Collagen reorganization at the tumor-stromal interface facilitates local invasion

Mouse mammary tissues and tumors

This study was approved by the institutional animal use and care committee and meets N.I.H. guidelines for animal welfare. To study non-tumor bearing mammary glands, tissue was obtained from B6129SF2/J mice or Col1a1^tmJae mice (The Jackson Laboratory, Bar Harbor, ME, USA). To study tumor-stromal interactions in intact tissue, two mouse breast tumor models were utilized: MMTV-Wnt-1 (colony founder mice provided by Dr. Caroline Alexander, University of Wisconsin, Madison, WI, USA) and MMTV-polyoma middle-T (abbreviated PyVT following the Jackson Laboratory title but is also commonly abbreviated as PyMT or PyV-MT; colony founder mice originally obtained from Jackson Laboratory were provided by Dr. Amy Moser, University of Wisconsin, Madison, WI, USA).

Tumor explants and collagen gel culture

To study tumor-mediated collagen reorganization and tumor cell invasion in vitro, tumor explants were obtained and cultured in a manner similar to a previous report by Friedl and co-workers [22]. Small pieces of tumor were harvested from the central region of palpable PyVT tumors (that were confirmed by histology) with a 3 mm biopsy punch and were cultured in type I collagen gels. Following removal, tumors were rinsed in DMEM containing penicillin/streptomycin/fungizone solution (Cellgro, Herndon, VA). A single tumor explant was then cultured within a 2.0 mg/ml collagen gel (8.0 mg/ml rat-tail collagen solution (BD Biosciences, San Jose, CA) neutralized with 100 mM HEPES in 2 × PBS). Following Gel polymerization for 1 hr, the tumor explant containing collagen gels were released from the culture dish and floated in DMEM containing penicillin/streptomycin solution supplemented with 10% heat inactivated FBS. Imaging was performed on live (non-fixed) cells in intact three-dimensional collagen gels.

Multiphoton microscopy and second harmonic generation

Multiphoton excitation (MPE) with MPLSM allows imaging of endogenous fluorophores from deep inside live biological tissues with the fluorescence excitation primarily restricted to the plane of focus due to a quadratic dependence on the laser light intensity and a low probability of multiple low-energy photons being absorbed outside the focal plane [16, 17, 23]. For the case of a pulsed laser excitation, the time averaged fluorescent intensity is a function of the molecular cross-section δ ₂(λ) and the square of the laser intensity, I(t)², and can be expressed as [23, 24]:

$〈{\text{I}}_{\text{f},\text{p}}(\text{t})〉={\delta }_2\frac{{\text{P}}_{\text{ave}}^2}{{\tau }_{\text{p}}{\text{f}}_{\text{p}}}{\left[\pi \frac{{(\text{NA})}^2}{\text{hc}\lambda }\right]}^2\left(1\right)$

In Equation 1, δ ₂ is defined as a molecular cross section that represents the dependence for the probability of two-photon excitation on the square of photon density, P is the laser power, τ _p is the laser pulse width, f _p is the laser repetition rate, NA is the numerical aperture, h is Planck's constant, c is the speed of light, and λ is the wavelength.

In contrast to multiphoton excitation, which obeys the fundamental physical relationship of energy loss following excitation, SHG is a conserved polarization process that follows the relationship [21, 25, 26]:

P = χ ⁽¹⁾ * E + χ ⁽²⁾ * E * E + χ ⁽³⁾ * E * E * E + …     (2)

where the polarization (P) and electric field (E) are vectors, and the nonlinear susceptibilities, χ⁽ⁱ⁾, are tensors. Therefore, SHG arises from the laser field suffering a conserved nonlinear, second-order, polarization when passing through non-centrosymmetric ordered structures that is described by term 2 of Equation 2.

