Main

To understand the mechanisms that underlie the formation of a pro-metastatic niche in the liver, we used the LSL-KrasG12D/+;LSL-Trp53R127H/+;Pdx1-cre (KPC) mouse model of pancreatic ductal adenocarcinoma (PDAC)4,5. We looked for features of a pro-metastatic niche in the livers of over-16-week-old tumour-bearing KPC mice and 8- to 10-week-old non-tumour-bearing (NTB) KPC control mice, which lack PDAC but harbour pancreatic intraepithelial neoplasia (PanIN)6. Compared to control mice, the livers of KPC mice contained increased numbers of myeloid cells, accompanied by an increase in the deposition and expression of fibronectin and type I collagen (COL1) (Fig. 1a, Extended Data Fig. 1a–d). Orthotopic implantation of KPC-derived PDAC cells into wild-type mice recapitulated these changes (Extended Data Fig. 1e–i). As shown previously7,8, matrix deposition did not require myeloid cells (Extended Data Fig. 1j–l). These results are consistent with evidence that myeloid cell accumulation and extracellular matrix deposition are key components of a pro-metastatic niche7,8,9,10.

Fig. 1: Primary PDAC development induces a pro-metastatic niche in the liver.
figure 1

a, Images and quantification of myeloid cells, fibronectin (FN), and COL1 in the liver. Arrows indicate Ly6G+ cells. Numbers in parentheses on plots indicate the number (n) of mice. Data pooled from two experiments. TB, tumour-bearing; NTB, non-tumour-bearing. b, Images of the liver and quantification of PDAC–YFP cells. Control mice (n = 14) and NTB KPC mice (n = 10) were intrasplenically injected with PDAC–YFP cells, and the liver was analysed after 10 days. Data representative of two independent experiments. c, Scatter plot of transcriptome data. FPKM, fragments per kilobase of exon per million mapped fragments (n = 5 for both groups). Scale bars, 50 μm (a) and 1 cm (b). Statistical significance calculated using one-way analysis of variance (ANOVA) with Dunnett’s test (a) and two-tailed Mann–Whitney test (b). Data represented as mean ± s.d.

Source data

Full size image

We next evaluated the susceptibility of the liver to metastatic colonization. Yellow fluorescent protein (YFP)-labelled KPC-derived PDAC cells (PDAC–YFP)6 were injected into control mice and KPC mice. The metastatic burden was threefold higher in KPC mice, and metastatic lesions were detected in the livers of KPC mice at increased frequency and size with enhanced proliferation (shown using Ki-67) (Fig. 1b, Extended Data Fig. 2a, b). Similar findings were observed using a YFP-negative KPC-derived cell line (Extended Data Fig. 2c, d). Orthotopic implantation of PDAC cells also increased the susceptibility of the liver to metastatic colonization, and this finding was independent of the presence of CD4+ and CD8+ T cells (Extended Data Fig. 2e–s).

We next performed mRNA sequencing on RNA isolated from the livers of control and KPC mice. We identified 275 differentially expressed genes (Extended Data Fig. 3a, b, Supplementary Data 1) and found that genes upregulated in KPC mice were associated with immune-related processes (Extended Data Fig. 3c). Notably, genes encoding myeloid chemoattractants, including SAA and members of the S100 family, were upregulated in KPC mice11,12,13 (Fig. 1c, Extended Data Fig. 3d, e). We also found enrichment of immune-related pathways, particularly the IL-6–JAK–STAT3 signalling pathway (Extended Data Fig. 3f, Supplementary Table 1). We validated our results by examining the livers of KPC mice for the presence of phosphorylated STAT3 (pSTAT3). STAT3 was activated in 80–90% of hepatocytes from KPC mice, compared to less than 2% of hepatocytes in control mice (Extended Data Fig. 3g, h). By contrast, we did not detect activation of STAT1 signalling (Extended Data Fig. 3i). Orthotopic implantation of PDAC cells also induced phosphorylation of STAT3 in hepatocytes (Extended Data Fig. 3j, k).

