Identifying cell-to-cell variability in internalization using flow cytometry

Alexander P. Browning, Niloufar Ansari, Christopher Drovandi, Angus P. R. Johnston, Matthew J. Simpson, Adrianne L. JennerJournal of The Royal Society Interface 2022 · DOI

TL;DR

This paper introduces a novel mathematical and computational framework to quantify cell-to-cell variability in biological processes like internalization, a crucial step for drug delivery and viral entry. By developing a sophisticated model that accounts for both biological differences between cells and technical noise from flow cytometry, the researchers can better understand how individual cells internalize materials, leading to more accurate predictions and experimental design guidance.

What Problem Does This Paper Address?

Understanding how cells internalize materials is vital for developing effective therapeutics and comprehending disease mechanisms. However, a significant challenge in this area is "biological heterogeneity" – the fact that even genetically identical cells in the same population behave differently. This variability can profoundly impact how cells respond to drugs or pathogens. While flow cytometry offers a high-throughput way to measure these processes at the single-cell level, it's often plagued by technical noise (e.g., instrument limitations, background fluorescence) that can mask or be mistaken for true biological differences. Many existing studies tend to overlook this cell-to-cell variability, focusing instead on population averages, which can lead to misleading conclusions. This paper aims to bridge this gap by providing a robust method to disentangle and quantify biological heterogeneity from technical noise in internalization assays.

Methods

The core of this work lies in a two-pronged approach:

  1. A Novel Dynamical Model: The researchers developed a mathematical model that describes the internalization of antibodies by cells. This model goes beyond simple rate equations by explicitly incorporating:
  2. Cell-to-cell variability: Key parameters like the number of receptors, internalization rate, and recycling rate are treated as random variables that vary across individual cells.
  3. Extrinsic noise: The model accounts for noise introduced by the flow cytometry process itself, including photomultiplier tube (PMT) noise and cellular autofluorescence.
  4. Distribution-Matching Approximate Bayesian Computation (ABC): To calibrate this complex model to experimental data, the authors devised a new ABC algorithm. Instead of matching individual data points, this method focuses on matching the distributions of the observed data to simulations from the model. This approach offers several advantages:
  5. Robustness to noise: It's less sensitive to the exact signal-to-noise ratio of the data.
  6. Agnostic to sample size: It can handle the large datasets typical of flow cytometry.
  7. Handles model complexity: It can accommodate intricate observation models without requiring explicit likelihood functions.
  8. Provides uncertainty quantification: It allows for estimation of uncertainty in model parameters.

Datasets

The study utilized experimental data from an internalization assay involving a dual-labelled anti-transferrin receptor (anti-TFR) antibody in a human B-cell lymphoblastoid cell line (C1R cells). This assay involved incubating cells with the labelled antibody for various time points and then measuring fluorescence using flow cytometry. Measurements were taken both with and without a quencher dye to differentiate between surface-bound and internalized antibody. The paper does not introduce new benchmark datasets but rather uses this specific experimental setup to validate their methodology.

Key Insights and Ideas

  • Biological variability is key: The study strongly emphasizes that neglecting cell-to-cell variability can lead to misleading interpretations of internalization dynamics. Their heterogeneous model provides a much richer picture than a simple homogeneous model.
  • Flow cytometry noise is significant: While the data showed high correlation between different fluorescent signals (indicating biological variability), the model explicitly needed to account for technical noise (PMT noise, autofluorescence) to accurately fit the data.
  • ABC for distribution matching is powerful: The novel ABC approach, focusing on matching distributions rather than individual points, proves effective for complex models and large datasets, offering a flexible alternative to traditional likelihood-based methods.
  • Parameter non-identifiability is common but manageable: While some parameters (like the skewness of certain distributions) were difficult to precisely identify from the data, the overall model still provided valuable insights into the key drivers of variability.
  • Heterogeneity impacts predictions: The study demonstrates that predictions from a heterogeneous model can differ significantly from those of a homogeneous model, particularly in estimating the proportion of internalized material over time.

Key Results

  • The developed heterogeneous model successfully captured both the marginal and joint distributions of fluorescence measurements from the flow cytometry data.
  • Evidence for heterogeneity in the internalization rate was found, with $\sigma_\lambda > 0.057$ (lower bound on a 95% CrI).
  • The estimated probability of antibody dissociation and receptor recycling ($p$) was lower in the heterogeneous model ($4.7\%$) compared to the homogeneous model, with greater uncertainty.
  • The model could predict unobservable statistics, such as the time evolution of the fraction of internalized antibody, and provided credible intervals for these predictions.
  • The study highlighted that homogeneous model predictions can lie within the lower tail of the heterogeneous model's predictions, suggesting potential biases when heterogeneity is ignored.

Limitations and Future Work

  • Parameter non-identifiability: The study acknowledges that some parameters, particularly those related to the shape (skewness) of distributions for internalization and recycling rates, were not well-identified by the experimental data.
  • Snapshot data limitations: The use of snapshot data from flow cytometry limits the ability to infer the ordering of individual cell trajectories over time. Future work could benefit from single-cell trajectory data (e.g., from microscopy) to better validate inferred distributions and correlations.
  • Non-Gaussian dependence: The current model assumes a Gaussian copula for dependence structures. For more complex, non-Gaussian relationships between parameters, a finite-mixture approach might be necessary, though this would increase model complexity and parameter non-identifiability.
  • Experimental design: The authors suggest that future experimental designs could focus on inhibiting recycling, pre-sorting cells to reduce variability in receptor numbers, or collecting single-cell trajectory data to improve parameter identification.
  • Broader applicability: The framework is broadly applicable to other internalization assays and cellular dynamical processes where single-cell variability is a key factor.

Summary updated

Abstract

Biological heterogeneity is a primary contributor to the variation observed in experiments that probe dynamical processes, such as the internalization of material by cells. Given that internalization is a critical process by which many therapeutics and viruses reach their intracellular site of action, quantifying cell-to-cell variability in internalization is of high biological interest. Yet, it is common for studies of internalization to neglect cell-to-cell variability. We develop a simple mathematical model of internalization that captures the dynamical behaviour, cell-to-cell variation, and extrinsic noise introduced by flow cytometry. We calibrate our model through a novel distribution-matching approximate Bayesian computation algorithm to flow cytometry data of internalization of anti-transferrin receptor antibody in a human B-cell lymphoblastoid cell line. This approach provides information relating to the region of the parameter space, and consequentially the nature of cell-to-cell variability, that produces model realizations consistent with the experimental data. Given that our approach is agnostic to sample size and signal-to-noise ratio, our modelling framework is broadly applicable to identify biological variability in single-cell data from internalization assays and similar experiments that probe cellular dynamical processes.
BibTeX
@article{browning2022identifying,
  author = {Alexander P. Browning and Niloufar Ansari and Christopher Drovandi and Angus P. R. Johnston and Matthew J. Simpson and Adrianne L. Jenner},
  title = {{Identifying cell-to-cell variability in internalization using flow cytometry}},
  journal = {Journal of The Royal Society Interface},
  year = {2022},
  doi = {10.1098/rsif.2022.0019},
}
Figure