Estimation of three-dimensional chromatin morphology for nuclear classification and characterisation

Priyanka Rana, Arcot Sowmya, Erik Meijering, Yang Song

Abstract

Classification and characterisation of cellular morphological states are vital for understanding cell differentiation, development, proliferation and diverse pathological conditions. As the onset of morphological changes transpires following genetic alterations in the chromatin configuration inside the nucleus, the nuclear texture as one of the low-level properties if detected and quantified accurately has the potential to provide insights on nuclear organisation and enable early diagnosis and prognosis. This study presents a three dimensional (3D) nuclear texture description method for cell nucleus classification and variation measurement in chromatin patterns on the transition to another phenotypic state. The proposed approach includes third plane information using hyperplanes into the design of the Sorted Random Projections (SRP) texture feature and is evaluated on publicly available 3D image datasets of human fibroblast and human prostate cancer cell lines obtained from the Statistics Online Computational Resource. Results show that 3D SRP and 3D Local Binary Pattern provide better classification results than other feature descriptors. In addition, the proposed metrics based on 3D SRP validate the change in intensity and aggregation of heterochromatin on transition to another state and characterise the intermediate and ultimate phenotypic states.

Introduction

The spatial organization of chromatin (DNA and proteins) inside the nucleus is a sensitive indicator of a cell's health and identity. When cells transition from a healthy to a diseased state—such as in cancer—the chromatin undergoes structural remodeling. These changes often manifest as alterations in "nuclear texture" long before large-scale morphological changes are visible.

While deep learning has seen success in medical imaging, it often requires massive datasets and lacks interpretability—a "black box" problem that makes it hard for biologists to understand why a cell was classified as cancerous. Existing handcrafted texture features are often limited to 2D or "pseudo-3D" approaches that ignore the rich information stored between the slices of a 3D microscope stack. This paper seeks to bridge that gap with a robust, interpretable 3D texture descriptor.

Methodology

The authors propose an extension of the Sorted Random Projections (SRP) algorithm into the third dimension. The proposed approach includes true 3D information by utilizing hyperplanes to capture spatial correlations across volumetric image slices.

Overview of the 3D nuclear texture analysis pipeline.

Key components of the methodology include:

  1. 3D SRP Descriptor: Extracts features from cubic patches of volumetric data using hyperplanes and 3D Bresenham’s line algorithms to capture intensity distributions across X, Y, and Z planes.
  2. Dimensionality Reduction: Employs Random Projections (RP) to compress high-dimensional data while preserving distances between data points for computational efficiency.
  3. Feature Representation: Uses a Bag of Visual Words (BOVW) approach to cluster local 3D features into a global "vocabulary" representing the entire nucleus.
  4. Biological Metrics: Derived specific metrics for Heterochromatin (HC) intensity and aggregation to quantify the ratio of condensed (HC) to open (Euchromatin, EC) DNA.

The study utilized 3D image datasets from the Statistics Online Computational Resource (SOCR), focusing on Human Fibroblasts (proliferating vs. quiescent) and Human Prostate Cancer (PC3) cells (epithelial vs. mesenchymal transition).

Results and Discussion

The inclusion of true 3D information (interslice correlations) drastically improves classification performance compared to "pseudo-3D" methods. While Local Binary Patterns (LBP) are a standard in texture analysis, SRP proved more robust by encoding intensity distributions in multiple patterns rather than relying solely on local pixel differences.

ROC curves comparing 3D SRP with other feature descriptors for (a) Human Fibroblasts and (b) PC3 cells.

3D SRP achieved near-perfect classification results, with an AUC of 0.997 for fibroblasts and 0.999 for PC3 cells. The method significantly outperformed previous state-of-the-art shape-based models and proved competitive with deep learning features (FV-CNN) while requiring significantly less training data and offering higher interpretability.

Beyond classification, the tool successfully quantified biological phenomena. It demonstrated that fibroblasts undergoing serum starvation experience chromatin condensation, while prostate cancer cells undergoing EMT experience chromatin decondensation. Furthermore, the metrics were sensitive enough to identify "Intermediate Cell States" (ICS) within the EMT population, which are critical for understanding drug resistance.

Conclusion

While 3D SRP is sensitive to Z-resolution and initial segmentation accuracy, it provides a powerful new "biomarker" for epigenetic changes. Future work may involve applying these HC/EC quantification metrics to 2D fluorescence images from liquid biopsies to detect similar epigenetic signatures in a less invasive manner.

Summary updated

Abstract

Classification and characterisation of cellular morphological states are vital for understanding cell differentiation, development, proliferation and diverse pathological conditions. As the onset of morphological changes transpires following genetic alterations in the chromatin configuration inside the nucleus, the nuclear texture as one of the low-level properties if detected and quantified accurately has the potential to provide insights on nuclear organisation and enable early diagnosis and prognosis. This study presents a three dimensional (3D) nuclear texture description method for cell nucleus classification and variation measurement in chromatin patterns on the transition to another phenotypic state. The proposed approach includes third plane information using hyperplanes into the design of the Sorted Random Projections (SRP) texture feature and is evaluated on publicly available 3D image datasets of human fibroblast and human prostate cancer cell lines obtained from the Statistics Online Computational Resource. Results show that 3D SRP and 3D Local Binary Pattern provide better classification results than other feature descriptors. In addition, the proposed metrics based on 3D SRP validate the change in intensity and aggregation of heterochromatin on transition to another state and characterise the intermediate and ultimate phenotypic states.
BibTeX
@inproceedings{rana2021estimation,
  author = {Priyanka Rana and Arcot Sowmya and Erik Meijering and Yang Song},
  title = {{Estimation of three-dimensional chromatin morphology for nuclear classification and characterisation}},
  booktitle = {Scientific},
  year = {2021},
  doi = {10.1038/s41598-021-82985-9},
}
Figure