As IL-6 is fundamental to STAT3 signalling in hepatocytes14, we examined the livers of control mice (Il6+/+) and IL-6 knockout mice (Il6−/−) orthotopically injected with PBS or PDAC cells. Tumour-implanted Il6−/− mice displayed a decrease in STAT3 activation, particularly in hepatocytes (Fig. 2a, Extended Data Fig. 4a). This loss in STAT3 activation was accompanied by reductions in myeloid cell accumulation and extracellular matrix deposition without alterations in the morphology and density of liver sinusoids (Fig. 2a and Extended Data Fig. 4a-d). We also observed reduced expression of SAA, other chemoattractants, and extracellular matrix proteins (Fig. 2b, Extended Data Fig. 4e). Genetic ablation of Il6, however, did not alter proliferation, vascular density, or primary tumour growth (Extended Data Fig. 4f, g). Il6−/− mice were also less susceptible than control mice to metastatic colonization, and blockade of the IL-6 receptor (IL-6R) similarly inhibited the formation of a pro-metastatic niche in the liver (Fig. 2c–e, Extended Data Fig. 4h–s). Notably, genetic ablation of Il6 or blockade of IL-6R did not completely inhibit STAT3 signalling, suggesting that IL-6-independent mechanisms may contribute to STAT3 activation.

Fig. 2: IL-6 is necessary for the establishment of a pro-metastatic niche in the liver.
figure 2

a, b, n = 5 and 6 for Il6+/+ mice and n = 4 and 5 for Il6−/− mice orthotopically injected with PBS or PDAC cells, respectively. a, Quantification of pSTAT3+ cells, myeloid cells, and fibronectin. b, mRNA levels of Saa1 and Saa2 in the liver. c–e, n = 4 and 5 for Il6+/+ mice and n = 4 for Il6−/− mice orthotopically injected with PBS or PDAC cells, respectively. All groups were intraportally injected with PDAC–YFP cells on day 10. c, d, Images of liver and flow cytometric analysis. e, Quantification of PDAC–YFP cells. Data representative of two independent experiments (a–e). Scale bars, 1 cm. Statistical significance calculated using one-way ANOVA with Dunnett’s test. Data represented as mean ± s.d.

Source data

Full size image

IL-6 promotes the development and progression of PDAC15,16,17,18. To identify the source of IL-6, we orthotopically injected PBS or PDAC cells into Il6+/+ and Il6−/− mice and measured the concentration of IL-6 at distinct anatomic sites (Extended Data Fig. 5a). We detected IL-6 only in tumour-implanted Il6+/+ mice, with the highest concentration of IL-6 found in the primary tumour (Extended Data Fig. 5b, c). Although Il6 mRNA was undetectable in the liver, lung, and malignant cells, we observed Il6 mRNA in host cells adjacent to CK19-expressing PDAC cells (Extended Data Fig. 5d–g). Human primary tumours displayed a similar expression pattern (Extended Data Fig. 5h). Moreover, Il6 mRNA was detected in α-SMA+ stromal cells located adjacent to PanIN and PDAC cells in KPC mice (Extended Data Fig. 5i–k). We also found that primary pancreatic tumour supernatant activated STAT3 signalling in hepatocytes, and this was reduced in the presence of anti-IL-6R antibodies (Extended Data Fig. 6a, b). These results show that IL-6 released by non-malignant cells within the primary tumour is a key mediator of STAT3 signalling in hepatocytes.

To study a role for hepatocytes in directing liver metastasis, we generated mice that lacked Stat3 in hepatocytes (Stat3flox/flox Alb-cre). Compared to control mice (Stat3flox/flox), tumour-implanted Stat3flox/flox Alb-cre mice lacked features of a pro-metastatic niche (Fig. 3a–c, Extended Data Fig. 6c) and failed to produce SAA (Fig. 3d–f). However, deletion of Stat3 in hepatocytes did not affect liver sinusoid density or morphology and did not alter the size, proliferation, or vascular density of the primary tumour (Extended Data Fig. 6d–f). The livers of tumour-implanted Stat3flox/flox Alb-cre mice were also less susceptible to metastatic colonization (Extended Data Fig. 6g–l). In addition to its expression in hepatocytes (Extended Data Fig. 6m), mRNA for SAA was detected in colonic cells19 and in cells present in the periphery of the primary tumour (Extended Data Fig. 6n). However, both cell types maintained comparable levels of SAA mRNA despite deletion of Stat3 in hepatocytes.

