymph node metastasis, BMI, and pathological stage). For
the parameter organization, CI limits were too extensive (EL), thus were not indicated in this Table.

Association with clinicopathological features When the analyses were restricted to the luminal
Higher intratumoral COL uniformity was associated breast cancer, higher intratumoral COL quantity was
with other clinical parameters related to unfavorable out- associated with an approximately eightfold increased risk
come: BMI above 30 Kg/m2 (p = 0.019), with high of death (p = 0.017; CI: 1.45–42.48) and a fourfold risk
pathological stage III (p = 0.004), and with axillary of recurrence (p = 0.031; CI: 1.14–15.01). When axillary
lymph node metastasis (p = 0.034). No other association lymph node metastasis, BMI, and pathological stage were
was found for intratumoral COL parameters. Higher included for multivariate analysis, there was still a signifi-
ratio between intra and peritumoral COL quantity was cantly higher risk of death for higher intratumoral COL
associated with BMI above 30 Kg/m2 (p = 0.036) and quantity (p = 0.031) but not with relapse (p = 0.109).
death related to disease (p = 0.048). No associations Higher intratumoral COL uniformity was associated with
were found when peritumoral COL parameters (quan- an approximately threefold increased risk of relapse
tity, uniformity, and organization) were analyzed accord- (p = 0.009; CI: 1.33–7.72); when the same other clinico-
ing to patients’ age, histological grade, pathological pathological parameters were included for multivariate
stage, IHC-based subtype, relapse, and death. analysis, higher intratumoral COL uniformity was still
significantly associated with relapse (p = 0.015; Table 3).
The multivariate response recursive partitioning model
Prognostic implication and pathological value determined the cut-off points of 12.857 (p = 0.013) and
11.894 (p = 0.036) for intratumoral COL quantity and
RFS and OS calculated with the Cox proportional intratumoral COL uniformity, respectively, for death
hazards model for the whole group of patients showed and relapse endpoints (Figure 4). Applying these cut-offs,
that no peritumoral COL parameters presented prog- 34 cases presented lower COL quantity (21 luminal A, 1
nostic value. In contrast, higher intratumoral COL uni- luminal B, and 12 luminal HER-2) and 8 cases presented
formity was associated with an approximately twofold higher (5 luminal A and 3 luminal-HER-2). For intratu-
increased risk of relapse (p = 0.043; CI: 1.02–3.92). moral COL uniformity, 35 had lower (23 luminal A, 1
When other clinicopathological parameters were luminal B and 11 luminal HER-2) and 7 had higher val-
included (axillary lymph node metastasis, BMI, patho- ues (4 luminal A and 3 luminal HER-2). Kaplan–Meier
logical stage, and molecular subtype) for a multivariate curves for death and RFS with regard to intratumoral
analysis, this significance was not observed (p = 0.219).
8 Tumor Biology

which may reflect different interactions between COL

and the microenvironment inside and outside breast
cancer. Therefore, although there was a moderate cor-
relation between peri and intratumoral areas, only
intratumoral parameters achieved prognostic value.
Thus, the importance of studying tumor ECM in eval-
uating patients with breast cancer should be stressed.
These considerations appeared especially valuable in
the luminal subtype, for which quantitative analysis by
SHG rendered the most considerable results, suggest-
ing that such parameters could be part of a biologically
based classification system.
Previous studies on COL evaluation in breast cancer
stroma have demonstrated that the relation of spatial
and quantitative parameters could be related to out-
come.17,18,20,33,34 However, this study was innovative,
in that we have compared COL parameters with IHC-
based subtypes of breast cancer. For example, in the
study by Conklin et al.,18 COL parameters were consid-
ered for each one of the tumor features, histological
grade, expression of hormone receptor, and HER-2 sta-
tus. The IHC-based molecular subtyping used herein
considers groups of biologically related types of breast
cancer, integrating the features which were individually
analyzed before, as explored by Brabrand et.al.35 This
subtyping is clinically relevant for both prognostic eva-
luation and therapeutic decision in breast cancer patients.
