Theranostics 2021, Vol.

11, Issue 7 3229

Ivyspring
International Publisher
Theranostics
2021; 11(7): 3229-3243. doi: 10.7150/thno.55921
Research Paper

Large-scale tumor-associated collagen signatures

identify high-risk breast cancer patients
Gangqin Xi1†, Wenhui Guo2†, Deyong Kang3†, Jianli Ma4†, Fangmeng Fu2, Lida Qiu1,5, Liqin Zheng1, Jiajia
He1, Na Fang1,6, Jianhua Chen1,7, Jingtong Li8, Shuangmu Zhuo1, Xiaoxia Liao9, Haohua Tu10, Lianhuang
Li1, Qingyuan Zhang8, Chuan Wang2, Stephen A. Boppart10, and Jianxin Chen1
1. Key Laboratory of OptoElectronic Science and Technology for Medicine of Ministry of Education, Fujian Provincial Key Laboratory of Photonics
Technology, Fujian Normal University, Fuzhou, China
2. Breast Surgery Ward, Department of General Surgery, Fujian Medical University Union Hospital, Fuzhou, China
3. Department of Pathology, Fujian Medical University Union Hospital, Fuzhou, China
4. Department of Radiation Oncology, Harbin Medical University Cancer Hospital, Harbin, China
5. College of Physics and Electronic Information Engineering, Minjiang University, Fuzhou, China
6. Department of Ophthalmology and Optometry, Fujian Medical University, Fuzhou, China
7. College of Life Science, Fujian Normal University, Fuzhou, China
8. Department of Medical Oncology, Harbin Medical University Cancer Hospital, Harbin, China
9. National Center for Supercomputing Applications, University of Illinois at Urbana-Champaign, Urbana, USA
10. Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, USA
†These authors contributed equally to this work.

 Corresponding authors: Lianhuang Li, College of Photonic and Electronic Engineering, Fujian Normal University, Fuzhou 350007, China, Telephone number:
86-591-83465818, E-mail: lhli@fjnu.edu.cn; Qingyuan Zhang, Department of Medical Oncology, Harbin Medical University Cancer Hospital, Harbin 150081,
China, Telephone number: 86-13313612989, E-mail: zhma19650210@163.com; Chuan Wang, Breast Surgery Ward, Department of General Surgery, Fujian
Medical University Union Hospital, Fuzhou 350001, China, Telephone number: 86-13515020716, E-mail: dr_chuanwang@fjmu.edu.cn; Stephen A. Boppart,
Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA, Telephone number:
001-217-244-7479, E-mail: boppart@illinois.edu; Jianxin Chen, College of Photonic and Electronic Engineering, Fujian Normal University, Fuzhou 350007, China,
Telephone number: 86-591-22686078, E-mail: chenjianxin@fjnu.edu.cn.

© The author(s). This is an open access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/).
See http://ivyspring.com/terms for full terms and conditions.

Received: 2020.11.14; Accepted: 2020.12.20; Published: 2021.01.01

Abstract
The notion of personalized medicine demands proper prognostic biomarkers to guide the optimal
therapy for an invasive breast cancer patient. However, various risk prediction models based on
conventional clinicopathological factors and emergent molecular assays have been frequently limited by
either a low strength of prognosis or restricted applicability to specific types of patients. Therefore, there
is a critical need to develop a strong and general prognosticator.
Methods: We observed five large-scale tumor-associated collagen signatures (TACS4-8) obtained by
multiphoton microscopy at the invasion front of the breast primary tumor, which contrasted with the
three tumor-associated collagen signatures (TACS1-3) discovered by Keely and coworkers at a smaller
scale. Highly concordant TACS1-8 classifications were obtained by three independent observers. Using
the ridge regression analysis, we obtained a TACS-score for each patient based on the combined
TACS1-8 and established a risk prediction model based on the TACS-score. In a blind fashion, consistent
retrospective prognosis was obtained from 995 breast cancer patients in both a training cohort (n= 431)
and an internal validation cohort (n = 300) collected from one clinical center, and in an external validation
cohort (n = 264) collected from a different clinical center.
Results: TACS1-8 model alone competed favorably with all reported models in predicting disease-free
survival (AUC: 0.838, [0.800-0.872]; 0.827, [0.779-0.868]; 0.807, [0.754-0.853] in the three cohorts) and
stratifying low- and high-risk patients (HR 7.032, [4.869-10.158]; 6.846, [4.370-10.726], 4.423,
[2.917-6.708]). The combination of these factors with the TACS-score into a nomogram model further
improved the prognosis (AUC: 0.865, [0.829-0.896]; 0.861, [0.816-0.898]; 0.854, [0.805-0.894]; HR
7.882, [5.487-11.323]; 9.176, [5.683-14.816], and 5.548, [3.705-8.307]). The nomogram identified 72 of
357 (~20%) patients with unsuccessful 5-year disease-free survival that might have been undertreated
postoperatively.

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3230

Conclusions: The risk prediction model based on TACS1-8 considerably outperforms the contextual
clinical model and may thus convince pathologists to pursue a TACS-based breast cancer prognosis. Our
methodology identifies a significant portion of patients susceptible to undertreatment (high-risk patients),
in contrast to the multigene assays that often strive to mitigate overtreatment. The compatibility of our
methodology with standard histology using traditional (non-tissue-microarray) formalin-fixed
paraffin-embedded (FFPE) tissue sections could simplify subsequent clinical translation.
Key words: Breast cancer, multiphoton imaging, tumor-associated collagen signatures, disease-free survival

Introduction
Breast cancer prognosis after diagnosis plays a by signs and symptoms on physical exam (rather than
central role for patients and oncologists to choose screening) at relatively late stages. In this situation,
optimally personalized therapies, and thus have wherein under-treatment is common and medical
motivated the development of a wide variety of budgets are often limited, there is a critical need to
survival prognostic biomarkers (prognosticators) [1] develop a strong and general imaging prognosticator
and risk prediction models [2]. Beyond tumor size and compatible with standard histology that outperforms
nodal status, the most popular prognosticators are the conventional clinicopathological factors.
tubule formation, nuclear pleomorphism, and mitotic Since this type of prognosticator has rarely been
count that dictate the histological tumor grade [3]. The clinically validated beyond the infiltrating immune
pursuit of similar structural or imaging cells in colorectal cancer [12], we aim to test the
prognosticators has focused on automatic image extracellular structures of interstitial (non-basement-
analysis and the tumor microenvironment [4]. membrane) collagen (bulk biomolecule) that have
However, the overall trend is to shift from prognosis prominently defied the above trend of prognosticator
based on structural tissue features to more functional development and highlighted the frequently
prognostic features, which, for example, has delivered underappreciated role of the tumor microenviron-
stronger prognosis for specific types of patients via ment [16-23]. Although the corresponding imaging
multigene assays [2, 5]. In some cases, structural prognosticators have exhibited independent
prognosticators have been developed to prognostic values over individual clinicopathological
“complement” the functional prognosticators of gene factors, they have not been put into a clinical context
expression [6, 7], rather than the other way around. of a pathological alternative (i.e. multivariate risk
One inadvertent consequence of this trend is an prediction model of the clinicopathological factors
increased emphasis on rare biomolecules such as that makes a computer decision support system
certain mRNAs and proteins [8], over bulk available for pathologists) or have not unambiguously
biomolecules such as lipid, despite the important role demonstrated their differential values over this
of lipid in tumor development [9]. Another context, possibly due to their insufficient prognostic
inadvertent consequence is to focus on cells and strength and general applicability (Table S1).
intracellular components over extracellular matrix In this study, we expand the three
constituents, even though the latter is known to form tumor-associated collagen signatures (TACS1-3)
a dynamic niche in cancer progression [10]. This discovered by Keely and coworkers using
imbalance also holds true in the context of the tumor multiphoton microscopy (MPM) [24], by defining and
microenvironment, as emergent prognosticators recognizing five new tumor-associated collagen
beyond tumor cells themselves have relied more on signatures (TACS4-8) at a larger (~7×) scale and at the
the stromal cells [6, 11] or infiltrating immune cells invasion front of the primary tumor. We surprisingly
[12] than the extracellular matrix constituents. find that TACS1-3 (well-established biomarkers) at a
As one prominent example of this trend, various scale of 0.4 mm (typical scale for MPM) have
multigene assays have been compared with computer significantly lower prognostic strength in comparison
decision support systems based on conventional to TACS4-8. For the disease-free survival (DFS)
clinicopathological factors [13, 14], and have gained prognosis of 995 Chinese patients, the risk prediction
prognostic strength for various early-stage patients model based on TACS1-3 competes poorly with a
with young ages (or premenopausal status), few contextual clinical model developed in our systematic
positive lymph nodes, or small tumor sizes [5]. study, while its counterpart based on TACS4-8 (or
However, the restriction to these patients decreases TACS1-8) considerably outperforms this contextual
the general applicability of these multigene assays, clinical model and may thus convince pathologists to
limiting their usefulness in less developed countries pursue a TACS-based prognosis. Our methodology of
[15] where breast cancer is predominantly diagnosed prognosis identifies a significant portion of patients