Fig. 3: STAT3 signalling in hepatocytes orchestrates the formation of a pro-metastatic niche in the liver.
figure 3

a, Study design for b–f (n = 4 for Stat3flox/flox mice injected with PBS or PDAC cells; n = 8 and 7 for Stat3flox/flox Alb-cre mice injected with PBS and PDAC cells, respectively). b, c, Quantification of pSTAT3+ cells, myeloid cells, and fibronectin. d, mRNA levels of Saa1 and Saa2 in the liver. e, Images of Saa1 and Saa2 mRNA in liver cells. Dashed lines and asterisks indicate sinusoids and hepatocytes, respectively. f, Concentration of circulating SAA. Data representative of two independent experiments (a–f). Scale bars, 50 μm. Statistical significance calculated using one-way ANOVA with Dunnett’s test. Data represented as mean ± s.d.

Source data

Full size image

SAA proteins are acute phase reactants20. Consistent with elevated levels of circulating SAA in tumour-implanted mice (Fig. 3f), patients with PDAC displayed elevated levels of circulating SAA (Extended Data Fig. 7a). Overexpression of SAA and pSTAT3 by hepatocytes was also observed in five of seven patients with liver metastases (Fig. 4a, Extended Data Fig. 7b). Notably, high levels of circulating SAA correlated with worse outcomes (Extended Data Fig. 7c). Elevated levels of circulating SAA were also observed in patients with non-small-cell lung carcinoma (NSCLC) with liver metastases, and overexpression of SAA by hepatocytes was detected in the livers of patients with colorectal carcinoma (CRC) (Extended Data Fig. 7d, e). In addition, compared to tumour-implanted control mice (Saa+/+), double-knockout Saa1−/− Saa2−/− mice (hereafter referred to as Saa−/− mice) implanted with PDAC or MC-38 CRC cells failed to show features of a pro-metastatic niche in the liver, though genetic ablation of Saa1 and Saa2 had no effect on primary tumour growth (Fig. 4b–e, Extended Data Fig. 7f–s). SAA was also necessary for IL-6-mediated formation of a pro-metastatic niche and for fibrosis and myeloid cell recruitment in the setting of liver injury (Extended Data Fig. 8).

Fig. 4: SAA is a critical determinant of liver metastasis.
figure 4

a, Images of SAA in the livers of healthy donors (top) and patients with PDAC with liver metastases (bottom). Dashed lines and asterisks indicate sinusoids and hepatocytes, respectively. Data representative of one experiment. b, Quantification of pSTAT3+ cells, myeloid cells, and fibronectin (n = 5 for all groups orthotopically injected with PBS or PDAC cells). For c–e, n = 4 and 5 for Saa+/+ mice and n = 5 and 6 for Saa−/− mice orthotopically injected with PBS and PDAC cells, respectively. All groups were intraportally injected with PDAC–YFP cells on day 10. c, d, Images of liver and flow cytometric analysis. e, Quantification of PDAC–YFP cells. Data representative of two independent experiments (b–e). Scale bars, 50 μm (a) and 1 cm (c). Statistical significance calculated using one-way ANOVA with Dunnett’s test. Data represented as mean ± s.d.

Source data

Full size image

Tissue inhibitor of metalloproteinases 1 (TIMP1)7,8 and macrophage migration inhibitory factor (MIF)9,10 have been implicated in the promotion of metastasis. However, expression of these molecules was not affected by IL-6–STAT3–SAA signalling (Extended Data Fig. 9). We next determined whether formation of a pro-metastatic niche in the liver is dependent on the anatomical proximity of the pancreas to the liver. To this end, we looked for features of a pro-metastatic niche in the livers of CD45.1 and CD45.2 mice that were parabiotically joined (Extended Data Fig. 10a). Although only CD45.2 mice were implanted with PDAC cells, both mice displayed myeloid cell accumulation and fibrosis in the liver (Extended Data Fig. 10b–g), suggesting that formation of this niche is not dependent on the anatomical distance between the tumour and the liver. We also investigated whether SAA has a role in establishing a pro-metastatic niche in the lung. Development of PDAC in KPC mice induced accumulation of Ly6G+ myeloid cells and deposition of fibronectin within the lung, but IL-6–STAT3–SAA signalling was not required for the formation of a pro-metastatic niche in the lung (Extended Data Fig. 10h–o).

