Module 7 · When Stem Cells Go Wrong

Cancer Stem Cells

Tumours are not clonal in the stem-cell sense. A human breast carcinoma contains cells that look alike under the microscope but differ greatly in their tumour-initiating capacity: only a fraction, when transplanted into an immunocompromised mouse, will seed a new tumour. That observation — that functional hierarchy exists within a genetically clonal tumour — is the cancer stem cell (CSC) hypothesis. It has therapeutic implications for why cancers relapse and how they resist treatment, and it sits at the crossroads of stem-cell biology and oncology.

1. Bonnet & Dick (1997): AML Hierarchy

John Dick’s lab in Toronto pioneered the human-AML xenotransplantation assay. In 1997, Dominique Bonnet and John Dick published the first rigorous demonstration of a cancer-stem-cell hierarchy in human cancer: only CD34⁺CD38⁻cells, a small subset of AML blasts, could reconstitute the leukaemia in NOD/SCID mice. The bulk AML cells, despite being malignant, could not. Those that could were phenotypically similar to normal HSCs.

The paper reshaped thinking about leukaemia — and opened the question of whether solid tumours also contain CSC hierarchies. They do (Al-Hajj 2003 breast; Singh 2003 brain; Ricci-Vitiani 2007 colon), though in variable proportions and with disputed universality.

2. CSC vs Clonal Evolution: Two Models of Heterogeneity

Two mechanisms generate tumour heterogeneity:

Clonal evolution (Nowell 1976): ongoing mutation and selection produce subclones with different behaviours. Each subclone is driven by one set of mutations; any cell in a subclone is equally tumorigenic.
CSC hierarchy: tumorigenic potential resides in a minority population of “stem-like” cells; bulk cells are transient amplifying / differentiated, limited in proliferation.

The two are not mutually exclusive. Modern understanding (Kreso & Dick 2014): most solid tumours have both. Genetic subclones coexist, each with its own CSC hierarchy, and the CSC identity is often plastic — non-CSC cells can convert to CSC state under stress (chemotherapy, hypoxia, radiation). This plasticity makes the CSC population a moving target, not a fixed subpopulation.

Simulation: Limiting-Dilution Xenografts & Relapse

Python

script.py56 lines

import numpy as np, matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt

# Two competing models of tumor heterogeneity:
# (a) Clonal evolution: all cells can seed a tumor, heterogeneity from mutation
# (b) CSC hierarchy: only rare CSCs seed; bulk are non-tumorigenic
# Simulate serial transplantation: both models predict tumor formation,
# but diluted cell assays distinguish

fractions = np.array([1.0, 0.1, 0.01, 0.001, 0.0001])
cell_numbers = 1e5 * fractions
# CSC model: tumor forms if any CSC is transferred, P(tumor) = 1 - (1 - f_csc)^N
# Assume f_csc = 0.001 (1 in 1000 cells is a CSC)
f_csc = 0.001
p_tumor_csc = 1 - (1 - f_csc)**cell_numbers
# Stochastic model: assume every cell has 10% chance per tumor seeding
p_tumor_stoch = 1 - (1 - 0.1)**cell_numbers

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(14, 5.5), facecolor='#0a0a1a')
for ax in (ax1, ax2):
    ax.set_facecolor('#111827'); ax.tick_params(colors='#cbd5e1')
    for s in ax.spines.values(): s.set_color('#334155')
    ax.grid(True, color='#334155', alpha=0.3)

ax1.semilogx(cell_numbers, p_tumor_csc, 'o-', color='#fbbf24', lw=2.6, markersize=10, label='CSC model (f=0.001)')
ax1.semilogx(cell_numbers, p_tumor_stoch, 'o-', color='#2dd4bf', lw=2.6, markersize=10, label='Stochastic/clonal evolution')
ax1.set_xlabel('Cells transplanted', color='#cbd5e1')
ax1.set_ylabel('P(tumor formation)', color='#cbd5e1')
ax1.set_title('Limiting-dilution transplantation assay',
              color='#fda4af', fontweight='bold')
ax1.legend(facecolor='#1e293b', edgecolor='#334155', labelcolor='#cbd5e1')

# Panel 2: treatment resistance schematic
t = np.linspace(0, 30, 300)
# Bulk cells (sensitive): die with half-life 3 days under therapy
# CSCs (resistant): survive, re-populate
N_bulk = 1e6 * np.exp(-t/3)
N_csc = 1e3 * np.ones_like(t)
# After therapy stops (t>20), CSCs regrow
regrow = np.where(t > 20, 1e3 * np.exp((t-20)/2), 1e3)
N_total = N_bulk + regrow
ax2.semilogy(t, N_bulk, color='#60a5fa', lw=2, label='Bulk tumor (sensitive)')
ax2.semilogy(t, regrow, color='#f87171', lw=2, label='CSCs (resistant + regrowth)')
ax2.semilogy(t, N_total, color='#fbbf24', lw=2.6, label='Total (apparent response + relapse)')
ax2.axvspan(0, 20, color='#94a3b8', alpha=0.1, label='Therapy')
ax2.set_xlabel('Days', color='#cbd5e1')
ax2.set_ylabel('Cell number (log)', color='#cbd5e1')
ax2.set_title('CSCs drive relapse after conventional therapy',
              color='#fda4af', fontweight='bold')
ax2.legend(facecolor='#1e293b', edgecolor='#334155', labelcolor='#cbd5e1')