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3231

susceptible to undertreatment, in contrast to the Nottingham histologic grade. The expression of

multigene assays that often strive to mitigate estrogen receptor (ER), progesterone receptor (PR),
overtreatment. The compatibility of our methodology HER2, and the Ki67 were detected by
with standard histology using traditional (non-tissue- immunohistochemistry (IHC). Tumors with HER2,
microarray) formalin-fixed paraffin-embedded (FFPE) IHC staining scores of 0 or 1+ were defined as HER2
tissue sections could simplify subsequent clinical negative, while IHC staining scores of 3+ were
translation (Table S1). considered HER2 positive. For IHC staining scores of
2+, molecular detection (in situ hybridization (ISH))
Methods was needed to further confirm that the unamplified
results of ISH were negative for HER2, and the
Patients
amplified result of ISH were positive for HER2.
This research used anonymous data for According to the detection results of ER, PR, HER2
retrospective study and was conducted under a and Ki67 expression by IHC or ISH, patients were
protocol approved by the Institutional Review Boards classified into four molecular subtypes as follows:
of Fujian Medical University Union Hospital and Luminal A: ER positive and/or PR positive, HER2
Harbin Medical University Cancer Hospital. We negative and Ki67 low expression; Luminal B (HER2
collected 1223 FFPE tissue blocks (Figure 1A) from negative): ER positive and/or PR positive; HER2
1223 patients (aged 21–87 years) who underwent negative and Ki67 high expression; Luminal B (HER2
surgical resection, in which 228 patients were positive): ER-positive and/or PR positive, and HER2-
excluded and 995 patients passed quality control for positive; HER2-enriched: ER negative, PR negative
subsequent analysis (Figure 1B). The inclusion criteria and HER2 positive; Triple-negative: negative for ER,
were: (i) patients had pathologically confirmed IBC PR and HER2. In this study, age, molecular subtypes,
without distant metastasis and underwent surgical tumor size, nodal status, clinical stage and histological
resection; and (ii) clinicopathological characteristics grade were categorical variables. Median DFS
and follow-up information were complete. For the follow-up was 70.0 months (IQR 37.0-81.0).
training and internal validation cohorts, 731 samples
were obtained between Nov. 2003 and Jun. 2017 from Multiphoton imaging system
patients treated at the Fujian Medical University The imaging system was built on a commercially
Union Hospital (Fuzhou, China). We used laser scanning microscope platform (LSM 880 Zeiss,
computer-generated random numbers to assign 431 of Germany) using a mode-locked femtosecond
these patients to the training cohort and 300 patients Ti:Sapphire laser that emitted linearly polarized 810
to the internal validation cohort. The external nm excitation light. The backscattered signals from
validation cohort comprised 264 patients collected tissue samples were simultaneously obtained via two
between Jul. 2009 and Dec. 2014 at Harbin Medical independent channels. One channel detected second
University Cancer Hospital (Harbin, China) (Figure harmonic generation (SHG) signal (green color)
1B). between 395 nm and 415 nm while the other channel
detected two-photon excitation fluorescence (TPEF)
Clinicopathologic characteristics and follow-up
signal (red color) between 428 nm and 695 nm. A
information
Plan-Apochromat ×20 objective (NA = 0.8, Zeiss,
Tumor size was defined as the maximal diameter Germany) was employed to acquire large field
of the tumor in a resected specimen (T1: the long images. Larger scale imaging was enabled by a
diameter of the tumor mass is less than 2 cm, T2: the motorized stage under computer control (ZEN 2.3 SP1
long diameter of the tumor mass is greater than 2 cm software). The lateral resolution was ~0.8 µm while
and less than or equal to 5 cm; T3: the long diameter of the imaging field of view was about 0.5×0.5 mm2,
the tumor mass is greater than 5 cm). Nodal status typically.
was classified into three categories according to the
number of positive lymph nodes (N0: 0 positive Sample preparation and TACS quantification
lymph node; N1: 1 to 3 positive lymph nodes; N2: For each patient, an archived ~2-cm sized FFPE
more than or equal to 4 positive lymph nodes). tissue block containing the invasive front [12, 16, 19]
Clinical stage (I, II, III) of the tumor was obtained by (i.e. the perceived tumor boundary that separates the
two surgeons with more than 10 years of clinical primary tumor and adjacent normal appearing tissue)
experience through reviewing clinical data. Two was selected for serial sectioning. Two consecutive
pathologists with more than 10 years of experience in 5-μm thick serial sections were cut from each paraffin
the diagnosis of breast tumor evaluated the block using a semiautomatic microtome in one
histological grade (G1 to G3) according to the pathology laboratory (Figure 1A). One section was

