Co-evolution of tumor cells and hepatocytes fostered by SLIT2-ROBO1 axis facilitates liver metastasis of pancreatic ductal adenocarcinoma

To explore the mechanism of co-evolution and potential driver of which in pancreatic ductal adenocarcinoma (PDAC) metastasis to liver, we studied key molecules involved in this progress and their translational values. Pre-metastatic niche (PMN) and macro metastatic niche (MMN) formation in mouse model were recognized via CT combined 3D organ reconstruction bioluminescence imaging. We next conrmed the expressions and distributions of SLIT2 and ROBO1 in 35 cases of human matched liver metastasis and primary PDAC samples, 14 case human PDAC liver metastasis transcriptional analysis, intrasplenic mouse models and Kras G12D /Trp53 R172H /Pdx1-Cre (KPC) mouse models. Translational value was assessed on Slit2 / /Alb1-Cre (Slit2 CKO) mice, KPC mouse model and Ex vivo tests via administration of neutralizing antibody targeting ROBO1. We also analyzed prognosis of 266 cases human PDAC tissue with or without SLIT2-ROBO1 fostered co-evolution and demonstrated the dependence receptor (DR) characteristics of ROBO1 in the following-up mechanism study. Experiments on Slit2 CKO, Slit2 CKO-RE and KPC mouse models demonstrated that disturbing SLIT2-ROBO1 mediated co-evolution in liver microenvironment via preventing their interaction could signicantly attenuate liver metastasis of PDAC. We have demonstrated that co-evolution took advantage of DR characteristics in PMN and MMN. Targeting SLIT2-ROBO1 axis could be a therapeutic strategy towards metastatic PDAC.


Introduction
Patients diagnosed with PDAC bearing surgically resectable tumor account for less than 20% mainly due to its devastating metastatic nature, and liver is the most preferred destination of distant metastasis.
Distant metastasis to speci c target organ requires co-evolution between disseminated tumor cells (DTCs) derived from primary tumor and target organ microenvironment resident cells [1][2] . It has been illustrated by the hypothesis of "seed and soil" theory that the "seed" (DTCs) and "soil" (microenvironment of target organ) both adapted to each other for gaining the advantage to achieve metastasis 3 . According to this theory, co-evolution may take place between DTCs and hepatocytes in every step of PDAC liver metastasis to facilitate metastatic niche formation and expansion, while the driving power throughout this progress remain to discover. To uncover underlying mechanism, key events of continuous coevolution involved in liver metastasis including PMN and MMN were selected for further study ( Figure   1a) [4][5] . Recent studies have de ned that PMN underwent complex tumor microenvironment (TMN) to adapt to "seeds" for their successful colonization, and formation of which was recognized as the initial step of metastasis progress. For metastasis, native cells at target organ altered original secretion protein pro les and began to recruit TMN cells to generate breeding ground before extravasation and implantation of tumor cells at PMN stage, though few morphological changes could be observed at these sites. Thus, it is hard to gain clinical samples of PMN formed liver of patients, and the means of PMN detection and de nition were also limited. Here we utilized intrasplenic injection mouse model with relatively permanent PMN and MMN time points of PDAC liver metastasis and further veri ed our discoveries in KPC mouse model and clinical samples.
Members of Axon Guidance (AG) family have been considered to play vital roles in a variety of physiological processes such as embryonic development and neuron growth 6-7 . While accumulating evidence also revealed their indispensable functions in cancers [8][9] . Intriguingly, several important receptors of AG were reported to share common characteristics in different cancers, termed as DRs 10-12 . DRs are kinds of special receptors acting as a two-way switch to trigger cell proliferation and survival in presence of their ligands, or induce apoptosis once ligands are moved off, rather than simply performing on and off. In normal conditions, it is generally supposed that DRs are responsible to maintain tissue homeostasis and prevent escape of cells from where they were 11,13 . Therefore, it is interesting to explore the mechanisms involved in DRs mediated metastasis. Though discussed in many previous studies, the meaning of existence of DRs in tumorigenesis, especially in metastasis remain largely unknown.
Here we ltered out AG molecular SLIT2 in our models which were overexpressed by hepatocytes through PMN to MMN and demonstrated for the rst time that its classic receptor ROBO1 was a DR. We have also unveiled that hepatocytes derived SLIT2 formed a breeding ground for implantation of ROBO1 positive DTCs thus triggered co-evolution between hepatocytes and tumor cells for further MMN achievement. Interference of the interaction of these two molecules via neutralizing antibody could signi cantly hamper the progress of liver metastasis. Our results revealed a previously unknown role of DRs in metastasis of cancer and put forward that SLIT2-ROBO1 axis mediated co-evolution of "seed" and "soil" was critical in implantation and outgrowth on PMN till MMN. We also provided a potential therapeutic strategy to inhibit co-evolution via targeting DRs to not only prevent proliferation of tumor cells but also trigger their apoptosis.