Therefore, the heterogeneity in the luminal subtype,
which had already been evidenced,25–27 was further con-
Figure 4. Graphic representation of a multivariate response firmed by our data, using the evaluation of COL fibers.
recursive partitioning model to determine the best cut-offs for
The results of earlier reports on the utility of exam-
intratumoral collagen (COL) fibers quantity (top) and uniformity
(bottom). In 34 patients with intratumoral COL quantity up to
ining tumor stromal reactions using a standard visual
or equal 12.857, none died, and 3 had relapse, while in 8 microscopic analysis of SHG images (TACS) are sup-
patients with intratumoral COL quantity above 12.857, 4 died, ported by our study and continue to be recognized.36–38
and 3 had relapse. In 35 patients with intratumoral COL This approach is also different from ours, as TACS rep-
uniformity up to or equal 11.894, 1 died, and 2 had relapse, resents a semiquantitative model.18 Our model gener-
while in 7 patients with intratumoral COL uniformity above ated independent numerical values of COL parameters,
11.894, 3 died, and 4 had relapse. like quantity and uniformity, which were sufficient to
determine RFS and OS. The feasibility of the technique
was reinforced herein by the inter-rater agreement in
COL quantity and uniformity were produced using the SHG imaging, as subjectivity in the process of demar-
recursive partitioning model cut-offs (Figure 5). Patients cating tumor areas to be analyzed was irrelevant, so
whose tumors had intratumoral COL quantity above high reproducibility can be expected.
12.857 experienced less favorable outcomes: death Moreover, the loss of information due to the lack of
(p \ 0.001) and RFS (p = 0.023). Similarly, for intratu- detection of backward SHG signals was shown mini-
moral COL uniformity, patients with values above mal by Abraham et al.,39 and there should be no loss of
11.894 had significantly higher probability of death information with clinicopathological impact. These
(p = 0.013) and relapse (p = 0.008). authors have evaluated forward/backward signals in
COL fibers associated to breast cancer cell cultures and
obtained similar implications for both evaluations,
Discussion
which inclined them to conclude that either of them
Higher intratumoral COL quantity and uniformity sig- presented comparable value in relation to tumor
nificantly correlated with unfavorable RFS and OS aggressiveness.39
among our cases of IHC-based luminal breast cancer. A subsequent study using IHC to detect ECM com-
In addition, our results suggest that all COL para- ponents (e.g. tenascin, type IV COL, and fibronectin)
meters differ between peri and intratumoral areas, has confirmed its prognostic value in cancer patients,
Natal et al. 9

Figure 5. (a) and (b) demonstrate an unfavorable overall and relapse-free survival, respectively, in patients with intratumoral
collagen (COL) fibers quantity above 12.857. (c) and (d) demonstrate an unfavorable overall and relapse-free survival, respectively, in
patients with intratumoral COL fibers uniformity above 11.894. Cut-offs for COL quantity and uniformity were determined using a
recursive partitioning model (for details, see text).

according to the predominant element.40 Furthermore, responsibility for this outcome. The prognostic value of
it has been demonstrated that breast cancer stroma dis- COL organization was not proven in this study; only
organization could predict poor response to neoadju- quantity and uniformity presented clinical relevance.
vant chemotherapy, using H&E and Heidenhain’s It is important to emphasize that not only molecular
AZAN trichrome staining,41 but this technique could mechanisms within cancer cells respond for cancer
not identify the isolated role of COL as the major behavior.42 Among the other hallmarks of oncogenesis,
10 Tumor Biology

tumor microenvironment, its cellular and molecular images is the area occupied by cells (neoplastic, endothe-
components, as well as its three-dimensional structure lial, stromal cells). However, a previous report addressing
represent important features explaining tumor progres- this issue demonstrated that the area of the cellular com-
sion, with potential predictive and prognostic impact.43 ponent does not represent a confounding variable.53
A recent study demonstrated that breast cancer micro- The relatively small numbers of patients in each
environment, in special increased proportion of tumor group represent a limitation of this study, which war-
stroma, was associated with tumor budding, a feature rants further confirmation by independent researchers.
