Morphometric analysis of autophagy-related structures in Saccharomyces cerevisiae

ABSTRACT

When macroautophagy (autophagy) is induced by nutrient starvation or rapamycin treatment, Atg (autophagy-related) proteins are assembled at a restricted region close to the vacuole. Subsequently, the phagophore expands to form a closed autophagosome. In Saccharomyces cerevisiae cells overexpressing precursor Ape1 (prApe1), a specific autophagosome cargo protein, the phagophore can be visualized as a cup-shaped structure labeled with green fluorescent protein (GFP)-tagged Atg8. Previously, our group has shown that the maximum length of GFP-Atg8-labeled structures reflects the magnitude of bulk autophagy. In that study, the morphological parameters of the autophagy-related structures were extracted manually, requiring a great deal of time. Moreover, only well-expanded phagophores were subjected to further analysis. Here we report Qautas (Quantitative autophagy-related structure analysis system), a high-throughput and comprehensive system for morphological analysis of autophagy-related structures using a combination of image processing and machine learning. We describe both the manual method and Qautas in detail.

1. Introduction

Macroautophagy (hereafter, autophagy) is a highly conserved bulk degradation system that maintains cellular homeostasis in eukaryotes.^1,2 When autophagy is induced by starvation or rapamycin treatment in the yeast Saccharomyces cerevisiae, autophagy-related (Atg) proteins accumulate to a restricted region near the vacuole; subsequently, the phagophore elongates, enclosing targets to be degraded, and ultimately becomes a mature autophagosome.^3,4 After the outer membrane of the autophagosome fuses with the vacuolar membrane, the inner membrane structure, termed the autophagic body, is released into the vacuolar lumen, where it is degraded together with its cargoes by catabolic enzymes in the vacuole.^5,6

Under a fluorescence microscope, the phagophore of S. cerevisiae can be detected as a dot labeled with GFP-Atg8.⁴ Our group has developed a method for visualizing the phagophore as a cup-shaped structure in cells overexpressing prApe1, a precursor form of a selective cargo of autophagosomes.⁷ We have analyzed the maximum length of the GFP-Atg8-labeled structures in wild-type and atg1 mutant cells, and find that this morphological parameter correlates closely with autophagosome formation activity, as quantitatively determined by the alkaline phosphatase activity.^8,9 However, in that study we measure this morphological parameter manually based on intensity profiles of GFP-Atg8-labeled structures, an approach that is difficult to apply to large-scale analysis due to the long processing time required. Moreover, the morphological parameters are extracted only from well-expanded phagophores; the majority of autophagy-related structures, which are shorter, are discarded and not used for further analysis.

Here we report Qautas (Quantitative autophagy-related structure analysis system), a high-throughput and comprehensive system for morphological analysis of autophagy-related structures using a combination of image processing and machine learning (Fig. 1). In this system, morphological parameters are extracted from fluorescence microscopy images by image processing (Fig. 1A). Based on these morphological parameters, autophagy-related structures are classified into dot-shaped or elongated structures by machine learning (Fig. 1B). Statistical analysis of the elongated structures revealed that the detection power of this new system is comparable to that of manual quantification. Below, we describe both the manual protocol and Qautas.

Figure 1.

Data processing using Qautas. (A) Images are filtered with a bandpass filter to remove autofluorescence (i), and binarized to extract GFP-labeled structures (ii). Nine morphological parameters (Area, Major axis length, Minor axis length, Perimeter, Angle, Circularity, Aspect Ratio, Roundness, and Solidity) were extracted from each autophagy-related structure by particle analysis (iii). (B) Quantified data were classified as dot-shaped or elongated structures with a machine learning algorithm, random forest.