Detection and veri cation of PMN and MMN in mouse models
To set off, we utilized the intrasplenic injection mouse model to study alteration of liver microenvironment from PMN to MMN. In this model, injected tumor cells could generate liver metastatic lesions via passing through the portal veins to simulate PDAC liver metastasis which is highly reproducible and make it possible for us to analyze the step-by-step changes of expression pro les in liver [14][15] . formed with few GFP + cells, while the reported markers LOX, SAA1/2, TIMP-1, and MIF or CD68, CD11b and CD163 for macrophages and Ly-6G for neutrophils expressions were observed, which was de ned as PMN group. D15 for Kpc1199 or D12 for Panc02 witnessed IHC-detectable metastatic clones planting on liver which was recognized as metastasis group (Figure 1f-1h, Figure 2f; Extended Data Figure 1c-1k, Figure 3a) [16][17][18][19][20][21][22] . Additionally, we selected D3 instead of D0 as control group to eliminate the interference of non-tumor in ammation of surgery among groups. The time point and location of metastases formed by PDAC cells on left hepatic lobule were relative reproducible due to the hemodynamic character of portal veins. In summary, we could thus gain liver tissues undergoing PMN and MMN in mouse models for further exploration.
Enriched hepatocytes derived SLIT2 is located with ROBO1 positive tumor cells in liver metastasis progress Hence, to comprehensively investigate the differentially expressed genes of hepatocytes during process of PDAC metastasis, we set out to slice 2×2×2 mm 3 of the liver tissue without tumor and close to portal veins in left hepatic lobule of mice at D3, D8 or D15 for Kpc1199 or D3, D6 or D12 for Panc02 after intrasplenic injection respectively. These specimens were further con rmed by positive staining of PMN markers before they were used for further studies (Figure 2a Figure 3a). To further con rm, we used KPC mice, which could spontaneously generate liver metastasis of PDAC. We rst detected DTCs in left hepatic lobule by measuring the PDAC DTCs markers CK19, CD133 and CD44 from KPC mice before 12 weeks old at which metastasis took place in frequency [23][24] (Extended Data Figure   3b). The PMN markers mentioned above in intrasplenic injection model were also detected in the livers of KPC mouse with DTCs (Extended Data Figure 3c). Furthermore, highly expressed SLIT2 was also found in DTCs-detected PMN or MMN formed liver of KPC mice (Figure 2g-2h). To further con rm the source of SLIT2, staining of which on liver metastasis pathological slice of PDAC patients was performed. Adjacent liver of PDAC patients' metastatic niches exhibited enhanced SLIT2 staining, indicating that SLIT2 was derived from the local hepatocytes but not tumor cells (Figure 3a-3b), which was further con rmed by costaining of SLIT2 and albumin in both mice or human derived tissues (Figure 3c-3d).
It has been demonstrated that aberrant high expression of SLIT2 could participate the co-evolution of hepatocyte-tumor cells in liver metastasis progress. The roundabout guidance (ROBO) family, except ROBO4, was considered as the most predominant receptor of SLIT2. We further discovered that it was ROBO1, but not ROBO2 or ROBO3, mainly expressed in metastatic CK19 + PDAC cells (Figure 3e, Extended Data Figure 3d). Interestingly, we have discovered that much more ROBO1 positive metastatic niches could be detected in SLIT2 enriched metastasized livers of KPC mice and strong correlation existed between these two molecules in livers bearing metastasis (Figure 3f-3g). What's more, we have analyzed the expression of ROBO1 and SLIT2 in our own datasets containing 14 cases liver metastasis tissues of PDAC patients (OEP000481 in National Omics Data Encyclopedia, NODE) and GEO DataSets (GSE151580 and GSE71729) and the results have revealed strong correlation between these two molecules in liver metastasis which further emphasized the vital role of SLIT2-ROBO1 in liver metastasis progress (Extended Data Figure 3e-3g).
To comprehensively analyze the relationship of prognosis and the expressions of these two molecules, we examined the functions of SLIT2-ROBO1 in primary PDAC via evaluating prognosis of 266 patients. To our interest, we have discovered that high level of ROBO1 would lead to poor prognosis in SLIT2-rich TME; while the results would be just the contrary if TME SLIT2 was de cient (Figure 3h-3i). These prognosis data have deciphered that this axis would be of value for further study.
Taken together, results above have revealed that hepatocytes could upregulate SLIT2 secretion for recruitment and aggregation of ROBO1 positive tumor cells, which could be responsible for co-evolution of them.
SLIT2-ROBO1 mediated co-evolution facilitates outgrowth of liver metastatic niches in vivo We next generated Slit2 hepatocytes-speci c conditional knockout mice (Slit2 / /Alb-Cre, CKO). To rescue, lentivirus carrying loxp-Slit2 was used to speci cally express SLIT2 in hepatocytes which was administrated 14 days before modeling (CKO-RE) (Extended Data Figure 4a To gain the direct evidence that SLIT2-ROBO1 mediated the co-evolution progress, we performed antibody treatment on KPC mice. In this model, high frequency of spontaneous liver MMN occurrence was observed at approximate 15-16 weeks old of mice. To prohibit the formation of PMN in liver, we thus started antibody administration from week12 ( Figure 4f). Data has demonstrated that antibody treatment signi cantly hampered the outgrowth of metastatic niches in left lobules and decreased the percent of ROBO1 + PDAC cells in niches (Figure 4g-4i). Interestingly, the accumulation of SLIT2 in adjacent hepatocytes signi cantly attenuated, indicating that disturbing ROBO1 function in this axis would also signi cantly affect SLIT2 expression. Furthermore, it was more effective to perform treatment from PMN rather than MMN, since the early administration would prevent more micro niches (≤500µm) formation in liver (Figure 4k-4l). Meanwhile, the interference of co-evolution loop could be of better e ciency when antibody was administrated from PMN to inhibit SLIT2 expression (Figure 4m).
These results have pointed out the signi cance of SLIT2-ROBO1 axis in liver metastasis, while the role it played in co-evolution need to be further investigated.

