Migration Phenotype of Brain-Cancer Cells Predicts Patient Outcomes

Abstract

Glioblastoma multiforme is a heterogeneous and infiltrative cancer with dismal prognosis. Studying the migratory behavior of tumor-derived cell populations can be informative, but it places a high premium on the precision of in vitro methods and the relevance of in vivo conditions. In particular, the analysis of 2D cell migration may not reflect invasion into 3D extracellular matrices in vivo. Here, we describe a method that allows time-resolved studies of primary cell migration with single-cell resolution on a fibrillar surface that closely mimics in vivo 3D migration. We used this platform to screen 14 patient-derived glioblastoma samples. We observed that the migratory phenotype of a subset of cells in response to platelet-derived growth factor was highly predictive of tumor location and recurrence in the clinic. Therefore, migratory phenotypic classifiers analyzed at the single-cell level in a patient-specific way can provide high diagnostic and prognostic value for invasive cancers.

Figure 1. Phenotypic screening of heterogeneous cell populations recapitulates the microenvironment of migrating cells

(A) Cells with heterogeneous phenotypes are isolated from a patient’s tumor (MRI of tumor for sample GBM612). (B) The cells are seeded on the platform that has a multi-well structure allowing testing of multiple conditions, and is an on-glass technology allowing direct imaging of migration and morphology with single-cell resolution. (C) Images show that GBM612 cells migrating on the platform have similar morphology and migration speed compared to GBM612 cells migrating in ex vivo human brain tissue and 3D Matrigel. In comparison, cells migrating on flat surfaces, are not polarized and have reduced migration speeds compared to the other cases (see supplemental figures) (scale bars 25 µm, duration between each frame 1 hr). (D) ROCK inhibitor (Y-27632) effect on migration of GBM612 cells on tissue-mimetic substrates and flat surfaces. (n = 30 cells, mean + s.e.m., ***p < 0.0001, *paired against Control group, #paired against 3 µM group, Kruskal-Wallis One Way ANOVA on Ranks, Dunn’s Method). (E) Platform provides important information on the migration response of heterogeneous cell populations. GBM612 samples show a subpopulation of cells that migrate fast and stable over time.

Figure 2. Migratory response to PDGF correlates with tumor characteristics both in vitro and in vivo

(A) RT-PCR analysis of PDGFRα mRNA levels in responding sample GBM 276 and non-responding sample GBM 253 (n = 3, *p = 0.006, Student’s t-test). (B) Western blot for PDGFRα protein expression in GBM samples grown as adherent or spheroid cultures. (C–D) Quantification of migration speed of cell lines GBM253 and GBM276 in the presence of PDGF-AA (50ng/mL) and Imatinib (30 µM) (n ~ 80 cells, mean + s.e.m., *p < 0.05, **p < 0.01, ***p < 0.001, *paired against Control group, #paired against PDGF group, Kruskal-Wallis One Way ANOVA on Ranks, Dunn’s Method). (E–F) Migration speed of GBM276 for the slowest and fastest quartile of the cells (i.e., 25% of the slowest and fastest moving cells), showing that only a subpopulation responds to PDGF. (G) Quantification of migration measured by alignment of GBM276 cells (*p < 0.05, Kruskal-Wallis One Way ANOVA on Ranks, Dunn’s Method. Asterisks indicate comparisons to all other conditions). (H) Survival curves of mice injected with GBM276 cells cultured in control spheroid conditions or in the presence of PDGF-AA (n = 4 mice each group, *p = 0.0097, Gehan-Breslow-Wilcoxon Test). (I–J) Kaplan-Meier plots based on clinical TCGA data of GBM patients, comparing survival between high and low expression of PDGF-AA (I) and PDGFRα (J), respectively. The cohorts were divided at the median of the expression level of the respective gene.

Figure 3. Information on migration speed reveals important differences between patient samples in response to PDGF

(A) Analyzing the fastest quartile (GBM499) reveals that the subpopulations display a significant response to PDGF while for the whole population there is no significant response. (B) Migration speed timelapse demonstrates that sample GBM501 does not respond to PDGF at all times (compared to GBM609); on average, however, both samples respond significantly to PDGF. (C) Both GBM630 and GBM544 samples display significant increase in average migration speed (*p < 0.05, **p < 0.01, ***p < 0.001, Wilcoxon rank-sum test). However, GBM630 has significantly larger number of fast outliers, while GBM544 displays a uniform increase in speed. (D) The platform allows patient samples classification based on multiple characteristics, permitting a better description of the heterogeneity of the samples. We compare 14 patients GBM cell lines. The samples were grouped based on whether there is a significant increase (p < 0.05, Wilcoxon rank-sum test) in average migration speed in response to PDGF (Group I), and whether this significant increase is persistent over time (Group II). Furthermore, samples are grouped based on the number of outliers (cells faster than the fastest cells in the control group) with a threshold of four cells (5%, for a total cell number of 80) (Group III). For reference, the PDGFRα protein expression level (Group IV), and the subclass of the GBM cells is also provided (Group V).

Figure 4. GBM migratory response to PDGF correlates with patient tumor characteristics

(A–B) Kaplan-Meier plots comparing recurrence between PDGF-responsive and unresponsive groups (n = 11 patient tumors, consensus group) (A) and based on the criteria in Figure 3D (n = 14 patient tumors, weak and strong responders) (B). P-values were calculated using a two-tailed log-rank (Mantel-Cox) test. (C–D) Magnetic Resonance Imaging (MRI) scans of patients with tumors from the unresponsive group (C) and responsive group (D). (E) Distribution of PDGF-responsive and unresponsive tumors that formed in specific locations in the brain (n = 11 patient tumors, Barnard’s exact test). (F) Time to recurrence for GBM samples separated into low directionality (< 3.25) and high directionality (≥ 3.25) groups. (n ≥ 4, mean + s.e.m, Wilcoxon rank-sum test. Threshold of 3.25 was determined using linear discriminant analysis).