Lymphatic vessel remodeling and invasion in pancreatic cancer progression

2.1. Human pancreatic specimens

Collection and use of human tissues were approved by the Institutional Review Board of National Taiwan University Hospital. Normal human pancreases were obtained from the cadaveric donors. Specimens from three donors with normal HbA_1c, amylase, and lipase levels were used to establish the pancreatic lymphatic network. Pancreatic specimens with PanIN lesions were obtained from pancreatectomy for treatment of PDAC. Immediately after the operation, the dissected specimens were perfused with 1000 ml of saline through the splenic artery, which was cannulated with a Fr. 21-peripheral venous catheter, to remove the residual blood. Afterward, the tissues were fixed in 4% formaldehyde for 2 days and then washed in saline for 4 days at 4 °C [25]. Specimens were later sectioned to 350 μm in thickness by vibratome and transferred to 0.1% paraformaldehyde for preservation at 4 °C. Specimens from three PDAC patients (sex/age (years)/staging: male/77/T1N0, male/52/T2N0, and male/67/T3N0) were used to identify the peri-lesional lymphatic network.

2.2. Genetically engineered and syngeneic mouse models of pancreatic cancer

Normal pancreases and pancreases with PanIN lesions were harvested from the 14-week-old wild-type C57BL/6 (B6) mice (control), elastase-CreER;LSL-Kras^G12D mice (or EK mice; elastase-CreER × LSL-Kras^G12D mice) [18], and elastase-CreER;LSL-Kras^G12D;p53^+/− mice (or EKP mice; EK × B6.129S2-Trp53^tm1Tyj/J mice, Jackson Laboratory; stock number: 002101; this mouse line develops liver metastasis at the advanced stage). To examine the early-stage duct lesion formation, the EK mice were injected with tamoxifen at age 6 weeks (Sigma, St. Louis, MO, USA; 2 mg/injection, three injections in 1 week to induce Cre-mediated recombination) and then allowed for duct lesion development for the next 7 weeks. To examine the advanced duct lesion formation, the EK and EKP mice were injected with tamoxifen at age 6 weeks and followed by 3-week cerulein treatment (starting at age 7 weeks; Sigma, 0.25 mg/kg body weight, six injections per week) and 4-week tissue regrowth to induce large-scale development of PanIN lesions. B6 mice with 3-week cerulein treatment and 4-week tissue regrowth were used as the control. Overall, six normal and cerulein-treated B6 mice, six EK mice with early-stage duct lesions, and seven EK and EKP mice with advanced lesion formation were used to generate the representative images.

To establish the syngeneic mouse model of pancreatic tumor, we developed a PDAC cell line (clone Pan18) from the pancreatic tumors in the EKP mice (induced with tamoxifen and cerulein injections described above). In the process, the EKP tumors were dissected and treated with the digestion solution (10 mM HEPES, 1 mg/ml collagenase P, 1 U/ml Dispase, and 2.5 mM CaCl₂) to disassociate the tumor cells. The cells were then seeded in low density and cultured in DMEM with 4.5 g/L glucose, 100 U/ml penicillin and streptomycin, and 10% fetal bovine serum to monitor their growth. Clone Pan18 was later picked, preserved, and magnified on the basis of its aggressive proliferation. Next, the Pan18 cells were labeled via transduction with the lentivirus carrying the IRES-based dual expression cassette of luciferase (for in vivo monitoring/confirmation) and EGFP (for cell tracing) and then purified three times with the FACSAria II cell sorter (BD Biosciences, San Jose, CA, USA) to select the EGFP⁺ cells. Finally, to engraft the tumor cells, ~1 × 10³ EGFP⁺ Pan18 cells in Matrigel (40 μl; BD Biosciences) were injected to the B6 mouse pancreas (6–8 weeks of age) through an abdominal incision (control: injection of Matrigel only). Once the Matrigel solidified, the incision was closed with 0.1 mm (5/0) Safil absorbable surgical suture (B. Braun Medical, Barcelona, Spain). The development of the pancreatic tumors was monitored using the Xenogen IVIS Spectrum System (Caliper, Waltham, MA, USA) for 1 month before killing the animals for examination. Six Pan18-injected and four control mice were examined with 3-D histology to characterize the tumor cell-induced lymphangiogenesis and lymphatic vessel invasion.