previously reported as indicative of unfavorable OS.44 The technique used in this study as well as the under-
In this context, high COL uniformity could be trans- standing of the physical principles required is not
lated as higher approximation of fibers,30 which was widely applied by the pathologists for now. However,
correlated with poorer RFS and OS. The presence of these novel relevant results in human specimens could
dense clusters of COL was already shown to indicate evolve to the development of more accessible instru-
increased matrix stiffness, which correlated with poor ments, as it has happened with the molecular pathol-
survival.45,46 Increased tissue rigidity or matrix stiffness ogy. These results present potential clinical value in
seems to play a significant role during tumor progres- prognostic stratification, including microenvironmental
sion, mainly by the activation of the Hippo pathway.47 features together the classical clinicopathological para-
In this case, yes-assciated protein (YAP) could regulate meters. In analogy to the inclusion of an ‘‘immuno-
the expression of the same genes in cancer-associated score’’54 to the classification of cancer, data on the
fibroblasts, like ANLN, DIAPH3 and MYL9. This tumor matrix could complete the information needed
could reinforce the circle that helps to maintain the for a given neoplasia to explain and treat tumor pro-
fibroblast phenotype, inducing matrix stiffening.47 gression and metastatization.
Moreover, high matrix stiffness promotes, in cancer In summary, our findings demonstrate that the eva-
cells, nuclear translocation of TWIST1 by releasing this luation of SHG images is a straightforward and reason-
molecule from its cytoplasmic binding G3BP2; loss of able method for the assessment of COL parameters in
G3BP2 leads to a constitutive TWIST1 nuclear locali- the ECM associated to breast cancer, which might ulti-
zation and synergizes with increasing matrix stiffness to mately be clinically useful to evaluate prognosis. This is
induce epithelial-mesenchymal transition and promote especially true for the luminal subtype breast cancer,
tumor invasion and metastasis.48 which could benefit from the inclusion of COL para-
On the other hand, studies using diverse biological meters of prognostic relevance.
specimens have shown that SHG detect primarily type
I and II COL (both form aligned fibers) and myosin
within acto-myosin complexes, but not type III and IV Acknowledgements
COL, as these do not produce adequate SHG signals Rodrigo A Natal and José Vassallo contributed equally to this
for imaging.12,49 In this context, SHG signals detected article.
in breast cancer tissue originate mostly in type I COL,
since type II COL is only deposited in metaplastic Declaration of conflicting interests
breast cancer, specifically matrix-producing breast can-
The author(s) declared no potential conflicts of interest with
cer.50 As for myosin and microtubules, only in muscles
respect to the research, authorship, and/or publication of this
and neurons they present enough organization to gen- article.
erate an SHG signal comparable to COL fibers signal.
Also, these elements can be observed mostly in the
absence of COL.51,52 Funding
SHG image analysis of COL represents a clinically The author(s) disclosed receipt of the following financial sup-
relevant and promising approach to breast pathology, port for the research, authorship, and/or publication of this
as it renders possible to distinguish diverse pathological article: The authors thank the National Institute of Science
entities.53 Our findings help to characterize the changes and Technology on Photonics Applied to Cell Biology
in the ECM associated with tumor RFS and OS. The (INFABiC) at the State University of Campinas for access to
equipment and assistance; INFABiC is co-funded by
description of COL in tumor ECM might facilitate the
Fundac xão de Amparo à Pesquisa do Estado de São Paulo
identification of tumor invasion because alterations in (FAPESP) (08/57906-3) and Conselho Nacional de
ECM parallel breast tumor relapse.1 Thus, SHG signals Desenvolvimento Cientı́fico e Tecnológico (CNPq) (573913/
could be used to (1) differentiate between healthy and 2008-0).
