基于 DeMemSeg 的重叠孢膜分割与定量分析的详细操作流程

Shodai Taguchi; Keita Chagi; Hiroki Kawai; Kenji Irie; Yasuyuki Suda

doi:10.21769/BioProtoc.5520

Improve Research Reproducibility A Bio-protocol resource

提交稿件
订阅
登录
/
注册
- 个人主页
- 编辑个人信息
- 修改密码
- 退出
CN
- EN - English
- CN - 中文

Peer-reviewed

Detailed Protocol for Segmentation and Quantification of Overlapping Prospore Membranes using DeMemSeg

基于 DeMemSeg 的重叠孢膜分割与定量分析的详细操作流程

Shodai Taguchi email

KC Keita Chagi

HK Hiroki Kawai

KI Kenji Irie

YS Yasuyuki Suda email

发布: 2025年12月05日第15卷第23期 DOI: 10.21769/BioProtoc.5520 浏览次数: 1588

评审: Hemant Kumar PrajapatiSAPTARSHI MAJIAnonymous reviewer(s)

PDF

Q&A

引用

Cited by

参见作者原研究论文

The authors used this protocol in:

Cover of Cell Structure and Function, featuring study using the protocol.

Oct 2025

Special Issue on Protocols for Organoid Research

实验方案合集

Cell Imaging - A Special Collection for Cell Bio 2023

相关实验方案

通过纯化优化和结构解析提高柑桔螺原体MreB5的稳定性

Vani Pande and Pananghat Gayathri

2024年10月20日 1941 阅读

利用 Maneval 染色双重染色法检测和区分微生物生物膜

B Nirmala and Balram Ji Omar

2025年03月05日 3023 阅读

利用GFP报告蛋白在活细胞中成像PI(4,5)P₂

Mariam Alkandari [...] Mahtab Tavasoli

2025年06月05日 2150 阅读

Abstract

Quantitative analysis of biological membrane morphology is essential for understanding fundamental cellular processes such as organelle biogenesis and remodeling. While manual annotation has been the standard for complex structures, it is laborious and subjective, and conventional automated methods often fail to accurately delineate overlapping objects in 2D projected microscopy images. This protocol provides a complete, step-by-step workflow for the quantitative analysis of overlapping prospore membranes (PSMs) in sporulating yeast. The procedure details the synchronous induction of sporulation, acquisition of 3D fluorescence images and their conversion to 2D maximum intensity projections (MIPs), and the generation of a custom-annotated dataset using a semi-automated pipeline. Finally, it outlines the training and application of our mask R-CNN-based model, DeMemSeg, for high-fidelity instance segmentation and the subsequent extraction of morphological parameters. The primary advantage of this protocol is its ability to enable accurate and reproducible segmentation of individual, overlapping membrane structures from widely used 2D MIP images. This framework offers an objective, efficient, and scalable solution for the detailed quantitative analysis of complex membrane morphologies.

Key features

• Provides a mask R-CNN-based pipeline to accurately segment individual, overlapping membrane structures resulting from 2D maximum intensity projections of 3D image stacks.

• Optimized for quantifying the dynamic morphology of yeast prospore membranes (PSMs), a key model system for studying de novo membrane biogenesis.

• Presents a complete workflow from single-cell isolation using a custom CellPose model to detailed manual annotation for creating high-quality training datasets.

• Enables robust quantitative phenotyping by extracting morphological parameters (e.g., length, roundness) to distinguish subtle differences between wild-type and mutant strains.

Keywords: Deep learning (深度学习)

Instance segmentation (实例分割)

Mask R-CNN (Mask R-CNN)

Overlapping objects (重叠结构)

Quantitative image analysis (定量图像分析)

Microscopy image processing (显微图像处理)

Yeast sporulation (酵母孢子形成)

Cellular morphology (细胞形态学)

Graphical overview

Graphical overview of the DeMemSeg pipeline for automated segmentation and quantitative analysis

Background

Understanding fundamental cellular processes like organelle biogenesis, remodeling, and trafficking often requires quantitative analysis of biological membranes. This involves measuring key morphological features—such as perimeter and roundness—to describe their geometric shape. A key challenge in this field is the accurate segmentation of individual membrane structures from microscopy images, which is often a prerequisite for obtaining reliable quantitative data. The budding yeast prospore membrane (PSM), a model for de novo membrane formation, presents a typical challenge where dynamically growing membranes frequently overlap when visualized in 2D maximum intensity projection (MIP) images.