2. Materials

2.1. Strains and culture

• 1.No particular strain is required to visualize cup-shaped phagophores. It is preferable to use a strain in which both TRP1 and URA3 markers are available, so that it can carry both the GFP-Atg8-expressing and prApe1-overexpressing plasmids. It is helpful to use a strain genomically expressing monomeric RFP (mRFP)-tagged prApe1, so that the formation of the giant Ape1 complex can be monitored by fluorescence microscopy. The mRFP-prApe1-expressing strain was constructed as described previously.¹⁰ The yeast strains used in this study are listed in Table 1.
• 2.Synthetic dextrose containing casamino acids (SDCA) medium: 0.17% Difco yeast nitrogen base w/o amino acids and ammonium sulfate, 0.5% ammonium sulfate, 0.5% bacto casamino acids, 2% glucose; sterilized by autoclaving. This medium is stable for months at room temperature.
• 3.CuSO₄ stock solution: 100 mM CuSO₄ dissolved in Milli-Q water. The working concentration is 250 μM (1:400 dilution). This solution is stable for months at room temperature.
• 4.Rapamycin stock solution: 1 mg/mL rapamycin (LC laboratories, R-5000) dissolved in dimethyl sulfoxide. The working concentration is 200 ng/mL (1:5000 dilution). This solution is stable for months at −20°C.
• 5.Difco yeast nitrogen base w/o amino acids and ammonium sulfate (Becton, Dickinson and Company, 233520)
• 6.Ammonium sulfate (Wako Pure Chemical Industries, 019–03435)
• 7.Bacto casamino acids (Becton, Dickinson and Company, 223050)
• 8.Glucose (Nacalai Tesque, 16806–54)
• 9.Copper (II) sulfate pentahydrate (Wako Pure Chemical Industries, 039–04412)
• 10.Rapamycin (LC Laboratories, R-5000)
• 11.Dimethyl sulfoxide (Wako Pure Chemical Industries, 043–07216)

Table 1.

Yeast strains used in this study.

Strain	Genotype	Reference
SEY6210	MATα lys2 suc2 his3 leu2 trp1 ura3	15
GYS638	SEY6210; mRFP-APE1:LEU2	16
YCK424	SEY6210; mRFP-APE1:LEU2 atg1D211A-kanMX	7
YCK422	SEY6210; mRFP-APE1:LEU2 atg1K54A-kanMX	7

2.2. Plasmids

• 1.Plasmids used in this study are listed in Table 2. Plasmids are amplified by culturing Escherichia coli cells in LB medium (1% bacto tryptone [Becton, Dickinson and Company, 211705], 0.5% bacto yeast extract [Becton, Dickinson and Company, 212750], 1% NaCl [Sigma-Aldrich, 28-2270-5]). Ampicillin (Wako Pure Chemical Industries, 016–23301) is added to the LB medium at a final concentration of 100 μg/mL.
• 2.The pRS314[GFP-ATG8] plasmid was created by inserting the BamHI cassette of the GFP sequence into the BamHI site in an Atg8-expressing plasmid, immediately after the start codon of the ATG8 open reading frame.
• 3.pYEX-BX[APE1], a prApe1-overexpressing plasmid, was generated by inserting the BglII cassette of the APE1 gene into the pYEX-BX plasmid (Clontech, 6199-1) after digesting the plasmid with BamHI.
• 4.Ampicillin stock solution: 100 mg/mL ampicillin sodium dissolved in Milli-Q water. This solution is sterilized with a Millex syringe-driven filter unit (Merck Millipore, SLGS033SS). The working concentration is 100 μg/mL (1:1000 dilution). This solution is stable for months at −20°C.

Table 2.

Plasmids used in this study.

Name	Properties	Marker	Reference
pYEX-BX	2μ plasmid for expression of a protein from the CUP1 promoter	URA3	Clontech
pRS314	Centromeric plasmid	TRP1	17
pYEX-BX[APE1]	2μ plasmid for expression of prApe1 from the CUP1 promoter	URA3	7
pRS314[GFP- ATG8]	Centromeric plasmid for expression of GFP-Atg8 from the ATG8 promoter	TRP1	7

