Supplementary Figure Legends

Supplementary Figure S1. Characterization of cancer spheroids. A, Bright field images of H1650 cancer cell spheroids (1,000 cells/well) were monitored at the indicated time-points. B, Growth curve of H1650 cancer cells in 2D and 3D culture system. On indicated days, 2D cells and 3D spheroids were dissociated to single cells by the cell detachment buffer and cell number was counted. C, Correlation between cell number seeded and spheroid area. The spheroid area was calculated using Nikon microscopy program. Correlation coefficient (r) was 0.98. D, Expression levels of the indicated proteins were compared between 2D and 3D culture system by Western blot analysis at the indicated time points. E, qRT-PCR analysis of hypoxia-induced gene expression. Relative gene expression level, including HIF-1a and VEGF, was normalized by that of 2D monolayer cells. F, Hypoxia status in cancer spheroids. Spheroids, cultured for 5 days in 3D plate, were incubated with LOX-1, the 3D permeable hypoxia probe, for 24 h (left). LOX-1 staining was also conducted on 2D monolayer cells as a control (right). Red fluorescence was monitored with EMCCD. G, Luminal cancer cell survival in spheroids. Luminal cell survival was analyzed by DAPI staining in cancer spheroid. The upper and lower panels are non-sectioned and sectioned images, respectively (10 mm).

Supplementary Figure S2. Characteristics of spheroids based on morphologic classification. A, Cell growth comparison of spheroid-type representative cancer cells in 2D and 3D culture. 2D and 3D cultured cells were disrupted by cell detachment buffer and cell number was counted at time interval. B and C, Hypoxia status in 3D (B) and 2D (C) culture system was observed with LOX-1 staining. Hypoxia status in 3D and 2D culture system was monitored on day 6 and 2, respectively, after cell seeding. Fluorescence signal was monitored by fluorescence microscopy. D, Bright field images of colon cancer cell line spheroids. Morphological type-representative cells (LoVo: R type, SW480: A type) were seeded on the 3D plate and then bright field images were monitored by microscopy at 6 days. E and F, Hypoxia status in colon cancer spheroids. LoVo and SW480 cells were cultured in 3D culture plate for 5 days. qRT-PCR analysis of hypoxia-induced gene expression, including HIF-1a E) and VEGF (F) relative to GAPDH level. Relative gene expression of spheroid was normalized to that of 2D cultured cells.

Supplementary Figure S3. Validation of spheroid morphology-dependent characteristics. Nine different cancer cell lines (lung cancer: H441, H23, Calu-6; colon cancer: LoVo, SW480, Colo205; GBM: LN18, U251MG, U251E4) were used to validate spheroid morphology-dependent characteristics. A and B, Cell growth was determined by cell counting in 2D (A) and 3D (B) culture system. C, Using 2D and 3D culture system, hypoxia status and drug sensitivity of 5-FU were determined by the LOX-1 staining and LDH cell viability assay, respectively. To quantify hypoxia, mean fluorescence intensity was measured by fluorescence microscopy. P values of hypoxia and drug sensitivity were determined by one-way and two-way (spheroid type x culture system) ANOVA, respectively. *P<0.01 (one-way ANOVA), **P<0.01 (two-way ANOVA)

Supplementary Figure S4. Validation of PLS-DA model. A, Selection of discriminatory genes. Among 22,285 genes with CNV, mutation, and mRNA expression data in 70 cell lines (1st Venn diagram), 2,202 discriminatory genes were selected by MPLS-DA (P<0.05; 2nd Venn diagram). Of them, 560 showing subtype-specific molecular patterns were further identified by individual PLS-DAs (3rd Venn diagram). B-D, mRNA expression, CNV and mutation profiles. Columns include 70 tested cancer cells (R, M, A, N for round, mass, aggregate and none spheroid types, respectively, across the top) and rows include tested genes for each data types (926, 65, and 1,333 genes for mRNA (B), CNV (C) and mutation (D) profiles, respectively, down the side). Colors indicate increase (red) and decrease (green) in the levels of molecular signatures relative to the median levels. Dendrograms were generated using 2-D hierarchical clustering (complete linkage and Euclidean distance as a similarity measure). E and F, Accuracy (E) and misclassification error rate (F) during leave-one-out-cross-validations (LOOCVs) of the PLS-DA model. The model was constructed with 560 subtype-specific discriminatory genes in the prediction of spheroid subtypes. G, Block contribution of MPLS-DA.

