Background
The tumor microenvironment or tumor stroma has a critical role in regulating cancer cells. It is involved in tumorigenesis by inducing stem cell-like properties and epithelial-to-mesenchymal transition (EMT) in cancer cells [
1]. The tumor-stroma ratio (TSR) is a prognostic tool that stratifies tumors into stroma-low and stroma-high based on the quantity of stroma scored in hematoxylin & eosin (H&E) stained sections [
2,
3]. Stroma-high tumors were shown to have a poor patient prognosis in colorectal cancer as well as in other solid epithelial tumors [
4‐
7]. Additionally to the quantity of the tumor stroma, the composition of the stroma may be an important determinant of cancer behavior. Several transcriptomic and immunohistochemical studies have shown that desmoplastic stroma is associated with poor patient outcome and can predict response to therapy [
8‐
11]. The desmoplastic reaction consists mainly of activated fibroblasts, also called cancer-associated fibroblasts (CAFs). CAFs are a heterogeneous cell population in terms of origin and biological function and derive mainly from mesenchymal cells that are resident or recruited by the tumor [
12]. They are situated close to cancer cells and other components of the stroma like immune cells, blood vessels and components of the extracellular matrix (ECM). Fibroblast is a general term which includes resident quiescent fibroblasts, CAFs, myofibroblasts and pericytes. The identification and nomenclature of fibroblasts present in the tumor remain challenging due to the lack of specific markers for known and still undefined subtypes.
The aim of this study is to investigate and quantify CAFs present in the tumor stroma in colorectal cancer (CRC) stratified by the TSR. The composition of the tumor stroma may add prognostic value to the TSR. In a cohort of 71 CRC patients, the difference in transcriptomic data of currently used fibroblast and CAF markers were first compared between stroma-low and stroma-high tumors. From the fibroblast markers investigated, the most commonly used in literature were stained in colon tumor sections of thirty-two stage III patients using multiplex immunofluorescence microscopy [
13,
14]. PDGFRβ, FAP, FSP1 and αSMA were selected as markers for CAF subtypes. During evaluation of the markers, it became clear that FSP1 showed a different staining pattern compared to the other fibroblast markers. Based on the preliminary data, the fibroblast markers were divided into two panels: the first panel included the markers FAP, PDGFRβ, αSMA and CD31. The panel was called ECM-related fibroblast markers as it identified activated fibroblast involved in creating and remodeling the ECM. The second panel consisted of FSP1, CD45 and αSMA and was called immune-related fibroblast markers as it included CD45 staining for immune cells. After quantifying the markers in an automated manner using the Vectra microscope, the staining results were compared between stroma-low and stroma-high tumors on the one hand and between the tumor center versus the most invasive part of the tumor on the other hand.
Methods
Patient material
The LUMC cohort consisted of 71 CRC tissues of patients diagnosed with TNM stage I – III who underwent surgery at the Leiden University Medical Center (LUMC) between 1991 and 2005. The patients were not treated with (neo-)adjuvant therapy. This cohort was previously analyzed as part of a larger cohort [
15]. Formalin fixed, paraffin-embedded (FFPE) whole sections of the tumors were not available for immunofluorescence staining.
A second cohort was therefore used to perform immunofluorescence staining on FFPE whole sections and consisted of 32 patients diagnosed with stage III colon cancer who underwent surgery at the LUMC. Stage III tumors were selected as these tumors have an increased proportion of stroma-high tumors with more CAFs compared to lower stages tumors. None of the patients received preoperative anticancer therapy. Detailed patient characteristics of both cohorts can be found in Additional file
1: Table S1.
Transcriptomic analysis
RNA was previously isolated from fresh frozen tissue, hybridized to a customized Agendia oligonucleotide array and normalized as described elsewhere [
15].
Immunofluorescence staining
The antibodies were first optimized using single fluorescence staining (Table
1). Various FAP antibodies (AF3715 R&D, ab53066, ab28244 Abcam) were evaluated (data not shown) due to inconsistency with the literature and the selected antibody was confirmed by immunohistochemistry and western blot (Additional file
1: Figure S3). Immunofluorescence staining was performed on two 5 μm histological tissue sections of each tumor. After deparaffinization and rehydration, the sections underwent antigen retrieval by heating during 10 min at 95 °C in pH high Target Retrieval Solution (Dako). Unspecific protein binding sites were blocked with protein block (Dako) for 15 min. Sections were incubated overnight with the primary antibodies at specific dilutions described in Table
1. The following day, appropriate AlexaFluor secondary antibodies (ThermoFisher Scientific, 1:200) were applied to the sections for one hour. Sections were counterstained with DAPI (2μg/ml, Sigma-Aldrich) and mounted in ProLong Gold Antifade Mountant (ThermoFisher Scientific). The immunohistochemical staining was performed according to a similar protocol. Following deparaffinization and rehydration, sections were incubated with 0.3% hydrogen peroxide solution (Millipore) to block endogenous peroxidase. Then, after incubation with the primary antibody FAP (AF3715 R&D, 1:400), the sections were incubated with a HRP-labelled secondary antibody donkey anti-sheep (Invitrogen, A16041), then they were developed with DAB chromogen (Dako) and finally the sections were counterstained with hematoxylin.
