Part VI · Heterogeneity, evolution, and metastatic biology · Chapter 31

Plasticity and non-genetic heterogeneity

Not everything that survives treatment has changed its genome. The difference is settled by an experiment, not by inspection.

1 · Epithelial to mesenchymal transition and hybrid states

Everything in this chapter turns on a single test, and it should be stated before any biology.

Take a population that survived treatment. Passage it without the drug for several generations. Challenge it again. A genetic change is heritable and persists through drug-free passage. A state change reverts, and sensitivity returns. That test separates the two situations. Inspecting the cells does not, because a resistant clone and a tolerant state can look identical at the moment of progression. Genetic evolution and reversible cell state change as different problems raised this distinction. This chapter develops the half of it that reverts.

Epithelial to mesenchymal transition is the best characterised of these states, and the most often misdescribed. It is not a switch between two conditions. Screening cell surface markers across skin and mammary primary tumours in mice identified multiple subpopulations spanning fully epithelial to fully mesenchymal, passing through intermediate hybrid states¹. Those subpopulations had similar tumour-propagating capacity. They differed in plasticity, invasiveness and metastatic potential.

Hybrid states are the interesting ones because they retain both capacities at once. A cell that keeps some cell-cell adhesion can move as part of a group rather than alone, which matters for the seeding argument in Circulating tumor cells and the inefficiency of seeding.

The clinical counterpart comes from circulating cells. Mesenchymal markers were rare in primary tumour cells and highly enriched in circulating tumour cells from patients with breast cancer². In one patient followed serially, the balance shifted back and forth with each cycle of response and progression.

2 · Cancer stem cell models and dedifferentiation

The original cancer stem cell model was hierarchical. A small self-renewing compartment sits at the top, generates non-renewing progeny, and the arrow points one way. If that were correct, eliminating the apex would empty the tumour.

Two experiments in breast cells broke the arrow. Basal-like human mammary epithelial cells spontaneously dedifferentiated into stem-like cells, and oncogenic transformation increased the rate of that conversion³. Separately, subpopulations sorted for a single phenotypic state returned towards the original equilibrium proportions over time in culture, behaviour captured by a model in which cells transition stochastically between states⁴.

The consequence is specific and it is not a quibble about nomenclature. A compartment that refills itself cannot be depleted by killing its current members. Any therapy directed at a stem-like state has to either kill the state faster than it regenerates or block the transition that regenerates it. Those are different drugs.

This also reframes what a tumour-initiating assay measures. It reports the probability that a cell in a given state initiates growth in a given host. It does not establish that the state is fixed, and the dedifferentiation data show that it is not. The epigenetic machinery that makes such states heritable over short timescales is set out in Epigenetic memory, heritable drug tolerance, and reversibility.

3 · Stem-like states, CD44 and CD24 phenotype and ALDH activity

Two markers dominate this literature and they do not identify the same cells.

The first is a surface phenotype. Cells sorted as CD44-positive and CD24-negative or low, with lineage markers absent, were tumorigenic in immunodeficient mice in eight of nine patient samples⁵. As few as 100 such cells formed tumours. Tens of thousands of cells with other phenotypes did not. The tumours that grew contained both the sorted phenotype and the phenotypically diverse non-tumorigenic populations of the original.

The second is an enzyme activity. High aldehyde dehydrogenase activity identified a tumorigenic fraction capable of self-renewal and of regenerating the heterogeneity of the parent tumour. In a series of 577 breast carcinomas, ALDH1 detected by immunostaining correlated with poor prognosis⁶.

Three measurement points belong with these findings, and they are why the markers have not become clinical tests.

The readout is engraftment in an immunodeficient mouse. That measures the capacity to form a tumour in a host without adaptive immunity. It is not the same quantity as the capacity to cause relapse in a patient whose immune system is intact, which Immune-mediated dormancy and natural killer cell control shows is an active constraint.

The gates are set by the operator. A continuous flow cytometry distribution is being cut into categories, and the cut point is a convention in the same sense as the one described in Definitions and thresholds, spatial, clonal, and the ten percent convention.

The two markers overlap only partially. A study reporting on one has not reported on the other.

4 · Drug-tolerant persister cells and reversible tolerance

Persisters are the clearest worked example of the test in Epithelial to mesenchymal transition and hybrid states, because they were defined by it.

Exposing drug-sensitive tumour cell lines to a range of anticancer agents reproducibly left a small subpopulation with more than 100-fold reduced drug sensitivity⁷. That subpopulation was maintained by IGF-1 receptor signalling and by an altered chromatin state requiring the histone demethylase KDM5A. The state was acquired and relinquished at low frequency by individual cells. It was not a fixed property of a lineage.

The mechanism was then tested directly in breast cancer. In triple-negative and basal-like models, drug-tolerant states induced by MEK and by PI3K and mTOR inhibitors arose through cell-state transitions rather than through selection of pre-existing subpopulations⁸. Those transitions involved remodelling of open chromatin. Co-treating with a BET inhibitor prevented the chromatin change, prevented acquisition of the tolerant state, and produced cell death and xenograft regression.

Persisters also acquire a dependency they did not previously have. Across a range of cancers and treatments, persister cells became dependent on the lipid hydroperoxidase GPX4, and loss of GPX4 function killed them by ferroptosis and prevented relapse in mice⁹.

5 · Lineage switching, ER loss, neuroendocrine and squamous transformation

Lineage switching is plasticity taken to its limit. The cell does not merely alter a program. It stops being the kind of cell that the therapy was designed against.

Loss of oestrogen receptor expression under endocrine pressure is the version a breast oncologist meets most often. The biology of losing endocrine dependence is developed in Lineage plasticity and loss of endocrine dependence, and the transcriptional reprogramming that permits it in Lineage plasticity and transcriptional reprogramming under therapy. What belongs here is the interpretive problem at the bedside.