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3232

used for H&E histology by whole slide imaging, after predictors to construct the nomogram and generate a
which a pathologist confirmed the presence of tumor comprehensive indicator for estimating disease-free
cells and their borders. Throughout the entire tissue survival (DFS). The performance of nomogram was
section, several (7-20) ~2.8-mm-sized non-overlap- estimated using discrimination and calibration. The
ping regions of interest (ROI) across the invasive calibration of the nomogram was evaluated by a
margin and adjacent tumor area were labeled calibration plot, which was a graphic representation
(numbered) in the H&E images (Figure 1A) by two of the relationship between the actual incidence and
imaging scientists (Lianhuang Li, Gangqin Xi) who the predicted probabilities. In a well calibrated model,
were blind to the final pathological outcomes of the the predictions should be close to the 45-degree
patients. ROIs with ductal carcinoma in situ (DCIS) diagonal line. We also constructed receiver operating
were preferably selected inside the tumor, while ROIs characteristic (ROC) curves and calculated the areas
were extensively sampled at the invasive front. The under the curves (AUCs) to evaluate prognostic
other section was deparaffinized by alcohol and accuracy. The ROC curve was used to calculate the
xylene. Label-free dual-modal multiphoton optimal cutoff value that was determined by
microscopy (MPM) simultaneously collecting second maximizing the Youden index in the training cohort,
harmonic generation (SHG) and two-photon excited and then, the same cutoff value was applied to the
fluorescence (TPEF) images [25] was performed for all validation cohorts. The Kaplan-Meier survival curves
labeled ROIs on this deparaffinized but unstained were further used to estimate the correlation between
section and co-registered with the H&E image variables and disease-free survival, and the log-rank
(Figures 2-3). This provided an average MPM test was used to compare differences in the survival.
sampling area of ~60 mm2/patient, much larger than
that in prior works often employed core needle biopsy Results
(Table S1). Subsequently, the MPM images were
Definition of TACS1-8 and patient-specific
visually examined by three independent observers
quantification
(Jiajia He, Gangqin Xi, Lianhuang Li) who were also
blind to the final pathological outcomes. For each ROI, TACS1-3 have been proposed as biomarkers in
an individual TACS (defined below) was determined mouse mammary carcinoma progression [24] and
to be “present” if at least two reviewers answered subsequently identified as survival prognosticators
“yes” (Figure 1A). The average MPM imaging time on for human invasive breast cancer (IBC) [20] and DCIS
one section (patient) was ~1 h, and typical [21], as well as human ovarian cancer [22] and canine
examination time for one fully trained reviewer to mammary gland carcinoma [23]. We examined five
extract TACSs was ~10 min/section. new TACS-like (TACS4-8) structures in 431 IBC
Intraclass correlation coefficients (ICCs) were patients (Figure 1A) intended as the training cohort
used to evaluate the intra-observer and inter-observer for our retrospective study (Figure 1B). It should be
discordance of feature determination. To ensure noted that the differentiation among TACS1-3 (Figure
reproducibility, inter-observer correlation coefficients 2) and TACS4-8 (Figures 2-3) occurred at a scale of 0.4
of the TACSs determined by the three observers were mm and 2.8 mm, respectively. The accumulation of
calculated. To assess the intra-observer discordance, numerous large-scale SHG-TPEF images and
we randomly selected MPM images of 30 patients for co-registered H&E images from the training cohort
TACS feature extraction. Each panelist performed the allowed us to recognize 8 major TACSs (Figure 1C,
extraction for 10 patients and repeated the same steps Figures 2-3), in a manner like how up to 17
a week later. The mean intra-observer and histopathological subtypes were identified as
inter-observer agreement of the TACS feature biomarkers from H&E images [1]. TACS1 and TACS2
extraction among three panelists were 0.921 (95% CI, resemble those defined in mouse mammary
0.896-0.945) and 0.857 (95% CI, 0.825-0.888). An ICC carcinoma [24] except for the lack of dense collagen
value of >0.8 is considered as highly consistent. (TACS1) and the lack of highly straightened (taut)
collagen fibers (TACS2). These two TACSs can thus be
Statistical analysis treated as the human counterparts of the two
Statistical analysis was implemented with R 3.5.2 DCIS-like structures. As reported previously [20, 21],
and IBM SPSS Statistics 24. All statistical tests were TACS3 reflects a transition from DCIS to IBC due to
two-sided, and a P-value of less than 0.05 was the breakdown of the basement membrane. Therefore,
considered statistically significant. Univariate and TACS1-3 can be attributed to the initiation stage of
multivariate Cox proportional hazard regression tumor development (Figure 2).
analyses were used to select independent predictors Importantly, TACS4 is defined by a reticular
by likelihood ratio test. We used independent distribution of collagen fibers adjacent to

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3233

continuously distributed tumor cells, which leads to a are located at the invasion front near the tumor
clear tumor boundary. While TACS5 is defined by boundary (Figure 1A, Figure 1C) and can only be
directionally distributed collagen fibers that enable easily recognized at a large imaging field-of-view of
unidirectional tumor cell migration without a clear ~2.8 mm. They might have escaped detection in
tumor boundary, TACS6 is defined by chaotically previous studies that limited imaging field of view to
aligned collagen fibers that enable multidirectional 1 mm in at least one dimension (dictated by the core
tumor cell migration without a clear tumor boundary. size of core needle biopsy) or did not differentiate
Finally, TACS7 is defined by densely distributed whether the imaging was conducted at the invasion
collagen fibers at the tumor invasion front largely free front or tumor center (Table S1) [16-23]. For a given
of tumors cells, in contrast to TACS8 defined by patient represented by one FFPE section, multiple
sparsely distributed collagen fibers at the tumor TACSs might be present in one ROI and one TACS
invasion front largely free of tumors cells. TACS4-8 might exist in multiple ROIs. This complexity was
apparently reflect different tumor-stroma interactions caused by the large size (~2.8 mm) of the ROIs.
at the invasion stage of tumor development after the Regardless of the complexity, TACS information was
DCIS-to-IBC transition (Figures 2-3). quantified as an 8-element vector of CTACSi, which
To a large degree, TACS1-3 are located inside the reflected the percentage of TACSi (i = 1, 2, ... or 8)
tumor or near the tumor boundary, while TACS4-8 present in all ROIs (Figure 1A).

Figure 1. (A) Extraction and quantification of TACSs for one exemplary patient among the training, internal validation, and external validation cohorts. For one H&E section of
a patient, a total of 9 regions of interest (ROIs) are located either at the invasive front (1-8) or inside the tumor (9). (B) Study flowchart to exclude patients with neoadjuvant
chemotherapy or radiotherapy, unknown pathological characteristics and follow-up, or damaged and tumor-free sections. The TACS-score is calculated for each patient using the
linear combination of TACS percentages weighted by their regression coefficients. (C) Illustration of the structural and organizational features of collagen in the TACSs. TACS1-3
are plotted in the tumor center for simplicity but may be present in the invasion front like TACS4-8.

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3234

Figure 2. Images of TACS1-3 (or TACS4) at the initiation (or expansion) stage of tumor development. TACS1: curved collagen fibers wrapped around emergent tumor foci;
TACS2: collagen fibers stretched due to tumor growth and aligned more parallel to tumor boundary; TACS3: collagen fibers aligned perpendicular to the tumor boundary in a
radiation pattern to facilitate tumor cell migration; TACS4: reticular distribution of collagen fibers adjacent to expanding tumor that leads to a clear tumor boundary. Scale bar:
500 μm.

The quantified TACS information completed the cohort (n = 264) (Figure 1B). In all cohorts, the high
patient-specific quantification in this study that also and low TACS-scores were consistently associated
included age, molecular subtype (Luminal A, with high and low recurrence rates, respectively
Luminal B, HER2-enriched, or Triple-negative), tumor (Figure 4A).
size, nodal status, histological grade, clinical stage, The calculated patient-specific TACS-score,
(chemo-, endocrine, radiation, and/or targeted) along with age, molecular subtype, tumor size, nodal
therapy, and DFS (Tables S2-S3). Recurrence was status, clinical stage, chemotherapy, and endocrine
defined as the regeneration of tumor at the local site therapy (Table S3), were significantly (P < 0.05)
or in regional or distant organs after surgical resection associated with DFS in the univariate Cox
of the primary tumor, while DFS was defined as the proportional hazard regression analysis of the
time from the date of diagnosis (within 3 months from training cohort (Table S4, part 3). As expected, older
primary tumor surgery) to the date of first recurrence age, Triple-negative subtype, larger tumor size, more
or death [26]. positive lymph nodes, higher clinical stage, absence of
chemotherapy, absence of endocrine therapy, and
TACS1-8 as a strong prognosticator higher TACS-score were associated with worse DFS.
Based on the quantified TACS and DFS data in These prognosticators, except for the endocrine
the training cohorts (Table S2), we used ridge therapy strongly correlated with the molecular
regression with cross validation to retrieve the subtype (Table S4, part 2), were selected in stepwise
coefficient of each TACS. The coefficients of all TACSs building of prognostic model. We used the forward
were fixed in a formula to calculate a patient-specific stepwise selection method as the stopping rule to
TACS-score (Figure 1B), which was a manifestation of select independent predictors by the likelihood ratio
the 8 TACSs. We then applied this formula to each test. In the corresponding multivariate analysis,
patient in the training cohort, as well as an internal TACS-score remained as an independent prognostic
validation cohort (n = 300) and an external validation factor, along with the well-known biomarkers of