Table 1
Characteristics of antibodies
Extracellular matrix-related fibroblast markers |
DAPI | | | Sigma-Aldrich | 2μg/ml | | |
PDGFR-β | | 3169S | Cell Signaling | 1:50 | Rabbit | AlexaFluor 488 |
FAP | | AF3715 | R&D Systems | 1:20 | Sheep | AlexaFluor 546 |
CD31 | | JC70A | Dako | 1:1000 | Mouse IgG1 | AlexaFluor 594 |
α-SMA | | 1A4 | Dako | 1:100 | Mouse IgG2a | AlexaFluor 680 |
Immune-related fibroblast markers |
DAPI | | | Sigma-Aldrich | 2μg/ml | | |
CD45 | | PD7/2 + 2B11 | Dako | 1:1000 | Mouse IgG1 | AlexaFluor 514 |
FSP1 | | D9F9F | Cell Signaling | 1:1000 | Rabbit | AlexaFluor 594 |
α-SMA | | 1A4 | Dako | 1:100 | Mouse IgG2a | AlexaFluor 680 |
Microscopical analysis
Spectral library and background correction
For each antibody panel, a single-stained section was prepared for every marker (including DAPI) to build a library containing the emitting spectral peak of each fluorophore in InForm (PerkinElmer, 2.2.1) (Additional file
1: Figure S1 A and B). Unstained colon cancer tissues were used as negative controls to perform background correction.
Multispectral imaging and spectral unmixing
Multiplexed stained sections were imaged using the VECTRA 3.0 multispectral sections imaging system (PerkinElmer, 3.0.4). Using PhenoChart software (PerkinElmer, 1.0.4), a maximum of 8 multispectral imaging fields of 334 × 250 μm (40x field) were selected in the tumor center and at the invasive part (Additional file
1: Figure S2B). The selection of the 40x fields was based on the annotations of a pathologist on H&E stained sections. Filter cubes used for multispectral imaging were DAPI (440–680 nm), FITC (520 – 680 nm), Cy3 (570 – 690 nm), Texas Red (580 – 700 nm) and Cy5 (670 – 720 nm). Next, spectral unmixing was performed on the extracted profiles using the spectral library by InForm (PerkinElmer, 2.2.1). Following background correction using two negative control sections, the unmixed images were quantified for pixel count or phenotype (see below).
Quantification of pixel count – ECM-related fibroblast markers
Following background correction, the absolute intensity of pixel counts per marker was determined in the spectrally unmixed images. An active learning algorithm was developed to segment the tissue (“Stroma”, “Tumor” and “Empty”) based on 40x fields of tumors originating from different patients (Additional file
1: Figure S2C). The total number of pixels present in the stroma was used to normalize the data.
Regarding the ECM-related fibroblast panel, 29 colon tumors and 392 40x fields were analyzed. The tumor center was analyzed in 27 colon tumors and the invasive part was analyzed in 25 tumors.
Quantification of cell count – immune-related fibroblast markers
A second learning algorithm was developed to count the number of DAPI-stained cells with different phenotypes. The algorithm was prepared as follows: background correction was performed, tissue segmentation (“Stroma”, “Tumor” and “Empty”) was performed on the 40x fields and cell segmentation was assessed on the DAPI-stained cells. The algorithm was then trained on 40x fields originating from various patients (Additional file
1: Figure S2C). Cells were phenotyped and categorized into one of the following classes CD45
+, FSP1
+, CD45
+/FSP1
+ and CD45
−/FSP1
− (Additional file
1: Figure S2D).
In the immune-related fibroblast panel, 27 colon tumors and 398 40x fields were analyzed. 25 tumors were analyzed in the tumor center and 26 at the invasive part. Only cells present in the stroma, assessed based on the tissue segmentation, were included in the analysis. The spindle-like shape of fibroblasts identified with the ECM-related fibroblast markers did not allow automated counting of the cells by nuclei. These markers were therefore quantified by pixel count while, for the immune-related fibroblast markers, the number of cells were counted using DAPI. The pixels or the cell numbers were averaged according to their location (tumor center versus invasive part) in each tumor.