Mouse pancreatic blood vessels were labeled with cardiac perfusion of the lectin-Alexa Fluor 488 conjugate (Invitrogen, Carlsbad, CA, USA) followed by 4% paraformaldehyde perfusion fixation [26]. Afterward, the pancreas was harvested and post-fixed in 4% paraformaldehyde solution for 40 min at 15 °C. Vibratome sections of the fixed pancreas were then prepared and the sections were transferred to 0.1% paraformaldehyde for preservation at 4 °C. The Institutional Animal Care and Use Committees at the Academia Sinica and the National Tsing Hua University approved all procedures with mice.

2.3. Tissue immunolabeling

The fixed human and mouse specimens were immersed in 2% Triton X-100 solution for 2 h at 15 °C for permeabilization before immunolabeling. Eight different primary antibodies were used to immunolabel the tissues following the protocol outlined below. For human tissue labeling, the antibodies used were mouse anti-podoplanin (clone D2-40, lymphatic endothelial marker; 916602, BioLegend, San Diego, CA, USA) [27,28], rabbit anti-PGP9.5 (neuroendocrine marker; 2932–1, Epitomics, Burlingame, CA, USA) [25], rabbit anti-glucagon (2810–1, Epitomics), and rabbit anti-α-SMA (stellate cell marker; ab5694, Abcam, Cambridge, MA, USA). For mouse tissue labeling, the antibodies used were rabbit anti-lymphatic vessel endothelial hyaluronan receptor 1 (Lyve1, lymphatic vessel endothelial marker; Lyve1 staining also labels a subpopulation of macrophages; ab14917, Abcam), rat-anti-substance P (sensory nerve marker, MAB356, Millipore, Billerica, MA, USA), rabbit anti-Ki-67 (cell proliferation marker; ab15580, Abcam), and mouse anti-nestin (cancer stem cell marker, MAB353, Millipore) [29] antibodies. Before applying the antibody, the tissue was rinsed in phosphate-buffered saline (PBS). This was followed by a blocking step, incubating the tissue with the blocking buffer (2% Triton X-100, 10% normal goat serum, and 0.02% sodium azide in PBS). The primary antibody was then diluted in the dilution buffer (1:100, 0.25% Triton X-100, 1% normal goat serum, and 0.02% sodium azide in PBS) to replace the blocking buffer and incubated for 1 day at 15 °C. Negative controls were prepared by omitting the primary antibody in the buffer.

Alexa Fluor 647 conjugated goat anti-rabbit secondary antibody was used in combination with Alexa Fluor 546 conjugated goat anti-mouse or anti-rat secondary antibody (1:200, Invitrogen) to reveal the immunostained structures. Propidium iodide or SYTO 16 (Invitrogen) staining was performed at room temperature for 1 h to reveal the nuclei. The labeled specimens were then immersed in the tissue clearing solution (RapiClear 1.52 solution, SunJin Lab, Hsinchu, Taiwan) before being imaged via confocal microscopy.

2.4. Deep-tissue confocal microscopy

Imaging of the tissue structure was performed with Zeiss LSM 510 Meta or LSM 800 confocal microscope (Carl Zeiss, Jena, Germany) equipped with 10× Fluar lenses and 25× LD Plan-Apochromat lenses (working distance: 570 μm) under a tile-scan mode with automatic image stitching. The laser-scanning process was operated under the multi-track scanning mode to acquire signals, including the transmitted light signals. The Alexa Fluor 647-labeled structures were excited at 633 nm and the fluorescence was collected by the 650–710-nm band-pass filter. The propidium iodide-labeled nuclei and Alexa Fluor 546-labeled structures were excited at 543 nm and the signals were collected by the 560–615-nm band-pass filter. The SYTO 16-labeled nuclei and lectin-Alexa Fluor 488-labeled blood vessels were excited at 488 nm and the fluorescence was collected by the 500–550-nm band-pass filter. Fluorescence signals in figures are pseudo-colored. After confocal imaging, the specimens were further processed with saline washing (removal of clearing reagent) and H&E histology to confirm the PanIN structure (e.g., Fig. 1c) if necessary.