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3235

tumor size, nodal status, and molecular subtype 72%, and 54%, respectively (Figure 4B, red line). The
(Table S4, part 1). This was consistent with the calibration curve of the nomogram for 5-year DFS rate
increased appreciation of molecular subtype in IBC showed good agreement between observed and
classification and prognosis [27]. The multicollinearity predicted rates in the training cohort (approximation
analysis yielded a variance inflation factor of the to 45-degree diagonal line is indicative of a well
prognosticators from 1.001 to 1.187 (values far less calibrated model), which was retained in the two
than 10), indicating that there was no collinearity validation cohorts (Figure 4C). This guaranteed the
among these prognosticators. repeatability and reliability of the established
The risk prediction model containing these nomogram.
independent prognosticators were established from As a reference for TACS-based prognosis, a
the training cohort and presented as a nomogram, contextual clinical model was developed based on
wherein the TACS-score considerably outweighed eight clinicopathological factors (age, molecular
molecular subtype, nodal status, and tumor size subtype, tumor size, nodal status, clinical stage,
(Figure 4B). Each subgroup of these prognosticators histological grade, chemotherapy, and radiation
was assigned a point in the corresponding point scale. therapy) as the computer decision support system
By summing up the points of all prognosticators in the available for pathologists. To assess the differential
scale of total points, we could determine the 1-year, values of TACS4-8 unique to this study, a multivariate
3-year and 5-year DFS rates for a given patient. For model of TACS1-3 was developed in parallel to the
example, a patient with a TACS-score of 0.23, nodal TACS1-8 model (TACS-score) discussed above, and
status of 0, tumor size of ≤2cm, and molecular subtype would reflect what extent of prognosis could be
of Triple-negative, would have a total point of 75 and achieved for these 995 patients from the
predicted 1-year, 3-year and 5-year DFS rates of 92%, methodologies of previous works [20-24].

Figure 3. Images of TACS5-8 at the invasion stage of tumor development. TACS5: directionally distributed collagen fibers that enables unidirectional tumor cell migration
without a clear tumor boundary; TACS6: chaotically aligned collagen fibers that enables multidirectional tumor cell migration without a clear tumor boundary; TACS7:
densely-distributed collagen fibers at the tumor invasion front largely free of tumors cells; TACS8: sparsely-distributed collagen fibers at the tumor invasion front largely free of
tumors cells. Scale bar: 500 μm.

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3236

Figure 4. (A) Recurrence histograms of TACS-score for three cohorts. (B) Nomogram of TACS-score, molecular subtype, tumor size and nodal status derived from the training
cohort. (C) Calibration curves of the nomogram to predict 5-year DFS rate for three cohorts.

The discriminatory accuracy of the clinical diagnosis year [26, 28]. The TACS1-3 model had low
model on 5-year DFS rate was reflected by the area prediction accuracy and led to a lower AUC of ~0.64
under the curve (AUC) of ~0.73 from the receiver for all cohorts, echoing the low prognostic strength of
operating characteristic (ROC) curve, which TACS1-3 observed in canine mammary gland
expectedly outperformed simpler models based on carcinoma [23]. In contrast, the TACS1-8 model had
some individual prognosticators (tumor size, nodal high prediction accuracy (Figure S1B) and led to a
status, and molecular subtype) (Figure 5A, Figure higher AUC of ~0.82 for all cohorts (Figure 5A, Figure
S1A), and approximated the high AUC (0.79-0.81) S1A). The nomogram model further improved the
obtained by including unconventional prognostic- AUC to ~0.86 and enhanced the discriminatory
cators such as blood test variables, race/ethnicity, accuracy (Figure 5A, Figure S1A). As to the
cancer detection mode, alcohol consumption, and corresponding stratification of high- and low-risk

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3237

patients, the TACS1-3 model, clinical model, TACS1-8 than the low TACS risk of histological grade G3
model, and nomogram model delivered an increasing (Figure S2). These results indicated that the
hazard ratio (HR) from ~2 to ~7 (Figure 5A), TACS-score risk stratification reclassified a significant
indicating an increasing ability for risk stratification. portion of patients classified by some well-known
Thus, the combined high discriminatory accuracy and clinicopathological factors (i.e. highly complemented
super risk stratification ability validated the these factors), and the TACS-score functioned as a
TACS-score as a strong DFS prognosticator. general prognosticator with no restriction to
specifically classified patients. The small deficiency in
TACS1-8 as a general prognosticator risk assessment of Luminal A patients (with a low
To demonstrate the general applicability of the AUC of 0.758) was more than compensated by the
TACS-score prognosticator, we combined all cohorts outstanding ability in risk assessment of
into 995 patients classified by various clinicopatholo- HER2-enriched and Triple-negative patients. To the
gical factors (Table 1). We found that the TACS-score opposite of, yet highly complementary with typical
remained statistically significant in all cases. In the multigene assays, the TACS-score performed better
Kaplan-Meier analysis, each classified group showed for estrogen receptor-negative patients than for
relatively large separation between the low and high estrogen receptor-positive patients, and attained a
TACS risk patients (Figure S2). The high TACS risk of higher AUC in highly node-positive (>3) patients than
tumor size ≤2cm or 2-5cm had a poorer prognosis in node-negative (0) patients (Table 1). Interestingly,
than the low TACS risk of tumor size >5cm; the high the TACS-score also performed well for the early
TACS risk of 0 or 1-3 positive lymph nodes had a stage IBC (Table 1) where the multigene assays have
poorer prognosis than the low TACS risk of ≥4 delivered the best value to avoid chemotherapy [2, 5].
positive lymph nodes; the high TACS risk of clinical A larger number of early stage IBC patients will be
stage I or II had a poorer prognosis than the low TACS needed to explore the ability of the TACS-score to
risk of clinical stage III; and the high TACS risk of mitigate cancer overtreatment.
histological grade G1 or G2 had a poorer prognosis

Table 1. Prognosis of clinicopathologically classified IBC patients by the TACS1-8 model.