For MPE and SHG imaging of live unfixed, intact (not sectioned), non-stained glands, and tumor explants within collagen gels, as well as hematoxylin and eosin stained histology slides, we used an optical workstation [27] that was constructed around a Nikon Eclipse TE300. For live tissue imaging, twenty mammary tissues including nine from the Col1a1^tmJae strain (three each of wild type, heterozygous and homozygous), and tumors from Wnt-1 (n = 10, plus wild type controls) and PyVT (n = 20) mice were harvested and live tissue maintained in buffered media at 37°C. All tissues were imaged immediately following tissue harvest and a Ti:sapphire laser (Spectra-Physics-Millennium/Tsunami, Mountain View CA) excitation source producing around 100 fs pulse widths and tuned to 890–900 nm was utilized to generate both multiphoton excitation (cellular autofluorescence from FAD) and SHG. The beam was focused onto the sample with either a Nikon 40× Plan Fluor oil-immersion lens (N.A. = 1.4) or a Nikon 60× Plan Apo water-immersion lens (N.A. = 1.2). All SHG imaging was detected from the back-scattered SHG signal [26], and the presence of collagen confirmed in our tissues using fluorescence lifetime imaging microscopy, or FLIM [12], on the same optical workstation, as the SHG from collagen has no lifetime. Additionally, due to the fundamental differences between MPE and SHG signals, filtering can separate the emission signals. Using a 464 nm (cut-on) long pass filter, MPE was discriminated from the total emission while a 445 nm narrow band pass filters was used to separate SHG (filters from TFI Technologies, Greenfield, MA). For 3D imaging in intact tissues, 2D (x-y) images were acquired at various serial depths (z) into the samples.

Acquisition was performed with WiscScan [28] a software acquisition package developed at LOCI (Laboratory for Optical and Computational Instrumentation, University of Wisconsin, Madison, WI, USA). Image analysis for combined MPE-SHG was performed with ImageJ [29] and VisBio [30] software. Using ImageJ, differences in collagen density were quantified by measuring the area of collagen signal following density slicing from a constant threshold, and local and mean intensity were measured within these normalized areas. For TACS-1 image analysis additional surface rendering plug-ins for ImageJ were utilized. For TACS-2 and -3, ImageJ was used to quantify the collagen fiber angle relative to the tumor. The tumor boundary was defined and the angle relative to the tangent of tumor boundary was measured every 10 microns.

Histology and electron microscopy

For histology, formalin-fixed paraffin-embedded samples from eight B6129SF2/J mice and eight Col1a1^tmJae mice were sectioned and stained for hematoxylin and eosin, trichrome, and picrosirius red using standard techniques. Additionally, all tissues imaged with MPLSM were subsequently fixed and processed for histology to confirm the presence of tumors and characterize the tumor morphology. Sample preparation for scanning and transmission electron microscopy (SEM and TEM, respectively) was performed by fixing whole mammary glands in 2.5% formaldehyde/2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer for 1 hr at room temperature (RT), after which sample were placed in fresh fixative overnight at 4°C. Samples were then washed in cacodylate buffer and postfixed in 1.5% osmium tetroxide at RT for 1.5 hrs. Samples (SEM n = 8 glands and TEM n = 6 glands) from B6129 mice were again washed in buffered solution and dehydrated, fractured, critical point dried, sputter coated, and imaged with SEM, or stained, dehydrated and cleared, embedded, and then sectioned for imaging with TEM [31].