Historically, segmenting such complex structures has relied on meticulous manual annotation. While often considered the gold standard for accuracy, this method is extremely time-consuming, laborious, and prone to inter-observer variability, making it impractical for large-scale studies. Traditional automated methods based on intensity thresholding [1] or edge detection [2] often fail to accurately delineate boundaries in images with low contrast or, critically, to separate overlapping instances. The advent of generalist deep learning tools, most notably CellPose [3] and ilastik [4], has revolutionized biological image analysis by enabling robust, automated segmentation of individual cells, thereby greatly facilitating single-cell image analysis from dense populations. However, while these tools are highly effective for delineating whole-cell boundaries, their utility is often limited when analyzing more detailed and complex intracellular structures. Specifically, they frequently struggle to resolve extensively overlapping subcellular components like the PSMs within a single cell. Similarly, while large-scale foundation models like the Segment Anything Model (SAM 2) [5] show impressive zero-shot capabilities, our findings indicate they may not achieve the instance-level precision required for resolving the specific, artificial overlaps created by MIP imaging without specialized prompting or fine-tuning.

The DeMemSeg protocol presented here offers several advantages over these existing methods for this specific task. Its primary strength lies in the custom-trained mask R-CNN model, which is specifically optimized to recognize and separate overlapping PSMs in 2D MIP images with high fidelity, a capability where generalist models may falter. By providing a complete, automated workflow from single-cell isolation to instance segmentation, it dramatically improves efficiency and reproducibility compared to purely manual methods. The protocol's utility is demonstrated by its ability to yield quantitative data statistically comparable to manual annotation and to successfully characterize the complex morphological phenotype of an untrained mutant strain. The main limitation of this protocol is its supervised nature, requiring the upfront, labor-intensive creation of a high-quality annotated dataset. Additionally, as a 2D-based analysis, it cannot resolve the true 3D spatial relationships between structures, a general limitation of MIP image analysis.

Beyond yeast PSMs, the principles of this protocol are broadly applicable. The workflow of using a general cell segmentation model (like CellPose) for preprocessing and isolation, followed by training a specialized instance segmentation model (like mask R-CNN) on cropped single-cell images, can be adapted to analyze other challenging, overlapping subcellular structures, such as mitochondrial networks, the Golgi apparatus, or endoplasmic reticulum subdomains in various cell types. It is particularly well-suited for quantitative phenotyping in genetic or chemical screens where precise morphological measurements are required.

Materials and reagents

Biological materials

1. Saccharomyces cerevisiae strains, SK1 background [6]

a. WT strain, YSIY1218

MATa/MATα his3ΔSK/his3ΔSK ura3/ura3::pRS306-mCherry-Spo20^51–91 trp1::hisG/trp1::hisG leu2/leu2 arg4-NspI/ARG4 lys2/lys2 ho∆:: LYS2/ho∆:: LYS2 rme1::LEU2/RME1 AUR1::P_ACT1-LexA-ER-haVP16::AUR1-C/AUR1::P_ACT1-LexA-ER-haVP16::AUR1-C

NDT80::hphNT1::P4×lexA-9×Myc-NDT80 /NDT80::hphNT1::P4×lexA-9×Myc-NDT80

b. gip1∆ strain, YSIY793

MATa/MATα his3ΔSK/his3ΔSK ura3/ura3::pRS306-mCherry-Spo20^51–91 trp1::hisG/trp1::hisG leu2/leu2 arg4-NspI/ARG4 lys2/lys2 ho∆: :LYS2/ho∆: :LYS2 rme1::LEU2/RME1