tumor tissues, since peri and intratumoral areas pre-
sented distinct COL parameters in this study and (2)
ORCID iD
determine tumor prognosis (RFS and OS) simultane-
ously. A problem mentioned in the evaluation of SHG Rodrigo A Natal https://orcid.org/0000-0001-7843-3314
Natal et al. 11

References 17. Provenzano PP, Inman DR, Eliceiri KW, et al. Collagen
1. Burke K, Tang P and Brown E. Second harmonic gener- density promotes mammary tumor initiation and progres-
ation reveals matrix alterations during breast tumor pro- sion. BMC Med 2008; 6: 11.
gression. J Biomed Opt 2013; 18: 31106. 18. Conklin MW, Eickhoff JC, Riching KM, et al. Aligned
2. Han X, Burke RM, Zettel ML, et al. Second harmonic collagen is a prognostic signature for survival in human
properties of tumor collagen: determining the structural breast carcinoma. Am J Pathol 2011; 178: 1221–1232.
relationship between reactive stroma and healthy stroma. 19. Burke KA, Dawes RP, Cheema MK, et al. Second-har-
Opt Express 2008; 16: 1846–1859. monic generation scattering directionality predicts tumor
3. Sund M and Kalluri R. Tumor stroma derived biomar- cell motility in collagen gels. J Biomed Opt 2015; 20:
kers in cancer. Cancer Metastasis Rev 2009; 28: 177–183. 051024.
4. Weigel MT and Dowsett M. Current and emerging bio- 20. Kakkad SM, Solaiyappan M, Argani P, et al. Collagen I
markers in breast cancer: prognosis and prediction. fiber density increases in lymph node positive breast can-
Endocr Relat Cancer 2010; 17: R245–R262. cers: pilot study. J Biomed Opt 2012; 17: 116017.
5. Bergamaschi A, Tagliabue E, Sørlie T, et al. Extracellular 21. Walsh AJ, Cook RS, Lee JH, et al. Collagen density and
matrix signature identifies breast cancer subgroups with alignment in responsive and resistant trastuzumab-treated
different clinical outcome. J Pathol 2008; 214: 357–367. breast cancer xenografts. J Biomed Opt 2015; 20: 26004.
6. Keely PJ, Wu JE and Santoro SA. The spatial and tem- 22. Goldhirsch A, Wood WC, Coates AS, et al. Strategies
poral expression of the alpha 2 beta 1 integrin and its for subtypes—dealing with the diversity of breast cancer:
ligands, collagen I, collagen IV, and laminin, suggest highlights of the St. Gallen international expert consen-
important roles in mouse mammary morphogenesis. Dif- sus on the primary therapy of early breast cancer 2011.
ferentiation 1995; 59: 1–13. Ann Oncol 2011; 22: 1736–1747.
7. Lagacé R, Grimaud JA, Schürch W, et al. Myofibroblas- 23. Voduc KD, Cheang MCU, Tyldesley S, et al. Breast can-
tic stromal reaction in carcinoma of the breast: variations cer subtypes and the risk of local and regional relapse. J
of collagenous matrix and structural glycoproteins. Clin Oncol 2010; 28: 1684–1691.
Virchows Arch A Pathol Anat Histopathol 1985; 408: 24. Yanagawa M, Ikemot K, Kawauchi S, et al. Luminal A
49–59. and luminal B (HER2 negative) subtypes of breast cancer
8. Campagnola P. Second harmonic generation imaging consist of a mixture of tumors with different genotype.
microscopy: applications to diseases diagnostics. Anal BMC Res Notes 2012; 5: 376.