plt.tight_layout()
plt.savefig('output.png', dpi=120, bbox_inches='tight', facecolor='#0a0a1a')
print('Dilution assay: CSC model predicts plateau, stochastic predicts linearity')
print('Resistance: CSCs survive chemotherapy due to quiescence + drug efflux + DNA repair')

Click Run to execute the Python code

Code will be executed with Python 3 on the server

3. CSC Markers Across Tumour Types

Tumour	CSC Markers	Reference
AML	CD34⁺ CD38⁻	Bonnet & Dick 1997
Breast	CD44⁺ CD24^−/lo ESA⁺	Al-Hajj 2003
Glioblastoma	CD133⁺	Singh 2003
Colon	CD133⁺, Lgr5⁺	Ricci-Vitiani 2007; Barker 2009
Pancreas	CD44⁺ CD24⁺ ESA⁺	Li 2007
Melanoma	ABCB5⁺ (contested)	Schatton 2008; Quintana 2008

Caveats: markers are tumour-type-specific, often unreliable across patients, and frequently mark both CSCs and some normal stem cells. Quintana 2008 argued that in melanoma, phenotypic heterogeneity is largely stochastic — improved xenograft conditions reveal tumour-forming potential in a large fraction of cells. The CSC concept is clearest in AML and glioblastoma; it is weakest in melanoma and some sarcomas.

4. CSCs and Therapy Resistance

CSCs resist conventional therapy through multiple mechanisms:

Quiescence: cytotoxic chemotherapy targets dividing cells; a quiescent CSC is simply not present for the drug to hit.
Drug efflux: ABC-family transporters (ABCG2/BCRP, ABCB1/MDR1) export chemotherapeutic drugs. High efflux is the basis of the side-population assay for CSC enrichment (Goodell 1996).
DNA repair: CSCs in glioblastoma show enhanced ATM/Chk1 responses to radiation (Bao 2006) — the basis of radio-resistance.
Metabolic flexibility: CSCs often rely on OxPhos and fatty-acid oxidation rather than the Warburg glycolysis of bulk tumour cells — different metabolic targets.
Niche protection: CSCs often reside in niches (perivascular in glioma, hypoxic/hyaluronan-rich in breast) that shelter them from therapy.

This explains the clinical observation that cytotoxic chemotherapy produces dramatic tumour shrinkage followed, frequently, by regrowth that is more aggressive and more resistant — the original bulk tumour has been debulked; the CSCs have not. Modern oncology treats this as the central therapeutic problem.

5. Therapeutic Targeting of CSCs

Several approaches:

Pathway inhibition: WNT, Hedgehog, and Notch inhibitors kill CSCs specifically (in principle). Vismodegib (Hedgehog) and Glasdegib are approved for basal-cell carcinoma and AML respectively. WNT inhibitors are in trials but generally toxic because normal intestinal and hair-follicle stem cells share the same signalling.
CD47 “don’t eat me” signal: the Weissman lab developed anti-CD47 antibodies (magrolimab) that license macrophage phagocytosis of CSC-enriched populations in AML and myelodysplastic syndrome.
Metabolic targeting: CSCs’ OxPhos dependence makes them vulnerable to complex I inhibitors (IACS-010759, in trials) and fatty-acid oxidation blockers (etomoxir derivatives).
Differentiation therapy: forcing CSCs to differentiate removes their self-renewal. ATRA in APL (M3 AML) is the historical prototype; similar approaches under development for other malignancies.
CAR-T & immunotherapy: CAR-T cells can target CD19 on leukaemic CSCs; the durability of CD19 CAR-T cures in B-ALL is consistent with their elimination of a CSC population.

6. Plasticity & the Dedifferentiation Route

Recent lineage-tracing work in mouse models (Blanpain, Fuchs, Wahl labs) has shown that differentiated cancer cells can revert to a CSC state under stress. In skin squamous cell carcinoma, differentiated tumour cells dedifferentiate during chemotherapy and resume CSC function. This plasticity complicates clinical predictions: even if one eliminated all current CSCs, the surviving bulk cells could generate new ones. The implication is that therapies targeting CSCs must simultaneously block the plasticity transition — often via EMT/MET regulators (ZEB1, SNAI1, TWIST) or YAP/TAZ-dependent mechanotransduction. This is one of the most active frontiers of translational cancer biology.

← Module 6 Module 8: Organoids →