Measuring Single-Cell Aging with an Imaging-based Biomarker of Chromatin and Epigenetic Aging

Summary

The protocol presents the imaging and computational workflow to extract and validate imaging-based chromatin and epigenetic age (ImAge).

Abstract

To advance personalized healthcare and assess potential rejuvenation strategies, biomarkers reflecting biological age are crucial. We have introduced imaging-based chromatin and epigenetic age (ImAge), an imaging-based biomarker derived from the spatial organization of chromatin and epigenetic marks within single nuclei, capturing intrinsic age-related changes. ImAge effectively captures and quantifies the effects of interventions modulating biological age, showing sensitivity through significant decreases following rejuvenation strategies (such as caloric restriction and partial reprogramming through transient Oct4, Sox2, Klf4, and Myc (OSKM) expression) and increases after treatments like chemotherapy. Furthermore, lower ImAge correlated with higher locomotor activity, suggesting it captures functional improvements alongside biological age reversal, offering a promising single-cell tool to evaluate longevity interventions. As ImAge is an integrative method requiring good-quality imaging and meticulous computational analysis, we aim to provide thorough guidance to extract and validate it. This comprehensive protocol will detail the workflow for both wet laboratory procedures (sample preparation and imaging) and the proper utilization of our open-sourced computational pipeline (environment setup and processing of the imaging data) to extract and validate ImAge.

Introduction

The increasing average lifespan and an aging global population underscore the critical need for effective biomarkers of functional or biological age^1,2,3. These biomarkers are essential for accurately predicting disease risk and lifespan, which can significantly enhance geriatric care and reduce associated costs. While DNA methylation (DNAm) clocks have made notable advancements in quantifying biological aging^4,5,6, they often rely on linear regression and necessitate large cohorts, potentially overlooking nuanced, context-dependent biological components specific to individuals^7,8. Furthermore, these sequencing-based techniques require cell lysis, complex sample preparation, and expensive reagents for each cell, making them more costly and challenging for high-throughput, large-scale, or longitudinal studies where sample destruction is a disadvantage⁹.

To overcome these limitations, we developed imaging-based chromatin and epigenetic age (ImAge), a novel technique designed to quantify aging and rejuvenation at single-cell resolution. This technique was applied and validated in mouse peripheral blood and various solid organs as described by Alvarez-Kuglen et al.¹. This publication describes the optimized protocols in detail such that they can be adapted and adopted by other researchers for various applications.

This approach fundamentally differs from traditional DNAm clocks by capturing intrinsic age-related trajectories based on the spatial organization of chromatin and epigenetic marks within single nuclei¹. Imaging-based methods offer distinct advantages: it is inherently non-destructive, meaning cells are not destroyed during processing, unlike sequencing-based methods. This non-destructive nature crucially allows for the preservation of the spatial structure of chromatin and epigenetic marks at the single-cell level, maintaining the spatial context of cells within tissues-a perspective often lost in conventional sequencing workflows. Furthermore, imaging techniques are generally more cost-effective for large-scale or longitudinal studies, as they do not require expensive reagents or complex library preparation per cell, making them suitable for repeated observations over time^10,11.

A key advantage of ImAge is its ability to reveal age-related trajectories as intrinsic principal features of the data itself, without requiring regression on chronological age. ImAge has demonstrated a strong correlation with chronological age in mouse peripheral blood mononuclear cells and various solid organs¹. Crucially, ImAge effectively captures expected perturbations to biological age, showing an increase following chemotherapy treatment and a decrease with caloric restriction and partial reprogramming by transient OSKM expression in liver and skeletal muscle. Furthermore, ImAge readouts have been observed to inversely correlate with locomotor activity in chronologically identical mice, indicating their utility in measuring aspects of biological and functional age.

