Technical Summary: Inheritance Entropy in Cell Lineage Trees

1. Problem Statement

Quantifying the impact of hereditary transmission within cell lineage trees remains a fundamental challenge in biology. While single-cell profiling allows for the characterization of cell states, combining this with classic lineage tracing to track hereditary ramifications is difficult. A specific challenge arises with inactivity (cells that stop dividing, including senescence, differentiation, or quiescence). Unlike active proliferation, where hereditary mutations amplify the phenotype in progeny, inactivity is self-limiting: once a cell becomes inactive, its branch terminates, erasing downstream information. This makes it difficult to determine if the emergence of inactivity is driven by random stochastic events or by hereditary (epigenetic) factors that regulate cell-cycle exit. Furthermore, distinguishing between genetic mutations and epigenetic inheritance is crucial, particularly in stem cell colonies where inter-colony heterogeneity affects therapeutic potential.

2. Methodology

The authors introduce a model-independent method based on Shannon entropy applied to the topology of lineage trees, specifically focusing on the distribution of inactive cells.

A. Data Acquisition

The study utilizes high-definition single-cell lineage tracing of 32 human Bone Marrow Stromal Cell (BMSC) clonal colonies. Experiments were conducted via phase-contrast time-lapse microscopy until all replicating cells reached the seventh generation ($k_{max}=7$). Cells were classified as:

• Active: Observed mitosis.
• Inactive (G0/Dead): No division for 84 hours or cell death.
• Leaves: Cells at $k=7$ tracked only to birth.

B. Inactivity Imbalance

To quantify the topological impact of inactivity, the authors define the inactivity imbalance ($I_m$) for every mitosis $m$ in the tree:
$$I_m = |M_{m}^{left} - M_{m}^{right}|$$
where $M$ represents the number of missing $k_{max}$ leaves in the left and right branches. A large imbalance at a specific node suggests a divergence in the probability of inactivity between the two daughter branches, potentially indicating a mutation event upstream.

C. Inheritance Entropy

The core metric is the inheritance entropy ($S$), calculated from the normalized weights ($w_m$) of the imbalances across all mitosis in the lineage:
$$w_m = \frac{I_m}{\sum_n I_n}, \quad S = -\sum_m w_m \log w_m$$

• High Entropy: Imbalances are scattered homogeneously across the tree, suggesting random, non-hereditary fluctuations.
• Low Entropy: Imbalances are concentrated on a few nodes, indicating a specific hereditary event (mutation) that altered the probability of inactivity for a specific sub-branch.

D. Null Ensembles and Validation

To determine statistical significance, the biological tree's entropy is compared against three non-hereditary null ensembles:

1. Galton-Watson (GW): Constant probability of inactivity ($q$) for all cells.
2. Non-Homogeneous GW (NHGW): Generation-dependent probability of inactivity ($q_k$).
3. Random Reshuffling (RR): The preferred method. This involves scrambling the positions of cells within each generation while preserving the total number of inactive cells per generation. This severs all mother-daughter hereditary links while maintaining the tree's structural constraints.

The P-value is defined as the fraction of null ensemble trees with entropy $S \leq S_{biological}$. A low P-value indicates strong evidence of inheritance.

E. Model Fitting (SKH)

The authors propose a Stochastic Kirkwood-Holliday (SKH) model to describe the data. In this model:

• Wild cells mutate to a "mutated" state with probability $p$.
• Mutated cells (and their descendants) have a probability $q$ of becoming inactive.
• Crucially, the model incorporates a mutation lag ($l$), the number of generations between the mutation event and the phenotypic expression of inactivity.

3. Key Contributions

1. Inheritance Entropy: A novel, model-agnostic metric to detect hereditary structure in lineage trees based solely on the topology of truncated (inactive) branches.
2. Validation of Null Models: A rigorous benchmarking study demonstrating that while GW and NHGW models fail to distinguish hereditary trees from non-homogeneous non-hereditary trees (producing false positives), the Random Reshuffling (RR) method is robust, sensitive, and model-independent.
3. Mutation Lag Quantification: The ability to locate the specific mitosis where the probability of inactivity changed and to measure the mutation lag (the delay between the epigenetic change and phenotypic expression).
4. Stochastic Kirkwood-Holliday Model: A generalized stochastic model that successfully captures the hereditary features of BMSC lineages, including the lag between mutation and inactivity.

4. Results

• Hereditary Structure: In the experimental dataset of 32 BMSC colonies, 21 colonies (75%) showed a significant hereditary character ($P < 0.05$) using the RR test. The remaining colonies had too few inactive cells to yield a statistically significant result, not necessarily a lack of inheritance.
• Entropy Analysis: Biological trees consistently exhibited significantly lower entropy than their randomly reshuffled counterparts, proving that the distribution of inactive cells is not random but follows a hereditary pattern.
• Mutation Lag: The average mutation lag was found to be approximately 2 generations. This confirms that the epigenetic change leading to inactivity occurs upstream of the actual phenotypic expression.
• Model Fit: The SKH model, fitted to the data, yielded parameters $p \approx 0.1$ (mutation probability) and $q \approx 0.46$ (inactivity probability for mutated cells). The synthetic trees generated by SKH with these parameters closely resembled the topological structure of the experimental BMSC lineages.
• Spatial vs. Topological: Analysis confirmed that the clustering of inactive cells is not an artifact of physical proximity (spatial correlation) on the culture dish but is a topological feature of the lineage.

5. Significance and Claims

The paper claims that the observed heterogeneity in human BMSC colonies is driven by epigenetic hereditary factors regulating cell-cycle exit, rather than purely stochastic or genetic mechanisms.

• Robustness: The entropy-based test provides a simple, robust, and model-agnostic way to assess the hereditary origin of differential growth in any lineage tree.
• Clinical Relevance: The findings suggest that the inter-colony heterogeneity observed in BMSC transplants (a major hurdle for bone regeneration therapies) has an epigenetic hereditary origin. Understanding and controlling these hereditary mechanisms could improve the consistency of stem cell therapies.
• Methodological Advancement: By establishing a "mutation lag," the method allows researchers to back-track the origin of phenotypic changes even when the phenotype (inactivity) is self-erasing.
• Future Directions: The authors note that while the current method focuses on inactivity, the correlation between lineage topology and division times suggests that cell-cycle duration may also be regulated by epigenetic hereditary mechanisms, though this requires further study.

The authors remain modest, acknowledging that the SKH model is a simplification and that more complex models could better capture the variance in mutation lags. They emphasize that the method detects changes in probability; if a lineage has no variation in inactivity expression, the test will not detect inheritance, even if a hereditary mechanism exists.