RASP: Optimal Single Puncta Detection in Complex Cellular Backgrounds

Bin Fu, Emma E. Brock, Rebecca Andrews, Jonathan C. Breiter, Ru Tian, Christina E. Toomey, Joanne Lachica, Tammaryn Lashley, Mina Ryten, Nicholas Wood, Michele Vendruscolo, Sonia Gandhi, Lucien E. Weiss, Joseph S. Beckwith, Steven F. LeeThe Journal of Physical Chemistry B 2024 · DOI

TL;DR

Detecting tiny fluorescent signals (puncta) in complex biological images is tough due to noisy backgrounds. This paper introduces RASP, a new method that uses image gradients to accurately identify these signals, significantly improving precision and speed. This advancement is crucial for advancing high-throughput biological imaging and understanding cellular processes.

What Problem Does This Paper Address?

Super-resolution and single-molecule microscopy allow us to see biological processes at unprecedented detail. However, a major hurdle is reliably detecting the faint fluorescent signals (puncta) that represent individual molecules or structures. These signals are often buried in complex cellular backgrounds, which can be noisy and have varying levels of autofluorescence. Traditional methods struggle to distinguish true signals from these false positives, leading to inaccurate analyses. This paper aims to solve this by developing a method that can accurately detect these single fluorescent puncta even in challenging, heterogeneous environments.

Methods

The core of RASP (Radiality Analysis of Single Puncta) lies in its ability to distinguish true fluorescent puncta from false positives by analyzing their "radiality" – how symmetrical they are.

  • Feature Enhancement: RASP first enhances image features, similar to standard methods.
  • Spot Detection: It then identifies potential puncta.
  • Radiality Metrics: This is where RASP innovates. It calculates two key metrics for each detected spot:
  • Steepness: Measures the intensity drop-off from the peak to a surrounding ring.
  • Integrated Gradient: Sums the image gradient within a ring around the spot.
  • Decision Boundary: RASP learns a decision boundary based on these radiality metrics from "negative control" images (images known to contain only background noise and no true signals). This boundary is then used to filter out false positives from actual experimental images.
  • Computational Efficiency: RASP is designed to be computationally efficient, making it suitable for high-throughput analysis.

Datasets

  • Simulated Data: Used to test RASP's performance in controlled conditions.
  • TetraSpeck Beads: Standard fluorescent beads used for evaluating performance in the absence of structured background across varying contrast-to-noise ratios (CNRs).
  • FFPE Human Brain Slices: Formalin-fixed paraffin-embedded human brain tissue samples, stained for a-synuclein, were used to evaluate RASP's performance in complex, structured biological environments. These included both negative control images (no primary antibody) and stained images.

Key Insights and Ideas

  • Radiality as a Key Feature: The paper highlights that the symmetrical nature of true fluorescent puncta is a powerful feature for distinguishing them from artifacts caused by structured background.
  • Data-Driven Filtering: RASP's decision boundary is learned from experimental data (negative controls), making it adaptable to different imaging conditions and sample types without requiring extensive manual parameter tuning or complex simulations.
  • Computational Efficiency is Crucial: The need for high-throughput imaging in modern biology means that analysis methods must be fast. RASP's speed is a significant advantage.
  • Generalizability: RASP is presented as a filtering step that can be applied after initial spot detection, making it compatible with existing detection algorithms and potentially improving their performance.

Key Results

  • Improved Precision: RASP significantly reduces false positives in images with structured background, increasing detection precision. For example, in FFPE human brain slices, RASP detected 53 ± 42 false positives compared to 1278 ± 497 for ThunderSTORM and 1716 ± 41 for PeakFit in confocal imaging.
  • Maintained Sensitivity: RASP achieves this precision improvement without sacrificing the detection of true positives, performing comparably to state-of-the-art methods in low-noise conditions.
  • Speed: RASP is computationally efficient, requiring approximately 30% of the time of ThunderSTORM for processing similar image sizes.
  • Biological Insights: RASP enabled the researchers to perform a large-scale correlation analysis between a-synuclein aggregates and neuronal/microglial cells in human brain tissue, leading to robust biological conclusions.

Limitations and Future Work

  • Parameter Tuning: While RASP is designed to be data-driven, the "stringency" of the decision boundary (how many false positives are accepted) is a user-controlled parameter that might require some optimization for specific applications.
  • Future Work: The authors suggest that implementing more advanced decision boundaries could further improve RASP's performance. They also anticipate its use in a wider range of high-throughput single-molecule experiments.

Summary updated

Abstract

Super-resolution and single-molecule microscopies have been increasingly applied to complex biological systems. A major challenge of these approaches is that fluorescent puncta must be detected in the low signal, high noise, heterogeneous background environments of cells and tissue. We present RASP, Radiality Analysis of Single Puncta, a bioimaging-segmentation method that solves this problem. RASP removes false-positive puncta that other analysis methods detect and detects features over a broad range of spatial scales: from single proteins to complex cell phenotypes. RASP outperforms the state-of-the-art methods in precision and speed using image gradients to separate Gaussian-shaped objects from the background. We demonstrate RASP's power by showing that it can extract spatial correlations between microglia, neurons, and α-synuclein oligomers in the human brain. This sensitive, computationally efficient approach enables fluorescent puncta and cellular features to be distinguished in cellular and tissue environments, with sensitivity down to the level of the single protein. Python and MATLAB codes, enabling users to perform this RASP analysis on their own data, are provided as Supporting Information and links to third-party repositories.
BibTeX
@article{fu2024rasp,
  author = {Bin Fu and Emma E. Brock and Rebecca Andrews and Jonathan C. Breiter and Ru Tian and Christina E. Toomey and Joanne Lachica and Tammaryn Lashley and Mina Ryten and Nicholas Wood and Michele Vendruscolo and Sonia Gandhi and Lucien E. Weiss and Joseph S. Beckwith and Steven F. Lee},
  title = {{RASP: Optimal Single Puncta Detection in Complex Cellular Backgrounds}},
  journal = {The Journal of Physical Chemistry B},
  year = {2024},
  doi = {10.1021/acs.jpcb.4c00174},
}
Figure