A receptor-negative result on a biopsy taken at progression has at least three explanations, and they carry different consequences.

It may be a selected pre-existing receptor-negative compartment, which is the mechanism in Temporal heterogeneity and clonal evolution. It may be a genuinely reprogrammed population that previously expressed the receptor. It may be a measurement artefact of fixation, antibody, block or field, which is the subject of True biological conversion, assay variability, and sampling error.

No routine assay separates these. The first and second differ in whether the population was ever sensitive. The third differs in whether anything happened at all.

Transformation to neuroendocrine or to squamous morphology occurs in breast cancer and is described in Mesenchymal, metaplastic, and rare biological states. Reported frequencies vary with how hard anyone looked, which in practice means with how often a progressing lesion was rebiopsied and examined by a pathologist expecting the finding. A frequency estimated from case reports is a frequency of reporting.

The practical rule is narrow. A morphology or receptor result at progression that contradicts the original diagnosis is worth confirming and worth acting on, and it is not by itself evidence that the original diagnosis was wrong.

6 · Transcriptional noise, bet-hedging, and stochastic resistance

If a state is entered and left stochastically, then a clonal population is never uniform, and the variation is the point rather than the error term.

The cleanest demonstration is in melanoma. Single-cell analysis showed profound transcriptional variability that predicted which cells would later resist treatment, with infrequent semi-coordinated high-level transcription of resistance markers in a very small percentage of cells¹⁰. Adding drug then induced epigenetic reprogramming in those cells, converting the transient transcriptional state into a stable resistant one. Resistance was a two-stage process, and only the second stage was durable.

That is bet-hedging in the ecological sense. A population maintains a minority in a state that is costly under normal conditions and protective under a shock it cannot predict. The equilibrium behaviour of breast cancer populations is consistent with the same logic⁴.

The separation between biology and measurement is unusually sharp here. Single-cell data contain technical variability from capture efficiency and dropout, so observed heterogeneity in a single-cell experiment is not by itself evidence of biological heterogeneity. What makes the melanoma result a biological claim is that the variability predicted a later outcome. Noise that predicts nothing is noise.

The implication for A framework for heterogeneity is that this axis of heterogeneity is invisible to every assay that averages. Bulk sequencing of a tumour in which one cell in a thousand is primed to survive reports the other 999.

7 · Targeting the plastic state rather than the resistant clone

If the problem is a state, the therapeutic options divide into two, and they are usually confused.

The first is to kill cells while they occupy the state. High-throughput screening against breast cells driven into a stem-like state identified salinomycin¹¹. It reduced the proportion of those cells by more than 100-fold relative to paclitaxel in the same assay. In mice it inhibited mammary tumour growth. The GPX4 dependency of persisters is the same strategy against a different state⁹.

The second is to prevent entry. Blocking the chromatin remodelling that accompanies the transition stopped the tolerant state from forming rather than killing cells that had reached it⁸.

These differ in timing and in what they are combined with. A state-killing agent is given alongside or just after the pressure that creates the state. A transition-blocking agent has to be present before it.

What has changed practice is a timing argument rather than an agent. Acting when the surviving population is smallest is the principle behind post-neoadjuvant escalation and behind the residual disease strategies in Dormancy, residual disease, and late recurrence and Dormancy-directed strategies.

References

1. Pastushenko I, Brisebarre A, Sifrim A, et al. Identification of the tumour transition states occurring during EMT. Nature 2018 556:463-468. PMID 29670281
2. Yu M, Bardia A, Wittner BS, et al. Circulating breast tumor cells exhibit dynamic changes in epithelial and mesenchymal composition. Science 2013 339:580-584. PMID 23372014
3. Chaffer CL, Brueckmann I, Scheel C, et al. Normal and neoplastic nonstem cells can spontaneously convert to a stem-like state. Proc Natl Acad Sci U S A 2011 108:7950-7955. PMID 21498687
4. Gupta PB, Fillmore CM, Jiang G, et al. Stochastic state transitions give rise to phenotypic equilibrium in populations of cancer cells. Cell 2011 146:633-644. PMID 21854987
5. Al-Hajj M, Wicha MS, Benito-Hernandez A, Morrison SJ, Clarke MF. Prospective identification of tumorigenic breast cancer cells. Proc Natl Acad Sci U S A 2003 100:3983-3988. PMID 12629218
6. Ginestier C, Hur MH, Charafe-Jauffret E, et al. ALDH1 is a marker of normal and malignant human mammary stem cells and a predictor of poor clinical outcome. Cell Stem Cell 2007 1:555-567. PMID 18371393
7. Sharma SV, Lee DY, Li B, et al. A chromatin-mediated reversible drug-tolerant state in cancer cell subpopulations. Cell 2010 141:69-80. PMID 20371346
8. Risom T, Langer EM, Chapman MP, et al. Differentiation-state plasticity is a targetable resistance mechanism in basal-like breast cancer. Nat Commun 2018 9:3815. PMID 30232459
9. Hangauer MJ, Viswanathan VS, Ryan MJ, et al. Drug-tolerant persister cancer cells are vulnerable to GPX4 inhibition. Nature 2017 551:247-250. PMID 29088702
10. Shaffer SM, Dunagin MC, Torborg SR, et al. Rare cell variability and drug-induced reprogramming as a mode of cancer drug resistance. Nature 2017 546:431-435. PMID 28607484
11. Gupta PB, Onder TT, Jiang G, et al. Identification of selective inhibitors of cancer stem cells by high-throughput screening. Cell 2009 138:645-659. PMID 19682730