2.5. Image projection and analysis

The Avizo 6.2 image reconstruction software (VSG, Burlington, MA, USA), Zen software (Carl Zeiss), and LSM 510 software (Carl Zeiss) were used for projection, signal segmentation, noise reduction, and analysis of the confocal images. Signal segmentation for quantification of the tissue network (or cell) density is illustrated in Juang et al. [30]. Briefly, feature extraction and image segmentation for calculation of the lymphatic vessel density were performed by the Label Field function of Avizo to collect the voxels of the area of interest (parenchyma, lesion, or peri-lobular space; the last one is within 500 μm off the edges of lobules, excluding the intra-lobular signals) and the associated Lyve1⁺ lymphatic vessels. Voxels of the vessels were divided by those of the area of interest ×100% to estimate the lymphatic vessel density. Quantitation of the Lyve1⁺ macrophage density follows the same approach except that the cell numbers instead of the voxels were used in the numerator, and the voxels of the lesion, peri-lesional parenchyma (within 500 μm of the lesion), or normal parenchyma were used in the denominator. The same immunolabeling, imaging, and quantitation processes were conducted on the comparable pancreatic middle sections to characterize the tissue densities on the same basis.

2.6. Statistical analysis

The quantitative values are presented as means ± SD or with the distribution of data points. Statistical differences were determined by the unpaired Student's t-test. Differences between groups were considered statistically significant when P < .05.

3.1. Human pancreatic lymphatic network in health and around PanIN lesions

The pancreas consists of the endocrine islets and the exocrine acini and ductal network to control and participate in the body's metabolic and digestive activities. From the map of human pancreas (Fig. 1a, b), we see both the endocrine and exocrine tissues and the infiltrated adipocytes contact with the podoplanin (D2-40) -labeled lymphatic network [27]. The association underscores the surveillance feature of lymphatics. Note that we used tissue clearing to prepare transparent human pancreases for penetrative 3-D imaging [24,25]. The transparent specimens allowed us to use both the transmitted light and fluorescence signals to identify and confirm the pancreatic microstructure and tissue networks (ducts, nerves, and vasculature) in space.

Importantly, in addition to the normal pancreas, this 3-D imaging approach can be used to examine the human pancreatic cancer specimens with high definition. Fig. 1c shows the map of a human PDAC tissue acquired from pancreatectomy. Deposits of high-grade PanIN lesions are detected in the tumor bulk with tissue scanning. In the map, patches of nuclear signals highlight the inflammatory areas, which are associated with the PanIN lesions. A side-by-side comparison of the H&E and fluorescence micrographs identifies the close association between the high-grade PanINs and the lymphatic network (insets of Fig. 1c and Supplementary Video S1). In addition, in the diseased pancreas distal to the tumor bulk (Fig. 1d and Supplementary Fig. S1), the lesion (low grade) -lymphatic vessel association is also seen in the lobule with apparent acinar atrophy (vs. Fig. 1a; n = 3 surgical biopsies).

Although the human PDAC analysis provides direct evidence of the lesion-lymphatic network association, it does not allow us to trace the development of the association, which requires examination of the PanIN lesions at their early stage (i.e., when the pancreas appears to be normal) as well as the advanced stage. In the next six sections we systematically analyze the mouse PanIN and PDAC models to characterize the lymphatic vessel remodeling and invasion in pancreatic duct lesion progression.

3.2. Mouse pancreatic lymphatic system detected by 3-D histology

Similar to the human pancreatic lymphatic system [2], the mouse pancreatic lymph nodes reside in the peri-lobular space and around the artery (Fig. 2a, Area 1). In mice, the lymph nodes and lymphatic vessels are visualized with Lyve1 (lymphatic vessel endothelial hyaluronan receptor 1) staining of the lymphatic endothelium [[31], [32], [33]] (note that Lyve1 also labels a subpopulation of F4/80⁺ macrophages, Supplementary Fig. S2 [34,35]). Morphologically, outside the lymph node, the Lyve1⁺ endothelium forms a torturous network which follows the artery and arteriole in extension (Fig. 2a, inset, and b). Inside the lymph node, the Lyve1⁺ endothelium lines the floors of the medullary and subcapsular sinuses (Fig. 2c; positive control of Lyve1 staining) [36]. Interesting, although the endothelial lining of the lymph node has been known, its extension across the subcapsular space has not been reported (Fig. 2c, insets, and d). This unique bridge structure is only seen inside the lymph node in normal mice but appears outside the lymph node in lymphangiogenesis, which will be discussed later.