SLIT2-ROBO1 axis is critical for co-evolution between hepatocytes and tumor cells in liver metastasis progress
We then performed IHC-P staining of ROBO1 on 35 cases PDAC liver metastasis tissues with their matched primary tumors. Results have illustrated high ROBO1 expression in most liver metastasis tissues, even if the positive rate of which was much lower in matched primary tumors respectively (Figure 5a-5b). Further results in KPC mice have also con rmed the conclusion (Extended Data Figure 7a). To explore why ROBO1 enriched at liver metastasis and how SLIT2-ROBO1 manipulated the selection of tumor cells in co-evolution, we next performed cell mixture composed of Panc02 Robo1−FL and Panc02 CTRL each half (PG0) in intrasplenic injection mouse model. Tumor cells derived from separated liver metastasis formed by PG0 were then cultured (PG1) before next modeling and PG2 cells were gained from separated liver metastasis formed by PG1 (Figure 5c). Flow cytometry analysis have demonstrated the increasing of Panc02 Robo1−FL cells in population from PG0 to PG2 in tumor cell mixture, indicating that ROBO1 has facilitated survival and outgrowth ability of tumor cells for selection in liver (Figure 5d-5e). To explore whether SLIT2 exerted the driving power, we analyzed the protein levels of ROBO1 and SLIT2 in PMN and MMN of mouse models experienced intrasplenic injection of PG0 and PG1 respectively via IHC-P. The results have revealed that ROBO1 staining at MMN was much stronger in PG1 than those in PG0, while SLIT2 staining could be detected in all groups, displaying an increasing manner from PMN of PG0 to MMN of PG1 (Figure 5f; Extended Data Figure 7b). WB on PG1 and PG2 cells have also provided evidence (Extended Data Figure 7c). Especially, this co-evolution effect could also be disturbed by antibody treatment (Figure 5g). Blocking of SLIT2 and ROBO1 interaction signi cantly attenuated the staining of both these two molecules ( Figure 5h). These data indicated that DTCs-induced SLIT2-rich microenvironment exerted selective pressure on DTCs themselves and provides ROBO1 + ones with a growth advantage in liver thus ful lls the co-evolution. To consolidate, another tumor cell mixture Kpc1199 Mix−I containing equal amounts of Kpc1199 Robo1−FL/GFP and Kpc1199 ΔRobo1/mCherry or Kpc1199 Mix−II containing equal amounts of Kpc1199 Robo1−FL/GFP and Kpc1199 CTRL/mCherry were utilized on Slit2/CKO or WT mice modeling (Figure 5i). In WT mice, ROBO1-FL expressed tumor cells displayed predominant population in liver metastatic niches especially in Kpc1199 Mix−I modeled ones during coevolution with hepatocytes; while in Slit2/CKO mice Kpc1199 CTRL cells outcompeted Kpc1199 Robo1−FL in absence of SLIT2 (Figure 5j-5k). What's more, the outgrowth of both Kpc1199 Robo1−FL cells and Kpc1199 ΔRobo1 cells were signi cantly hampered in SLIT2-de cient TME. Data gained from ow cytometry analysis were also in consistence with the results before (Figure 5l; Extended Data Figure 7d).
These phenomena have indicated that ROBO-FL expressed tumor cells and SLIT2 expressed hepatocytes could achieve co-evolution, losing SLIT2 in TME or blocking ROBO1 not only led to failure of ROBO1 + tumor cells in competition, but led to their elimination in liver.