The normal mammary gland

Examination of the ECM surrounding mammary epithelial cells revealed that the majority of the stroma is fibrillar collagen, which was seen surrounding epithelial cells (Figure 1). Figure 1A, in which picrosirius red (a selective staining agent for collagen) was used indicates the presence of collagen around the mammary ductal structure (Figure 1A,B). Observations at high magnification with SEM showed collagen fibers wrapping around the ductal structure in an organized manner surrounding the epithelial cells (Figure 1C), in a fashion similar to that seen in Figure 1A. Additionally, collagen can be seen wrapping individual cells in multiple directions (Figure 1C–E), suggesting a role for collagen as a containing and anchoring structure. Of further importance, higher magnification images with both SEM (Figure 1F,G) and TEM (Figure 1H) confirmed that the collagen fibers are composed of fibrillar collagen (Figure 1F; as indicated by their rod-like structure and presence of the ~67 nm banding pattern [32]) that are adjacent to the epithelial cells (Figure 1G,H). These data, in combination with published works indicating an important interaction between the stroma and the epithelial cell (see [4, 5, 33]), could suggest that these fibers apply physical restraint and mechanical signals across a thin basement membrane (<200 nm), or that the thin mesh-like basement membrane may contain some small gaps or membrane spanning cell processes and therefore may not completely isolate the epithelial cell; resulting in micro-regions of direct collagen-epithelial cell interactions [34]; or, most likely, a combination of both scenarios.

Although histology and electron microscopy provide valuable and detailed information regarding the composition and morphology of the epithelial-stromal interaction, these techniques are destructive to the sample and the capability for three-dimensional imaging is limited. Therefore, we have sought non-destructive imaging techniques (e.g. MPE/SHG) that allow imaging in four dimensions (x, y, z, and time). Combined MPE-SHG produced clear images of collagen (Additional file 1) as well as endogenous fluorophores, such as NAD(H) and FAD (Additional file 1, part C), in intact non-treated tissues [13, 15, 21]. Application of MPE/SHG to intact, non-fixed, non-stained mammary gland captured the morphology of the fibrous stroma seen in Figure 1. For example, images captured "above" the gland (Figure 2A: a and 2A: g) clearly showed the presence of wavy (crimped) collagen as well as the presence of taut (straight) fibril bundles (Figure 2A: a,e,f,g). Taut fibers showed a periodicity of ~250 nm (Figure 2A: c), consistent with previous reports for SHG resolution [14], which may represent an approximately fourfold super-periodicity of the basic ~67 nm banding pattern of collagen fibrils noted with SEM (Figure 2A: d). Additionally, collagen was observed wrapping around the epithelial duct (consistent with Figure 1A,C), as well as radiating away from the duct (Figure 2A: b). Furthermore, in addition to detecting unique features from normal mammary glands, examination of the epithelial-stromal interaction in Wnt-1 mice [35], which possessed ductal hyperplasia (Figure 2B: a–c), revealed abnormal mammary duct development with irregular collagen organization (Figure 2B: b,c). The ability to clearly discern these pathologic changes in tissues that were not fixed, sectioned, or stained (Figure 2B: d–f) validates the application of MPE/SHG imaging to detect normal mammary gland and collagen structure as well as abnormalities in mammary ducts and collagen distribution in situ (Figure 2B: f).