Supplementary Figure S5. Network models describing cellular processes associated with spheroid subtypes. A, Network associated with round type-specific genes. B and C, Network models denoting cellular processes overrepresented by mass type-specific genes (B) and aggregate type-specific genes (C) are also provided. In the networks, the nodes with the same GOBPs or KEGG pathways were grouped together. The asterisk symbols were attached to the labels of the GOBPs or KEGG pathways overrepresented by round (186), mass (158) and aggregate (152) type-specific genes. Node colors represent VIPs of mRNA expression (center) and CNVs (boundary). Node labels represent the presence of mutations (bold and purple). Underline text represents cancer drug targets reported in NCI chemical index. Blue edges represent protein-protein interactions selected from HotNet and obtained from the databases with experimental evidence (HitPredict, STRING, and GeneGO). Purple lines denote signaling pathways obtained from KEGG pathway database.

Supplementary Figure S6. Drug penetration test in spheroids using bimolecular fluorescence complementation (BiFC). A, Significance of the JAK-STAT signaling in the round type spheroids. Bright field images of the round type spheroids of H226 and H441 cells are shown in the absence (con) or presence of AG490 (JAK2 inhibitor, 150 mM) for 3 days. B, 293T cells expressing either or both of VN-FKBP12 and VC-FRB were culture for 3 days to form spheroids. The proteins were then extracted from the spheroids and the expression levels of VN- FKBP12 and VC-FRB were determined by immunoblotting. C, 293T cells expressing each BiFC vector for FKBP12 and FRB in 2D system were seeded on 3D plate (5,000 cells/well). After 3 days incubation, the spheroids were incubated with or without rapamycin (1 mM) and the BiFC signal was monitored with fluorescence microscopy. D, 293T cells were transfected with BiFC vectors. Using these cells, spheroids were formed and then treated with or without AG490. After 2 days incubation, spheroids were treated with rapamycin. S6K phosphorylation was monitored by immunoblotting.

Supplementary Figure S7. Molecular signature of GBM patient-derived round type spheroid. A, Selection of the discriminatory genes specific to the round type spheroids. For the 186 round type-specific genes, the arrays include 167 genes (164 and 150 genes on mRNA expression and CNV arrays, respectively; 1st Venn diagram). Among them, 36 (each 20 identified by mRNA expression and CNV) were selected as the round type-specific discriminatory genes in GBM patients-derived cells (2nd Venn diagram). B, mRNA expression (left panel) and CNV (right panel) profiles. Columns represent GBM patient-derived spheroids, and rows represent differential levels of molecular signatures. Colors indicate increase (red) and decrease (green) in the levels of molecular signatures relative to the median levels. C, The genes overlapped between the JAK-STAT signature from cell lines and the 36 genes from patient-derived R type spheroids (a) and their interactors (b). The significance (P value from the hypergeometric test) of the overlaps is indicated. D, Integration of the 36 genes identified from patient-derived R type spheroids into JAK-STAT pathway in the R type-specific network (Fig. 2C). Round and diamond nodes are the genes in the R type-specific network and the 36 genes, respectively. Node colors represent the increase (red) and decrease (green) in mRNA or CNV levels, as well as the decrease in CNV levels, but the increase in mRNA levels (blue). Gray edges represent protein-protein interactions, and purple lines denote activations (arrows) and inhibitions (inhibition symbols) obtained from KEGG pathway database.

Supplementary Figure S8. Xenograft model to validate JAK-STAT pathway for drug sensitivity. A, H1650 xenografted mouse weights were measured every 2 days. B, After sacrificing mice, two representative tumors isolated from each group were shown.


Supplementary Table S1. Summary of cancer cell lines and their spheroid morphology. Spheroid type, tissue origin, and genomic data of 100 different cancer cell lines were represented.

Supplementary Table S2. Percent variance explained by PLS model of 560 type specific discriminatory genes. We determined the number of latent variables (LVs) to 6 so that the PLS model captures more than 80% of variance of Y-block; LV1-10, 1st to 10th PLS LVs we tested; X1-, X2- and X3-blocks, mRNA expression data, mutation data and CNV data, respectively; Y-block, data matrix containing the subtypes of the 70 cancer cells. ‘This LV’ includes a percent variance captured by an LV in X1-, X2-, X3- or Y-blocks which reflects the amount of the information in X1-, X2-, and X3-block used to explain the amount of the separation explained by the LV among the binary values in Y-block. ‘Total’ includes cumulative sum of percent variance captured by LVs from LV1 to the specific LVs we tested.

Supplementary Movie S1. Spheroid formation of cancer cells. A-C, H1650 (A), H460 (B) and A549 cells (C) (1,000 cells/well) were seeded on the 3D culture plate. The cells were imaged in 2 h after seeding on live cell imager equipped with x10 objective lens. Images were taken every 10 min for 36 h, and the movie is played 15 frames per second.

1