2.3. Equipment and reagents

• 1.Fluorescence microscope system #1: Fluorescence microscopy is performed using an IX81 TIR-FM system (Olympus, Tokyo, Japan) equipped with a UPlanSApo100 × /1.40 Oil objective (Olympus Tokyo, Japan) and a CoolSNAP HQ charge-coupled device (CCD) camera (Nippon Roper, Tokyo, Japan). Blue (Sapphire 488–20, Coherent, Tokyo, Japan) and yellow (85-YCA-010, Melles Griot, Rochester, NY, USA) lasers are used for excitation of GFP and mRFP, respectively. For simultaneous observation of GFP and mRFP, both lasers are combined to obtain a fluorescence image. Fluorescence is filtered with a DM488-561 dichroic mirror and an EM505-540&572IF bandpass filter (Olympus, Tokyo, Japan) and split into 2 channels using a U-SIP splitter (Olympus, Tokyo, Japan) equipped with a 565DCLP dichroic mirror (Chroma, Bellows Falls, VT, USA). For each channel, the fluorescent light is further filtered with an FF495-Em02-25 bandpass filter (Semrock, Rochester, NY, USA) for the GFP channel and an FF593-Em02-25 bandpass filter (Semrock, Rochester, NY, USA) for the mRFP channel. Both channels are imaged on different parts of the CCD camera. Fluorescence images are acquired and processed using the MetaMorph software (Molecular Devices, USA).
• 2.Fluorescence microscope system #2: An IX83 inverted system microscope (Olympus, Tokyo, Japan) equipped with an UPlanSApo100 × /1.40 Oil (Olympus) and a CoolSNAP HQ CCD camera (Nippon Roper) is used. U-HGLGPS (Olympus, Tokyo, Japan), a mercury light source system for microscopes, is used for excitation of fluorescent proteins. U-FGFP and U-FMCHE filter sets (Olympus, Tokyo, Japan) are used for GFP and mRFP visualization, respectively. Images are acquired with the MetaVue software (Version 7.8.2.0., Molecular Devices, Japan).
• 3.76 × 26 mm slide glass (Matsunami, S1225) and 18 × 18 mm micro cover glass (Matsunami, No. 1-S)
• 4.Low autofluorescence immersion oil (Type-F, Olympus)
• 5.Microsoft Excel for Mac 2011
• 6.ImageJ software (version 1.48v) (https://imagej.nih.gov/ij/download.html)
• 7.R software (version 3.1.3) (https://www.r-project.org) with the randomForest package (version 4.6–10)¹¹

3. Methods

3.1. Cell culture and fluorescence microscopy

• 1.Culture cells in SDCA medium at 30°C overnight in test tubes on a rotator.
• 2.Add 1/20^th volume of the culture to fresh SDCA medium containing 250 μM CuSO₄ and culture cells at 30°C for 1 day. prApe1 is overexpressed from the Cu²⁺-inducible CUP1 promoter in the presence of CuSO₄ (see Note 1).
• 3.Dilute cells again and culture at 30°C to log phase (∼2 × 10⁷ cells/mL; OD₆₀₀ = ∼1) in SDCA liquid medium containing 250 μM CuSO₄.
• 4.Add 1/5000^th volume of rapamycin (stock concentration, 1 mg/mL) to a final concentration of 200 ng/mL and culture cells at 30°C (see Note 2).
• 5.Collect cells from 500 μL of the culture by centrifugation at 2,000 × g for 1 min in a microcentrifuge (HIMAC CF15RX with the T15AP31 rotor; Hitachi, Tokyo, Japan) at room temperature.
• 6.Discard 450 μL of supernatant, and resuspend pellet in the remaining medium. The final concentration should be ∼ 2 × 10⁸ cells/mL.
• 7.Place 2.5 µL cell suspension onto a slide glass and cover with a glass coverslip (see Note 3).
• 8.Place one drop of Type-F low auto-fluorescence immersion oil onto the glass coverslip and subject the samples to microscopy observation.
• 9.Focus on the plane where the edges of vacuoles are clearly observed by differential interference contrast microscopy. Observe cells by fluorescence microscopy (either fluorescence microscope system #1 or #2 can be used) and acquire images (see Note 4).