Detecting dense mammary tissue

High breast tissue density (due to increased collagen; [36]) is one of the single largest risk factors for developing breast cancer [37], yet the molecular mechanisms behind this high risk are not known. One reason for this deficit has been the lack of adequate animal model systems for studying the effects of increased collagen density in vivo. As such, we sought to identify an animal model system possessing collagen-dense breast tissue. Analysis of the col1a1^tmJae mouse model [38] revealed such a system. These mice possess a type I collagen mutation in the α1(I) chain, making them resistant to human collagenase, resulting in fibrosis of the skin and uterus due to excessive collagen accumulation [38]. However, increases in collagen deposition in the mammary glands of these mice have not been previously described. Analysis with standard histology clearly revealed increased collagen surrounding the mammary ducts in both homozygous and heterozygous female mice regardless of parous status (Figure 3A vs 3C; Additional file 2). By utilizing histology and imaging hematoxylin and eosin (H&E) stained sections with MPLSM (SHG signal did not exist after formalin fixation; data not shown), the presence of increased collagen was confirmed (Figure 3B vs 3D), as well as an apparent hyperplasia in Col1a1 mice (Figure 3C,D; Additional file 2). Interestingly, this hyperplasia was associated with invasive-looking epithelial cells that resemble an epithelial-mesenchymal transition at the ductal end (Figure 3C,D). Therefore, in an approach opposite to that taken by Jain and co-workers [39] where a decrease in collagen in vivo was detected with SHG, we detected increased collagen in live intact mammary glands. Analysis of glands from col1a1 heterozygous and homozygous mice, and associated wild type littermates, noticeably revealed increased collagen surrounding epithelial ducts from transgenic mice (Figure 3E–H), demonstrating our ability to detect differences in collagen density. Furthermore, collagen was locally dense adjacent to (wrapped around) the epithelium, as well as increased in the space extending from the duct resulting in a 2.53- (± 0.24) and 2.79- (± 0.34) (mean ± SEM) fold increase in the area of collagen signal, as well as increased signal intensity around the duct, for heterozygous and homozygous col1a1^tmJae mice relative to wild type animals, respectively. Therefore, the col1a1 mouse model appears to be a promising candidate for studying the effects of increased collagen density in normal as well as transformed epithelial cells as appropriately crossed heterozygous col1a1 mice are capable of forming tumors without inhibition from the collagenase resistant matrix (P.P.P. and P.J.K unpublished observations). Moreover, we could discern changes in density and obtain images from deep (maximum Depth: 440 nm) within intact mammary tissue (Figure 3H; Additional file 4) without fixing, sectioning, or staining the tissues, suggesting that this approach could have wide application in live animal or tissue models for understanding stromal changes associated with pathologic conditions.

Tumor-associated collagen signatures (TACS)

Tumor-associated collagen signature-1

Analysis of tumor bearing Wnt-1 mice, which progress through hyperplasia, mammary adenocarcinoma, and invasive (metastatic) ductal carcinoma, revealed multiple epithelial clusters, containing hemorrhagic regions, intermixed and surrounded by increased collagenous stroma (i.e. desmoplasia), consistent with previous reports [35]. Importantly, the presence of increased collagen allowed us to identify pre-palpable tumors (Figure 4A: a–c, Additional file 5; and confirmed with histology – data not shown) by the existence of what we are classifying as one of three tumor-associated collagen signatures (TACS). Namely, TACS-1: the presence of dense collagen (Figure 4A: a), indicated by increased signal intensity (see surface map in Figure 4A) at a region near the tumor (Figure 4A: a; Additional file 5) that served as a reliable hallmark for locating small tumor regions. Although increased collagen has been reported near tumors, the structure of such collagen has previously been largely unknown, and the additional focal localization shown in Figure 4A indicates increased local areas of high density within the globally increased collagen concentration surrounding tumors. However, it is unclear whether: (1) a pre-tumor dense region is present that serves to stimulate tumor formation, (2) the dense region of collagen is pulled into a grouped cluster through increased contraction of an epithelial tumor mass or motile cells at the tumor boundary, or (3) fibroblast activation results in increased local collagen deposition. Further analysis into this fundamental characteristic should provide useful information on tumor initiation and progression.

Tumor-associated collagen signatures-2 and -3

As the size of the tumor increased, we identified a second collagen signature, TACS-2: "taut" collagen fibers stretched around the tumor (Figure 4B). This collagen morphology likely arose from stretching of the stroma due to tumor growth, which may act to constrain portions of the tumor (i.e. compressive restraint) as well as provide a stretch induced tensile stress in expanded fibrils (and larger resistance to cell contraction in the stroma) that stimulates and activates fibroblasts. Evidence for tumor restraint in Wnt-t mice can be seen in Figure 4B, a–c, where collagen fibers are stretched around a relatively smooth tumor boundary as indicated by the fact that the fiber angle is primarily distributed tangentially (0° relative to the tumor boundary) along the tumor boundary (see histogram in Figure 4B).

