Multi-region biopsies and patient-derived neurosphere cultures reveal spatial divergence in glioblastoma.

This study demonstrates that multi-region biopsies and derived neurosphere cultures from glioblastoma patients exhibit significant spatial heterogeneity in phenotypes, proliferation, and drug responses, highlighting the necessity of using multi-region models to accurately capture intratumor diversity for effective therapeutic testing.

The Gist

Imagine a Glioblastoma (GBM) tumor not as a single, uniform lump of bad cells, but as a bustling, chaotic city. In this city, different neighborhoods have different cultures, economies, and rules. Some areas are rich and crowded (the tumor center), while others are on the fringes, infiltrating the surrounding "normal" land (the tumor edges).

For decades, doctors trying to treat this "city" have made a critical mistake: they've only looked at one small neighborhood to understand the whole city. They take a single biopsy (a tiny sample), grow those cells in a lab dish, and test drugs on them. The problem? That single neighborhood might be totally different from the rest of the city. If a drug works on the "downtown" cells but fails on the "suburban" cells, the patient's tumor will keep growing in those other areas, leading to treatment failure.

This paper is like a team of urban planners who decided to map the entire city instead of just one block. Here is what they discovered, explained simply:

1. The "City Map" Approach (Multi-Region Biopsies)

Instead of taking just one sample, the researchers used MRI-guided surgery to take 40 different samples from 6 different patients. They grabbed pieces from the center of the tumor, the edges, and deep inside the brain tissue. They treated each sample like a unique neighborhood in the city.

2. The "Transplant" Experiment (Neurospheres)

They took cells from each of these 40 neighborhoods and tried to grow them in the lab (creating "neurospheres," which are tiny 3D balls of cells).

• The Old Belief: Scientists thought that once you put these cells in a lab dish, they would all forget where they came from and turn into the same generic type of cell (like everyone in a city suddenly wearing the same uniform).
• The Reality: The researchers found that the cells remembered their roots. The cells from the "downtown" tumor center acted differently than the cells from the "suburban" edges. They grew at different speeds, looked different, and even reacted differently to a special dye used during surgery (5-ALA).

3. The "Fluorescent Dye" Mystery (5-ALA)

During surgery, patients are given a special drink (5-ALA) that makes cancer cells glow under a blue light, helping surgeons see where to cut.

• The Surprise: The researchers found that some parts of the tumor glowed brightly, while other parts (often the deep, dangerous edges) were dim or didn't glow at all.
• The Danger: Because these "dim" cells don't glow, surgeons might leave them behind, thinking the tumor is gone. The study showed that these "invisible" cells are actually very aggressive and have different biological needs than the glowing ones.

4. The Drug Test (Why One Size Doesn't Fit All)

The team tested several drugs on these different cell lines.

• The Finding: A drug that killed the "downtown" cells might do nothing to the "suburban" cells.
• The Big Insight: Here is the most important part. The researchers tried to predict which drugs would work by looking at the cells in the lab dish. It was a bit of a mess; the lab cells had changed too much.
• The Solution: However, when they looked back at the original tissue samples (the "city map" before the cells were moved to the lab), they could predict the drug response much better. It's as if the "memory" of the tumor's original environment was still encoded in the cells, even if they looked different in the dish.

The Takeaway: Why This Matters

Think of treating GBM like trying to put out a fire in a massive, multi-story building.

• Old Way: You send a firefighter to the lobby, see the fire is small, and use a small hose. You miss the fire raging in the basement and the attic. The building burns down anyway.
• New Way (This Paper): You send firefighters to the lobby, the basement, the attic, and the side wings. You realize the fire in the basement needs water, but the fire in the attic needs foam. You also realize that the "invisible" smoke in the attic (the non-glowing cells) is the most dangerous part.

In simple terms: This study proves that to cure Glioblastoma, we can't just test drugs on one type of cell. We need to test them on a diverse mix of cells that represent the whole tumor. Furthermore, the best way to predict if a drug will work is to look at the original tumor's blueprint, not just the cells after they've been moved to a lab. This approach could help doctors choose the right "fire extinguisher" for the specific parts of the tumor that are most likely to survive and come back.

Technical Summary

1. Problem Statement

Glioblastoma (GBM) is characterized by extreme intratumor heterogeneity (ITH), which is a primary driver of therapeutic failure and recurrence. Current precision medicine approaches often rely on single-biopsy genomic profiling and drug screening using established cell lines (e.g., U87-MG) or single-region patient-derived cultures. These methods fail to capture the spatial diversity of the tumor ecosystem. Furthermore, it remains unclear whether in vitro neurosphere cultures derived from different spatial regions of the same tumor converge into a single phenotype due to culture selection pressures or if they retain the distinct phenotypic and transcriptional diversity of their spatial origins. This gap hinders the development of effective targeted therapies that account for the full spectrum of tumor heterogeneity.