Cell culture, lysis and western blot
The human colon fibroblast cell line CCD-18Co (ATCC) was cultured at low passage in EMEM medium with 10% fetal calf serum (FCS). The fibrosarcoma cell line HT1080 and the FAP-transfected HT1080 (a kind gift of R. Kontermann and O. Seifert, University of Stuttgart, Germany) were cultured in DMEM medium supplemented with 10% FCS. The cell lines were mycoplasma tested before use. 1.5 × 10^5 cells of CCD-18Co fibroblasts were seeded in 6 well-plates. The following day, the CCD-18Co cells were stimulated with recombinant human TGFβ1 (5 ng/ml, HEK293 derived, Peprotech) in serum free medium for three days and compared with non-stimulated cells. HT1080 and FAP-transfected HT1080 were used as negative and positive controls, respectively. Both fibroblasts and HT1080 cells were lysed with RIPA buffer (150 mM NaCl, 1% Triton X-100, 0.5% SDS, 50 mM Tris) and a western blot was performed.
Proteins were separated using electrophoresis on a denaturing 10% polyacrylamide gel and transferred to a polyvinylidene difluoride (PVDF) membrane using TurboBlot (Biorad). The primary antibodies against FAP (1:10 000 for HT1080; 1:2500 for CCD-18Co, AF3715 R&D) and beta-actin (1:2000, DM1A Cell Signaling) were applied to the membrane overnight at 4 °C. The membrane was incubated for 1 h with HRP-labeled secondary antibodies anti-sheep HAF016 (1:3000, R&D Systems) or anti-mouse (1:2000, Cell Signaling). The membrane was developed using enhanced chemiluminiscence and exposed to hyperfilm (GE Healthcare Life Sciences).
Tumor-stroma ratio
The primary tumors of both cohorts were scored for TSR on 5 μm H&E stained tissue sections as described previously [
2,
5]. The tissue samples selected were those defined as the most invasive part of the primary tumors as used by the pathologist to determine the T-status. To determine the TSR, the region with the highest stroma was selected using an 2.5x or 5x objective. A microscopy field was scored where tumor cells were present at all borders of the image field (north, south, east, west) of the 10x objective. Scoring percentages were given in 10-fold percentage per image field and the field with the lowest percentage determined the final score. The TSR of the LUMC cohort was scored previously [
16] and one investigator (T.P.S.) estimated the stromal percentage of the stage III cohort in a blinded manner.
Statistical analyses and data analyses
Statistical analyses were performed using IBM SPSS Statistics software (version 23) and the figures were made using GraphPad Prism 7 (version 7.02). A normality test was performed in the transcriptomic data and the mean of each gene in the stroma-low and stroma-high groups was compared using multiple independent t-tests followed by multiple testing correction using False Discovery Rate (FDR) (q = 0.05). Regarding the immunofluorescence data, the pixel count (ECM-related fibroblast markers) and the phenotyping (immune-related fibroblast markers) of each 40x field were normalized by the total number of stromal pixels or the total number of stromal cells, respectively. The normalized data was then averaged by location (tumor center or invasive part) and transformed using the following formula log(p/(1-p)). It was not possible to score all tumors at both the tumor center and the invasive part because of poor quality of the tissue or because the tumor center or invasive part was not present on the tissue (Additional file
2: Supplementary data). Independent student’s t-tests were therefore performed to compare the averaged markers between the location as well as between the stroma-low and stroma-high groups. The mean (M) and the standard error (SE) were reported for each t-test.
P values lower than 0.05 were considered significant.
Discussion
In the transcriptomic data, almost all CAF markers were higher expressed in stroma-high tumors compared to stroma-low tumors. Of the markers investigated, FAP was the only marker that showed an increased expression in stroma-high tumors using immunofluorescence. FAP expression was also increased at the invasive part compared to the tumor center.
Previous research showed that FAP
+ CAFs contribute to the invasive behavior of cancer cells [
17]. FAP is a cell surface protease expressed predominantly by CAFs. The protein is able to suppress anti-tumor immune response [
18]. FAP is involved in ECM remodeling with collagen being a key substrate, herewith facilitating tumor migration [
19]. The ability of FAP to affect cancer cell behavior might explain why the present study found an increased expression of FAP at the invasive part and in stroma-high tumors, known to be more aggressive than stroma-low tumors. Furthermore, FAP expression has shown prognostic significance in colon cancer [
20]. It would therefore be interesting to investigate the prognostic value of FAP expression using the automated method of this study. Combining the quantity of the stroma as defined by the TSR with the activity level of CAFs using FAP may add prognostic value to the TSR.
