Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2016 Apr 11;29(4):563-573.
doi: 10.1016/j.ccell.2016.03.012.

Single-Cell Phosphoproteomics Resolves Adaptive Signaling Dynamics and Informs Targeted Combination Therapy in Glioblastoma

Affiliations
Free PMC article

Single-Cell Phosphoproteomics Resolves Adaptive Signaling Dynamics and Informs Targeted Combination Therapy in Glioblastoma

Wei Wei et al. Cancer Cell. .
Free PMC article

Abstract

Intratumoral heterogeneity of signaling networks may contribute to targeted cancer therapy resistance, including in the highly lethal brain cancer glioblastoma (GBM). We performed single-cell phosphoproteomics on a patient-derived in vivo GBM model of mTOR kinase inhibitor resistance and coupled it to an analytical approach for detecting changes in signaling coordination. Alterations in the protein signaling coordination were resolved as early as 2.5 days after treatment, anticipating drug resistance long before it was clinically manifest. Combination therapies were identified that resulted in complete and sustained tumor suppression in vivo. This approach may identify actionable alterations in signal coordination that underlie adaptive resistance, which can be suppressed through combination drug therapy, including non-obvious drug combinations.

Conflict of interest statement

No potential conflicts of interest were disclosed by the other authors.

Figures

Figure 1
Figure 1
Characterization of GBM39 in vivo mouse model. (A) Tumor growth curve for control (sample size n=11), and treated (n=14) xenografts (n=7 for responsive and resistant groups respectively; variations expressed as mean +/− SD; *p< 0.05, with far right comparing sizes at day 19 versus day 39). (B) 18F-FDG PET (left), PET-CT (middle), and CT (right) scans for the three conditions (n=4 per condition). The arrow indicates the localization of the tumor. PET and CT images demonstrate significantly decreased metabolic burden and tumor volume in the responsive state. (C) IHC results for the three conditions. Scale bar: 100 μm. See also Figure S1 and Table S1.
Figure 2
Figure 2
Genomic analyses of GBM39 PDXs. (A) Whole genome SNPs and WES analyses reveal characteristic CNVs and SNVs in untreated GBM39 PDXs. There is no genetic variation between control and resistant samples for the three major TCGA pathways (RTK, TP53 and RB1). (PIK(3)K*: PIK3R1 M56I/M26I/M326I; PIK3C2G P146L (damaging, deleterious), A261E/A39E, L669F/L710F/L488F (damaging, deleterious), non-frameshift deletion exon 2 c.385_387; PIK3C2B P273H (damaging); PIK3R2 S313P, A727T (damaging)). (B) Copy number plot across the chromosomes for control (C) and resistant (R) PDXs, highlighting their genome profile similarity. See also Figures S2 and S3 and Tables S2–S5.
Figure 3
Figure 3
Assay protocols and single cell proteomic analysis of GBM39 tumors. (A) EGFR+ cells were separated from the GBM39 models and loaded onto an SCBC. Two SCBCs were run in parallel for each test condition. (B) Background subtracted SCBC data represented as one-dimensional scatter plots (mean +/− SEM was overlaid for each protein and is indicated by the black horizontal bar). Grey bars indicate the background level of each protein assayed. Statistical uniqueness is evaluated by two-tailed Mann-Whitney test for pairwise comparison (black stars) and Kruskal-Wallis test for comparison among three groups (blue stars, NS, not significant; *p < 0.05; **p < 0.005; ***p < 0.0005). (C) Digitized IHC results for selected proteins assayed from tumor tissue slides (mean +/− SD). (D) Immunoblots of various proteins from bulk assays of GBM39 PDXs. See also Figure S4.
Figure 4
Figure 4
Statistical analysis of single cell data informs targeted combination therapies. (A) Protein-protein correlation networks, extracted from SCBC data. Average protein levels are reflected in the sphere diameters, while correlation strengths are reflected in the thickness of the edges (see key). For the resistant state, existing, new, and lost correlations, relative to control, are indicated (see key). The relative change in average protein abundance from SCBC data is shown in the bar graphs above (p-P70S6K level is scaled by 1/4 on this graph). (B) Correlations between key functional proteins and the PC1 for the control and responsive states from in vivo and in vitro drug treatment tests. In both cases, two independent signaling modes are identified. (C) Quantification of the heterogeneity of GBM39 PDXs cells at the three stages. See also Figure S5.
Figure 5
Figure 5
In vivo validation of 7 mono- or combination therapies. (A) In vivo test results for the 7 mono- or combination therapies based upon the predictions from the SCBC data analysis. All 7 predictions proved correct (data are shown as mean +/− SD; n=11 for vehicle, n=6 for C, n=4 for D, n=4 for U, n=4 for each combinatorial treatment group. **p<0.005 relative to samples after treatment stop versus responsive samples, ***p<0.001 relative to responsive samples versus vehicle samples). (B) IHC images of drug targets for the combinatory treatments of CC214-2 and ERK and/or Src inhibitors,. Scale bar: 100 μm. See also Figures S1 and S6 and Table S1.
Figure 6
Figure 6
PLS modeling confirms the presence of independent signaling modes. (A) Validation of the PLS modeling: the calibration phase of the model was constructed by randomly choosing part of observations (orange part). The first two PCs were used to perform a leave-one-out cross-validation to assess the model stability. The established model was then employed to predict the TGR at sacrifice and cell cycle metric τ for the remaining observations (blue part). (B) Correlations of the IHC assayed proteins, as well as the functional observations, with PC1 and PC2. TGR strongly correlates with mTOR effectors (mode 1: red), while p-Src and p-ERK1 (mode 2: blue) largely dominate PC2, constituting an independent signaling mode that accounts for adaptive response to mTOR kinase inhibitor. (C) PLS modeling shows that for effective drug combinations projections are qualitatively different from linear superposition of individual drugs, which in turn implies that synergistic drug combinations do not simply act in a linearly additive manner. See also Table S6.
Figure 7
Figure 7
Analysis of a low-passage pcGBM2 patient-derived GBM cell line. (A) Single-cell data represented as one dimensional scatter plots for control, lapatinib and XL-765 + trametinib (XL-765+T) treated samples. The cells were stimulated with 100 ng/mL EGF at 37°C for 10 min before on-chip cell lysis. The average fluorescence intensity with SEM is overlaid for each protein (black horizontal bars). Statistical uniqueness is evaluated by two-tail Mann-Whitney test for pairwise comparison (*p<0.05; **p<0.005; ***p<0.0005; NS: not significant). The down regulated p-EGFR levels shown in the inset indicate that lapatinib has successfully engaged its target. (B) Correlation between key functional proteins and the PC1 for control, lapatinib and XL-765+T treated samples. (C) The in vitro validation confirms that adapted tumor cells are more resistant to lapatinib treatment (data are shown as mean +/− SD). Student’s T-test is used to evaluate the statistical significance (*p<0.05). (D) Cell viability test shows that XL-765+T can induce significant cell death after 60 hr in vitro treatment. See also Figure S7.

Comment in

Similar articles

See all similar articles

Cited by 52 articles

See all "Cited by" articles

Publication types

MeSH terms

Feedback