Rapid and label-free detection of gastrointestinal stromal tumor via a combination of two-photon microscopy and imaging analysis

This study was approved by the Institutional Review Board of Fujian Medical University Union Hospital, and all patients signed an informed consent before the research. In this work, 30 fresh samples were collected including 15 gastrointestinal stromal tumors and 15 normal tissues. Moreover, 15 gastric adenocarcinomas were also collected for the sake of comparison. After removal by surgeons, every fresh sample without any processing (unfixed, unstained) was sent to pathology department immediately, and we cut two serial slices with 10 μm thickness by a cryostat microtome (Thermo Scientific CryoStar NX50, USA), where we used one section for two-photon imaging and stained another section with hematoxylin and eosin (H&E) to confirm experimental results. All the digital images of H&E-stained sections were obtained by a commercial whole slide scanner (VM1000, Motic, China).

We used a commercial imaging instrument (LSM 880, Zeiss, Germany), which was combined with a mode-locked femtosecond Ti: sapphire laser (Chameleon Ultra, Coherent, USA), for obtaining two-photon images. In this work, a Plan-Apochromat 10× objective (NA = 0.45, Zeiss) was firstly used for whole-slide imaging and then a Plan-Apochromat 63× oil immersion objective (NA = 1.4, Zeiss) was selected to obtain high-resolution two-photon images of the regions of interest (ROIs). TPAF and SHG signals were detected in two-track channel mode under linearly polarized 810 nm excitation beam: one channel covered the wavelength range of 430–759 nm was used for collecting TPAF signal (color-coded with red) via a 32-channel GaAsP PMT array, while another channel covered the wavelength range of 394–416 nm was used for the collection of the backward SHG signal (color-coded with green) by a GaAsP photomultiplier tube (PMT). This imaging system takes ∼1.8 µs for collecting every pixel, and large-area images are obtained by a ZEN imaging software that could automatically assemble an array of two-photon images with 512*512 pixels.

An automated image analysis method which was performed using MATLAB 2016b was developed to extract the morphological features of collagen fibers from SHG images. As shown in Fig. 1, we selected the region of interest (ROI) of 1500*1500 pixels for quantitative analysis. For each sample, SHG images as the input images were first filtered by the Frangi filter to enhance the collagen fiber structures from noisy background, and then the enhanced images were segmented into collagen fibers and background by a segmentation algorithm based on Gaussian mixture models [13, 14]. The morphological closing and hole-filling were performed to smooth the binary mask of the collagen fibers, and any segment with less than 5 pixels was removed. In this work, a well-established fiber network extraction algorithm called “Fire” [15] was used to track and identify all potential collagen fibers reflected in SHG images, and then a series of ordered vertex sequences [16] were constructed to calibrate the skeletons of collagen fibers (any common vertices of different sequences were defined as cross-link points between collagen fibers). As a structured representation of collagen fiber networks, the ordered vertex sequences were used to quantify collagen morphological features including collagen fiber area (a.u.), density (fiber number per mm²), length (µm), width (µm), orientation (a.u.), straightness (a.u.), cross-link space (µm), cross-link density (a.u.) as previously described [13]. Quantitative results were presented using means with standard deviations (SD).

GIST is the most common malignant mesenchymal neoplasm, and accurate diagnosis of this lesion is currently based on histopathologic examination of endoscopic biopsy specimens. However, this procedure is costly, labor-consuming and time-costing. Numerous studies have shown that two-photon imaging method is rapid, sensitive, reproducible, and especially SHG imaging offers a new way to guide region of interest selection for quantifying collagen fibers in different biological tissues [17, 18]. In this work, TPAF and SHG imaging were combined to image healthy gastric tissue, GIST and gastric adenocarcinoma ex vivo samples to perform a morphological characterization. As shown in Fig. 2, two-photon images clearly present the tissue architecture details of GIST. Specifically, SHG imaging (Fig. 2 A, E) allows users to visualize collagen distribution in tumor microenvironment, and TPAF imaging (Fig. 2B) shows that tumor cells appear with dark nuclei (white arrows in Fig. 2 F), and composite image (Fig. 2 C) could let users visually observe the spatial distribution of tissue components such as the cellular environment within the collagen matrix. All these features correspond to the H&E-stained image (Fig. 2D). Hence, this technique may provide new opportunities for revealing the relationship between tumor cell behavior and the role of collagen fibers in regulating such cell behaviors.

