(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .



Organelle landscape analysis using a multiparametric particle-based method [1]

['Yoshitaka Kurikawa', 'Department Of Biochemistry', 'Molecular Biology', 'Graduate School Of Medicine', 'The University Of Tokyo', 'Tokyo', 'Ikuko Koyama-Honda', 'Norito Tamura', 'Seiichi Koike', 'Noboru Mizushima']

Date: 2024-09

The applicability of the multiparameter particle-based analysis was further validated in cell lines other than HeLa cells. HEK293T cells expressing mTagBFP2–SEC61B and GFP–OMP25 were loaded with Alexa647–EGF, and their organelle particles were labeled with anti-PMP70–Alexa594 and anti-LAMP1–Alexa680 antibodies ( S5 Fig ). The analysis of UMAP embedding revealed clusters based on the marker proteins ( S6 Fig and S5 Data ). Thus, organelle particles in HEK293T cells were correctly detected, and their original identities were preserved during the detection process. These findings suggest that the multiparameter particle-based analysis is applicable across different cell types.

Next, we determined whether exogenous markers truly represented endogenous organelles in our multiparametric particle-based analysis. Organelle particles were prepared from HeLa cells expressing BFP–SEC61B and either GFP–VAMP7 or PEX3–GFP and were labeled with Alexa405–NHS and both anti-PMP70–A594 and anti-LAMP1–A680 antibodies ( S3 Fig ). Particles were extracted from the captured images, and the five- or six-dimensional data were embedded into two-dimensional planes using UMAP. The analysis revealed clusters based on the marker proteins ( S4A and S4B Fig and S3 and S4 Data) (note that although most of these organelle markers were used in the experiment in Fig 2 , independent UMAP generated different distribution patterns). In these UMAP spaces, exogenous GFP–VAMP7 and endogenous LAMP1 were found in Cluster 3 ( S4C Fig and S3 Data ). The distribution of GFP–VAMP7 in Cluster 3 was slightly different from that of LAMP1 in Cluster 3. This may reflect preferential localizations of VAMP7 to late endosomes and of LAMP1 to mature lysosome [ 28 , 29 ]. In addition, exogenous PEX3–GFP colocalized well with endogenous PMP70 in Cluster 2, representing peroxisomes ( S4D Fig and S4 Data ). These results suggest that even when exogenous organelle markers are overexpressed, UMAP embedding accurately reflects the authentic distribution of these markers.

( A ) UMAP embedding of the data obtained from 8-color fluorescent images of particles of 7 typical organelles derived from HeLa cells. The numbers of particles classified in each cluster were as follows: Cluster 1, 21,522; Cluster 2, 5,028; Cluster 3, 2,517; Cluster 4, 2,481; Cluster 5, 1,806; Cluster 6, 1,702; and Cluster 7, 1,139. ( B ) The intensities of the fluorescent markers. Particles were colored according to the fluorescence intensity of each marker. The maximum fluorescence intensity in each marker was set to 100%. ( C ) Violin plots of signal intensities (arbitrary unit) of the organelle markers on particles in each cluster. Data obtained from 8-color fluorescent images of particles of 7 typical organelles derived from HeLa cells can be found in S1 Data ( Fig 2A ) and S2 Data ( Fig 2C ).

To analyze the fluorescence images, the sum of pixel intensities of each organelle marker was determined for each particle, and eight-dimensional data were thus acquired ( Fig 1A and S1 Data ). Fluorescent signals of 36,195 particles from 3 independent experiments were combined and dimensionally reduced to be visualized in a two-dimensional plane by principal component analysis (PCA) or uniform manifold approximation and projection (UMAP). While PCA resulted in 1 cluster overall ( S2A Fig ), UMAP separated the data into 7 clusters ( Fig 2A ). The numbers of particles classified in each cluster were as follows: Cluster 1, 21,522; Cluster 2, 5,028; Cluster 3, 2,517; Cluster 4, 2,481; Cluster 5, 1,806; Cluster 6, 1,702; and Cluster 7, 1,139. In UMAP, cluster size is dependent on various factors, including the number of data points, their variability, and hyperparameters (a parameter used to tune dimensionality reduction). The particles in each cluster showed different properties of organelle markers: Cluster 1 corresponded to the ER (SEC61B), Cluster 2 to peroxisomes (PMP70), Cluster 3 to mitochondria (GFP–OMP25 and SNAP–OMP25), Cluster 4 to early endosomes (EGF), Cluster 5 to the plasma membranes (Alexa405–NHS), Cluster 6 to the Golgi (GS27), and Cluster 7 to lysosomes (LAMP1) ( Fig 2B and 2C and S2 Data ). The clusters with high fluorescence intensity of the mitochondrial markers GFP–OMP25 and SNAP–OMP25 coincided with Cluster 3, verifying the specificity of this method. Additionally, as particles from 3 independent experiments were included in all clusters and no cluster unique to specific experiments was found, the classification of these particles was considered reproducible ( S2B Fig ). These results suggest that organelle particles retain their original membranous components and were accurately detected by this multiparametric particle-based method.