Implementing the ImAge methodology necessitates expertise in both wet-lab experimental procedures¹² and computational analyses¹ (Figure 1); therefore, this paper provides comprehensive, step-by-step protocols for both aspects. This optimized protocol aims to facilitate the widespread adoption of ImAge, enabling researchers to investigate age-associated changes in chromatin and epigenetic organization at single-cell resolution across various cell types and tissues. In this protocol, H3K27ac and H3K27me3, two markers that provide sufficient contrast and resolution for determining ImAge, were selected as representative marks. Researchers can replicate this protocol to quantify biological aging, assess the impact of longevity interventions, and explore the inherent heterogeneity within aging and rejuvenation processes at the individual organism level.

Protocol

All animal experiments are conducted according to guidelines and protocols approved by the Institutional Animal Care and Use Committee (IACUC) of The Scripps Research Institute (Protocol 09-0004-6). Data presented within this manuscript were obtained using male i4F mice, ranging in age from 2 to 14 months of age. Organs of interest (liver and skeletal muscle) were harvested from 13.8-month-old i4F and littermate control mice treated for 1 week with a low dose of doxycycline (0.2 mg/ml)¹⁴. Nuclei were isolated from samples of young (3.2 months), aged (13.8 months), and aged treated with doxycycline to overexpress OSKM factors (aged-OSKM).

1. Imaging of samples

1. Sample preparation
   1. Collect organs of interest (liver and skeletal muscle) from freshly dissected mice (or other species of interest) and flash freeze by liquid nitrogen. Store samples at -80 °C until ready to use.
   2. Transfer organs and tissues to a pre-chilled mortar and lay over dry ice. Pour liquid nitrogen over the frozen tissue and grind using a pre-chilled pestle. Pulverize the sample using the pre-chilled mortar and pestle until a uniformly fine powder is obtained.
   3. Aliquot ground-up tissue using a pre-chilled metal spatula into multiple tubes and store unused samples at -80 °C.
      NOTE: Ground tissue samples in the tubes can be stored at -80 °C or proceed with nuclei isolation.
   4. Extract nuclei from frozen aliquots, using a commercially available Nuclei Lysis Buffer kit as per the manufacturer's instructions.
   5. Count nuclei using Trypan Blue. Mix nuclei and Trypan Blue in equal ratio in a separate tube. Transfer 10 µl of the nuclei-Trypan Blue solution to a hemocytometer or an automated cell counter.
   6. Dilute samples in PBS + 0.5% BSA such that 10,000 nuclei can be plated in a volume of 20-30 µL/well.
2. Plating preparation
   1. Enter all relevant metadata for plating preparation and computational analysis into the Platemap (provided in Supplementary File 1). Record the sample identifiers (e.g., numbers or characters) in the same layout as the actual well plate in the sampleID tab. Record experimental conditions, including the sample's age, in the same layout as the actual well plate in the condition tab.
      NOTE: The information provided in the condition tab will be utilized to identify the age group of each sample (i.e., young or old). These identified age groups will subsequently serve as references for constructing the aging (ImAge) prediction model. Therefore, the condition tab must contain data that can be associated with age, along with other relevant experimental conditions as necessary.
   2. Record the target biomarkers (e.g., H3K27me3) in the same layout as the actual well plate in the channel# (# is replaced by a number) tab. Verify that all combinations of metadata match and are correctly reflected in one consecutive table format in the platemap tab.
      NOTE: Each column in the table should correspond to a designated tab name (e.g., "sampleID"). An additional tab labeled "metadata" is provided for documenting project objectives and staining details; this information will not be utilized in the data analysis. Utilize this tab as deemed necessary.
   3. Export the tab platemap as csv file format with the short, descriptive project name (e.g., project.csv).
      NOTE: The metadata tab provides a template to summarize project aims and immunostaining information. The use of the metadata tab in the Platemap template is optional for the plating process itself. However, other tabs, including sampleID, channel# (# is replaced by a number), are required for computational analysis.
3. Plate 10,000 nuclei/well in 20 - 30 µL volume in a well plate (e.g., 384-well plates). Spin plate at 1,000 x g (acceleration nine, deceleration 4) for 10 min at 4 °C.
4. Fix with 1:1 volume of 8% paraformaldehyde (PFA) in PBS for 10 min at room temperature.
   CAUTION: Conduct PFA steps in a fume hood and dispose of waste as per institutional safety procedures.
5. Remove solution. Add 30 µL of 0.1 M glycine in PBS for 5 min at room temperature. Remove solution. Wash with 100 µL of PBS 2x and add 30 µL of PBS for storage.
   NOTE: Cells can be stored at 4 °C or proceed immediately with immunolabeling.
6. Immunolabeling
   1. Remove PBS. Add 30 µL of blocking solution (2% BSA and 0.5% Triton X-100 in PBS) and incubate for 1 h at room temperature. Remove the blocking solution.
   2. Add 25 µL of primary antibodies: Mouse anti-H3K27ac (1:250) and Rabbit anti-H3K27me3 (1:250) diluted in blocking buffer (0.4 - 2 µg/mL). Incubate overnight at 4 °C.
   3. Remove primary antibody solution. Wash with 100 µL of PBS 3x for 3 min at room temperature.
   4. Add 25 µL of secondary antibodies: Alexa 488 Donkey Anti-Mouse (1:250) and Alexa 555 Donkey Anti-Rabbit (1:250) diluted in blocking buffer (1 - 10 µg/mL). Incubate for 2 h at 4 °C or 1 h at room temperature.
   5. Wash with 100 µL of PBS 3x for 3 min at room temperature. Stain with DAPI/Hoechst in 30 µL of 0.01 mg/mL concentration. Wash with 100 µL of PBS.
      NOTE: Store plate at 4 °C in foil or proceed immediately with imaging. DAPI/Hoechst is required for the nuclei segmentation step.
7. Imaging
   1. Image cells on a high-content imager for maximum resolution (for this data set, 20x was used, but 63x is recommended).
      NOTE: Although imaging can be performed with any high content imager, this protocol assumes a flat folder structure of acquired images and a generic image format (e.g., .tiff) as the acquired image output (Figure 2).
   2. Collect multiple fields per well and multiple z-stacks, ideally covering from top to bottom of cell nuclei. A 10 µm coverage along with the z-axis (e.g., 20 z-slices at a 0.5 µm z-step) typically covers most mouse cell nuclei. The size and required coverage depend on the species and cell types of interest. Representative imaging is shown in Figure 3.
   3. Transfer the imaging data to the computer where computational analysis will be performed, following the folder structure presented in Figure 2.
   4. Add the square-bracket-annotated prefix of the project name defined in step 1.2.2 above to the image data folder (e.g., [project]original_folder_name)