Chem 2011; 83: 3224–3231. 25. Cancer Genome Atlas Network. Comprehensive molecu-
9. Campagnola PJ and Loew LM. Second-harmonic ima- lar portraits of human breast tumours. Nature 2012; 490:
ging microscopy for visualizing biomolecular arrays in 61–70.
cells, tissues and organisms. Nat Biotechnol 2003; 21: 26. Ellis MJ and Perou CM. The genomic landscape of breast
1356–1360. cancer as a therapeutic roadmap. Cancer Discov 2013; 3:
10. Zoumi A, Yeh A and Tromberg BJ. Imaging cells and 27–34.
extracellular matrix in vivo by using second-harmonic 27. Ignatiadis M and Sotiriou C. Luminal breast cancer:
generation and two-photon excited fluorescence. Proc from biology to treatment. Nat Rev Clin Oncol 2013; 10:
Natl Acad Sci U S A 2002; 99: 11014–11019. 494–506.
11. Georgiou E, Theodossiou T, Hovhannisyan V, et al. Sec- 28. Serra KP, Sarian LO, Rodrigues-Peres RM, et al. Expres-
ond and third optical harmonic generation in type I col- sion of cyclooxygenase-2 (COX-2) and p53 in neighbor-
lagen, by nanosecond laser irradiation, over a broad ing invasive and in situ components of breast tumors.
spectral region. Opt Commun 2000; 176: 253–260. Acta Histochem 2012; 114: 226–231.
12. Cox G, Kable E, Jones A, et al. 3-dimensional imaging of 29. Chen W-L, Chou C-K, Lin M-G, et al. Single-wavelength
collagen using second harmonic generation. J Struct Biol reflected confocal and multiphoton microscopy for tissue
2003; 141: 53–62. imaging. J Biomed Opt 2009; 14: 54026.
13. Adur J, Pelegati VB, de Thomaz AA, et al. Optical bio- 30. Rezakhaniha R, Agianniotis A, Schrauwen JTC, et al.
markers of serous and mucinous human ovarian tumor Experimental investigation of collagen waviness and
assessed with nonlinear optics microscopies. PLoS ONE orientation in the arterial adventitia using confocal laser
2012; 7: e47007. scanning microscopy. Biomech Model Mechanobiol 2012;
14. Adur J, Carvalho HF, Cesar CL, et al. Nonlinear optical 11: 461–473.
microscopy signal processing strategies in cancer. Cancer 31. Anchang B, Do MT, Zhao X, et al. CCAST: a model-
Inform 2014; 13: 67–76. based gating strategy to isolate homogeneous subpopula-
15. Boudaoud A, Burian A, Borowska-Wykre˛t D, et al. tions in a heterogeneous population of single cells. PLoS
FibrilTool, an ImageJ plug-in to quantify fibrillar struc- Comput Biol 2014; 10: e1003664.
tures in raw microscopy images. Nat Protoc 2014; 9: 32. Hallgren KA. Computing inter-rater reliability for obser-
457–463. vational data: an overview and tutorial. Tutor Quant
16. Provenzano PP, Eliceiri KW, Campbell JM, et al. Col- Methods Psychol 2012; 8: 23–34.
lagen reorganization at the tumor-stromal interface facili- 33. Burke K, Smid M, Dawes RP, et al. Using second harmo-
tates local invasion. BMC Med 2006; 4: 38. nic generation to predict patient outcome in solid tumors.
BMC Cancer 2015; 15: 929.
12 Tumor Biology

34. Riching KM, Cox BL, Salick MR, et al. 3D collagen 44. Gujam FJA, McMillan DC, ,Mohammed ZMA, et al.
alignment limits protrusions to enhance breast cancer cell The relationship between tumour budding, the tumour
persistence. Biophys J 2014; 107: 2546–2558. microenvironment and survival in patients with invasive
35. Brabrand A, Kariuki II, Engstrøm MJ, et al. Alterations ductal breast cancer. Br J Cancer 2015; 113: 1066–1074.
in collagen fibre patterns in breast cancer. A premise for 45. Paszek MJ, Zahir N, Johnson KR, et al. Tensional home-
tumour invasiveness? APMIS 2015; 123: 1–8. ostasis and the malignant phenotype. Cancer Cell 2005; 8:
36. Ni Y-B, Tsang JYS, Chan SK, et al. A novel 241–254.
morphologic-molecular recurrence predictive model 46. Levental KR, Yu H, Kass L, et al. Matrix crosslinking
refines traditional prognostic tools for invasive breast forces tumor progression by enhancing integrin signaling.
carcinoma. Ann Surg Oncol 2014; 21: 2928–2933. Cell 2009; 139: 891–906.