Variable 5-yr DFS Low risk HR 95% CI P value AUC Sensitivity Specificity
Age
≤50 369 (66.5%) 329 (59.3%) 5.91 4.32-8.08 <0.0001 0.825 0.747 0.764
>50 269 (61.1%) 231 (52.5%) 6.29 4.42-8.95 <0.0001 0.829 0.819 0.747
Molecular subtype
Luminal A 180 (81.1%) 135 (60.8%) 3.88 2.13-7.05 <0.0001 0.758 0.738 0.689
Luminal B 255 (60.3%) 229 (54.1%) 5.64 4.03-7.91 <0.0001 0.822 0.780 0.757
HER2-enriched 118 (61.1%) 116 (60.1%) 8.35 4.95-14.1 <0.0001 0.871 0.773 0.839
Triple Negative 85 (54.1%) 80 (51.0%) 8.02 4.38-14.7 <0.0001 0.875 0.819 0.788
Tumor size
≤2cm 326 (73.3%) 264 (59.3%) 5.94 4.02-8.77 <0.0001 0.820 0.790 0.736
2-5cm 293 (59.2%) 272 (54.9%) 6.64 4.82-9.14 <0.0001 0.841 0.782 0.778
>5cm 19 (34.5%) 24 (43.6%) 3.62 1.71-7.66 0.001 0.804 0.750 0.789
Nodal status
0 380 (77.7%) 328 (67.1%) 6.07 4.15-8.88 <0.0001 0.828 0.743 0.789
1-3 157 (63.3%) 130 (52.4%) 4.46 2.82-7.05 <0.0001 0.776 0.758 0.688
≥4 101 (39.1%) 102 (39.5%) 5.34 3.57-7.99 <0.0001 0.840 0.822 0.743
Clinical stage
Ⅰ 216 (81.5%) 178 (67.2%) 4.41 2.61-7.47 <0.0001 0.803 0.714 0.759
Ⅱ 317 (68.0%) 278 (59.7%) 6.06 4.25-8.64 <0.0001 0.817 0.758 0.763
Ⅲ 105 (39.8%) 104 (39.4%) 5.37 3.59-8.02 <0.0001 0.833 0.824 0.733
Histological grade
G1 98 (74.2%) 73 (55.3%) 5.48 2.60-11.6 <0.0001 0.780 0.794 0.684
G2 397 (66.3%) 342 (57.1%) 6.45 4.73-8.80 <0.0001 0.828 0.792 0.756
G3 143 (54.2%) 145 (54.9%) 6.09 4.05-9.16 <0.0001 0.853 0.760 0.811
Estrogen receptor
positive 434 (67.8%) 363 (56.7%) 5.06 3.77-6.78 <0.0001 0.798 0.767 0.728
negative 204 (57.5%) 197 (55.5%) 8.47 5.71-12.6 <0.0001 0.878 0.801 0.819
Young early stage 120 (83.3%) 102 (70.8%) 4.96 2.40-10.2 <0.0001 0.862 0.708 0.792
General early stage 216 (81.5%) 178 (67.2%) 4.41 2.61-7.47 <0.0001 0.803 0.714 0.759
Note: Young early stage invasive breast cancer (Age ≤50, Tumor size ≤2cm, Nodal status 0, Clinical stage I); General early stage invasive breast cancer (Tumor size ≤2cm,
Nodal status 0, Clinical stage I).

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3238

We also generalized the binary risk stratification two high-risk quartiles (Figure 5A, right panel). This
of the 995 patients to the corresponding quaternary feature was also observed for the nomogram model
stratification into equal sized quartiles (Figure 5A, but not for the TACS1-3 model or clinical model
Table S5). In the TACS1-8 model, the Kaplan-Meier (Figure 5A, right panel), consistent with the dominant
curves of the two low-risk quartiles approximated role of TACS-score in the nomogram (Figure 4B).
each other but differed significantly from those of the

Figure 5. (A) Kaplan-Meier curves of DFS according to the TACS1-3 model, clinical model, TACS1-8 model, and nomogram model in four cohorts, with AUC values at the
5-year time point, HR intervals with 95% confidence level, and numbers of patients in two or four risk groups (colored numbers). (B) Numbers of likely undertreated
(red-highlighted) and overtreated (purple-highlighted) patients according to Chinese treatment guideline and the four models. (C) Distribution histogram of TACS1-3 score and
TACS-score among 995 patients in the TACS1-3 model and TACS1-8 model, respectively.

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3239

Treatment implication explored in future studies.

The correlation between individual TACSs and
To assess postoperative adjuvant treatment, we
DFS was also assessed by Spearman's rank correlation
used the 5-year prognosis obtained by the clinical,
coefficient. While TACS-specific coefficients exhibited
TACS, and nomogram models to evaluate the
small difference among different cohorts, TACS6, 5, 8
recurrence risk groups classified by the treatment
(TACS4, 1, 7) were consistently correlated with poor
guideline in China, which is consistent with the 10th
(good) prognosis (Figure 6B). The significant
St Gallen expert consensus (often considered as the
univariate correlation of TACS5 (TACS7) was
best treatment approach for primary IBC) [29]. The
obscured in the multivariate analysis (Table S6) and
execution of this guideline for the 995 patients
ridge regression (Figure 6A), plausibly by the larger
allowed 626 patients to be stratified as minimum risk
effect of TACS6 (TACS4). In contrast, the small
(n = 32) or moderate risk (n = 594) (defined together as
univariate correlation of TACS3, likely due to its
low risk) for less aggressive treatment, and 369
low-percentage presence (Figure 6C), emerged as a
patients as high risk for more aggressive treatment
large coefficient in the (multivariate) ridge regression
(Figure 5B). The guideline caused likely
(Figure 6A). We noted that the external validation
undertreatment to 150 patients and likely
cohort (northern Chinese patients, Figure 1B) had
overtreatment to 162 patients, which served as the
rather different percentages of TACS1-8 presence
references for the four models: (i) the TACS1-3 model
from those of the training or internal validation cohort
lowered the undertreated patients by 94 at the cost of
(southern Chinese patients) (Figure 6C). The
240 more overtreated patients, reflecting a worsened
consistent results from the two disparate cohorts
performance; (ii) the clinical model lowered the
(Figure 5) reinforced the general applicability of the
overtreated patients by 49 at the cost of 24 more
TACS-score.
undertreated patients, reflecting a small
We compared the 5-year DFS prognosis using
improvement; (iii) the TACS1-8 model lowered the
individual TACS models, the clinical model, TACS1-3
undertreated patients by 72 and the overtreated
model, and TACS1-8 model. We found that the
patients by 6, reflecting a larger improvement; and (iv)
TACS1-8 model showed significantly higher AUC
the nomogram model lowered the undertreated
than individual TACS models (Table S7), which was
patients by 72 and the overtreated patients by 36,
associated with the more even distribution of
reflecting the best overall improvement (Figure 5B,
TACS-score among all 995 patients than a specific
Figure S3). Thus, our TACS-score-based prognosis
CTACSi (Figure 5C vs. Figure 6D). Thus, the effects of all
showed great potential to complement the
TACSs should be comprehensively considered. On
standard-of-care treatment guideline in China and
the other hand, one model that integrated three
improve the choices in tailored adjuvant treatment. In
independent TACS structures (TACS4, TACS6, and
particular, the patients with poor TACS-sore
TACS8) outperformed the TACS1-3 model or the
prognosis but low risk under the guideline (72 of 357
clinical model, and approximated the performance of
patients with unsuccessful 5-year DFS) could have
the TACS1-8 (full) model, indicating the
been treated more aggressively. Thus, the TACS-score
disproportional contribution of TACS4-8 over
could be valuable to address the undertreatment in
TACS1-3 to the TACS-score (Table S7). In this sense,
less developed countries, which accounts for 45%
our study demonstrates the limitation of previous
(and increasing) of the global burden of breast cancer
studies based on TACS1-3 alone [20-24].
[15].
Individual TACS effects Discussion
To assess the individual effects of the 8 TACSs, Stephen Paget’s “Seed and Soil” hypothesis
we performed similar univariate and multivariate suggested that breast cancer metastasis originated
Cox proportional hazard regression analysis as from the intricate interaction between tumor cells and
discussed above. While the combinations of their microenvironment [30]. Breast tumor
independent TACSs exhibited small difference among microenvironment can regulate tumor progression
different cohorts, TACS1, 3, 4, 6, 8 were identified as through extracellular matrix remodeling in
independent prognosticators for the combined cohort tumor-associated stroma [10]. From the viewpoint of
(Table S6). Consistently, the coefficients associated cellular gene expression, the role of the stroma in
with TACS5 and TACS7 from the ridge regression regulating breast tumor development and clinical
(training cohort) were small in comparison to those outcome has also been recognized [11, 31]. Improved
associated with other TACSs (Figure 6A). Whether the understanding of tumor-associated extracellular
TACS prognosticator can be simplified from an matrix and embedded stromal cells at the molecular
8-structure model to a 6- or 5-structure model will be level has motivated new cancer therapies to target the