3.2. Manual image processing

• 1.Choose well-expanded phagophores for analysis. The fluorescence intensity of each phagophore is extracted with MetaVue software. Select the “Linescan” operation in the “Measure” menu. Push the “Open Log” button and check the “A text file” checkbox and generate a LOG file, which will be in a comma-separated values (CSV) format. Select the “Average” function for the Y-axis setting in the “Linescan” window. Select the “Single Line” or “Traced Line” button in the toolbox and trace the center of the elongated autophagic structure. Check the intensity profile displayed in the “Linescan” window and press the “F9: Log Data” button to record this profile. When the “Single Line” and “Traced Line” buttons are hidden, select the “Region Tools…” operation in the “Regions” menu.
• 2.The intensity profile is drawn with Microsoft Excel. The LOG file is opened from Excel as a CSV format. You can see intensity values of consecutive pixels recorded in the “Pixel (Avg)” column. Insert a blank column just left of the “Pixel (Avg)” column and put sequential integral numbers into the column using the “Fill Series” function. We tentatively call this column the “Pixel” column. Draw the “Scatter” diagram by choosing the “Insert Scatter” operation in the “Insert” menu using the “Pixel” column as the X-axis and “Pixel (Avg)” column as the Y-axis.
• 3.Modify the intensity profile for measurement of the maximum length of GFP-Atg8-labeled structures. Move the graph to a new sheet by selecting the “Move Chart” operation in the “Design” menu. Check “New sheet” and press the “OK” button. Select a marker of the data series and select the “Format” menu, followed by the “Format Selection” operation, and the “Solid line” option in the “LINE” tab. Set the line color to black and the line width to 0.5 points. The markers of the data series can be hidden by choosing “None” in the “MARKER OPTIONS” in the “MARKER” tab. Select the Y-axis, the “AXIS OPTIONS” and the rightmost graph icon, set minor units to 1–10, and select “Add Minor Gridlines” by right-clicking the Y-axis. Select the X-axis, the “AXIS OPTIONS,” and the rightmost graph icon; set minor units to 0.1–0.2 and select “Add Minor Gridlines” by right-clicking the X-axis.
• 4.Print the chart and measure the length (Fig. 2). Draw a background line (line A in Fig. 2A), and a line parallel to the background line (line A) through the maximum value of the intensity profile (line B in Fig. 2A). Draw a line vertical to the X-axis (line C in Fig. 2B) from the point of contact (point A in Fig. 2B). The line C intersects the background line at the point B (Fig. 2B). Calculate the middle point of the segment AB (point C in Fig. 2C). Draw a line parallel to the background line (line A) through point C (line D in Fig. 2D). Line D intersects the intensity profile at 2 points (points D and E in Fig. 2D). Measure the distance on the X-axis between points D and E, obtaining the maximum length of GFP-Atg8-labeled structures as pixels. Finally, the length in pixels is converted to μm by multiplying the length (pixels) by the value (μm/pixel) calculated from a previously acquired objective micrometer image (see Note 5).

Figure 2.

Schematic diagram for measurement of the length of GFP-Atg8-labeled structures by the manual procedure. Details are described in Section 3.2 in Methods.