Our data provide insight into the mechanisms that direct liver metastasis. Although recent studies have suggested a role for tumour-intrinsic factors in driving metastatic spread of cancer7,8,9,10,21,22,23, we provide evidence that inflammatory responses mounted by hepatocytes are critical to liver metastasis. Mechanistically, hepatocytes orchestrate this process through activation of IL-6–STAT3 signalling and the subsequent production of SAA, which alters the immune and fibrotic microenvironment of the liver to establish a pro-metastatic niche (Extended Data Fig. 10p). Our findings suggest that therapies that target hepatocytes might prevent liver metastasis in patients with cancer.

Methods

Mice

CD45.2 (wild type, C57BL/6J), CD45.1 (B6.SJL-Ptprca Pepcb/BoyJ), Il6 knockout (Il6−/−, B6.129S2-Il6tm1Kopf/J), Stat3flox/flox (B6.129S1-Stat3tm1Xyfu/J), and Alb-cre+/+ (B6.Cg-Tg(Alb-cre)21Mgn/J) mice were obtained from the Jackson Laboratory. Stat3flox/flox mice were bred to Alb-cre+/+ mice to generate Stat3flox/+ Alb-cre+/− mice, which were backcrossed onto Stat3flox/flox mice to generate Stat3flox/flox Alb-cre+/− mice. These mice were then bred to each other to create Stat3flox/flox Alb-cre+/+ and Stat3flox/flox Albumin-cre+/− mice (Stat3flox/flox Alb-cre), and Stat3flox/flox Albumin-cre−/− mice (Stat3flox/flox). KrasLSL-G12D/+ Trp53LSL-R172H/+ Pdx1-cre (KPC) mice and Trp53LSL-R172H/+ Pdx1-cre (PC) mice were as previously described4,5. Saa1 and Saa2 double-knockout (Saa−/−) mice were as previously described24 and provided by the University of Kentucky College of Medicine. Saa−/− mice used for experiments had been bred to obtain a 99.9% C57BL/6 background using the Jackson Laboratory Speed Congenic Service24. All transgenic mice were bred and maintained in the animal facility of the University of Pennsylvania. Animal protocols were reviewed and approved by the Institute of Animal Care and Use Committee of the University of Pennsylvania. In general, mice were monitored three times per week for general health and euthanized early based on defined endpoint criteria including tumour diameter ≥1 cm, ascites, lethargy, loss of ≥10% body weight, or other signs of sickness or distress.

Clinical samples

All patient samples were obtained after written informed consent and were de-identified. Studies were conducted in accordance with the 1996 Declaration of Helsinki and approved by institutional review boards of the University of Pennsylvania and the Mayo Clinic. To obtain plasma from healthy donors, patients with PDAC patients, and patients with NSCLC, peripheral whole blood was drawn in EDTA tubes (Fisher Scientific). Within 3 h of collection, blood samples were centrifuged at 1,600g at room temperature for 10 min with the brake off. Next, the plasma was transferred to a 15-ml conical tube without disturbing the cellular layer and centrifuged at 3,000g at room temperature for 10 min with the brake off. This step was repeated with a fresh 15-ml conical tube. The plasma was then stored at –80 °C until analysis. Biopsy results, computed tomography, and/or magnetic resonance imaging records were used to determine sites of metastasis in patients with PDAC or NSCLC whose plasma samples were used for assessment of SAA levels. Liver specimens from healthy donors were obtained by percutaneous liver biopsy, and acquisition of liver specimens from patients with liver metastases was as previously described25. Liver specimens from patients with CRC with liver metastases were obtained from the Cooperative Human Tissue Network (CHTN). Patient characteristics are shown in Supplementary Table 2.

Cell lines

PDA.69 cell line (PDAC cells) was used for intrasplenic and orthotopic injection, and PDA.8572 cell line (PDAC–YFP cells) was used for intrasplenic, intraportal, and retro-orbital injections. These cell lines were derived from PDAC tumours that arose spontaneously in KPC mice, as previously described4,26. The MC-38 cell line, which was used for orthotopic implantation, was purchased from Kerafast. Cell lines were cultured in DMEM (Corning) supplemented with 10% fetal bovine serum (FBS, VWR), 83 μg/ml gentamicin (Thermo Fisher), and 1% GlutaMAX (Thermo Fisher) at 37 °C, 5% CO2. Only cell lines that had been passaged fewer than 10 times were used for experiments, and trypan blue staining was used to ensure that cells with >95% viability were used for studies. Cell lines were tested routinely for Mycoplasma contamination at the Cell Center Services Facility at the University of Pennsylvania. All cell lines used in our studies tested negative for Mycoplasma contamination.