ROBO1 acts as a DR to exert dual effects on tumor cells
Previous studies have demonstrated that several receptors of axon guidance family including DCC, Plexin-D1, UNC5 or EphA4 were DRs 10, 12, 25-26 . Our clinical prognosis data also provided evidence to support that ROBO1 might own dual functions in PDAC progress (Figure 3i-3k) To con rm, we constructed Robo1 lacking rst two Ig domains (ΔRobo1), which was required for SLIT2 and ROBO1 interaction (Extended Data Figure 4c Based on the results above, we postulated that ROBO1 might belong to DRs. Examinations of the SLIT2 and ROBO1 expression levels in 10 pancreatic cancer cell lines have elucidated that SLIT2 seldom displayed high expression in PDAC cell lines, especially in liver metastasis-derived CAPAN-1 and cFPAC-1 cells with relatively higher ROBO1 expression (Extended Data Figure 10a-10b). SW-1990 with dual low expressions of ROBO1 and SLIT2 was selected out to transfected with ΔROBO1, ROBO1-FL or empty vector respectively for further study. Results have revealed that proliferation capability of SW-1990 ROBO1−FL was much higher than that of SW-1990 ΔROBO1 when additional recombinant SLIT2 (rSLIT2) existed. Notably, in absence of rSLIT2, both of two groups presented slower growth rate than that of SW-1990 CTRL (Figure 6f). Similar results were also obtained in Panc02 (Extended Data Figure 10c Taken together, these data indicated that ROBO1, recognizing as a DR, would lead to cell apoptosis without ligand binding, while facilitate cell growth when ligands existed. This dual function of ROBO1 mechanically laid foundation for its selective pressure in metastatic tumor-host co-evolution. Activated ROBO1 facilitates cell growth via enhancing MEK3/6-p38α MAPK interaction Given that Mitogen-Activated Protein Kinases (MAPK) pathway controls cell behaviors by modulating cell proliferation, migration, survival and apoptosis, we wondered whether MAPK pathway was involved in SLIT2-ROBO1 axis triggered survival signal for PDAC cell metastasis to liver [29][30] . The results revealed that binding of rSLIT2 to ROBO1-FL signi cantly induced the phosphorylation of p38αMAPK while exerted little effects on other key molecules involved in this pathway. Intriguingly, the phosphorylation levels of MEK3 and MEK6, the p38αMAPK-speci c kinases, were also not altered after rSLIT2 stimulation (Extended Data Figure 13a). Furthermore, PH-797804 and VX-702, the speci c inhibitors targeting p38αMAPK, could e ciently abolished the SLIT2-ROBO1 mediated cell growth advantage (Figure 7a, Extended Data Figure b). Additionally, VX-702 e ciently decreased the cell viability of SW-1990 CTRL and rSLIT2 stimulated SW-1990 ROBO1−FL to the same level, indicating that the SLIT2-ROBO1 axis induced growth advantage was meditated by p38αMAPK (Figure 7b, Extended Data Figure 13c). Since creative work have been achieved for demonstrating that p38α MAPK was vital in PMN formation and there also has been reported that p38 MAPKs could be rapidly activated by SLIT2-contained conditional medium in 5 min in Xenopus retinal growth cones, we hypothesized that p38αMAPK pathway was closely related to SLIT2-ROBO1 mediated co-evolution in PMN and MMN [31][32] . Consistent with previous report, our results also showed that 5-10 min treatment of rSLIT2 elevated the phosphorylation level of p38αMAPK and peaked at 30min (Figure 7c). This phenomenon could be observed most obviously in PANC-1 cells with relatively higher expression of ROBO1 (Figure 7d). We further demonstrated that it was ROBO1 that mediated rSLIT2 induced phosphorylation of p38αMAPK in SW-1990 ROBO1−FL (Figure 7e). IF staining was also performed to further con rm the increase of P-p38 and nuclear translocation induced by rSLIT2 (Figure 7f, Extended Data Figure 13d), while the level of P-MEK3/6 almost unchanged as before ( Figure  7C-7D, Extended Data Figure 13d). Furthermore, patient derived metastasized liver specimens were used for IHC-P staining of P-p38 and the results illustrated that tumor cells with activated P-p38 were surrounded by abundance of SLIT2 (Extended Data Figure 13e). We then proceeded to further seek for the direct evidence that p38αMAPK was phosphorylated in the presence of SLIT2 by ex vivo test. The mouse livers metastasized by Panc02 Robo1−FL were separated and sliced into two pieces for exposure in medium with or without rSLIT2 exposure for 1 hour respectively. The results displayed that the more signi cant phosphorylation of p38αMAPK and nuclear accumulation was observed in rSLIT2 treated group than those in control group ( Figure 7G, Extended Data Figure 13f). Moreover, livers bearing metastasis of KPC mice were also performed ex vivo, data have elucidated that neutralizing antibody against ROBO1 could su ciently prevent p38αMAPK from phosphorylation and nuclear translocation ( Fig. 4H; Extended Data Figure 13g).
Together, these data indicated that the phosphorylation of p38α MAPK triggered by SLIT2-ROBO1 was rapid and less sensitive to alteration of MEK3/6, which inspirited us to speculate that there could be direct interactions between activated ROBO1 and p38α MAPK or MEK3/6. The Co-IP elucidated that ROBO1 performed directly binding of p38αMAPK and MEK3/6 (Figure 7i-7j). And further study con rmed that binding of SLIT2 enhanced the phosphorylation of p38αMAPK precipitated by ROBO1 (Figure 7k). Together, these results have elucidated that SLIT2 bound ROBO1 could increase the opportunity for interaction of p38αMAPK and its kinase MEK3/6 to activated the downstream pathway.
Our study has unveiled a novel mechanism of "seed" and "soil" co-evolution, which was driven and pushed forward by DR derived selection pressure through p38αMAPK pathway (Figure 7l).