AUR1::P_ACT1-LexA-ER-haVP16::AUR1-C/AUR1::P_ACT1-LexA-ER-haVP16::AUR1-CNDT80::hphNT1::P4×lexA-9×Myc-NDT80 /NDT80::hphNT1::P4×lexA-9×Myc-NDT80 gip1∆:: kanMX6/gip1∆:: kanMX6

2. The plasmid used in this study [6] was pRS306-mCherry-Spo20^51–91URA3, integration, pTEF1-mCherry-Spo20^51–91

Reagents

1. Yeast nitrogen base without amino acids (Difco YNB, catalog number: 291940)

2. D(+)-glucose (FUJIFILM Wako Pure Chemical Corporation, catalog number: 045-31167)

3. STAR agar L-grade 01 (Rikaken Co., catalog number: RSU-AL01-500G)

4. Bacto peptone (Thermo Fisher Scientific, catalog number: 211677)

5. Bacto yeast extract (Thermo Fisher Scientific, catalog number: 212750)

6. Adenine (6-Aminopurine) (FUJIFILM Wako Pure Chemical Corporation, catalog number: 012-11512)

7. Potassium acetate (FUJIFILM Wako Pure Chemical Corporation, catalog number: 167–3185)

8. β-Estradiol (Sigma-Aldrich, catalog number: E8875)

9. Amino acid mixture (five amino acid drop-out: arginine, leucine, uracil, histidine, tryptophan) (mixed manually in our lab)

Solutions

1. SD-RLU (see Recipes)

2. YPD medium (see Recipes)

3. YPA medium (see Recipes)

4. Sporulation medium (see Recipes)

5. Amino acid mixture (see Recipes)

6. 2 mM β-Estradiol (see Recipes)

Recipes

1. SD-RLU

Reagent	Final concentration	Quantity or volume
Yeast nitrogen base without amino acids	6.7 g/L	3.35 g
Amino acid mixture	1 g/L	0.5 g
L-Histidine	0.16 g/L	0.08 g
L-Tryptophan	0.16 g/L	0.08 g
20% D(+)-glucose solution	2% D(+)-glucose	50 mL
Deionized water	n/a	450 mL
Total	n/a	500 mL

a. Dissolve all ingredients except 20% D(+)-glucose solution in 450 mL of deionized water.

b. Adjust the pH to 6.5 if necessary and then autoclave.

c. Allow medium to cool to ~55 °C and then add 50 mL of 20% dextrose (glucose) solution to 2% final concentration.

d. For making plates, add STAR agar L-grade 01 (2% at final concentration) before autoclaving.

2. YPD medium

Reagent	Final concentration	Quantity or volume
Bacto yeast extract	10 g/L	5 g
Bacto peptone	20 g/L	10 g
Adenine	0.03 g	0.015 g
20% D(+)-glucose solution	2% D(+)-glucose	50 mL
Deionized water	n/a	450 mL
Total	n/a	500 mL

a. Dissolve all ingredients except 20% D(+)-glucose solution in 450 mL of deionized water and then autoclave.

b. Allow medium to cool to ~55 °C and add 50 mL of 20% dextrose (glucose) solution to 2% final concentration.

c. For making plates, add STAR agar L-grade 01 (2% at final concentration) before autoclaving.

3. YPA medium

Reagent	Final concentration	Quantity or volume
Bacto yeast extract	10 g/L	5 g
Bacto peptone	20 g/L	10 g
Adenine	0.03 g	0.015 g
20% potassium acetate solution	2% potassium acetate	50 mL
Deionized water	n/a	450 mL
Total	n/a	500 mL

a. Dissolve all ingredients except the potassium acetate solution in 450 mL of deionized water and then autoclave.

b. Allow medium to cool to ~55 °C and add 50 mL of 20% potassium acetate solution to 2% final concentration.

4. Sporulation medium

Reagent	Final concentration	Quantity or volume
20% potassium acetate solution	2% potassium acetate	50 mL
Deionized water	n/a	450 mL
Total	n/a	500 mL

a. Dissolve all ingredients and autoclave.