Animal experiments

For all animal studies, mice of similar age and gender were block randomized in an unblinded fashion. Male and female mice aged between 8 to 12 weeks were used unless indicated otherwise. Mice were age- and gender-matched with appropriate control mice for analysis. Sample sizes were estimated based on pilot experiments and were selected to provide sufficient numbers of mice in each group for statistical analysis.

For orthotopic and intrasplenic injections of pancreatic tumour cells, mice were anaesthetized using continuous isoflurane, and their abdomen was sterilized. After administering analgesic agents and assessing the depth of anaesthesia, we performed a laparotomy (5–10 mm) over the left upper quadrant of the abdomen to expose the peritoneal cavity. For orthotopic injection, the pancreas was exteriorized onto a sterile field, and sterile PBS or pancreatic tumour cells (5 × 105 cells suspended in 50 μl of sterile PBS) were injected into the tail of the pancreas via a 30-gauge needle (Covidien). Successful injection was confirmed by the formation of a liquid bleb at the site of injection with minimal fluid leakage. The pancreas was then gently placed back into the peritoneal cavity. For intrasplenic injection, 150 μl sterile PBS was drawn into a syringe and then sterile PBS or pancreatic tumour cells (5 × 105 cells suspended in 100 μl sterile PBS) was gently drawn into the same syringe in an upright position as previously described27. After the spleen was exteriorized onto a sterile field, pancreatic tumour cells were injected into the spleen via a 30-gauge needle. Successful injection was confirmed by whitening of the spleen and splenic blood vessels with minimal leakage of content into the peritoneum. Splenectomy was then performed by ligating splenic vessels with clips (Horizon) then cauterizing them to ensure that there was no haemorrhage. Afterwards, the remaining blood vessels were placed back into the peritoneal cavity. For both procedures, the peritoneum was closed with a 5-0 PDS II violet suture (Ethicon), and the skin was closed using the AutoClip system (Braintree Scientific). Following surgery, mice were given buprenorphine subcutaneously at a dose of 0.05-0.1 mg/kg every 4–6 h for 12 h and then every 6–8 h for 3 additional days. Mice that were orthotopically injected with pancreatic tumour cells were analysed after 20 days, unless indicated otherwise in study designs. Mice that were intrasplenically injected with PDAC cells were analysed after 10 days.

For intraportal injection of pancreatic tumour cells and hydrodynamic injection of expression vectors, mice were anaesthetized using continuous isoflurane, and their abdomen was sterilized. After administration of analgesic agents, median laparotomy (10 mm) was performed, and the incision site was held open using an Agricola retractor (Roboz). After exposure of the peritoneal cavity, the intestines were located and exteriorized onto a sterile field surrounding the incision site to visualize the portal vein. Throughout the procedure, the intestines were kept hydrated with sterile PBS that was pre-warmed to 37 °C. For intraportal injection, sterile PBS or pancreatic tumour cells (5 × 105 cells suspended in 100 μl sterile PBS) were injected into the portal vein via a 30-gauge needle. Successful injection was confirmed by partial blanching of the liver. For hydrodynamic injection, 1 μg of pLIVE expression vectors was suspended in sterile saline corresponding to 8% of mouse body weight as previously described28. Vectors were injected into the portal vein via a 27-gauge needle within 5–8 s. Successful injection was confirmed by complete blanching and swelling of the liver. For both procedures, a sterile gauge was then held over the injection site for 1 min to ensure that no injected contents would leak into the peritoneal cavity. Afterwards, the intestines were placed back into the peritoneal cavity, and the peritoneum and skin were closed with a suture and autoclips, respectively. Following surgery, mice were given buprenorphine subcutaneously as described above. Intraportal injection of pancreatic tumour cells was performed on day 10, and metastatic burden in the liver was evaluated on day 20, unless indicated otherwise in study designs.

For orthotopic implantation of colorectal tumour cells, wild-type mice were first subcutaneously injected with MC-38 (1 × 106 cells suspended in 100 μl of sterile PBS) into the right flank. After 10 days, mice were euthanized, and subcutaneous tumours were collected. Tumours were then cut into small pieces, each 3 × 3 mm in size, and placed in sterile PBS on ice until implantation. Mice were anaesthetized using isoflurane, and their abdomen was sterilized. Following administration of analgesic agents, median laparotomy was performed as described above. Implantation of colorectal tumour tissues into the caecum was then performed as previously described29. After we placed the intestines back into the peritoneal cavity, the peritoneum and skin were closed with a suture, and mice were given buprenorphine as described above. Mice were analysed after 10 days.