Discussion
Numerous DTCs derived from primary tumors were released into circulations as "seeds" for distant metastasis, while only few of them achieved MMN in target organs. It is also confusing that why different kinds of tumor cells prefer speci c target organs 5 . It is reasonable to gain the explanation that DTCs and the "soil" (native microenvironment at target organ) can ful ll running-in for reciprocal TMN formation. The necessities involved in this progress are alterability of TMN cells and adaptability of DTCs. In our study, we have unveiled that hepatocytes would provide SLIT2 for PMN formation for only ROBO1 positive PDAC DTCs, supporting the survival of which at their arrival. The implantation and outgrowth of metastatic niches also stimulated more hepatocytes for sustained supplement of SLIT2.
PMNs are built up by aberrantly activated native cells such as tumor associated macrophages (TAMs), cancer associated broblasts (CAFs) or some organ speci c cells like hepatocytes 16,22,[33][34][35] . These PMN cells exerts metastasis-friendly effects mainly by producing extracellular matrix (ECM) proteins to form niches for DTCs. PMNs are lacking well-established de nitive detecting means since they usually generate without histological alteration, and different cancers or different distant metastasis formed by same cancers always requires for different PMN proteins 5 . It is also di cult to gain clinical samples containing PMNs due to the reasons mentioned before: for example, it is often possible that PDAC patients bearing hepatic PMNs may own seemingly healthy liver diagnosed by known tests. While intrasplenic injection mouse models could generate liver metastasis at relative permeant time points once other experiment conditions (e.g. kinds and amounts of injected tumor cells) are kept stable, it is possible for us to decide the approximate time point of PMN. Moreover, we performed examinations on tissues via utilizing reported PMN markers to ensure reliability before further explorations were kept on. We nally ltered out that hepatocytes derived Slit2, members of AG family, enriched at PMN and MMN for tumor cells to loom large.
Receptors have long been de ned as switches which could ful l their functions only if triggered by their ligands. While DRs could generate and conducts contrary signals according to the presence or absence of their own ligands 11 . Furthermore, it is also interesting to discover that many known DRs are encode by neuroendocrine or development related genes and share some common characteristics including that their ligands are usually only abundant in speci c organs 10,12,26,36 . Hypothesis postulates that DRs which are regarded as important guardians of tissue homeostasis perform as safe lock against heterogenous cell implantation including cancer metastasis to other organs 11 . In normal conditions, DRs are used to restrict cells in the given organ with abundant ligands and prevent cell outgrow into alien microenvironment without ligand by inducing cell apoptosis. As to pathological settings such as metastatic cancers, it is vital for DRs to cause suppressive effects on tumor cells to inhibit metastasis. Here we have uncovered that PDAC tumor cells just utilized this mechanism for their colonization and outgrowth in liver by inducing hepatocytes to secrete SLIT2 from PMN to MMN of liver metastasis.
Metastasis requires for changes of both "seed" and "soil" to cater to needs of each other and erases unquali ed ones, termed as co-evolution. In this study we tried to explain that the selective power or driving power involved in co-evolution was brought in by the DR properties of ROBO1. SLIT2-rich liver microenvironment not only supported survival and outgrowth of disseminated ROBO1 + tumor cells, but also exerted selective pressure on DTCs to enrich ROBO1 + cells.
Furthermore, antibody against the SLIT2-ROBO1 axis has provided superior therapeutic effect, since blocking of SLIT2 binding to ROBO1 not only hampered the proliferation advantage brought by ROBO1-p38MAPK pathway, but also triggered ROBO1 induced cell apoptosis, indicating the therapeutic strategy involving DRs might be of potential prospect. Considering that surgery could not be executed on PDAC patients bearing liver metastasis which limits effective treatment, our study has proposed a promising treatment strategy for these cases by targeting SLIT2-ROBO1 axis. For further study, it would be interesting to explore whether this is a common mechanism for cancer metastasis.

Materials And Methods
Constructs & Reagents.