Generally speaking, tumor invasion will cause desmoplastic reaction, and some studies demonstrated that histological categorisation of the desmoplastic reaction is significantly associated with the prognosis of colorectal cancer patients with or without preoperative chemoradiotherapy [19, 20]. The presence of collagen fibers is therefore considered to influence the development of cancer. Our experimental results show that two-photon imaging has the ability to accurately and quickly monitor desmoplastic reaction induced by GIST, and even could directly discern different levels of response (Fig. 3 A, D, G). Specifically, for the type of mild reaction (Fig. 3B, C), collagen fibers are sparse, disordered, and fragmented in the tumor microenvironment; by contrast, there are abundant and directionally distributed collagen fibers for the severe desmoplastic reaction (Fig. 3 H, I). On the basis of these two reactions, we could recognize the moderate response (Fig. 3E, F) in which some collagen fibers are chaotic (left side of the red line in Fig. 3E) and some are orderly aligned (right side of the red line in Fig. 3E). Although it is not clear whether desmoplasia (tumor fibrosis) is a direct or an indirect indicator of GIST, the results of our research suggest that SHG imaging is of value in identifying different kinds of desmoplastic reaction.

For the comparison and analysis, we further investigated the structural characteristics of different samples, finding differences between gastric adenocarcinoma and GIST. The histological structure of stomach is mucosa, submucosa, muscularis and serosa from the outside to the inside. Figure 4 presents two-photon images of gastric mucosa and submucosa with the invasion of adenocarcinoma. Imaging data reveals that GIST and gastric adenocarcinoma have completely different tissue architectural features. The muscularis mucosae (white arrows in Fig. 4B) separate the mucosa from submucosa, and in particular, adenocarcinoma cells surrounded by collagen fibers (Fig. 4 A) appear with nest-like architecture and tightly pack together (blue arrows in Fig. 4D). These tumor cells are totally different from GIST cells which are widely spread in the process of infiltration. Interestingly enough, it is found that adenocarcinoma cells with nest-like structure (cyan arrows in Fig. 4E) have infiltrated into the submucosal layer and were surrounded by collagen fibers. These structural characteristics are in excellent agreement with the digital image of H&E-stained adjacent tissue section (Fig. 4 C).

In Fig. 5, it is clear that the collagen content within gastric muscularis propria is altered with the tumor invasion in the SHG images (Fig. 5 A, D). A lot of collagen fibers (yellow arrows in Fig. 5D) emerge because of desmoplastic response, and thus the detection of SHG signal from collagen fibers could provide us an approach to differentiate normal from abnormal tissues. Cancer progression is often associated with the destruction of normal tissues. TPAF imaging has the power to make us directly visualize the infiltrating adenocarcinomas and broken muscular tissues (cyan arrows in Fig. 5E). Two-photon images (Fig. 5B) demonstrate that normal muscularis has been damaged, becomes fragmented, and are gradually replaced by gland-like tumors (white arrows in Fig. 5E). These tissue architecture details readily correlate with the H&E-stained image (Fig. 5 C).

As displayed in Fig. 6 A, SHG image reveals that gastric serosa mainly composes of collagen fibers. However, there are many tumors with nest-like structure (white arrows in Fig. 6D) too in the serosal layer by the detection of TPAF signal, indicating serosa invasion in gastric cancer. The H&E-stained image of adjacent section (Fig. 6 C) is used for confirming these experimental results (Fig. 6B) obtained by two-photon microscopy. It is surprising that the proliferation of elastic fibers turns up in residual muscular tissues (Fig. 6E). The abnormal elastic fibers are fractured or gather together (cyan arrows in Fig. 6E), which are different from normal elastic fibers with a long rope-like morphology. Changes in tumor microenvironment may accompany disease progression, and maybe TPAF imaging will become an alternative tool for monitoring unusual conditions induced by tumor invasion without labeling. It seems that the tissue structures of GIST are entirely different from those of gastric adenocarcinoma based on the imaging results, and therefore could be identified by two-photon imaging.

Morphological characteristics alone are not sufficient to precisely identify GIST as they do not provide quantitative information about the changes within the extracellular matrix. It has been suggested that there is a relation between cancer risk and collagen alterations [17, 21]. Hence, in this work, we analyzed and extracted 8 morphological features of collagen fibers in the tumor microenvironment of GIST by automatic image processing strategy. GIST often originates from the muscle layer, and therefore we also extracted the 8 collagen characteristics from normal muscularis propria for comparison. The means with standard deviations and data distributions of these collagen features, including fiber area, density, length, width, orientation, straightness, cross-link space and cross-link density, are presented in Fig. 7 A and 7B respectively. Quantitative analysis shows obvious difference in collagen fiber area, density, and cross-link density between the normal muscularis (NM) and GIST. Collagen structure and organization in the microenvironment are potentially key determinants of tumor cell behavior, and the three parameters may be treated as indicators to distinguish healthy from diseased tissues. In summary, our method of two-photon imaging coupled with automated image analysis is sensitive, robust, and offers a new opportunity to quantify collagen fibers in different tissue samples.