Moreover, we found a decreased number of cells co-expressing FSP1/CD45 in stroma-high compared to stroma-low tumors. FSP1 stained round-shaped cells instead of stellate-shaped cells and a part of the FSP1
+ round cells co-expressed with CD45+ but not with αSMA cells (Table
2). The simplest explanation might be that these FSP1
+ CD45
+ cells are a subset of immune cells, most likely macrophages, as has been reported previously in different tissue types [
21‐
23]. The increased presence of this cell subtype in stroma-low tumors might correspond to the better prognosis associated with stroma-low tumors.
Table 2
Expression and co-expression in different location and cell types of fibroblast and stromal markers analyzed in this study
CD31 | Blood vessels | Endothelial cells |
αSMA | Cancer stroma Blood vessels | CAFs, Smooth muscle cells Myofibroblasts in healthy colon |
PDGFRβ | Cancer stroma Blood vessels | CAFs, myofibroblasts, pericytes |
FAP | Cancer stroma | CAFs |
CD45 | Immune cells | Immune cells, hematopoietic stem cells |
FSP1 | Fibroblasts | Quiescent fibroblasts, cancer cells |
PDGFRβ αSMA | Around CD31+ endothelial cells | Pericytes, smooth muscle cells? |
Cancer stroma | CAFs |
FAP αSMA | Cancer stroma | CAFs |
PDGFRβ αSMA FAP | Cancer stroma | CAFs |
CD45 FSP1 | Cancer stroma | Macrophages / monocytes / fibroblast precursors or quiescent fibroblasts (?) |
The TSR assessed on H&E tissue sections was associated with the transcriptomic expression of fibroblast and CAF markers. However, the transcriptomic results were not exactly in line with the immunofluorescence data, which can be explained by the fact that transcriptomic data was not normalized by the amount of stroma in contrast to the immunofluorescence data. Consequently, an increased expression of a marker in the transcriptomic data can be attributed to either an increased gene expression level or to an increased presence of specific cells (or both). The transcriptomic and immunofluorescence methods should therefore be considered as complementary.
The main limitation of this study was that the algorithms made to quantify the markers still need further optimization. For instance, regarding the ECM-related fibroblast panel, no tumor stroma segmentation step could be implemented in the initial algorithm used for pixel count. An extra algorithm had to be made to normalize the pixel count which generally leads to inaccuracy. The phenotyping in the immune-related fibroblast algorithm was not optimal and underestimated the number of CD45+ cells, which might have influenced the results. The reported CD45+ cells in the tumor should be considered as relative numbers rather than absolute numbers. The InForm software to develop the algorithms should therefore be further optimized.
The question remains whether the markers used in the present study define subtypes of CAFs performing unique functions. Different functional subtypes of fibroblasts have been suggested ranging from tumor-promoting to tumor-suppressing depending on the context [
13,
24]. Novel promising technologies such as the mass cytometry CyTOF allow combining a large number of markers and allow isolating different phenotypic fibroblast subsets in order to perform functional experiments.
Conclusions
Considering that the different CAF markers tested stained different cells, this suggests that a single fibroblast marker cannot recapitulate the heterogeneous composition of CAFs in the tumor stroma. Gene expression data showed an increased expression of CAF markers in stroma-high compared to stroma-low tumors. Histologically, the expression of FAP+ fibroblasts and the number of FSP1+/CD45+ cells were dependent on the spatial distribution. Stroma-high tumors showed a decreased number of FSP1+/CD45+ cells and an increased expression of FAP compared to stroma-low tumors. The prognostic relevance of FAP+ CAFs should be further explored as a stromal marker alone or to expand the TSR. Therapeutically targeting FAP+ CAFs may eventually promote anti-tumor growth in CRCs with high stromal content.
Acknowledgments
We would like to thank G.W. van Pelt for providing the stage III cohort database, prof. dr. H. Morreau for providing the gene expression data of the LUMC cohort, dr. N.F. de Miranda for setting up de VECTRA at the institute, S. Bhairosing for performing the immunohistochemistry staining for FAP, dr. A.F. Sarasqueta for annotating the H&E sections and D. Krijgsman for critical revision of the manuscript.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (
http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.