5. Amino acid mixture

Reagent	Final concentration	Quantity or volume
Adenine (012-11512)	n/a	0.5 g
L-Alanine (010-01042)	n/a	2 g
L(+)-Arginine (017-04612)	n/a	(2 g)*
L-asparagine monohydrate (019-04812)	n/a	2 g
DL-aspartic acid (010-04842)	n/a	2 g
L-Cysteine hydrochloride monohydrate (033-05272)	n/a	2 g
L(+)-Glutamine (074-00522)	n/a	2 g
L-Glutamic acid hydrochloride (071-02092)	n/a	2 g
Glycine (073-00732)	n/a	2 g
L-Histidine (084-00682)	n/a	(2 g)*
myo-Inositol (092-00282)	n/a	2 g
L(+)-Isoleucine (121-00862)	n/a	2 g
L-Leucine (124-00852)	n/a	(10 g)*
L(+)-Lysine monohydrochloride (121-01462)	n/a	2 g
L-Methionine (133-01602)	n/a	2 g
p-Aminobenzoic acid (015-2332)	n/a	0.2 g
L(-)-Phenylalanine (161-01302)	n/a	2 g
L(-)-Proline (161-04602)	n/a	2 g
L-Serine (199-00402)	n/a	2 g
L(-)-Threonine (204-01322)	n/a	2 g
L-Tryptophan (204-03382)	n/a	(2 g)*
L-Tyrosine (202-03562)	n/a	2 g
Uracil (212-00062)	n/a	(2 g)*
L-Valine (228-00082)	n/a	2 g

*Quantities enclosed in parentheses indicate dropouts.

All amino acids are from FUJIFILM Wako Pure Chemical Corporation.

6. 2 mM β-Estradiol

Reagent	Final concentration	Quantity or volume
β-Estradiol	2 mM	n/a
Ethanol	n/a

a. Dissolve β-Estradiol in 100% ethanol to a final concentration of 2 mM.

b. Aliquot into 1.5 mL tubes and store at -30 °C.

Laboratory supplies

1. Sterile Veritable Petri dishes (Shallow, catalog number: 36-3412)

2. 24 × 50 mm cover slip (Matsunami Glass Ind., Ltd., No.1, catalog number: C024501)

3. 18 × 18 mm cover slip (Matsunami Glass Ind., Ltd., No.1, catalog number: C218181)

4. 250 mL Erlenmeyer flask (Corning, catalog number: 431144)

5. Incubator (PHC Holdings Corporation, catalog number: MIR-154-PJ)

6. Shaking water bath (Personal-11, TAITEC CORPORATION, catalog number: 0069409-000)

7. Clean bench (PHC Holdings Corporation, catalog number: MCV-131BNF)

8. Vortex mixer (LMS Laboratory and Medical Supplies, Brigachtal, catalog number: VTX-3000L)

9. Low speed benchtop centrifuge (TOMY, catalog number: LC-200)

Equipment

1. THUNDER Imager Live Cell system (Leica Microsystems; includes a DFC9000 GTC sCMOS camera)

2. Workstation for computation (HP, OMEN Desktop; equipped with an NVIDIA GeForce RTX 3060 Ti GPU)

3. iPad Air (5th generation) (Apple Inc., model: MM9C3J/A)

4. Apple Pencil (2nd generation) (Apple Inc., model: A2051)

Software and datasets

1. CellPose 3.1.0 (Oct 30, 2024, BSD, free)

2. MMdetection, v3.3.0 (Jan 5, 2024, Apache, free)

3. Fiji/ImageJ, 2.16.0/1.54p (Feb 17, 2025, GPL, free)

4. spacedesk Driver, 2.1.43 free (non-commercial private license)

5. Docker Engine, 27.1.4 (Dec 17, 2024, Apache, free)

Procedure

English

中文翻译

文章信息

稿件历史记录

提交日期: Aug 25, 2025

接收日期: Oct 22, 2025

在线发布日期: Nov 4, 2025

出版日期: Dec 5, 2025

版权信息

如何引用

Taguchi, S., Chagi, K., Kawai, H., Irie, K. and Suda, Y. (2025). Detailed Protocol for Segmentation and Quantification of Overlapping Prospore Membranes using DeMemSeg. Bio-protocol 15(23): e5520. DOI: 10.21769/BioProtoc.5520.