While multicolored particles can be detected by either flow cytometry or fluorescence microscopy, we chose fluorescence microscopy owing to its higher sensitivity for small particles. Using a confocal fluorescence microscope equipped with spectrometers ( S1B Fig ), lambda scanning was performed between wavelengths of 411 nm and 736 nm ( S1C Fig ). Single-color organelle particles labeled with different fluorescent dyes were used to acquire spectral data, and linear unmixing was performed to obtain 8-color fluorescence images ( Fig 1C ). To identify organelle particles in the obtained fluorescence images, Gaussian mixture modeling was applied to the fluorescence intensity of each pixel, thus determining the threshold for separating organelle signals from the background. Using the determined threshold, each image was binarized, and the images of the 8 colors were merged to obtain an image of all particles.

( A ) Workflow of multiparametric single-particle analysis of typical organelles. Markers used for labeling organelles are indicated in red. A405, A594, A647, and A680 indicate the fluorescent dyes Alexa Fluor 405, Alexa Fluor 594, Alexa Fluor 647, and Alexa Fluor 680, respectively. Fluorescence intensities of organelle particles for each marker were obtained as eight-dimensional data, as shown in the matrix, which were subjected to dimension reduction for visualization in a two-dimensional map. ( B ) CLEM of organelle particles. Magnified images are shown in the right panels. Blue, GFP–OMP25 (mitochondria); green, BFP–SEC61B (ER); and red, Alexa405–NHS (plasma membrane). Red arrows indicate ER fragments associated with mitochondria. Scale bar, 10 μm and 2 μm (magnified images). ( C ) Unmixing of 8-color fluorescent spectral images and their merged image. Scale bar, 50 μm. CLEM, correlative light and electron microscopy; EGF, epidermal growth factor; ER, endoplasmic reticulum; NHS, N-hydroxy-succinimidyl esters.

To detect multiple organelles with high resolution, we isolated organelle particles from cells and performed multicolor imaging. HeLa cells expressing markers for the ER (mTagBFP2 (BFP)–SEC61B) [ 24 ], mitochondria (GFP–OMP25 and SNAP–OMP25) [ 25 ], and the Golgi (Venus–GS27) [ 26 , 27 ] were used. Early endosomes were labeled by 5-min incubation with Alexa Fluor 647 (Alexa647)-conjugated epidermal growth factor (EGF) (hereafter, Alexa647–EGF) at 37°C. The plasma membrane was labeled by incubating cells for 15 min with Alexa405-conjugated N-hydroxy-succinimidyl (NHS) ester (hereafter, Alexa405–NHS) at 4°C just before cell homogenization. After homogenization by gentle sonication, peroxisomes and lysosomes were stained with Alexa594-conjugated anti-PMP70 antibody and Alexa680-conjugated anti-LAMP1 antibody, respectively. Thus, the organelle particles were labeled with eight colors in total (Figs 1A and S1A ). Correlative light and electron microscopy (CLEM) confirmed that membranous structures positive for the markers of these organelles, such as the mitochondria and ER, were contained in these samples ( Fig 1B ).

We then plotted the GFP signals derived from the ER–mitochondria contact sites onto the UMAP space. Particles with high GFP intensities were detected within both the mitochondrial and ER clusters ( Fig 3D ). These particles were found in the areas with high BFP–SEC61B signal in the mitochondrial cluster and high SNAP–OMP25 signal in the ER cluster (red arrows in S9A Fig ). Thus, those GFP-positive particles were considered to be ER–mitochondria contact sites. Moreover, in our CLEM data on organelle particles, ER fragments were often detected on the mitochondrial surface ( Fig 1B ). In most cases, ER fragments were much smaller than associated mitochondria particles. Therefore, it is reasonable that the signals of the ER–mitochondria contact sites were detected mainly in the mitochondrial cluster. In summary, the present method can be applied to detect small organelle populations like organelle contact sites in the organelle landscape.