3.3. Image processing using Qautas

3.3.1. Extraction of morphological parameters from GFP-Atg8-labeled structures

• 1.Images of GFP-labeled structures are acquired by fluorescence microscopy as described in Section 3.1. Take images of more than 100 elongated structures to create training data.
• 2.Launch the ImageJ software. Open image(s) to be analyzed by dragging-and-dropping files into the application.
• 3.GFP-labeled structures are extracted from the microscopy image with the ImageJ software. To process multiple images seamlessly, select “Image > Stack > Images to Stack.” Make sure that “Use Titles as Labels” is checked, and then press “OK.” Select “Process > Enhance Contrast…,” set “Saturated pixels” to 0%, and check “Normalize” and “Process all X slices.” After pressing “OK,” image fluorescence will be normalized as relative average and standard deviation.
• 4.The processed images are filtered with a bandpass filter to remove autofluorescence (Fig. 1Ai). Select “Process > FFT > Bandpass Filter…” Set “Filter large structures down to” as 5 pixels, “Filter large structures up to” as 2 pixels, “Suppress stripes” as “None,” and “Tolerance of direction” as 5% (See Note 6). After checking “Autoscale after filtering” and “Process entire stack,” press “OK.”
• 5.Images were binarized to extract GFP-labeled structures (Fig. 1Aii). Select “Process > Binary > Make Binary.” Set “Method” as “RenyiEntropy,” “Background” as “Dark,” and check “Calculate threshold for each image.” GFP-labeled structures are extracted after pressing “OK.”
• 6.Binarized images are analyzed with a particle analysis operation (Fig. 1Aiii). Select “Analyze > Set Measurements…” Check “Area,” “Perimeter,” “Fit ellipse,” and “Shape descriptors”; and then select “Redirect to” as “None” and “Decimal places” as 3. Select “Analyze > Analyze Particles…” Set “Size” as 12-Infinity, “Circularity” as 0.00–1.00, and “Show” as “Outlines.” Check “Display results” and “Exclude on edges,” and then press “OK” and “Yes” buttons in the “Process all X images” window. Select “File > Save as…” in the Results window, replace “xls” with “csv” in the filename extension, and press the “Save” button. Save parameters—Area, Perim. (perimeter), Major (major axis length), Minor (minor axis length), Angle, Circ. (circularity), AR (aspect ratio), Round (roundness) and Solidity—extracted from each GFP-labeled structure in CSV format (see Note 7). Save the “Drawing of Stack” images that outline extracted structures and the ID of each structure as a TIFF file.
• 7.Calculate the μm/pixel value for your microscopy system (see Note 5). The parameters in pixels (Area, Perim., Major, and Minor) are converted to μm by multiplying the parameters (pixels) by this value (μm/pixel) for Perim., Major, and Minor, and by the square of this value—(μm/pixel)²—for Area. Save the file.
• 8.This procedure can be processed semi-automatically by making an ImageJ macro (see Note 8).

3.3.2. Creation of a discriminator with the random forest algorithm

• 1.Extract morphological parameters as described in Section 3.3.1.
• 2.Open the CSV file containing the parameters with Excel and remove the parameters extracted from out-of-focus, abnormally shaped, or complexed structures by referring to the outlined images saved in 3.3.1.6 (please refer to supplemental online material, 3.3.1.6_Outlines.tif in the OutputDataExample folder).
• 3.Create training data by referring to the outlined images saved in 3.3.1.6. In these images, extracted structures are numbered. Judge whether each structure is classified as “dot-shaped” or “elongated.” Append “Label” column next to “Solidity” column and input “dot-shaped” or “elongated” in the corresponding cells of the “Label” column.
• 4.Create a discriminator with the R software. A sample script is provided in SupplementalText2.R. Briefly, set your working directory and load training data (SupplementalText2.R; line 26 [A]), and import the randomForest library (SupplementalText2.R; line 32 [B]). Sample half of the training data to create the discriminator and the remaining half (here, referred to as test data) for estimating the accuracy of discrimination (SupplementalText2.R; line 35 [C]). Create the discriminator with the “randomForest” function (SupplementalText2.R; line 40 [D)).¹¹
• 5.Calculate the accuracy of the discriminator by checking error rates with the test data (SupplementalText2.R; line 43 [E)). The error rate is displayed on the R console. If the average error rate of all structures is higher than 10%, go back to step 3.1 and append more images to improve the discriminator (see Note 9).

3.3.3. Classification of GFP-Atg8-labeled structures into dot-shaped and elongated ones

• 1.Collect fluorescence images for your analysis as described in Section 3.1.
• 2.Extract morphological parameters as described in Section 3.3.1.
• 3.Modify the extracted data with Excel as described in Section 3.3.2.2.
• 4.Discriminate these unclassified structures with the R software. A sample script is provided in SupplementalText2.R. Briefly, import the parameters extracted from the unclassified structures (SupplementalText2.R; line 46 [F]). Discriminate the structures with the “predict” function of the randomForest library using your discriminator (SupplementalText2.R; line 50 [G]). The results of discrimination are stored in a column named “Label.” Save the data as a CSV file (SupplementalText2.R; line 55 [H]).
• 5.Analyze the data. Half of the Perimeter (Perimeter/2) can be used instead of the maximum length of GFP-Atg8-labeled structures in the manual method (see Results & Discussion).