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3240

Figure 6. (A) Ridge regression of training cohort to attain the coefficients of TACSs in DFS prognosis. (B) Correlation analysis between individual TACSs and DFS for four
cohorts. (C) Average quantified TACS data of one patient in training, internal validation, or external validation cohort. (D) Distribution histogram of CTACSi among 995 patients
in individual TACS models.

tumor microenvironment, rather than the tumor cells cells alone.

themselves [32, 33]. Parallel attempts have been taken The innovations in anatomic pathology
to shift postoperative survival prognosis from tumor (histology) that use structural biomarkers have not
cells and intracellular molecules or functions to kept pace with those of clinical pathology that use
stromal or extracellular matrix structures [4, 6, 16-23]. functional biomarkers. The invention of MPM in 1990
However, the corresponding tumor microenviron- [35] was thought to empower anatomic pathology by
ment-based structural prognosticators have not introducing new imaging contrasts such as SHG [36].
demonstrated compelling strength and general Although MPM has shown promise in in vivo optical
applicability (clinical validity) to justify clinical trials biopsy to improve cancer diagnosis [37] and
[2], in comparison to various multigene assays based interoperative tumor margin detection to improve
on tumor cells, intracellular biomarker molecules, and cancer surgery [38, 39], studies with more direct
biological functions [5]. In this study, the TACS-score impact to histology have rarely been clinically
emerges as a tumor microenvironment-based validated. One plausible reason is the lack of a
structural prognosticator with surprising strength of precision technology to co-register a virtual imaging
5-year prognosis in discriminatory accuracy (AUC) section in the intact and often fresh tissue sample
and risk stratification ability (HR), which competes (MPM) with one physical section in a FFPE tissue
favorably with that of the 10-year prognosis by block (histology). Thus, the benefit of additional MPM
popular multigene assays [34]. Unlike the multigene contrasts over those available from H&E histology
assays, the TACS-score is generally applicable to all and stained immunohistology (IHC) has remained
IBC subtypes, including the early stage IBC where the unclear. Our study differs from most clinical MPM
multigene assays have delivered the best value (Table studies by imaging a thin FFPE section, an
1). The unexpected identification of the TACS-score as underappreciated sample that is usually avoided to
a strong and general IBC prognosticator reinforces the take advantage of MPM in three-dimensional imaging
central role of the tumor microenvironment in tumor that obviates the need for tissue sectioning and
invasion and metastasis. It treats cancer metastasis as pretreatments. Thus, our MPM-based technology of
a complex tumor-stroma interaction and may cancer prognosis inherits some beneficial elements of
therefore produce a survival prognosticator that standard histology: (i) a retrospective study is enabled
outperforms the prognosticators based on the tumor due to the availability of archived FFPE tissue blocks

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3241

with well-documented clinicopathological informa- collagen fibers in different organs and epithelial
tion; (ii) image co-registration is readily achieved tissues. For one common cancer type (e.g. breast or
(Figure 1A, Figures 2-3) just as H&E-stained histology colorectal cancer), it will be interesting to compare the
and stained IHC are correlated; (iii) enabled by this prognosis by the TACS-score originated from breast
image co-registration, a pathologist can use his/her cancer and the Immunoscore originated from
valuable experience to select local MPM imaging colorectal cancer [43]. Third, a wide variety of other
regions near the tumor boundary with a global view nonlinear intrinsic/label-free contrasts (e.g. TPEF in
of the primary tumor(s) and adjacent normal Figure 3 that reveals cellular information) may
appearing tissue (Figure 1A), which is important for synergistically complement the SHG contrast to
the TACS-score to attain high prognostic value; and further improve the prognosis [44], allowing MPM to
(iv) the associated IHC assays allow the determination translate from bench-level laboratorial research to
of 4 IBC molecular subtypes, which complement the clinical practice. On the other hand, alternative
TACS-score to form a better performing nomogram. cost-effective and/or high-throughput methods to
However, the fundamental difference that image collagen fibers, such as the wide-field
distinguishes our technology from standard histology microscopy with NHS-ester fluorescent labeling [45]
is the replacement of histological grade with the or picrosirius red staining [46], may be explored to
TACS-score as the only structural prognosticator lower the cost of dedicated TACS extraction. Fourth,
(Table S4). It is striking that the TACS-score this retrospective study provides strong justification
overpowers the well-established histological grade [3] for future prospective studies to validate the clinical
in prognostic strength, highlighting the indispensable utility of the TACS prognosticator, which not only
role of the tumor microenvironment in cancer demonstrates the ability to mitigate undertreatment in
prognosis. A paradigmatic shift from viewing the less-developed countries (Figure 5B), but also holds
discrete cell nuclei stained by hematoxylin to imaging potential to mitigate overtreatment for early stage
the collagen fibers by SHG (MPM) in our “revamped” cancer in well-developed countries (Table 1).
histology has yielded a structural prognosticator that There are several strengths in our study. First,
can compete with the best functional prognosticators. our instrumentation to extract TACSs is highly
Our histology approach that focuses on fibrillar compatible with standard histology and can be
collagen (main component of extracellular matrix) implemented without perturbing the routine
bridges the gap between automated H&E histology [4, histological workflow in future prospective studies
6] that lacks molecular specificity and clinical (Figure 1A). Alternative technologies using tumor
molecular pathology that lacks structural information. microarrays or gene expression assays are typically
This bulk biomolecule apparently outperforms many less compatible. Second, the standardization of TACS
biomarkers of rare biomolecules in balancing the feature extraction is simple after optimizing and
sensitivity and specificity of cancer prognosis (Table fixing the settings of the label-free MPM microscope
1), defying the conventional wisdom that bulk along with the size and number of imaging regions.
biomolecules are slow or insensitive to respond to Alternative technologies using external
subtle cancerous changes. This finding has opened staining/labeling in the wet laboratory are typically
many new fronts of basic and translational research. more resistant to standardization. Third, the FFPE
First, a rational understanding is needed to explain sections allow easy transportation from remote
the disparate prognostic values of different TACSs. It hospitals or other sample collection sites to one
may not be surprising that TACS4-8 at invasive or central MPM facility and may thus lower the cost of
expansion stages and a larger imaging scale are more the TACS feature extraction. We also acknowledge
prognostic than TACS1-3 at tumor initiation stages some limitations in this study. First, our study is
and relatively small scales. The prognostic differences limited to Chinese IBC patients that have relatively
among TACS4-8 necessitate cell-migration low 5-year DFS rate (Table S3) than the IBC patients in
extracellular matrix models that have only been most Western countries, even though the TACS-score
elucidated recently using SHG [40-42]. Regardless of prognosis exhibits comparable effectiveness in
the imaging scale, TACS4,7,8 and TACS5,6 are more southern and northern China (Figure 1B). Also, our
subtle structures within the TACS framework and results need to be further validated with longer
may be treated as different variations of TACS2 and follow-up (10 years rather than 5 years), to evaluate
TACS3, respectively (Figures 2-3). Second, it will be the long-term effect of the TACSs. Second, the
important to examine whether the TACSs or similar extraction of TACSs necessitates three reviewers and
structures of fibrillar collagen can be generalized for may be prone to interobserver discordance, which
risk assessment of other cancer (prostate, lung, skin, may be mitigated by the automatic recognition and
liver, etc.) types, considering the broad role of quantification of TACSs [47]. Third, our prognosis is