For parabiotic joining of mice, female CD45.2 mice were orthotopically injected with sterile PBS or pancreatic tumour cells as described above and co-housed with age-matched female B6 CD45.1 mice. Each parabiotic pair was housed in a separate cage to maximize bonding between partners. After one week, parabiotic partners were anaesthetized using continuous isoflurane, and their flanks were sterilized. After administration of analgesic agents, longitudinal skin flaps from the lower limb to the upper limb were created, and everted skin flaps were sewn using a suture. In addition, the knees and olecranons of parabiotic partners were joined together using a suture for additional stabilization. Following surgery, mice were given buprenorphine subcutaneously at a dose of 0.05-0.1 mg/kg every 4-6 h for 5 days. Parabiotically joined mice were analysed after 20 days.

For administration of antibodies, the abdomen of mice was sterilized, and anti-CD4 antibodies (GK1.5, 0.2 mg), anti-CD8 antibodies (2.43, 0.2 mg), anti-IL-6R antibodies (15A7, 0.2 mg), or rat isotype control antibodies (LTF-2, 0.2 mg) were suspended in 100 μl sterile PBS. Antibodies were subsequently injected into the peritoneum via a 30-gauge needle. All antibodies used in in vivo experiments were obtained from BioXCell. To deplete F4/80+ myeloid cells, clodronate-encapsulated liposomes (Liposoma) were administered by intraperitoneal injection according to the manufacturer’s protocol. For induction of liver injury, mice were intraperitoneally injected with CCl4 (Sigma, 1 ml/kg body weight) dissolved in sunflower seed oil as previously described30. Detailed information on antibodies and reagents used in experiments can be in found in Supplementary Table 3.

Microscopic analysis

For preparation of formalin-fixed paraffin-embedded (FFPE) sections, dissected tissues were fixed in 10% formalin for 24 h at room temperature, washed twice with PBS, and then stored in 70% ethanol solution at 4 °C until they were embedded in paraffin and sectioned at 5 μm. For preparation of cryosections, dissected tissues were embedded in Tissue-tek O.C.T. (Electron Microscopy Sciences) and frozen on dry ice. Frozen tissues were stored at –80 °C until they were sectioned at 7 μm.

Automated immunohistochemistry, immunofluorescence, and RNA in situ hybridization were performed on FFPE sections using a Ventana Discovery Ultra automated slide staining system (Roche). Reagents were obtained from Roche and ACDBio (Supplementary Table 3) and used according to manufacturer’s protocol. Images were acquired using a BX43 upright microscope (Olympus), an Aperio CS2 scanner system (Leica), or an IX83 inverted multicolour fluorescent microscope (Olympus). Manual immunohistochemistry of mouse tissues for SAA was previously described31. For manual multicoloured immunofluorescence staining, O.C.T. liver cryosections were briefly air dried and fixed with 3% formaldehyde at room temperature for 15 min. For intracellular staining, sections were permeabilized with methanol at –20 °C for 10 min immediately after formaldehyde fixation. Sections were then blocked with 10% normal goat serum in PBS containing 0.1% TWEEN 20 for 30 min. For intracellular staining, 0.3% Triton X-100 was added to the blocking solution for permeabilization of cellular and nuclear membranes. Sections were incubated with primary antibodies (Supplementary Table 3) in the blocking solution for 1 h at room temperature or overnight at 4 °C, followed by washing with PBS containing 0.1% TWEEN 20. Sections were then incubated with secondary antibodies (Supplementary Table 3) in the blocking solution for 1 h at room temperature or overnight at 4 °C. After washing, sections were stained with DAPI to visualize nuclei and subsequently with Sudan Black B in 70% ethanol to reduce autofluorescence, as previously described32. Immunofluorescence imaging was performed on an IX83 inverted multicolour fluorescent microscope (Olympus). For quantification of cells and extracellular matrix proteins, five random fields were acquired from each biological sample.