Clinical samples
Specimens involved in this study mainly included 3 cohorts for respective experimental designs: Cohort I containing 35 cases liver metastasis tissues of PDAC patients together with paired primary tumor tissues was used for IHC-P or IF analysis; cohort II containing 14 fresh cases liver metastasis tissues of PDAC patients together with their paired primary tumor tissues was used for transcriptional analysis, which could be referred on https://www.biosino.org/node/project/detail/OEP000481; cohort III containing 266 cases PDAC primary tumor tissue of patients with their respective prognosis was performed for IHC-P and survival analysis.
Specimens of patients involved in our study were all from --hospital, ----University School of Medicine.
Specimen collection, experiments design and performance were all approved by local ethics committee in ----Hospital with patients' informed consents. All these patients were diagnosed by both clinical surgeons and professional pathologists. Approval letter of ----University School of Medicine, ----Hospital Ethics Committee is ----.
Animal Experiments lox-stop-lox-Kras G12D/+ ; lox-stop-lox-Trp53 R172H/+ ; Pdx1-Cre mice were purchased from The Jackson Laboratories (Bar Harbor, ME) to generate transgenic PDAC mouse model. Slit2 / mice and Alb1-Cre mice were purchased from Cyagen. All mice above were on the C57BL/6 genetic background. All C57BL/6 wildtype mice and nu/nu mice were purchased from ----University. Intrasplenic injection model provides repeatable and controllable liver metastasis at relative xed time in mice, which make it possible for research on PMN and MMN [37][38] . The disadvantages of this model involves that the progress of liver metastasis is more rapid than that in spontaneous ones such as KPC model, while it generates liver metastasis at a stable location and time point once using same experimental conditions 15 .
To start, we rst explored the time points at which MMN formed in most modeled mice using Kpc1199 cells or Panc02 cells at different concentrations through IHC-P on liver left lobule sections every 3 days.
We then chose Kpc1199 at 4 × 10 5 /mouse or Panc02 1 × 10 6 /mouse to ensure that the formation of MMN would not be too early or late which would make it di cult to settle down the time points of PMN.
Results of IHC-P and CT combined 3D organ reconstruction bioluminescence imaging have nally de ned the approximate time point of MMN (day 12-for Kpc1199 and day 9-for Panc02). In preliminary data involved in 50 mice for each cell line respectively, more than 90% mice would generate MMN on right time (98% for Kpc1199 cells on day 12 and 92% for Panc02 cells on day 9). Then we examined liver sections before MMN formation by staining GFP and reported PMN markers including LOX, TIMP-1, MIF, SAA1/2, and markers of M2 type macrophages. PMN recognized as GFP − /markers + liver sections were then determined (day 7-11 for Kpc1199 and day 5-8 for Panc02). In preliminary data involved in 50 mice for each cell line respectively, more than 85% mice would generate PMN on right time according to markers mentioned before (90% for Kpc1199 cells on day 8 and 86% for Panc02 cells on day 6). We thus recognized the time point of both these events ( Fig. 1b; Extended Data Fig. 1a).
In brief, 4 × 10 5 Kpc1199 cells or 1 × 10 6 Panc02 cells suspended in 20 µl DMEM without FBS were performed. Immunocompetent isogenic C57BL/6 mice under 2.5% iso urane inhalation anesthesia were performed 10-15 mm subcostal incision through abdomen skin and peritoneum for surgical exposure of spleen after sterilization. The speed of injection was slower than 5 µl/s followed by 2s needle retention to prevent leakage. Wound closure was then rapidly performed.
For retrieval of expression of SLIT2 in SLIT2 conditional knockout mice, we perform vector GV348 (Ubi-MCS-SV40-puromycin) containing loxp-NM_004787-loxp for lentivirus package. 200µl saline with or without a dose of 3×10 7 of HIV were then delivered into the tail vein of mice respectively. Mice were anesthetized with 2.5% vaporized inhaled iso urane and placed in restraint that positioned the mouse tail in a lighted, heated groove. The speed of injection was slower than 50 µl/s. The e ciency of restoration was evaluated 10 days later and the performance. Treatment was performed 2 weeks before the intrasplenic injection of tumor cells. For neutralizing antibody treatment, antibody towards Ig1-Ig2 domain of ROBO1 was performed intraperitoneal injection 1 week before hepatic metastasis modeling, and the administration of which was every 3 days until the mice were sacri ced. For administration of neutralizing antibody on KPC mice, intraperitoneal injection was performed every 3 day since week 12 or week 16 till week 18 before tissue harvsesting.

Subcutaneous Xenograft Model.
Athymic male nu/nu mice ages 6 to 8 weeks were used in subcutaneous xenograft model. For SW-1990, SW-1990 CTRL , SW-1990 ΔROBO1 and SW-1990 ROBO1−FL suspended in DMEM at a concentration of 1 × 10 7 cells/ml were injected, the injecting volume was 200µl, respectively, and tumor diameters were monitored with calipers every 4 days until the sacri ce of mice at day 24; for CAPAN-1, CAPAN-1 CTRL and CAPAN-1 SLIT2−oe suspended in DMEM at a concentration of 1 × 10 7 cells/ml were injected, the injecting volume was 150µl, respectively, and tumor diameters were monitored with calipers every 3 days until the sacri ce of mice at day 15; for PANC-1, PANC-1 shCTRL and PANC-1 shROBO1 suspended in DMEM at a concentration of 2 × 10 7 cells/ml were injected, the injecting volume was 200µl, and tumor diameters were monitored with calipers every 5 days until the sacri ce of mice at day 40. Inoculation was performed under the right inguinal skin of mice. Tumor volumes were calculated as volume = 0.5× length× width 2 . After mice were sacri ced, the tumors were separated from body and the weights of which were measured.
CT combined 3D Organ Reconstruction Bioluminescence Imaging.
Mice bearing hepatic metastasis composed of luciferase expressing tumor cells were intraperitoneal injected 150 mg D-luciferin at volume of 200µl. Mice were then anesthetized with 2.5% vaporized inhaled iso urane 2 min after injection before placed into IVIS Spectrum (Caliper Life Sciences, Waltham, MA). CT was then performed to merge with re y bioluminescence signals. For organ reconstruction, CT image was utilized to be placed into auto-reconstructed liver and spleen at opacity of 20%. The normalization and quanti cation of re y bioluminescence signals were performed according to the red cube surrounding area. Mice were sacri ced at the end of test, the tumor-bearing livers were immediately isolated before icy PBS washing. Photo of luciferin emission imaging of every liver was then taken. All the imaging or calculating were performed by living Image software, version 4.5.3.