2. Computational analysis of imaging data

1. System software requirement preparation (Required for Initial Setup Only)
   1. Install Windows Subsystem for Linux (WSL) if operating on a Windows machine by following the instructions at https://learn.microsoft.com/en-us/windows/wsl/install. Install Anaconda or Miniconda by following the instructions at https://www.anaconda.com/docs/getting-started/anaconda/main 
2. Preparation of computational environment (Required for Initial Setup Only)
   1. Download the workflow scripts from github repository https://github.com/terskikh-lab/ImAge_workflow. Unzip the downloaded file.
      NOTE: The data storage location may be designated arbitrarily; however, it is recommended to select a directory distinct from the system file storage location (e.g., the C drive on Windows).
   2. Organize the cloned ImAge repository to match the assumed folder structure for proper script execution (Figure 2).
   3. Open terminal app if operating on Linux or macOS. Open the WSL app if operating on Windows.
   4. Change folder to the cloned repository file. Henceforth, the term Command: denotes an instruction to copy and paste the subsequent code into the terminal application, followed by pressing the Enter key.
      Command: cd /PATH_TO_CLONED_REPOSITORY/ImAge-main
      NOTE: Replace "/PATH_TO_CLONED_REPOSITORY" with the user-specific setting. For example, when the repository is unzipped in the "analysis" folder of the D drive and accessed from WSL, then this will be "/mnt/d/analysis".
   5. Create a new computational environment with Python version 3.10 using conda (command line application installed with Anaconda/miniconda).
      Command: conda create -n ImAge_workflow python=3.10
   6. Activate the newly created environment.
      ​Command: conda activate ImAge_workflow
   7. Install poetry (a python package manager) using pip installer (python package installer installed when creating the ImAge_workflow environment).
      ​Command: pip install poetry
   8. Install prerequisite python packages using poetry.
      Command: poetry install
      ​CAUTION: Make sure the current folder is the ImAge_workflow folder, which can be verified by running a command: pwd.
   9. Configure CUDA within the Conda environment to enable graphic processing unit (GPU) functionality in the segmentation step for faster computation (only applicable for linux or Windows PC with Nvidia GPU).
      Command: mkdir -p $CONDA_PREFIX/etc/conda/activate.d
      echo 'NVIDIA_PACKAGE_DIR=$(dirname $(python -c "import nvidia;print(nvidia.__file__)"))' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
      echo 'for dir in $NVIDIA_PACKAGE_DIR/*; do if [ -d "$dir/lib" ]; then export LD_LIBRARY_PATH="$dir/lib:$LD_LIBRARY_PATH"; fi done' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
      ​NOTE: The scripts automatically detect GPU availability on the server. If no GPUs are available, it will default to running on the CPU.
3. Run computations as described below.
   NOTE: The computational analysis consists of four steps: nuclei segmentation, image feature extraction, aging (ImAge) axis construction, and prediction of ImAge. Each process is performed using python scripts; o1_illumination_correction, o2_segmentation, o3_extract_features, and o4_ImAge_validation. All these steps are included in one Python script, main.py. The extracted ImAge readouts can be visualized using a box and violin plot as presented in Figure 4. The ImAge readout can also be extracted in a straightforward table format (standard .csv file format), with samples represented as rows and corresponding readouts and metadata as columns. Users have the flexibility to perform additional numerical analyses on the ImAge readouts as needed. These plots and the CSV file can be obtained by using one Python script, visualization.py. The ImAge prediction process utilizes bootstrapped signals obtained from individual cells within the same tissue. These bootstrapped signals are averaged signals derived from random sampling and repeated iterations. This approach aims to ensure that the analysis captures the most reliable and representative readouts, with the uncertainty of the analysis proportional to the inherent heterogeneity of the tissue.