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3242

limited to the DFS that emphasizes quality of life, Y. Fang, X. H. Han, X. X. Huang and Z. J. Li for their
rather than the overall survival that dictates cancer support in the sample preparation and multiphoton
mortality (which will be further explored to test the imaging.
validity of the TACSs). Despite these limitations, we
believe our study has demonstrated the potential use Author Contributions
of TACSs for clinical practice. Conception and design were provided by L. Li,
Q. Zhang, C. Wang, S. A. Boppart, J. Chen (Jianxin
Conclusions Chen). Development of methodology were
In summary, the current study demonstrates a contributed by G. Xi, L. Qiu, L. Li. Acquisition of data
surprising biomarker (TACS-score) to predict were supplied by G. Xi, W. Guo, D. Kang, J. Ma, F. Fu,
individual disease-free survival rate for breast cancer L. Zheng, J. He, N. Fang, J. Chen (Jianhua Chen), J. Li.
patients, with great potential to identify high-risk Analysis and interpretation of data were done by G.
patients (mitigate undertreatment) and expand to Xi, J. He, H. Tu, L. Li, C. Wang, J. Chen (Jianxin Chen).
other human cancers. The corresponding models have Writing, review, and/or revision of the manuscript
high discriminatory accuracy, superior risk were imparted by G. Xi, H. Tu, L. Li, Q. Zhang, C.
stratification ability, good calibration performance, Wang, S. A. Boppart, J. Chen (Jianxin Chen).
and broad applicability. This finding strengthens the Administrative, technical, or material supports were
often-underappreciated role of the tumor provided by G. Xi, W. Guo, J. Ma, F. Fu, L. Zheng, J.
microenvironment in (breast) cancer prognosis, He, N. Fang, J. Chen (Jianhua Chen), J. Li, S. Zhuo, X.
introduces innovative features to anatomic pathology, Li. Study supervision was supplied by L. Li, J. Chen
promotes the transformation of multiphoton imaging (Jianxin Chen).
from laboratory-based research to clinical practice,
and enables more informed decisions on adjuvant
Competing Interests
systemic therapy. The authors have declared that no competing
interest exists.
Abbreviations
TACS: tumor-associated collagen signatures;
References
1. Weigelt B, Peterse JL, Van't Veer LJ. Breast cancer metastasis: markers and
IBC: invasive breast cancer; MPM: multiphoton models. Nat Rev Cancer. 2005; 5: 591-602.
microscopy; SHG: second harmonic generation; TPEF: 2. Engelhardt EG, Garvelink MM, de Haes JH, van der Hoeven JJ, Smets EM,
Pieterse AH, et al. Predicting and communicating the risk of recurrence and
two-photon excited fluorescence; H&E: hematoxylin death in women with early-stage breast cancer: a systematic review of risk
and eosin; DFS: disease-free survival; FFPE: prediction models. J Clin Oncol. 2014; 32: 238-50.
3. ELSO CW, Ellis IO. Pathological prognostic factors in breast cancer I. The
formalin-fixed paraffin-embedded; ICCs: Intraclass value of histological grade in breast cancer: experience from a large study with
correlation coefficients; DCIS: ductal carcinoma in long-term follow up. Histopathology. 1991; 19: 403-10.
4. Beck AH, Sangoi AR, Leung S, Marinelli RJ, Nielsen TO, Van De Vijver MJ, et
situ; AUC: area under the curve; ROC: receiver al. Systematic analysis of breast cancer morphology uncovers stromal features
operating characteristic; HR: hazard ratio. 5.
associated with survival. Sci Transl Med. 2011; 3: 108-13.
Kwa M, Makris A, Esteva FJ. Clinical utility of gene-expression signatures in
early stage breast cancer. Nat Rev Clin Oncol. 2017; 14: 595-610.
Supplementary Material 6. Yuan Y, Failmezger H, Rueda OM, Ali HR, Gräf S, Chin S-F, et al. Quantitative
image analysis of cellular heterogeneity in breast tumors complements
Supplementary figures and tables. genomic profiling. Sci Transl Med. 2012; 4: 143-57.
7. Reis-Filho JS, Weigelt B, Fumagalli D, Sotiriou C. Molecular profiling: moving
http://www.thno.org/v11p3229s1.pdf away from tumor philately. Sci Transl Med. 2010; 2: 43-7.
8. Cuzick J, Dowsett M, Pineda S, Wale C, Salter J, Quinn E, et al. Prognostic
Acknowledgments value of a combined estrogen receptor, progesterone receptor, Ki-67, and
human epidermal growth factor receptor 2 immunohistochemical score and
comparison with the Genomic Health recurrence score in early breast cancer. J
This work was supported by the National Clin Oncol. 2011; 29: 4273-8.
Natural Science Foundation of China (Grant Nos. 9. Baenke F, Peck B, Miess H, Schulze A. Hooked on fat: the role of lipid
synthesis in cancer metabolism and tumour development. Dis Model Mech.
81671730, 61972187), Fujian Major Scientific and 2013; 6: 1353-63.
Technological Special Project for “Social 10. Lu P, Weaver VM, Werb Z. The extracellular matrix: a dynamic niche in cancer
progression. J Cell Biol. 2012; 196: 395-406.
Development” (No. 2020YZ016002), Natural Science 11. Finak G, Bertos N, Pepin F, Sadekova S, Souleimanova M, Zhao H, et al.
Foundation of Fujian Province (Nos. 2019J01269, Stromal gene expression predicts clinical outcome in breast cancer. Nat Med.
2008; 14: 518-27.
2020J011008, 2020J01839). Also, this work was 12. Galon J, Costes A, Sanchez-Cabo F, Kirilovsky A, Mlecnik B, Lagorce-Pagès C,
supported, in part, by grants from the National et al. Type, density, and location of immune cells within human colorectal
tumors predict clinical outcome. Science. 2006; 313: 1960-4.
Institutes of Health, U.S. Department of Health and 13. Ravdin PM, Siminoff LA, Davis GJ, Mercer MB, Hewlett J, Gerson N, et al.
Human Services (R01 CA241618, R43 MH119979, and Computer program to assist in making decisions about adjuvant therapy for
women with early breast cancer. J Clin Oncol. 2001; 19: 980-91.
R41 GM139528 to H. T. and S.A.B.), Special Funds of 14. Wishart GC, Azzato EM, Greenberg DC, Rashbass J, Kearins O, Lawrence G, et
the Central Government Guiding Local Science and al. PREDICT: a new UK prognostic model that predicts survival following
surgery for invasive breast cancer. Breast Cancer Res. 2010; 12: R1.
Technology Development (No. 2020L3008). We thank 15. Ferlay J, Héry C, Autier P, Sankaranarayanan R. Global burden of breast
cancer, Breast cancer epidemiology, Springer, 2010: 1-19.
W. J. Ren, T. F. Shen, W. Wang, Z. Chen, X. W. Chen,

