1. Supplementary Figure S1. a.a. days Cell number Doubling time =14 hours CT26CT26GFP isotype CXCR4 isotype CXCR7 isotype CXCR4CXCR7 FSC b.b. Supplementary Figure S1. Characterization of GFP gene labelling on CT26 cells and expression of receptors for CXCL12. a. Isolation and characterization of CT26GFP from a single-cell clone. The doubling time of CT26GFP cells was similar to that seen in parent CT26 cells. b. FACS analyses showing the low expression of CXCR4 and intermediate expression of CXCR7 on CT26 cells.

2. Supplementary Figure S2. a. b. c. Supplementary Figure S2. Role of hyaluronidase receptor CD44, integrins and collagens on spheroid formation of murine (a-c) and human (d) colon cancers. a. FACS analyses showing the high expression of CD44 and integrin  1 and  5, but not  1,  2, and  4, on CT26 cells. b. Sphere formation assay in vitro for CT26 cells under fresh-RPMI and PLCF-RPMI. Neither hyaluronidase nor anti-CD44 neutralizing antibody affected the spheroid formation. c. Sphere formation assay in vitro for CT26 cells under anti-integrin  1(anti-  1), anti-integrin  4(anti-  4), anti-integrin  5(anti-  5) neutralizing antibodies, and their combination. Their corresponding control antibodies were also involved. No effect was observed. d. PLCF-facilitated spheroid formation of 4 different human colon cancer cell lines is sensitive to collagenase. EDTA was used as positive control. hyaluro- nidase hyaluro- nidase ( - ) anti-CD44 Isotype IgG 1h 2h 3h 4h 5h 6h 16h 24h 48h 72h Hours after floating culture fresh -RPMI PCLF -RPMI ( - ) anti-  1 IgM (isotype) anti-  4 IgG (isotype) anti-  5 IgG + IgM (isotype) anti-  4 + -  5 anti-  4 + -  5 + -  1 d. DLD-1 LoVo Colo-201 SW620 RPMI 5mM EDTA collage- nase hyaluro- nidase

3. a. b. Cell count CXCR4 DMSO (solv.) HIF-inhibitor PBS DMSO (solv.) HIF-inhibitor PBS isotype Normoxia 21% O 2 Hypoxia 2% O 2 CXCR4 expression (monolayer) Cell count CXCR4 pFN cFN pFN + cFN PBS isotype Cell count CXCR4 (-) 8 nM 10  M isotype CXCR4 expression (sphere) FAK inhibitor ) Supplementary Figure S3. Relative expression 24h / 0h (normalized to α-tubulin) Average ± S.E. n=3, each *P<0.01 * PAX3 HRE YY1 NF-kB GC-boxes NRF-1 TATA- box -2,237 0 e. f. Supplementary Figure S3. a-d. Each experiment was done more than three times, and showed similar results. a. Focal adhesion kinase (FAK), a downstream regulator for integrins, is not a candidate for spheroidal expression of CXCR4. b. CXCR4 was not inducible on CT26 cells in response to either FNs (left panel), cultivation under a hypoxic condition, or the use of an inhibitor for hypoxia-inducible factor (HIF). c. The stressor CoCl2, a chemical that induces hypoxic responses, induces CXCR4 on the CT26 monolayer that can be inhibited by a radical scavenger NAC (left panels), but NAC can not inhibit the spheroid-induced CXCR4 expression (right panel) d. FACS analyses showing the upregulation of insulin-like growth factor-1 receptor (IGF-1R) and CXCR4 on a CT26 monolayer under a serum-starvation. e. and f. Schematic structure of the 5’-flanking region of CXCR4 (Tarnowski et al., 2010) (a) and mRNA expression of each transcription factor and CXCR4 at 0 h (monolayer) and 24 h after floating cultivation as assessed by quantitative real-time RT-PCR (b). Cell count CXCR4 sphere sphere + NAC isotype (-) CoCl 2 + NAC Isotype 3.55%47.7%5.19%0.12% monolayer sphere Forward scatter CXCR4 Serum 10% SphereSerum free monolayer IGF-1R  c. d.

4. a. b. c. d. LoVo -GFP HCT116 -CFSE en face observation cross section LoVo -GFP HCT116 -CFSE untreatedAMG3100Mithramycin A 7 days after tumor inoculation e. Cell count CXCR4 monolayer sphere isotype HCT116 CXCR4 LoVo Cell count 0h 1h 3h 6h 9h 12h 24h 48h Hours after floating culture HCT116 - PCLF + Coll-L - PCLF + Coll-L LoVo f. Supplementary Figure S4.