37. Mujtaba SS, Ni Y-B, Tsang JYS, et al. Fibrotic focus in 47. Calvo F, Ege N, Grande-Garcia A, et al. Mechanotrans-
breast carcinomas: relationship with prognostic para- duction and YAP-dependent matrix remodelling is
meters and biomarkers. Ann Surg Oncol 2013; 20: required for the generation and maintenance of cancer-
2842–2849. associated fibroblasts. Nat Cell Biol 2013; 15: 637–646.
38. Hasebe T, Mukai K, Tsuda H, et al. New prognostic his- 48. Wei SC, Fattet L, Tsai JH, et al. Matrix stiffness drives
tological parameter of invasive ductal carcinoma of the epithelial-mesenchymal transition and tumour metastasis
breast: clinicopathological significance of fibrotic focus. through a TWIST1-G3BP2 mechanotransduction path-
Pathol Int 2000; 50: 263–272. way. Nat Cell Biol 2015; 17: 678–688.
39. Abraham T, Kayra D, McManus B, et al. Quantitative 49. Chen X, Nadiarynkh O, Plotnikov S, et al. Second har-
assessment of forward and backward second harmonic monic generation microscopy for quantitative analysis of
three dimensional images of collagen type I matrix remo- collagen fibrillar structure. Nat Protoc 2012; 7: 654–669.
deling in a stimulated cellular environment. J Struct Biol 50. Kusafuka K, Muramatsu K, Kasami M, et al. Cartilagi-
2012; 180: 17–25. nous features in matrix-producing carcinoma of the
40. Ioachim E, Charchanti A, Briasoulis E, et al. Immuno- breast: four cases report with histochemical and immuno-
histochemical expression of extracellular matrix compo- histochemical analysis of matrix molecules. Mod Pathol
nents tenascin, fibronectin, collagen type IV and laminin 2008; 21: 1282–1292.
in breast cancer: their prognostic value and role in 51. Psilodimitrakopoulos S, Artigas D, Soria G, et al. Quan-
tumour invasion and progression. Eur J Cancer 2002; 38: titative discrimination between endogenous SHG sources
2362–2370. in mammalian tissue, based on their polarization
41. Dekker TJA, Charehbili A, Smit VT, et al. Disorganised response. Opt Express 2009; 17: 10168–10176.
stroma determined on pre-treatment breast cancer biop- 52. Campagnola PJ, Millard AC, Terasaki M, et al. Three-
sies is associated with poor response to neoadjuvant che- dimensional high-resolution second-harmonic generation
motherapy: results from the NEOZOTAC trial. Mol imaging of endogenous structural proteins in biological
Oncol 2015; 9: 1120–1128. tissues. Biophys J 2002; 82: 493–508.
42. Hanahan D and Weinberg RA. Hallmarks of cancer: the 53. Ambekar R, Lau T-Y, Walsh M, et al. Quantifying col-
next generation. Cell 2011; 144: 646–674. lagen structure in breast biopsies using second-harmonic
43. Herrmann D, Conway JRW, Vennin C, et al. Three- generation imaging. Biomed Opt Express 2012; 3:
dimensional cancer models mimic cell-matrix interactions 2021–2035.
in the tumour microenvironment. Carcinogenesis 2014; 54. Galon J, Mlecnik B, Bindea G, et al. Towards the intro-
35: 1671–1679. duction of the ‘‘Immunoscore’’ in the classification of
malignant tumours. J Pathol 2014; 232: 199–209.