To obtain clusters of the 6 markers (except the split-GFP-based contact site reporter), we used the 6-color dataset obtained in the previous 8-color analysis (BFP–SEC61B, SNAP–OMP25, Alexa647–EGF, Alexa405–NHS, Alexa594–anti-PMP70 antibody, and Alexa680–anti-LAMP1 antibody) as reference data ( S6 Data ). We embedded these reference data in a new two-dimensional plane using UMAP, resulting in 5 clusters ( Fig 3B and S6 Data ). Examination of the fluorescence intensity of each marker in the UMAP space revealed that the clusters were formed to contain different organelle markers ( S8A Fig and S6 Data ). These clusters were detected in all 3 independent experiments, validating the reproducibility of this experiment ( S8B Fig ). When the query data from cells expressing the ER–mitochondria contact site reporter were annotated using metric learning with the reference data, all 5 of the clusters were mapped with the query data ( Fig 3C and S7 Data ). Monitoring the fluorescence intensity of each marker in the plotted query data revealed that each cluster primarily contained a distinct marker, as observed among the reference data ( S9A Fig and S7 Data ). As these clusters were also detected in all 3 independent experiments, the particle classification was considered reproducible ( S9B Fig ).

( A ) Schematic illustration of cells labeled with the ER–mitochondrial contact site reporter and 6 organelle markers (left) and the split-GFP-based ER–mitochondria contact site reporter (right). ERj1(1–200)–V5–GFP1–10 and TOMM70(1–70)–3×FLAG–GFP11 are only assembled on the ER–mitochondria contact site to form GFP. Thus, those reporters that are not assembled on the ER–mitochondria contact site do not produce GFP signals. ( B ) UMAP embedding and clustering of the data obtained from fluorescent images of particles labeled with 6 organelle markers as references. The numbers of particles classified in each cluster were as follows: ER, 22,233; peroxisome, 5,088; endosome and lysosome, 3,583; mitochondria, 2,243; the plasma membrane, 1,711. ( C ) Plot of the data of the experiments with the ER–mitochondrial contact site marker as query using metric learning with the UMAP results in (B) as reference. The numbers of particles plotted on the query were 17,479, and those of references were 34,858. ( D ) Plot of the GFP signal intensity derived from the ER–mitochondria contact site reporter. The maximum fluorescence intensity of GFP was set to 100%. For visualization, the square root of % max has been plotted. Data obtained from 8-color fluorescent images of particles labeled with 6 organelle markers as references can be found in S6 Data . Data of the experiments with the ER–mitochondrial contact site marker as query can be found in S7 Data .

Most organelle markers were enriched in one of the clusters, but the ER marker SEC61B was mixed into several clusters, including mitochondrial clusters (Cluster 3) ( Fig 2B ). This small population was posited to correspond to the ER–mitochondria contact sites. We therefore sought to visualize the ER–mitochondria contact sites using a split-GFP-based reporter [ 30 ] ( Fig 3A ). This reporter consists of the first to 10th β-sheets of GFP (GFP1–10) fused to the N terminus of the ER membrane protein ERj1 and the 11th β-sheet of GFP (GFP11) fused to the N terminus of the mitochondrial outer membrane protein TOMM70. When GFP1–10 and GFP11 were associated at the ER–mitochondria contact sites, GFP fluorescence was emitted ( S7A Fig ). We introduced the fluorescent markers for the ER (BFP–SEC61B) and mitochondria (SNAP–OMP25) into cells expressing this contact site reporter and labeled early endosomes and the plasma membrane with Alexa647–EGF and Alexa405–NHS, respectively. After the preparation of organelle particles, peroxisomes and lysosomes were labeled with Alexa594-conjugated anti-PMP70 and Alexa680-conjugated anti-LAMP1 antibodies, respectively, and 7-color fluorescence images were thus obtained (Figs 3A , S7B , and S7C ).

Applications to organelles in transition: A diagram of endosomes

The successful detection of the ER–mitochondria contact sites suggested that not only typical organelles but also organelles during transition or maturation could be investigated with this method. Accordingly, we analyzed the endocytic pathway, which contains various organelles in transition, after incorporation of EGF (a cargo representative of lysosomal degradation) and transferrin (a cargo representative of recycling to the plasma membrane).