Zooming out from the lymph node, Fig. 2e presents the high-definition image of an inter-lobular lymphatic vessel, which resides at the central location among the pancreatic lobules. This image shows the contact of lymphatic endothelium with all of the pancreatic major structures, including the endocrine islets, exocrine acini and ducts, and blood vessels (arterioles and venules). The association is comparable to that of the lymphatic network in the human pancreas (Fig. 1a).

3.3. Lymphangiogenesis and macrophage accumulation in early-stage duct lesion formation

Oncogenic Kras mutation often occurs in human PDAC [9,37]. Experimentally we used elastase-CreER;LSL-Kras^G12D mice (EK mice) to induce acinar oncogenic Kras expression, which causes acinar-to-ductal metaplasia and spontaneous induction of PanIN lesions (PanIN-1/2) to mimic the human condition [13,18]. Fig. 3a shows a gross view of the EK mouse pancreas at the early stage of duct lesion formation. At this stage, the majority of parenchyma (~98%) is anatomically normal [18]. However, scattered PanIN lesions appear at the edges of parenchyma, such as in Area 1 and 2 in Fig. 3a, which are enlarged in Fig. 3b–d and e–g for examination.

Specifically, the 2-D images and 3-D projections of the early-stage duct lesions identify the formation of peri-lesional lymphatic vessels (Fig. 3b, c, and Supplementary Fig. S3), which is confirmed by the paired Prox1 and Lyve1 staining (Supplementary Fig. S4), and the accumulation of Lyve1⁺ macrophages around the lesion (Fig. 3d, g). Quantitation of the Lyve1⁺ signals shows 61% (P < .05) and 263% (P < .01) increases in the density of Lyve1⁺ macrophages in the peri-lesional parenchyma and the lesion area, respectively, and a 218% (P < .05) increase in lymphatic vessel density of the influenced area, suggesting an ongoing inflammatory response around the lesion (Fig. 3h, i).

3.4. Lymphangiogenesis, endothelial invagination, and vessel invasion in large-scale duct lesion formation

We next used 3 weeks of cerulein injections to create repetitive pancreatic injuries (pancreatitis) and followed with 4 weeks of tissue regrowth to induce aggressive epithelial proliferation and large-scale duct lesion formation [18]. Fig. 4a–d presents the remodeling of the pancreatic microstructure and lymphatic network in this condition. Morphologically, the tissue injury and epithelial overgrowth lead to acinar atrophy and lesion/stromal adhesion to the surrounding organs, such as the intestine (Fig. 4d). This tissue remodeling is accompanied with marked lymphangiogenesis, in which the Lyve1⁺ endothelium occupies the peri-lobular space and extends along the interface between the lesion and the attached tissues, such as the lymph node (Area 1 in Fig. 4a; enlarged in Fig. 4b, c and Supplementary Video S2) and the intestinal serosa (Area 2 in Fig. 4a; enlarged in Fig. 4d). These drastic changes, however, are not seen in the wild-type B6 mice which went through the same process of 3-week cerulein pancreatitis and 4-week tissue regrowth (Supplementary Fig. S5). The B6 pancreas regenerates without apparent microstructural and lymphatic network remodeling, confirming that the genetic predisposition in the EK mice is essential for the large-scale duct lesion formation and the associated lymphangiogenesis.

Furthermore, when the large-scale duct lesion formation is induced under the background of p53 mutation (elastase-CreER;LSL-Kras^G12D;p53^+/−, the EKP mice) [38], we identify the lymphatic luminal invasion in five of the seven examined animals (Fig. 4e–i). Histologically, the lymphatic luminal invasion has the following two features: (i) it occurs at the peri-lobular spaces with lymphangiogenesis (Fig. 4e, f, and Supplementary Fig. S6), and (ii) it is associated with invagination of the Lyve1⁺ endothelium (Fig. 4f, g, and Supplementary Fig. S7).

Quantitation of the Lyve1⁺ signals shows that the density of the pancreatic lymphatic vessels in the EK and EKP mice increases 5.5-fold (P < .01) and 4.9-fold (P < .01) against the wild-type B6 mice, respectively. However, there is no statistical difference between the EK and EKP mice in the level of lymphangiogenesis (Fig. 4h), indicating that the increased incidence of the luminal invasion in the EKP mice compared with the EK mice (Fig. 4i) is likely due to the behavioral changes of the lesion and/or stromal cells, rather than the probability of contacts between the lesions and lymphatic vessels.

