Normalizing for individual cell population context in the analysis of high-content cellular screens

Bettina Knapp, Ilka Rebhan, Anil Kumar, Petr Matula, Narsis Aftab Kiani, Marco Binder, Holger Erfle, Karl Rohr, Roland Eils, Ralf Bartenschlager, Lars Kaderali

Abstract

High-content screening (HCS) is a cornerstone of functional genomics, yet its accuracy is often compromised by "neighborhood" effects—the influence of a cell's local environment on its phenotype. This paper introduces a robust normalization framework that accounts for individual cell population context, such as local density and cell size. By utilizing Multivariate Adaptive Regression Splines (MARS) for single-cell normalization and a GSEA-based ranking method, the authors significantly enhance screen sensitivity. In a Hepatitis C Virus (HCV) RNAi screen, this method identified ten times more biological pathways than traditional well-averaging techniques, providing a more accurate foundation for discovering host-pathogen interactions.

The Challenge of Population Context

In traditional high-content screening, the signal for a genetic knockdown is typically calculated by averaging the fluorescence of all cells within a well. However, cells are highly sensitive to their environment. A cell’s likelihood of infection or phenotypic expression is heavily influenced by factors like local crowding, its position at the edge of a colony, or its physical size.

Ignoring these factors introduces significant noise. For instance, a knockdown might appear to inhibit viral infection simply because it reduced cell density or size, rather than by directly interfering with the viral life cycle. Decoupling these environmental influences from true genetic effects is critical for identifying genuine biological hits.

Relative contribution of different population context features to the variance in HCV infection signal.

As shown in the analysis of HCV infection, population context features like cell size and cell number are major drivers of signal variation. In this model, cell size alone accounted for a substantial portion of the observed environmental variance, highlighting the need for a normalization strategy that looks beyond simple well averages.

A Two-Step Normalization Pipeline

The researchers developed a pipeline to move from well-based averaging to single-cell analysis:

  1. Single-Cell Normalization: The team extracted five population features (e.g., cell size, local density) and four technical features for every cell. They employed Multivariate Adaptive Regression Splines (MARS) to model the complex, non-linear relationships between these features and the infection signal. By subtracting these modeled effects, they derived a "corrected" signal for every individual cell.
  2. GSEA-based Hit Scoring: Rather than re-averaging the corrected signals, the authors adapted Gene Set Enrichment Analysis (GSEA). They used a statistical test to determine if the distribution of cells in a specific knockdown was significantly shifted toward a non-infected state compared to negative controls.
Microscopy images and signal intensity distributions for control (Scrambled) and knockdown (CD81) samples.

Key Results and Biological Insights

The method was validated across multiple datasets, including RNAi screens for Hepatitis C Virus (HCV) and Dengue Virus (DENV).

  • Superior Accuracy: The cell-based normalization achieved an Area Under the Curve (AUC) of 0.99 in identifying control samples, significantly outperforming standard averaging (AUC 0.95).
  • Increased Sensitivity: The method identified 54 host dependency factors for HCV and revealed 20 significantly enriched biological pathways. In contrast, the standard averaging method only identified 2 pathways.
  • New Biology: The study highlighted evolutionary similarities in how HCV and DENV hijack host machinery and re-confirmed critical host factors such as PI4KA and CSNK2A1.

Discussion and Limitations

While highly effective, the authors note a potential "causality dilemma": if a viral infection itself changes cell morphology (e.g., causing cell swelling), the normalization might "over-correct" and dampen the signal. Additionally, analyzing data at the single-cell level for millions of cells is computationally intensive compared to traditional methods.

Despite these challenges, accounting for the "neighborhood" provides a far more robust framework for functional genomics, potentially extending to complex phenotypes like protein localization or morphological changes.

Summary updated

Abstract

Background High-content, high-throughput RNA interference (RNAi) offers unprecedented possibilities to elucidate gene function and involvement in biological processes. Microscopy based screening allows phenotypic observations at the level of individual cells. It was recently shown that a cell's population context significantly influences results. However, standard analysis methods for cellular screens do not currently take individual cell data into account unless this is important for the phenotype of interest, i.e. when studying cell morphology. Results We present a method that normalizes and statistically scores microscopy based RNAi screens, exploiting individual cell information of hundreds of cells per knockdown. Each cell's individual population context is employed in normalization. We present results on two infection screens for hepatitis C and dengue virus, both showing considerable effects on observed phenotypes due to population context. In addition, we show on a non-virus screen that these effects can be found also in RNAi data in the absence of any virus. Using our approach to normalize against these effects we achieve improved performance in comparison to an analysis without this normalization and hit scoring strategy. Furthermore, our approach results in the identification of considerably more significantly enriched pathways in hepatitis C virus replication than using a standard analysis approach. Conclusions Using a cell-based analysis and normalization for population context, we achieve improved sensitivity and specificity not only on a individual protein level, but especially also on a pathway level. This leads to the identification of new host dependency factors of the hepatitis C and dengue viruses and higher reproducibility of results.
BibTeX
@article{knapp2011normalizing,
  author = {Bettina Knapp and Ilka Rebhan and Anil Kumar and Petr Matula and Narsis Aftab Kiani and Marco Binder and Holger Erfle and Karl Rohr and Roland Eils and Ralf Bartenschlager and Lars Kaderali},
  title = {{Normalizing for individual cell population context in the analysis of high-content cellular screens}},
  journal = {BMC Bioinformatics},
  year = {2011},
  doi = {10.1186/1471-2105-12-485},
}
Figure