To assign organelles more precisely, we used fluorescent cargos in addition to static organelle markers. GFP–RAB5 (early endosome marker), Venus–RAB11 (recycling endosome marker), and SNAP–RAB7 (late endosome marker) were expressed in HeLa cells. These HeLa cells were incubated with Alexa594-conjugated transferrin (hereafter, Alexa594–transferrin) and Alexa647–EGF at 4°C for 30 min. Then, excess cargos were washed away, while the remainder was internalized by endocytosis at 37°C (Fig 4A). At each time point of endocytosis, we collected organelle particles, stained them with antibodies against LAMP1, and obtained 6-color fluorescence images (S10A–S10D Fig and S8 Data).

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 4. Multiparametric particle-based analysis of endocytic vesicles containing EGF or transferrin. (A) Schematic illustration of cells used for the analysis of endocytic compartments. Green and magenta indicate A647–EGF and A594–transferrin signals, respectively. Cells were treated with wortmannin or left untreated, followed by treatment with A647–EGF and A594–transferrin for 30 min at 4°C, washing, and incubation at 37°C for 0, 5, 10, 15, 20, 30, and 40 min. (B) Signal intensities of Alexa647–EGF and Alexa594–transferrin on particles positive for either EGF or transferrin over the time course (after the shift to 37°C). For wortmannin-untreated cells, the numbers of particles plotted at each time point were as follows: 0 min, 1,820; 5 min, 2,281; 10 min, 2,758; 15 min, 4,394; 20 min, 1,946; 30 min, 2,681; 40 min, 1,046. For wortmannin-treated cells, the particle numbers were as follows: 0 min, 461; 5 min, 2,050; 10 min, 3,415; 15 min, 3,329; 20 min, 1,534; 30 min, 2,369; 40 min, 2,017. (C) UMAP embedding of 17,113 data points obtained from fluorescent images of particles of endosomes containing EGF or transferrin. The numbers of particles classified in each cluster were as follows: Cluster 1, 3,347; Cluster 2, 3,092; Cluster 3, 2,959; Cluster 4, 2,869; Cluster 5, 2,423; Cluster 6, 1,374; Cluster 7, 1,049. (D) The intensities of the indicated fluorescent markers on the UMAP space. Particles are colored according to the fluorescence intensity of each marker. The maximum fluorescence intensity of each marker was set as 100%. (E) Violin plots of the signal intensities of the organelle markers on particles in each cluster. (F) A plot of the 15,197 data points of the experiments with the wortmannin treatment as a query using metric learning with the UMAP results in (C) as reference. The numbers of particles classified in each cluster were as follows: Cluster 1, 6,405; Cluster 2, 897; Cluster 3, 623; Cluster 4, 3,637; Cluster 5, 3,145; Cluster 6, 295; Cluster 7, 173. Data obtained from fluorescent images of particles of endosomes containing EGF or transferrin can be found in S8 Data (Fig 4C and 4D) and S9 Data (Fig 4E). https://doi.org/10.1371/journal.pbio.3002777.g004

To analyze the temporal changes of the features of only EGF- or transferrin-positive particles, we strictly identified these particles as follows (because many endosomes contained neither EGF nor transferrin). We measured the background intensities of particles derived from cells not treated with Alexa594–transferrin and Alexa647–EGF and set the 99th percentile point for signal strength as the threshold for the Alexa594–transferrin and Alexa647–EGF signals (S10E Fig, bottom panels). We then extracted particles that were positive for EGF or transferrin and tracked the EGF and transferrin signals in this population at each time point. A bright EGF population appeared 15 min after culture conditions were shifted to 37°C, possibly owing to the fusion of early endosomes, which increased the EGF fluorescence intensity per particle [31] (Fig 4B, left upper panel). The bright population decreased after 30 min, suggesting that Alexa647–EGF was degraded in lysosomes. This observation was consistent with previous reports that EGF reaches lysosomes approximately 30 min after incorporation [32]. To examine the impact of perturbations on EGF endocytosis, we treated cells with wortmannin, an inhibitor of the class III PI3 kinase complex, for 15 min before EGF addition. In wortmannin-treated cells, a decrease in the bright EGF population was not observed, indicating a delay in degradation (Fig 4B, right upper panel). In contrast to EGF, transferrin showed only a slight decrease after 10 min. This probably indicates a low homotopic fusion rate in RAB11-positive recycling endosomes [33] and recycling into the plasma membrane, which is consistent with the reported 10-min half-life of transferrin recycling [34] (Fig 4B, left lower panel).