   1. Open terminal app if operating on Linux or macOS. Open the WSL app if operating on Windows.
      Command: cd /PATH_TO_CLONED_REPOSITORY/ImAge-main
      ​CAUTION: Make sure the current folder is the ImAge_workflow folder, which can be verified by running a command: pwd.
   2. Activate the environment.
      ​Command: conda activate ImAge_workflow
   3. Configure the analysis parameters on workflow/main.py file in the code or text editor (e.g., Notepad app, Microsoft Visual Studio Code, etc.).
      1. Set the project name in the variable p to a short, filesystem-safe identifier used to organize outputs (e.g., p = brain_3ages_k27me3).
      2. Set the list of channels for feature extraction in chs. Include the segmentation channel if features from that channel are required for the analysis (chs = ['DAPI', 'H3K27me3', 'H3K27ac']).
      3. Map image channel identifiers from filenames to plate map columns using imageIndex. Determine the channel label format used by the imaging system to acquire image data and use that label as the key. Map each label to the exact, case-sensitive column header in the platemap saved in step 1.2.2 (e.g., imageIndex = {'1': 'Channel1', '2': 'Channel2', '3': 'Channel3'}).
         CAUTION: Column names are case-sensitive and must correspond to your plate map file.
      4. Set data locations: orgDataLoadPath to the dataset root, orgDataSubFolder to the image subfolder if necessary, and resultsSavePath to the output root.
      5. Confirm the filename extension in imageFileFormat and edit imageFileRegEx to use Python named capturing groups for all required parts: row, col, field, zposition, and channel. Use the following syntax: (?P<GROUP>\d+), where GROUP is replaced by row, col, field, zposition, or channel.
         NOTE: Update the regex if the filenames differ; all named groups must parse correctly. For example, the file name r01c11f01p01-ch1sk1fk1fl1.tiff should be converted to the following: r(?P<raw>\d+)c(?P<col>\d+)f(?P<field>\d+)p(?P<zposition>\d+)-ch(?P<channel>\d+)sk1fk1fl1\.tiff.
      6. Set the segmentation channel in segmentation_ch to the nuclear channel used for object detection (e.g., segmentation_ch = 'DAPI').
      7. Decide whether to use illumination correction by setting illumiCorrection. Set to True to compute and apply BaSiC models; set to False to process raw images.
         NOTE: For plates with ≥ 25 wells, illumination correction is recommended to reduce systematic bias.
      8. Set the physical voxel size in micrometers in voxel_dim as [z, y, x]. Use microscope metadata to determine spacing (e.g., voxel_dim = [1.0, 0.6, 0.6]). Save the file and close the editor after verifying each parameter
   4. Run scripts to obtain ImAge readouts and perform validation. While running the script, do not close the terminal or WSL app.
      Command: Python workflow/main.py
      NOTE: The script (main.py) encompasses all steps, including segmentation, feature extraction, ImAge axis construction, and prediction by default. The script will generate an estimated completion time for each step (e.g., nuclei segmentation). Each process (e.g., o1_illumination_correction) will generate interim files in .pickle format that contain results from each process. However, this is intended solely for data transfer between processes to ensure standardized processing and is not intended for user manipulation in this article. Nevertheless, if users need to manipulate or adopt these files, see the documentation included in each Python function for the data shape and types¹³.
   5. Visualize results and export raw ImAge readouts for additional analysis on external software.Command: Python workflow/visualization.py
      NOTE: This step produces figures and tables of ImAge readouts, and the files are located at fig_o4_ImAge_validation_VIOLIN and export_o4_ImAge_validation, respectively (Figure 2).