4. Results and discussion

The manual procedure has been used to estimate activity of phagophore expansion, which closely correlates with the magnitude of the bulk autophagic activity in wild-type and atg1 mutant cells.⁷ Here we describe Qautas, a high-throughput procedure for quantification of GFP-Atg8-labeled structures. In the manual procedure, it takes several min to measure the length of one GFP-Atg8-labeled structure. We analyzed the fluorescence images from our previous study, which were acquired using fluorescence microscope system #1,⁷ using Qautas. In this case, 171 structures were extracted and measured within 10 min. Thus, Qautas has a much faster image processing speed than the manual procedure.

Atg1 is a protein kinase, and atg1^D211A and atg1^K54A encode absolute and partial kinase-defective mutants, respectively.^{12, 13} In our previous study, the maximum length of GFP-Atg8-labeled structures is shorter in both atg1 mutant cells than in wild-type cells. In addition, the length is significantly shorter in atg1^D211A cells than in atg1^K54A cells.⁷ We analyzed raw morphological parameters extracted with ImageJ (the Section 3.3.1 has been finished). The box plot shows that there is no significant difference between wild-type and atg1^D211A cells in Perimeter/2 of the GFP-Atg8-labeled structures (Fig. 3A), indicating that the detection power was insufficient.

Figure 3.

Analysis of data obtained by Qautas. (A) Box plot using raw morphological parameters without classification by machine learning. The lengths of GFP-Atg8-labeled structures in wild-type (N = 145), atg1^D211A (N = 144), and atg1^K54A (N = 207) cells were extracted with ImageJ and plotted. (B) Perimeter/2 of elongated structures of wild-type (N = 53), atg1^D211A (N = 57), and atg1^K54A (N = 61) cells classified by machine learning. (C) Major axis length of elongated structures. *P < 0.05, **P < 0.01, N.S. not significant (Mann-Whitney U test with Bonferroni correction).

Subsequently, we processed the raw parameters with a discriminator generated by machine learning with a random forest algorithm. The GFP-Atg8-labeled structures were divided into 2 groups by the classifier, i.e., dot-shaped and elongated structures. Focusing on the elongated structures, we detected significant differences in Perimeter/2 of elongated structures between wild-type, atg1^D211A, and atg1^K54A cells (Fig. 3B). This result shows that the combination of image processing and machine learning enables us to detect morphological differences of elongated structures in each strain with a power comparable to that of the manual procedure. We also found that the parameter Major (major axis length) explained the morphological differences between elongated structures of each strain (Fig. 3C).

Here, we have described the use of the manual procedure and Qautas to obtain parameters of GFP-Atg8-labeled structures. Because Qautas has a much faster processing speed than the manual procedure, this system will be applicable to large-scale screens for unidentified proteins involved in autophagosome formation, from a morphological perspective.