http://www.thno.org
Theranostics 2021, Vol. 11, Issue 7 3243
16. Drifka CR, Loeffler AG, Mathewson K, Keikhosravi A, Eliceiri KW. Highly 45. Doyle AD, Carvajal N, Jin A, Matsumoto K, Yamada KM. Local 3D matrix
aligned stromal collagen is a negative prognostic factor following pancreatic microenvironment regulates cell migration through spatiotemporal dynamics
ductal adenocarcinoma resection. Oncotarget. 2016; 7: 76197–213. of contractility-dependent adhesions. Nat Commun. 2015; 6: 1-15.
17. Sridharan S, Macias V, Tangella K, Melamed J, Dube E, Kong MX, et al. 46. Lattouf R, Younes R, Lutomski D, Naaman N, Godeau G, Senni K, et al.
Prediction of prostate cancer recurrence using quantitative phase imaging: Picrosirius red staining: a useful tool to appraise collagen networks in normal
Validation on a general population. Sci Rep. 2016; 6: 33818. and pathological tissues. J Histochem Cytochem. 2014; 62: 751-8.
18. Burke K, Smid M, Dawes RP, Timmermans MA, Salzman P, van Deurzen CH, 47. Bredfeldt JS, Liu Y, Conklin MW, Keely PJ, Mackie TR, Eliceiri KW.
et al. Using second harmonic generation to predict patient outcome in solid Automated quantification of aligned collagen for human breast carcinoma
tumors. BMC Cancer. 2015; 15: 929. prognosis. J Pathol Inform. 2014; 5: 28.
19. Desa DE, Bhanote M, Hill RL, Majeski JB, Brown EB. Second-harmonic
generation directionality is associated with neoadjuvant chemotherapy
response in breast cancer core needle biopsies. J Biomed Opt. 2019; 24: 1-9.
20. Conklin MW, Eickhoff JC, Riching KM, Pehlke CA, Eliceiri KW, Provenzano
PP, et al. Aligned collagen is a prognostic signature for survival in human
breast carcinoma. Am J Pathol. 2011; 178: 1221-32.
21. Conklin MW, Gangnon RE, Sprague BL, Van Gemert L, Hampton JM, Eliceiri
KW, et al. Collagen alignment as a predictor of recurrence after ductal
carcinoma in situ. Cancer Epidemiol Biomarkers Prev. 2018; 27: 138-45.
22. Adur J, Pelegati VB, De Thomaz AA, Baratti MO, Andrade LALA, Carvalho
HF, et al. Second harmonic generation microscopy as a powerful diagnostic
imaging modality for human ovarian cancer. J Biophotonics. 2014; 7: 37-48.
23. Ashley C, Brisson BK, Durham AC, Suzanne R, James M, Elizabeth B, et al.
Identification of prognostic collagen signatures and potential therapeutic
stromal targets in canine mammary gland carcinoma. PLoS One. 2017; 12:
e0180448.
24. Provenzano PP, Eliceiri KW, Campbell JM, Inman DR, White JG, Keely PJ.
Collagen reorganization at the tumor-stromal interface facilitates local
invasion. BMC Med. 2006; 4: 38.
25. Zoumi A, Yeh A, Tromberg BJ. Imaging cells and extracellular matrix in vivo
by using second-harmonic generation and two-photon excited fluorescence.
Proc Natl Acad Sci USA. 2002; 99: 11014-9.
26. Zhu Z, Li L, Ye Z, Fu T, Du Y, Shi A, et al. Prognostic value of routine
laboratory variables in prediction of breast cancer recurrence. Sci Rep. 2017; 7:
8135.
27. The Cancer Genome Atlas Network, Comprehensive molecular portraits of
human breast tumours. Nature. 2012; 490: 61-70.
28. Wu X, Ye Y, Barcenas CH, Chow WH, Meng QH, Chavez-MacGregor M, et al.
Personalized Prognostic Prediction Models for Breast Cancer Recurrence and
Survival Incorporating Multidimensional Data. J Natl Cancer Inst. 2017; 109:
djw314.
29. Goldhirsch A, Wood W, Gelber R, Coates A, Thürlimann B, Senn H-J, et al.
Progress and promise: highlights of the international expert consensus on the
primary therapy of early breast cancer 2007. Ann Oncol. 2007; 18: 1133-44.
30. Paget S. The distribution of secondary growths in cancer of the breast. Lancet.
1889; 133: 571-3.
31. Bergamaschi A, Tagliabue E, Sørlie T, Naume B, Børresen-Dale AL.
Extracellular matrix signature identifies breast cancer subgroups with
different clinical outcome. J Pathol. 2010; 214: 357-67.
32. Joyce JA. Therapeutic targeting of the tumor microenvironment. Cancer Cell.
2005; 7: 513-20.
33. Albini A, Sporn MB. The tumour microenvironment as a target for
chemoprevention. Nat Rev Cancer. 2007; 7: 139-47.
34. Sestak I, Buus R, Cuzick J, Dubsky P, Kronenwett R, Denkert C, et al.
Comparison of the performance of 6 prognostic signatures for estrogen
receptor–positive breast cancer: a secondary analysis of a randomized clinical
trial. JAMA Oncol. 2018; 4: 545-53.
35. Denk W, Strickler JH, Webb WW. Two-photon laser scanning fluorescence
microscopy. Science. 1990; 248: 73-6.
36. Campagnola PJ, Dong CY. Second harmonic generation microscopy:
principles and applications to disease diagnosis. Laser Photonics Rev. 2011; 5:
13-26.
37. Li A, Hall G, Chen D, Liang W, Ning B, Guan H, et al. A biopsy‐needle
compatible varifocal multiphoton rigid probe for depth‐resolved optical
biopsy. J Biophotonics. 2019; 12: e201800229.
38. Tao YK, Shen D, Sheikine Y, Ahsen OO, Wang HH, Schmolze DB, et al.
Assessment of breast pathologies using nonlinear microscopy. Proc Natl Acad
Sci USA. 2014; 111: 15304-9.
39. Sun Y, You S, Tu H, Spillman DR, Chaney EJ, Marjanovic M, et al.
Intraoperative visualization of the tumor microenvironment and
quantification of extracellular vesicles by label-free nonlinear imaging. Sci
Adv. 2018; 4: eaau5603.
40. Riching KM, Cox BL, Salick MR, Pehlke C, Riching AS, Ponik SM, et al. 3D
collagen alignment limits protrusions to enhance breast cancer cell persistence.
Biophys J. 2014; 107: 2546-58.
41. Han W, Chen S, Yuan W, Fan Q, Tian J, Wang X, et al. Oriented collagen fibers
direct tumor cell intravasation. Proc Natl Acad Sci USA. 2016; 113: 11208-13.
42. Ray A, Lee O, Win Z, Edwards RM, Alford PW, Kim DH, et al. Anisotropic
forces from spatially constrained focal adhesions mediate contact guidance
directed cell migration. Nat Commun. 2017; 8: 14923.
43. Mlecnik B, Bindea G, Kirilovsky A, Angell HK, Obenauf AC, Tosolini M, et al.
The tumor microenvironment and Immunoscore are critical determinants of
dissemination to distant metastasis. Sci Transl Med. 2016; 8: 327ra26.
44. Tu H, Liu Y, Turchinovich D, Marjanovic M, Lyngsø JK, Lægsgaard J, et al.
Stain-free histopathology by programmable supercontinuum pulses. Nat
Photonics. 2016; 10: 534.

http://www.thno.org