3.5. Formation of luminal bridge/valve-like tubules in lymphangiogenesis

In lymphangiogenesis in the EK and EKP mice, the lymphatic endothelium extends from the central area of the parenchymas (Fig. 2a) to the peri-lobular space (Fig. 4a, e), forming a tube network (Supplementary Fig. S7). Interestingly, inside the tube network, the endothelial expansion is accompanied with formation of the bridge/valve-like structures, or tubules (with a diameter at ~5 μm), across the walls of the Lyve1⁺ endothelium (Fig. 5a–e). The geometric feature of the tubules resembles the Lyve1⁺ endothelium in the subcapsular space (sinus) of the lymph node in the normal mice (Fig. 2c, d).

To visualize the lymph node and the nearby duct lesions in a continuous fashion, Supplementary Video S3 presents the Lyve1⁺ tubules in the EKP mice, which extend across the subcapsular sinus of the lymph node and bridge the peri-lobular space of the pancreas. The detection of the luminal tubules indicates the creation of the internal structure (frames/valves) as well as the walls of lymphatic vessels in the marked lymphangiogenesis in the EK and EKP pancreases. Furthermore, using the paired Lyve1 and Ki-67 staining, we confirm that the formation of the lymphatic luminal tubules, lymphangiogenesis, and cellular proliferation are parallel events (Fig. 5e–g). The Ki-67⁺ cells are identified in the lesion, stroma, and wall of lymphatic vessel.

3.6. Lymphatic vasodilation in large-scale duct lesion formation

Among the reactive vascular responses in large-scale duct lesion formation, vasodilation is a prominent feature in the PanIN microenvironment. In the EKP mice, Fig. 6a–c shows the peri-lesional association among the inter-lobular arteriole, the peri-arteriolar lymphatic vessels, and the sensory nerves (labeled with substance P) [39]. In particular, the magnified 2-D image and 3-D projection of sensory nerves identify their contacts with the two vascular systems. The result indicates the vascular sensory innervation and the likely influence of the neuromodulators/neurotransmitters (e.g., substance P) on the microvessels. Quantitative analysis shows a 91% (P < .001) and 98% (P < .001) increase in the arteriolar diameter and a 59% (P < .001) and 72% (P < .001) increase in the major axis of the peri-arteriolar lymphatic vessels in the EK and EKP mice, respectively, in the cerulein-induced large-scale duct lesion formation (Fig. 6d). There is no significant difference in the B6 pancreatic microstructure and vasculature 4 weeks post the cerulein treatment compared with those of the untreated pancreas (Supplementary Fig. S5).

3.7. Lymphatic vessel invasion in syngeneic mouse model of orthotopic cancer cell injection

To mimic the localized pancreatic lymphatic vessel invasion (vs. generalized pancreatic remodeling in EK and EKP mice, Fig. 4), we used the orthotopic injection of the syngeneic, EGFP⁺ pancreatic cancer cells to trace their location and association with lymphatic vessels via 3-D imaging. Fig. 7a shows the tumor formed locally around the pancreatic parenchyma and the peri-tumoral lymphangiogenesis induced by the cancer cell injection. Specifically, in the peri-tumoral domain, clusters of the invading cells were found inside the Lyve1⁺ lymphatic vessels with few cells appearing >500 μm distal to the tumor domain (Fig. 7a, insets), indicating an ongoing spread of cancer cells. Furthermore, Supplementary Fig. S8 presents the budding of EGFP⁺ cells against the lymphatic endothelium, highlighting the vascular plasticity and the invasive nature of injected cancer cells.

Surprisingly, using the panoramic tumor scan, we identify that the lymphangiogenesis and lymphatic vessel invasion are not limited to the peri-tumoral domain. Inside the tumor, lymphangiogenesis is clearly seen associated with the necrosis in the tumor core (Fig. 7b, c). At the necrosis boundary (Fig. 7d), the endothelium-enclosed nuclear fragments and cancer cells (vessel invasion, Fig. 7e) are specified with high-definition 3-D microscopy (Supplementary Video S4). Quantitatively, the localized lymphangiogenesis and vessel invasion (peri- and/or intra-tumoral) are visualized in six of six (100%) and five of the six (83%) animals with the cancer cell injection, respectively. Overall, the 3-D image data provide novel details of the lymphatic vessel remodeling and invasion, which otherwise could not be illustrated using the standard 2-D histology.