In regions of tumor masses that are undergoing growth and invasion (Figure 4C: a–c), a third tumor-associated collagen signature, TACS-3, was identified: collagen fibers aligned normal to tumor boundary regions that display an irregular shape – indicative of local invasion through collective epithelial cell migration [22, 40]. This invasive tumor morphology was seen in regions of tumors where collagen fibers are primarily aligned in the direction of cell invasion (see histogram in Figure 4C, in which the angle of the collagen fibers relative to the tumor boundary distributes around 90°). Furthermore, at regions where TACS-3 is noted, we observed groups of cells advancing from the tumor boundary in a collective manner that appear to be undergoing collective invasion [22, 40, 41], as well as individual invading cells that may relate to single cell migration similar to what has been observed along collagen fibers in an in vivo xenograft model [9].

To further investigate the behavior of the third collagen signature in a more invasive and metastatic cancer model, we utilized the well established and characterized PyVT mouse model. This model bears resemblance to many aspects of human cancer, is reliably invasive and metastatic following defined, progressive, and reliable histological grades from hyperplasia to adenoma, and then to early and late carcinoma, and therefore provides a good model for studying human disease [42]. Tumor cell behavior and collagen morphology in this model were similar to observations from the Wnt-1 mouse tumors. For instance, tracking the tumor-stromal boundary around individual tumors (Figure 5A–D) first revealed regions of TACS-2 morphology, in which restraining collagen wrapped around a relatively smooth boundary (Figure 5A,B, and histogram: TACS-2). In regions of TACS-2, cells had a mostly random alignment with a subset of cells aligned in the same direction as the collagen (Figure 5A,B). The second region revealed a TACS-3 boundary where collagen has been primarily aligned perpendicular to the tumor (Figure 5C,D, and histogram: TACS-3). A reasonable hypothesis for this behavior is that collagen fiber realignment occurs through morphogenesis and contractile events from cells at the tumor boundary, to organize the matrix and to prepare for local invasion (see Figure 6 for TACS-2 to -3 transition). Evidence for this hypothesis is seen by the collagen structure associated with regions of invading cells in the Wnt-1 tumors (Figure 4C), PyVT tumors (Figure 5A–D; Figure 6), and MPLSM analysis of invasive regions in tumor histology sections (Additional file 3). In all these cases local tumor invasion occurred where collagen fibers are aligned radially from the tumor in the direction of tumor cell invasion (see TACS-3 histogram in Figure 5, and the change in stromal alignment correlated with the change to invasive cell morphology in Figure 6). Additionally, some invading cells were in direct contact with fibers (Figure 5E,F; Figure 6C,D). In certain cases, some matrix disorganization was also observable (see Figure 4C: c and Figure 5E) possibly indicating proteolytic cleavage of collagen [40, 43].

To further test the hypothesis that tumor cells realign the collagenous matrix to facilitate local invasion, we determined whether tumor cells could reorganize a random collagen matrix. Analysis of tumor explants within type-I collagen gels revealed radial collagen alignment at regions of tumor cell invasion into previously randomly aligned collagen gels (Figure 7), consistent with previous reports analyzing fixed tumor explants in 3D reconstituted matrices [22], and supporting information about invasion within intact live tissues (Figures 4, 5, 6). Following polymerization, circular collagen gels (disks) displayed a random orientation of collagen fibers (Figure 7A) without an outside force to initiate structural reorganization. Tumor explants cultured within such random collagen gels contracted and reorganized these collagen gels, resulting in various local outcomes with collagen wrapped around the explant at non-invading regions (Figure 7B). In contrast, at regions of invasion into the gels, collagen was reorganized to a radial alignment (Figure 7C) with direct contact between collagen fibers and invading cells (Figure 7C,D). This demonstrates that tumor cells can de novo reorganize a random collagen matrix to facilitate invasion, supporting the hypothesis that matrix reorganization at the tumor interface facilitates local invasion in vivo. Hence, our results suggest that in order to facilitate local invasion, cells at the tumor boundary contract and align collagen fibers, perhaps with the assistance of proteolytic cleavage to facilitate matrix reorganization, and then invade along aligned collagen structure to expand the tumor and later metastasize.