Flow cytometry

Mice were euthanized, and the liver and lung were removed after the blood was drained by severing the portal vein and inferior vena cava. The liver and lung were rinsed thoroughly in PBS before mincing with micro-dissecting scissors into small pieces (<0.5 × 0.5 mm in size) at 4 °C in DMEM containing collagenase (1 mg/ml, Sigma-Aldrich), DNase (150 U/ml, Roche), and Dispase (1 U/ml, Worthington). Tissues were then incubated at 37 °C for 30 min with intermittent agitation, filtered through a 70-μm nylon strainer (Corning), and washed three times with DMEM. Cells were resuspended in ACK lysing buffer (Life Technologies) at room temperature for 15 min to remove red blood cells. After washing three times with DMEM, cells were counted and stained using Aqua dead cell stain kit (Life Technologies) following the manufacturer’s protocol. For characterization of immune cell subsets, cells were washed three times with PBS containing 0.2 mM EDTA with 2% FBS and stained with appropriate antibodies (Supplementary Table 3). For quantification of PDAC–YFP cells, cells were not stained with any antibodies. Lastly, cells were washed three times with PBS containing 0.2 mM EDTA with 2% FBS and examined using a FACS Canto II (BD Biosciences). Collection and analysis of the peripheral blood was as previously described26. FlowJo (FlowJo, LLC, version 10.2) was used to analyse flow cytometric data and generate 2D t-SNE plots.

Detection of IL-6, SAA, and TIMP1

Mice that were orthotopically implanted with PDAC cells were euthanized, and primary tumours were removed and weighed. In addition, blood samples were collected from the portal vein and left ventricle of the heart using a 27-gauge needle. Tumours were rinsed thoroughly in PBS and minced with micro-dissecting scissors into small pieces (<0.5 × 0.5 mm in size) at 4 °C in serum-free DMEM at 1 mg of tissue per 1 μl medium. Tumour suspensions were then centrifuged at 12,470g at 4 °C for 15 min, and tumour supernatant was collected and stored at –80 °C until analysis. A similar procedure was performed to obtain pancreas supernatant from mice that were orthotopically injected with PBS. To collect the serum, blood samples were allowed to clot at room temperature for 30 min. Samples were then centrifuged at 12,470g at 4 °C for 15 min, and the serum was collected and stored at –80 °C until analysis. IL-6 levels in tumour or pancreas supernatant and serum were assessed using a cytometric bead array (BD Biosciences) following the manufacturer’s protocol. Samples were examined using a FACS Canto II (BD Biosciences), and data were analysed using FCAP Array (BD Biosciences, version 3.0). SAA and TIMP1 levels in mouse serum samples were measured using a commercially available enzyme-linked immunosorbent assay kit (Thermo Fisher) following the manufacturer’s protocol. Similarly, SAA levels in plasma samples collected from healthy donors and patients with PDAC as described under ‘Clinical samples’ were measured using a commercially available human enzyme-linked immunosorbent assay kit (Thermo Fisher) following the manufacturer’s protocol.

RNA and quantitative PCR

Mouse organs and cells were stored in TRIzol (Thermo Fisher) at –80 °C until analysis. Samples were thawed on ice and allowed to equilibrate to room temperature before RNA was isolated using a RNeasy Mini kit (Qiagen) following the manufacturer’s protocol. cDNA synthesis was performed as previously described33. Primers for quantitative PCR were designed using the Primer3 online program34, and sequences were analysed using the Nucleotide BLAST (NCBI) to minimize non-specific binding of primers. Primers were synthesized by Integrated DNA Technologies, and their sequences can be found in Supplementary Table 4. Quantitative PCR was performed as previously described33. Gene expression was calculated relative to Actb (β-actin) using the ∆Ct formula, and fold change in gene expression was calculated relative to the average gene expression of control groups using the ∆∆Ct formula. Genes with Ct greater than or equal to 30 were considered not detected.

QuantSeq 3′ mRNA sequencing and data analysis

RNA was isolated from the livers of control mice and NTB KPC mice as described above and submitted to the Genomics Facility at the Wistar Institute. After the quality of RNA was assessed using a 2100 Bioanalyzer (Agilent), samples were prepared using a QuantSeq 3′ mRNA-Seq library prep kit FWD for Illumina (Lexogen) following the manufacturer’s protocol and analysed on a NextSeq 500 sequencing system (Illumina). FASTQ files were uploaded to the BaseSpace Suite (Illumina) and aligned using its RNA-Seq Alignment application (version 1.0.0), in which STAR was selected to align sequences with maximum mismatches set to 14 as recommended by Lexogen. Output files were analysed using Cufflinks Assembly & DE application (version 2.1.0) in the BaseSpace Suite to determine differentially expressed genes, which were used to generate an expression heatmap and a FPKM scatter plot. In addition, these genes were analysed using ClueGO (version 2.3.3)35 and CluePedia (version 1.3.3)36, which are applications of Cytoscape software (version 3.5.1)37. Functional grouping of biological processes was performed on the basis of kappa score. Gene Ontology data38,39 downloaded on 23 January 2018 were used for analysis. Gene set enrichment analysis (version 3.0)40 was used to determine biological processes that were differentially enriched in experimental groups.