5. Notes

• 1.To express a sufficient amount of prApe1 from the CUP1 promoter, cells should be cultured in the presence of CuSO₄ for more than 24 h before use. CuSO₄ should be present during observation by fluorescence microscopy.
• 2.Other autophagy-inducing conditions could be examined using this technique. For nitrogen starvation, culture cells with SDCA medium to log phase (∼2 × 10⁷ cells/mL; OD₆₀₀ = ∼1) at 30°C, wash cells twice with SD(-N) medium (0.17% yeast nitrogen base without amino acids and ammonium sulfate, 2% glucose) and transfer them to the same volume of SD(-N) medium. Images are acquired after cells are incubated for 3 h at 30°C. In our hands, rapamycin yields more reproducible results for analysis of macroautophagy.
• 3.The density of the cells is important for acquiring high-quality images. Take care to ensure that the cells are not overlapping with each other in the fluorescence microscopy field. If the cells are too dense, they should be diluted with the media used to induce autophagy. If the cells are too sparse, they can be concentrated by centrifugation. Images should be acquired as soon as possible (within 20 min) after preparation to avoid extracellular environmental changes, such as drying and reduction of oxygen.
• 4.Exposure times for fluorescence microscopy should be optimize for each system. In our conditions, the exposure time for GFP-Atg8-labeled structures are set to 300 ms.
• 5.The μm/pixel values differ among microscopy systems. To calculate this value, acquire an image of an objective micrometer (we use AX0001 OB-M 1/100 mm, Olympus) under your experimental conditions and calculate the size per pixel. In the sample data acquired in our environment, one pixel equals 0.0637 μm.
• 6.
  The optimal values of filter range and binarization were estimated by calculating precision and recall.¹⁴ Ground truth data were established by manually plotting each pixel in 228 structures. True positive, true negative, false negative, and false positive rates were calculated based on areas of extracted structures. Precision and recall were calculated as shown below.

  $$\text{Precision =}\frac{\text{True Positive}}{\text{False positive + True positive}}$$

  $$\text{Recall =}\frac{\text{True Positive}}{\text{False negative }+\text{ True positive}}$$

In our environment, the optimal parameters were as follows: bandpass filter with a range of 2 to 5 pixels, and the RenyiEntropy algorithm for binarization (precision: 0.807; recall: 0.513).

• 7.We recommend clearing the “Results” window every time the data are saved. Otherwise, new data are appended to the previous set. The contents of the “Results” window can be cleared by selecting “Edit” > “Select All” followed by “Edit” > “Clear.”
• 8.Parameter extraction can be processed semiautomatically by making an ImageJ macro. Sample files for these steps are attached (Note: these are 16-bit TIFF images and need to be opened with ImageJ software to display the fluorescence images). Open SamleImage-1 to SamleImage-10 in “3.3.1_SampleImagesTobeAnalyzed”. Then select “Plugins>Macros>Run…” and select “SupplementalText1.ijm” to run the imageJ macro that extract morphological parameters from microscopy images. Using this sample macro, you can skip from 3.3.1.2 to 3.3.1.6 and morphological parameters are extracted as “3.3.1.6_Result_px.csv” file in the directory that images locate. A stack of images is named “3.3.1.6_Stack.tiff”, a stack of merged images of original images and extracted region is named “3.3.1.6_Merge.tiff” and a stack of images with numbered outline of quantified structures named “3.3.1.6_Outline.tiff” are extracted. Convert pixels into μm described in 3.3.1.7. In this sample data, one pixel equals to 0.0637 μm. Remove abnormal-shaped and complexed structures with deleting laws in excel sheet, for example, structure 86 and 117 in sample data.
• 9.Under our conditions, 336 structures were used to create the training data, and the accuracy (100% – the error rate) was 91.5%.

Abbreviations

Ape1

aminopeptidase I

GFP

green fluorescent protein

mRFP

monomeric RFP

prApe1

precursor aminopeptidase I

Qautas

quantitative autophagy-related structure analysis system

SDCA

synthetic dextrose containing casamino acids

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Acknowledgements

We thank Ms. Eri Hirata (The University of Tokyo, Japan) for providing microscopy images and critically reading our manuscript, and Drs. Kei Kojo (Sophia University, Japan), and Mitsunori Yoshida (National Institute of Infectious Diseases, Japan) for advice on the ImageJ software. We also thank Dr. Hayashi Yamamoto (The University of Tokyo, Japan) and Dr. Yoshinori Ohsumi (Tokyo Institute of Technology, Japan) for materials.

Funding

This work was supported by the Takeda Science Foundation (to K.S.), the Yamazaki Spice Promotion Foundation (to K.S.), and Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan (25291040, 26111505, 16H01195 16K14691, and JP16H06280 to K.S.).