We then integrated the time-course data of untreated cells from 3 independent experiments, reduced the dimensions of the fluorescence intensity data of 4 endosomal markers (excluding EGF and transferrin) into a two-dimensional UMAP plane, and performed clustering. This resulted in 7 clusters (Fig 4C and S8 Data), all of which were detected in 3 independent experiments, supporting the reproducibility of this particle classification (S11 Fig). These clusters roughly reflected the fluorescence intensity of each endosomal marker (Fig 4D and 4E and S8 and S9 Data). Cluster 1 was considered to comprise early endosomes based on high RAB5 fluorescence intensity, and Clusters 2 and 3 were considered to comprise recycling endosomes based on high RAB11 fluorescence intensity. Cluster 5, with high RAB5 and RAB7 signal intensity, was considered to be a population of endosomes undergoing RAB conversion, during which RAB5 is replaced by RAB7. Clusters 4, 6, and 7, with high signal intensities of RAB7 and/or LAMP1, were considered late endosomes and lysosomes. Next, we embedded the data from the wortmannin-treated cells into the UMAP space by metric learning, using the data from untreated cells as a reference (Fig 4F). The data were separated into 7 clusters, resembling the clustering of the untreated cells. However, the number of particles in Clusters 2, 3, 6, and 7 was reduced, suggesting that the inhibition of the recycling and degradation pathways by wortmannin was adequately represented in the UMAP space.

Next, we extracted the fluorescence intensities of EGF and transferrin from the original data and plotted them with the embedded data at each time point (Fig 5A and 5B). EGF was predominantly present in Clusters 1 and 5 at 0 to 10 min and moved to Clusters 4, 6, and 7 after 15 min. Transferrin was predominantly present in Clusters 1 and 2 at 0 min and migrated to Cluster 3 at 5 to 10 min. After 20 min, there were fewer particles with high transferrin signals. These results suggest that our method successfully visualized the continuous diagram of the endocytic process of EGF and transferrin.

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 5. Changes in the distribution of clusters during the endocytic process. (A, B) Distribution of Alexa647–EGF and A594–transferrin intensities of particles along the time course with or without wortmannin. The colors indicate the signal intensity of EGF or transferrin. The maximum fluorescence intensity of each marker was set to 100%. For visualization, the square root of %max has been plotted. (C, D) Stacked bar graphs showing the proportion of clusters in EGF-positive particles (A) and transferrin-positive particles (B) at each time point (after the shift to 37°C) from wortmannin-treated (lower panels) and untreated (upper panels) cells. (E) Characterization of the clusters and putative pathways depicted in UMAP space. The magenta arrow indicates the putative recycling pathway, and the green arrow indicates the degradation pathway. Data obtained from fluorescent images of particles of endosomes containing EGF or transferrin and extracted EGF- or transferrin-positive particles can be found in S10 Data. https://doi.org/10.1371/journal.pbio.3002777.g005

To further characterize each cluster, we analyzed the temporal transition of the 2 cargos. We counted the number of EGF- or transferrin-positive particles in each cluster and plotted the proportion over time (Fig 5C and 5D and S10 Data). Regarding EGF-positive particles, Clusters 6 and 7 showed the highest proportions at 15 min (Fig 5C). These clusters were considered to be late endosomes positive for RAB7, as they did not show high RAB5 signal intensity (Fig 4D and 4E). The proportion of Cluster 4 increased over time, indicating that it represents lysosomes (Fig 5C). In Cluster 4, LAMP1 was more enriched than RAB7 (Fig 4D and 4E). The proportion of transferrin-positive particles in Clusters 6 and 7 increased at 10 min, and that of Cluster 4 increased at 15 min (Fig 5D). These data are consistent with Clusters 6 and 7 representing late endosomes and Cluster 4 representing lysosomes because some population of transferrin receptors are directed to lysosomal degradation [35–37].

Upon treatment with wortmannin, EGF was predominantly distributed in Clusters 1 and 5, while EGF in Clusters 6 and 7 was reduced, indicating inhibition of transport to late endosomes and lysosomes (Fig 5C). Transferrin was also predominantly distributed in Clusters 1 and 5, with a decrease in Clusters 2 and 3 (Fig 5D). This is consistent with previous research showing that transferrin localizes to RAB5-positive structures instead of RAB11-positive structures upon PI3 kinase inhibition [38–41]. By tracking the intensities of EGF and transferrin, we were able to classify endocytic organelles in detail, which could not be determined simply by detecting organellar markers (Fig 5E).

[END]
---
[1] Url: https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3002777

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.