2. Methodology

The study employed a multi-omics approach combining spatially resolved sampling, in vitro modeling, and computational biology:

• Patient Cohort & Sampling:
  • Cohort: 6 IDH-wildtype GBM patients.
  • Sampling: MRI-guided neuronavigation was used to collect 40 multi-region biopsies (4–8 per tumor) from distinct spatial locations (central, peripheral, deep margins).
  • 5-ALA Integration: Patients received 5-aminolevulinic acid (5-ALA) pre-surgery. Biopsies were categorized based on 5-ALA fluorescence intensity (5-ALAhigh vs. 5-ALAlow), a marker for heme biosynthesis defects in GBM cells.
• In Vitro Modeling:
  • Neurosphere Derivation: 30 primary neurosphere cultures were generated from the 40 biopsies using non-adherent conditions (Neurobasal media with EGF/FGF).
  • Phenotyping: Cultures were assessed for proliferation rates, morphology (spheres vs. adherent), and 5-ALA accumulation capacity.
• Genomic & Transcriptomic Profiling:
  • Single-Cell RNA Sequencing (scRNA-seq): Performed on sorted 5-ALAhigh and 5-ALAlow fractions from tumor OS1 to map cell type composition and infer Copy Number Alterations (CNAs).
  • Bulk RNA Sequencing: Performed on all 30 neurosphere lines and corresponding parental biopsies.
  • Whole Exome Sequencing (WES): Performed on biopsies to reconstruct clonal phylogenies and identify branching evolutionary events.
• Computational Analysis:
  • Deconvolution: Used CIBERSORTx to estimate tumor microenvironment (TME) composition and cell states (NPC-like, OPC-like, AC-like, Mesenchymal-like) from bulk data.
  • Phylogenetics: Used PhylogicNDT to reconstruct tumor evolution.
  • Machine Learning: Applied LASSO and Random Forest regression to predict drug response (specifically to the CDK inhibitor dinaciclib) based on gene expression profiles of both biopsies and neurospheres.
• Drug Screening:
  • Tested a panel of 7 drugs (including CDK inhibitors, DNA repair inhibitors, and PI3K inhibitors) on neurospheres to generate dose-response curves and Area Under the Curve (AUC) metrics.

3. Key Contributions

• Spatially Resolved Neurosphere Library: Created a unique resource of 30 patient-derived neurosphere lines representing distinct spatial locations within 6 GBM tumors.
• Validation of Phenotypic Retention: Demonstrated that in vitro neurospheres do not converge to a single phenotype; they retain spatial divergence in proliferation, 5-ALA metabolism, and transcriptional programs.
• TME vs. Intrinsic Signature: Showed that while bulk transcriptomic signatures of neurospheres diverge from parental biopsies (due to loss of TME), the drug response of the neurospheres remains strongly linked to the cellular composition and gene expression of the original tumor biopsy.
• Predictive Modeling: Established that machine learning models trained on parental tumor transcriptomes predict drug response (dinaciclib) more accurately (96.67% variance explained) than models trained on the neurosphere transcriptomes themselves (85.22%).

4. Key Results

• Clonal and Spatial Heterogeneity:
  • WES revealed 3–9 distinct clones per tumor with complex branching evolution. Specific clones (e.g., BRAF-mutant clones in OS21) showed significant spatial enrichment (central vs. peripheral).
  • scRNA-seq confirmed that deep tumor margins contain metabolically distinct cancer cells (5-ALAlow) with CNA evidence, which are often missed in standard resections.
• Phenotypic Divergence in Culture:
  • Neurospheres derived from different regions of the same tumor exhibited variable doubling times (4.8–18.6 days) and distinct morphologies.
  • 5-ALA Metabolism: The ability to accumulate 5-ALA was retained in culture. 5-ALA- lines showed downregulation of heme metabolism pathways, while 5-ALA+ lines showed enrichment in interferon-gamma response signatures.
  • Cell State Plasticity: While GBM subtypes (Proneural, Mesenchymal, Classical) became more homogenous in culture, specific cell states (e.g., Neural Progenitor-like vs. Astrocyte-like) remained spatially divergent.
• Transcriptional Shifts:
  • Cultures generally showed an expansion of the Mesenchymal phenotype and upregulation of EMT, stemness, and cell cycle pathways compared to biopsies.
  • However, the direction of transcriptional shift from biopsy to neurosphere was consistent within a patient but distinct across different patients, suggesting tumor-specific adaptations to in vitro conditions.
• Drug Response Correlations:
  • Drug sensitivity varied significantly between neurospheres from the same tumor.
  • TME Link: The abundance of specific TME cells in the original biopsy predicted drug response. For example, high microglia abundance correlated with poor response to temozolomide, while high pericyte abundance correlated with better response to PI3K inhibition (alpelisib).
  • Machine Learning: Models using parental tumor gene expression (specifically neuronal differentiation pathways) were superior predictors of in vitro dinaciclib resistance compared to models using neurosphere gene expression.

5. Significance

• Paradigm Shift in Drug Screening: The study challenges the standard practice of using single-biopsy-derived lines for drug screening. It argues that multi-region neurospheres are essential to faithfully model GBM heterogeneity and identify vulnerabilities across the entire tumor ecosystem.
• Clinical Relevance of Spatial Sampling: The findings suggest that the "parental" tumor phenotype (specifically its cellular composition and spatial gene expression) is a better predictor of therapeutic response than the in vitro phenotype of the derived cell line. This supports the use of multi-region biopsies for guiding precision medicine.
• Targeting Recurrence: The identification of 5-ALA- populations in deep margins (which retain stemness and specific metabolic profiles) highlights a reservoir of cells responsible for recurrence that are currently under-targeted by standard fluorescence-guided surgery and single-region profiling.
• Future Directions: The authors propose that future preclinical screens must account for spatial origin and 5-ALA status to improve the translation of drug discovery results to clinical outcomes. The provided neurosphere library serves as a critical resource for validating these spatially aware therapeutic strategies.