Transcriptional Analysis
Panc02 or Kpc1199 intrasplenically injected mouse models were performed. Then mice were sacri ced at speci c days and the livers of which were harvested. For Panc02 modeled mice, the time points were day 3, day 6 and day12; for Kpc1199 modeled ones, the time points were day 3, day8 and day15. The fresh livers gained were then washed by icy saline and left hepatic lobules of which were sliced into 2×2×2 mm 3 cubes. Especially, for tissues of Panc02 modeled mice on day12 and Kpc1199 modeled mice on day 15, obvious metastatic niches could be observed, while only adjacent livers without metastasis would be selected for further study. In the following-up, PMN and MMN markers were measured before transcriptional analysis. The analysis data is available in the Sequence Read Archive (SRA), the number of which is PRJNA590588 (for Kpc1199) and PRJNA590599 (for Panc02).
Metastatic Tumor Dissociation, Culture and Examination Intrasplenic injection model mice bearing liver metastasis were sacri ced after operation. The metastasized livers were then gained carefully and immediately followed by icy saline washing. Metastatic niches were carefully recognized and separated from liver. Fine-pointed forceps were used for detachment of remained liver parenchymal tissue, tumor associated brous or necrotic areas. Then "clean" liver metastatic tumors were washed by icy DMEM for 3 times and cut into 1×1×1mm cubes before transferred into 2.5 mL DMEM containing 100 µL of Enzyme D, 50 µL of Enzyme R, and 12.5 µL of Enzyme A provided in kit. After suspending the dissociated tumor tissues in the solution mentioned before, the tubes were tightly closed and put on constant temperature oscillator for 1h at 37℃, 120 rpm. The mixtures were then performed centrifugation at 1000 rpm for 5 min and the samples would be resuspended in DMEM for 3 times. The suspended tumor cells in DEMN were next for cell counting before planted on dishes. Brie y, approximate 5 × 10 4 cells in complete DMEM would be add into 6-hole plates for further culture. Examination of GFP or mCherry expression and activity of luciferase would be taken into consideration for tumor cell certi cation.
Histology and Immunohistochemistry.
The tissues were xed in 4% paraformaldehyde and embedded in para nized blocks. The blocks were cut into sections at a thickness of 5 µm, then depara nized and rehydrated for histopathological evaluation. For hematoxylin-eosin (HE) staining, the sections were dyed in hematoxylin for 5 mins and in eosin for 5-10 secs for HE staining. For immunohistochemical staining, the sections were placed in sodium citrate buffer solution (PH 6.0) for 20 minutes to carry out repair of tissue antigens. Then these sections were incubated with 0.3% hydrogen peroxide/phosphate-buffered saline for 30 mins and blocked with 10% (w/v) BSA (Sangon)/Phosphate-buffered saline (PBS). Slides were rst incubated using the antibodies at 4°C overnight with optimal dilution. After washed three times with 1×PBS, the slides were labeled by HRP second antibody at room temperature for 1 h. Then the sections were washed three times in 1×PBS, then treated with DAB substrate liquid (Thermo) and counterstained by Immuno uorescence Staining.
For tissue staining, the sections carried tissues were placed in PH 6.0 sodium citrate buffer solution for 20 minutes to subject heat-mediated antigen retrieval. Then these sections were blocked by 10% (w/v) BSA/PBS and co-incubated with primary antibodies at 4°C overnight with optimal dilution. For cell staining, SW1990 cells at 2×10 4 cells per well, PANC1 at 2×10 4 per well or MIA PaCa-2 at 3×10 4 cells per well were seeded on slides in 12-well plates (ibidi) respectively and cultured at 37°C. After washed three times with 1×PBS, the slides were labeled with Alexa Fluor 594-conjugated anti-rabbit antibody and Fluor 488 conjugated anti-mouse antibody at room temperature for 1 h. The nucleus was stained with DAPI (Sigma) after the redundant second antibody were removed. After washed three times with 1×PBS, the immuno urescene signals were captured using confocal-scopy (Carl Zeiss).

Ex vivo Living Tumor Tissue Culture
Liver tissues containing Panc02 Robo1−FL formed metastatic niches gained from intrasplenic injection mouse model or metastasized liver tissue of KPC mouse model were separated on ice for maintaining their structural integrity and activity. Then tissues were washed in icy saline within 15min after resection, followed by sliced into small cubes at an approximate volume of 5×5×5 mm 3 . Each cube must contain liver metastatic niches and their adjacent livers. Then these cubes were placed into complete medium with or without 30nM rSLIT2 administration/ ROBO1 neutralizing antibody respectively.
For intrasplenic model derived livers, these tissues were stimulated for 1h to measure the rapid phosphorylation of p38αMAPK. Importantly, a cube would be cut into two pieces in the middle and placed into medium with or without rSLIT2 respectively for ensuring the same tumor microenvironment (TME) of metastatic niches undergoing. After rSLIT2 administration, all tissues were immediately xed by 4% paraformaldehyde for further IHC-P test.
For KPC mice derived livers, these tissues were treated with antibody no more than 8h to maintain the tissue constructures. All tissues were then immediately xed by 4% paraformaldehyde followed by IF staining.