Results

Representative images should be well populated to obtain single cells, yet individual nuclei should not be touching each other, since it will degrade the quality of the cell nuclei segmentation. Figure 3 shows the representative imaging to obtain robust computational analysis results.

Results outlined in this methods manuscript are adapted from the ImAge manuscript by Alvarez-Kuglen et al.¹. ImAge successfully detected the reversal of aging...

Discussion

Sample Preparation
For sample preparation, promptly harvest tissues of interest and immediately flash-freeze them in liquid nitrogen. When grinding tissue with a mortar and pestle, a silicone cover for the mortar can be used to prevent sample loss. Add more liquid nitrogen if the tissue is difficult to pulverize. For more fragile tissues, samples can also be processed fresh (rather than frozen) to reduce mechanical disruption or tissue damage. Nuclei isolation protocols may need optimization depend...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Bovine Serum Albumin	Sigma-Aldrich	A9647
Donkey Anti-Mouse IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa 488	Invitrogen	A-21202
Donkey Anti-Rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa 555	Invitrogen	A-31572
Glycine	Bio-Rad	161-0718
Histone H3K27ac antibody (mAb)	Active Motif	39685
Histone H3K27me3 antibody (pAb)	Active Motif	39155
Hoechst 33342	Invitrogen	H3570
Nuclei Isolation Kit: Nuclei EZ Prep	Sigma-Aldrich	NUC101
Operetta CLS High Content Imager	Revvity	https://www.revvity.com/au-en/product/operetta-cls-system-hh16000020
Paraformaldehyde 16% Aqueous Solution EM Grade	Electron Microscopy Sciences	15710-S
PBS	Thermo Fisher	AM9624
PhenoPlate 384-well Microplate	Revvity	6057302
Triton-X	Sigma	T8787
Trypan Blue Solution, 0.4%	Gibco	15250061