In vitro studies

To isolate primary hepatocytes for in vitro studies, mice were anaesthetized using continuous isoflurane, and their abdomen was sterilized. After administering analgesic agents and assessing the depth of anaesthesia, we performed a laparotomy (10–15 mm) along the midline of the abdomen to expose the peritoneal cavity. The intestines were then located and exteriorized to visualize the inferior vena cava and portal vein. The inferior vena cava was cannulated via a 24 gauge Insyte Autoguard cathether (BD), and the liver was perfused using 50 ml liver perfusion medium (Thermo Fisher) at a flow rate of 8–9 ml/min using a peristaltic pump. At the start of perfusion, the portal vein was severed to drain the blood from the liver. Successful perfusion was confirmed by blanching of the liver, which was subsequently perfused using 50 ml liver digest medium (Thermo Fisher) at the same flow rate. Both liver perfusion medium and liver digest medium were pre-warmed to 42 °C in a water bath. After perfusion, the liver was carefully transferred to a Petri dish containing William’s E medium (Sigma) supplemented with 10% FBS, 83 μg/ml gentamicin, and 1% GlutaMAX.

To dissociate hepatocytes from the liver, cell scrapers were used to create small cuts (5 mm) on the surface of the liver, and the tissue was gently shaken. Dissociated cells were then filtered through a 100-μm nylon strainer (Corning) and centrifuged at 50g at 4 °C for 5 min. After the supernatant was discarded, cells were resuspended in a solution consisting of isotonic Percoll (Sigma) and supplemented William’s E medium (2:3 ratio). Cells were then centrifuged at 50g at 4 °C for 10 min to obtain a pellet enriched in hepatocytes. The supernatant was discarded, and hepatocytes were resuspended in supplemented William’s E medium. Cell viability and number were determined using trypan blue staining, and 5 × 104 hepatocytes were seeded in each well of a 48-well plate pre-coated with collagen. Hepatocytes were incubated in supplemented William’s E medium for 4 h at 37 °C, 5% CO2 to allow attachment to the plate. The medium was then switched to HepatoZYME-SFM (Thermo Fisher) supplemented with 83 μg/ml gentamicin and 1% GlutaMAX. Medium was replenished every 24 h for the next 48–72 h.

For hepatocyte activation assays, hepatocytes were incubated in supplemented HepatoZYME-SFM mixed with (i) serum-free DMEM, (ii) primary pancreatic tumour supernatant, or (iii) serum-free DMEM containing 250 ng/ml IL-6 (Peprotech) for 30 min at 37 °C, 5% CO2. All mixtures were made in a 1:1 ratio, and each condition was run in triplicate. For the in vitro IL-6R blockade experiment, hepatocytes were pre-incubated with 5 μg/ml anti-IL-6R antibodies for 2 h before being stimulated with tumour supernatant. After stimulation, medium was carefully removed, and formaldehyde and methanol were used to fix and permeabilize hepatocytes, respectively, as described above. Hepatocytes were then stained for pSTAT3 (Supplementary Table 3), and their nuclei stained with DAPI. Immunofluorescence imaging was performed on an IX83 inverted multicolour fluorescent microscope (Olympus).

Statistical analysis

Statistical significance was calculated using Prism (GraphPad Software, version 7) unless indicated otherwise. Multiple comparisons testing was performed using one-way ANOVA with Dunnett’s test. Paired group comparisons test was carried out using two-tailed Wilcoxon matched-pairs signed rank test. Unpaired group comparisons test was performed using two-tailed unpaired Student’s t test or two-tailed Mann–Whitney test. Comparison of Kaplan–Meier overall survival curves was performed using log-rank (Mantel-Cox) test. P values less than 0.05 were treated as significant. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment, unless stated otherwise.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.