Recombinant SLIT2 and sROBO Protein Expression and Puri cation
Episomal expression vector with pCEP-Pu-Strep II-tag were cloned in the SLIT2 or sROBO1 ORF. The reconstructed plasmids were transfected into 293T cell line for recombinant protein expression.
Puromycin for screening was administrated into complete DMEM at a concentration of 5-10 µg/ml 48h after transfection for 10 days and 2 µg/ml of which was performed to maintain the expression positive

Cell Transfection
For shRNA transfection, lentivirus carried shRNA sequences were in Table. 1 For negative control, scramble shRNA targeting no known genes were also designed and used.
All transfected cells were screened with puromycin at a concentration gradient of 1µg/ml to 10µg/ml for more than at least 10 days before the transfection e ciencies were evaluated.
Cell Viability Assay.
Cells were seeded into 96-well plates. The numbers of planted cells were as follow: for human cell lines: SW-1990: 3000/well; BxpC-3: 3000/well; PANC-1: 5000/well; CFPAC-1: 6000/well; CAPAN-1: 3000/cell, HPAC: 5000/well; for murine cell lines: Panc02: 3000/well. The measurement or treatment for these cells started after overnight incubation and recognized as day 0. The consistent measurements lasted from day 0 to day 5; for cell stress exertion, culture medium in absence of FBS were performed after day 2. For rSLIT2 treatment, the concentration was 30nM in culture medium. VX-702 treatment was performed at a concentration of 5µM for normal cell viability assay and a concentration gradient of 200nM, 1µM, 5µM, 25µM, 125µM for concentration dependence test. PH-797804 treatment was performed at a concentration of 50µM for normal cell viability assay and a concentration gradient of 80nM, 400nM, 2µM, 10µM, 50µM for concentration dependence test. sROBO was performed at a concentration of 5µM in culture medium. At indicated time point, diluted CCK-8 in culture medium at a concentration of 10% (v/v) was added 100µl in every well and incubated at 37℃ for 1 hour. Cell viability was monitored using Power Wave XS microplate reader (BIO-TEK) by measuring the absorbance at 450nm. Every group contained 5 repeated wells and every experience was performed three times.

Colony Formation Assay
Panc02, Kpc1199 and SW-1990 cell lines cultured at 70% con uence were utilized for colony formation assay to ensure them at phase of logarithmic growth. Cells were detached with 0.25% trypsin/0.01% EDTA and planted in 3mm dishes in complete medium with or without rSLIT2 treatment (10nM or 30nM) respectively for growth. The numbers of planted cells were as follow: SW-1990: 4000/dish; Panc02: 3000/dish; Kpc1199:3000/dish. The medium was changed every 3 days. All cells were then xed in 4% paraformaldehyde for 30 mins and stained by crystal violet. The measurement of the results was performed by Image J via calculating the pixels of staining area.

Cell Apoptosis Assay
Cell apoptosis measured by Caspase-3/7 Activity Kit: SW-1990 CTRL , SW-1990 ΔROBO1 and SW-1990 ROBO1−FL were planted at a concentration of 4000/well into 96-well plates. Serum starvation was performed for 48 hours after 3-day growth with or without 30nM rSLIT2 administration. Then the operation and measurement of the activity of Caspase -3/7 was performed strictly according to manufacturer's guidance.
Cell apoptosis measured by Annexin V & propidium iodide (PI) staining: SW-1990 CTRL , SW-1990 ΔROBO1 and SW-1990 ROBO1−FL were detached with 0.25% trypsin/0.01% EDTA in 1×PBS after treatment with or without rSLIT2 administration (10nM or 30nM). Then suspended cells were harvested in DMEM and centrifuged at 800rpm for 3 minutes. These cells were then stained with 3.5µl Annexin V and 3.5µl PI diluted in 100µl binding buffer after 1×PBS washing. Flow cytometry (BD) was performed to analyze after incubation of cells for 20 minutes at 25℃. Cell Apoptosis measured by Terminal deoxynucleotidyl transferase (TdT) dUTP nick-end labeling (TUNEL) assay: all steps were strictly conducted in accordance with instructions. Quantitative Real-time PCR. Total RNA extraction was performed using the Trizol reagent followed by reversely transcription to harvest the cDNA. 7500 Real-time PCR system (Applied Biosystems, USA) was used to performing the qPCR at recommended cycling settings. The conditions were 1 initial cycle at 95℃ for 2 min, 35 cycles of 5 sec at 95℃ and 31 sec at 60℃. 2 −ΔCT method was utilized for evaluating and normalization to 18S mRNA levels when relative mRNA expressions were calculated.
The number of metastatic niches in liver with diameters over or under 500μm in each group was performed (g) and the percent of ROBO1 + metastatic niches in each group was also assessed (h) (n=5 mice per group, mean±s.e.m.; two tailed unpaired t-test; ***, P<0.001).