5. Supplementary Figure S4. (continued) Supplementary Figure S4. Identification of SCF+/CXCL12+ niche-like cell sheets, accelerated sphere formation, and directional tumor metastasis in human mesentery. Each experiment in a-d was performed three or more times with similar results. a and b. Representative en face (A) and cross section (B) findings of human mesentery samples obtained from patients who had undergone colectomy. a: Thin mesenteric surface sections, approximately 2-mm thick, were prepared using a surgical knife (left panel), and placed on the slide glass with adhesive followed by observation with a dissecting microscope (middle panel). These sections were immunohistochemically stained with SCF (red) and counterstained with DAPI (blue). Note that the red dots (white arrows) were mainly located on or around perivascular adipose tissue (yellow bipolar arrow). b: Cross sections identifying the localization of CAR cells. Doubly positive CAR cells (green: CXCL12; red: SCF) were frequently seen at the border area of perivascular adipose tissue (white arrows). c. Upregulation of CXCR4 expression of HCT116 by sphere formation. LoVo and HCT116 cells were maintained by monolayer (blue line) or floating (red line) culture for 7 days, and these cells were propagated and CXCR4 expression was examined by FACS analyses. Note that spontaneous expression of CXCR4 was seen in the monolayer of LoVo cells, but not in that of HCT116 cells, and CXCR4 expression was seen in both cell lines by sphere formation (arrow in sphere of HCT116). d. Mouse PCLF facilitated sphere formation of human colon cell lines in vitro. Floating cultivation of each tumor was done in fresh medium (without PLCF) as well as in medium that was used after intraperitoneal irrigation of nude mice (with PLCF). e. Mesenteric dissemination of LoVo and HCT116 was AMG3100- and mithramycin A-sensitive. Seven days after inoculation of LoVo-GFP or HCT116-CFSE into the peritoneal cavity of nude mice, the mesentery was subjected to analysis with a fluorescent microscope. Note that dissemination was mainly seen at the margin of the perivascular adipose tissue (panels, inset A and B), which was similar to the results using CT26 cells. These disseminations were significantly abrogated by AMG3100 or mithramycin A. f. SCF+-cell sheet-targeted dissemination of LoVo-GFP and HCT116-CFSE cells was assessed in an organ culture of the human mesentery samples. (panels) En face findings of the fluorescent dissecting microscopy. LoVo-GFP (the four upper panels) and HCT116-CSFE (the four lower panels) tumors (green) were observed on the SCF-positive cell sheet (red). (right graph) Percentage of tumor nodules on the SCF-positive cell sheet.

6. Supplementary Figure S5. a. b. c. 0h 1h 2h 3h 4h 5h 6h 9h 12h 24h 48h Hours after floating culture PCLF ( - ) ( + ) Monolayer Sphere Isotype CXCR4 Cell count AMD3100 untreated CFSE Inset Mithramycin Inset AMD3100untreatedMithramycin nodules/ vascular loops *P<0.01 * * Supplementary Figure S5. Sphere-induced spontaneous overexpression of CXCR4, acceleration of sphere formation by PCLF, and CXCR4-dependent peritoneal Dissemination is common mechanisms for HRA human ovarian cancer cells. a and b were done in triplicate, and showed similar results. a.Detection of the upregulation of CXCR4 expression of HRA via spheroid formation in vitro. Twenty-four hours after floating culture, the spheroids were subjected to FACS analysis after propagation with EDTA (5 mM). b. Sphere formation assay in vitro for HRA ovarian cancer cells under fresh-RPMI and PLCF-RPMI. c. Effect of AMD3100 and mithramycin A on HRA ovarian cancer cell dissemination. Seven days after drug pretreatment andCFSE-labelled tumor cell intraperitoneal inoculation to nude mice, the mesenteric nodules were counted under the fluorescent microscope. Mesenteric metastasis of HRA cells was significantly inhibited by the pretreatment with AMD3100 or mithramycin A.

7. Supplementary Figure S6. (key molecules) (key events) Supplementary Figure S6. Schematic representation of the whole molecular and cellular mechanisms of directional tumor dissemination as revealed in this study. After penetrating into the peritoneal cavity, which is rich with Coll-IV and pFN, through the serosa, tumor cells start to form some cell cluster by incorporating these ECMs, and then grow into spheroids within 6 h. Then, the Sp1 transcription factor begins to be activated and to upregulate CXCR4 in spheroids, which are mainly directed to the margin of the mesenteric perivascular adipose tissue to the SCF + /CXCL12 + niche-like cells to form disseminated tumor nodules.

8. Supplementary Figure S7. angiogenesis (VEGF, etc.) anti-apoptosis (Mcl-1, etc) invasive growth (uPA/uPAR, etc.) hypoxia (HIF-1, etc) starvation (IGF-1, etc) anaerobic metabolism (lactate, etc) inflammatory reaction (NF-  B, etc) others - gain stem cell properties - Sp1 activation adhesion molecules （ E-cad, ZO-1, etc ） → forming ‘shell’ on the sphere surface drug resistance CAR cell-niche directed tumor dissemination Supplementary Figure S7. Summary of biological events after the formation of cancer spheroids. Soon after forming spheroids, cancer cells start to express several adhesion molecules, including E-cadherin as well as ZO-1, to form ‘shell’ on the sphere surface, possibly resulted in drug resistance. Subsequently, based on several studies in the literature, inner cells demonstrate dynamic changes, including hypoxia (activation of HIF-1, etc.), starvation (expression of IGF-1R, etc.), anaerobic metabolism (accumulation of lactate, etc.), and inflammatory reactions (activation of NF-kB), causing possible angiogenic, antiapoptotic, and invasive properties via inducing VEGF, Mcl-1, and uPA/uPAR. Importantly, spheroid formation also facilitate cancer cells to gain stem cell-like properties, including upregulation of multi-drug resistant genes, as well as Sp1 activation causing niche-directed tumour dissemination, as revealed in this study.