Count single mitochondrial DNA nucleoids and quantify mitochondrial network volume

Expected run time for this demo: few minutes.

In this tutorial we will count the number of mitochondrial DNA nucleoids and we will segment the mitochondrial network in 3D as a reference channel.

For details about the method used to visualize these structures see this publication.

Goals

  • Detect and separate highly connected spots

  • Segment network-like structures as reference channel (mitochondrial network)

  • Filter valid spots based on reference channel signal

Preliminary steps

The first step with every new dataset is to segment the objects. These typically are the single cells, but it can be any object where you want to detect spots.

While you can detect spots in the entire image, it is highly recommended to identify region of interests (ROIs) and segment them.

The dataset provided with this tutorial already contains the segmentation files with the ROIs of the single cells, but if you need to segment ROIs we recommend using out other software called Cell-ACDC.

Dataset

To follow this tutorial, download the dataset from here.

This dataset was published in this publication.

After unzipping the downloaded file, you will see the following folder structure:

Position_26
└── Images
    ├── ASY15-1_0nM-26_s26_acdc_output.csv
    ├── ASY15-1_0nM-26_s26_last_tracked_i.txt
    ├── ASY15-1_0nM-26_s26_metadata.csv
    ├── ASY15-1_0nM-26_s26_mKate.tif
    ├── ASY15-1_0nM-26_s26_mNeon.tif
    ├── ASY15-1_0nM-26_s26_phase_contr.tif
    ├── ASY15-1_0nM-26_s26_segm.npz
    └── ASY15-1_0nM-26_s26_segmInfo.npz

Some of these files are generated by Cell-ACDC and they will not be discussed here.

What we can see is that we have 3 .tif files corresponding to 3 channels whose filename end with mKate.tif, mNeon.tif, phase_contr.tif.

The phase_contr channel is the channel we used to segment the single cells with Cell-ACDC. The resulting segmentation masks are saved in the file ending with segm.npz.

The mNeon channel is the channel where we want to detect the spots.

The mKate channel is the labelling of the mitochondrial network and we can use it in SpotMAX as the reference channel (more details below).

../_images/tutorials_mtDNA_mito_yeast_figure.svg

A) Phase contrast channel used for segmentation of the cells. B) mNeon channel used to visualize the mitochondrial DNA nucleoids (spots channel). Arrows indicate areas of connected spots. C) mKate channel used to label the mitochondrial network (reference channel).

Tip

SpotMAX can take advantage of mother-bud (or sister cells) relationships. To annotate the relationship use the Cell-ACDC software. These annotations are saved in the file ending with acdc_output.csv.

Loading the dataset

Now that we have our dataset with the segmentation file of the cells, we can proceed with detecting the spots.

Important

In this tutorial we assume that you are already familiar with the analysis parameters. If not, please read about them here: Description of the parameters.

Click on the load-folder Load folder button on the top-right of the GUI. Select the Position_26 folder you downloaded and load the mNeon channel.

When the dataset is loaded, you will see on the Analysis parameters tab on the left that some of the parameters have already been filled out.

Now let’s see how we can determine the optimal parameters for this dataset.

Setting up the parameters

The parameters are grouped into separate sections so we will go one by one.

File paths and channels

Since we want to segment the mitochondrial network as a reference channel from the mKate channel, we write ‘mKate’ in the Reference channel end name parameter.

If we want to take advantage of the mother-bud (or sister cells) pairings we write ‘acdc_output.csv’ in the Table with lineage info end name parameter.

We can then decide on a Run number (in this case we leave it at 1), and, optionally, we can append a text at the end of the output files, for example we could write ‘tutorial’ at in the Text to append at the end of the output files.

Finally, we select ‘.csv’ for the File extension of the output tables.

METADATA

Since some of the metadata is already saved in the file ending with metadata.csv some of the entries were correctly loaded.

We need to correct the Spots reporter emmission wavelength (nm) to 509 since the fluorescence probe used to image the mitochondrial DNA is mNeonGreen.

Now we need to determine the optimal values for the Spot minimum z-size (μm) and Resolution multiplier in y- and x- direction parameters. These are important because if the resulting Spot (z, y, x) minimum dimensions (radius) is too low we will detect multiple peaks within the same spot. On the other hand, if it is too high, we risk to miss the smaller spots. For this tutorial we will use Spot minimum z-size (μm) = 1.0 and Resolution multiplier in y- and x- direction = 2.0.

Tip

The simplest way to determine these values is to use the tools available in the Tune parameters tab. See more instructions in this section Tune parameters tab and here Spot minimum z-size (μm).

Once you have inserted these values you should now see the following at the Spot (z, y, x) minimum dimensions (radius) parameter:

Spot (z, y, x) minimum dimensions (radius)  (1.0, 0.4357, 0.4357) μm
                                            (2.8571, 6.479, 6.479) pxl

Pre-processing

For the pre-processing activating or not the Aggregate cells prior analysis will not make any difference becasue we are working with a single mother-daughter pair. If we would be working with multiple cells and we already know that some of the cells in the image do not have spots activating this parameter might be very important (especially if we use Thresholding for the Spots segmentation method).

We do not need to activate Remove hot pixels because this specific dataset does not have any very bright isolated single pixel.

We leave the Initial gaussian filter sigma to 0.75 because we want to activate Sharpen spots signal prior detection. When sharpening is active, the gaussian filtered image is not used for detection but only for quantification. Using a small gaussian sigma is recommended to remove some of the background noise. With a higher sigma the smoothing would be to aggressive, especially because we are dealing with low signal-to-noise ratio.

Tip

You can visually inspect the result of every pre-processing filter by pressing on the compute compute button beside each filter.

Reference channel

In this tutorial, as well as detecting the spots, we also want to segment the mitochondrial network from the mKate signal as the reference channel. We will then use the resulting segmentation to remove those spots that are outside of the mitochondrial network (we are interested in detecting only the mitochondrial DNA that is inside the mitochondrial network).

To this purpose we activate Segment reference channel.

Another important parameter we want to activate is Use the ref. channel mask to determine background. This is because we want to keep only those spots that are significantly brighter than the mKate signal in the same area. However, we cannot compare absolute intensities between mNeon and mKate because they are two different fluorophores. Therefore, we need to normalize the two signals before comparing them. To do so, SpotMAX will divide the mean of the signal within each spot (using the minimum spot size determined in the metadata_mtdna_yeast_tutorial section) by the background median. To determine the background, we want to take the median of the pixels that are outside of the spots, but inside of the mitochondrial network, hence we activate this parameter.

Next, we turn our attention onto the filters that are available to enhance the segmentation accuracy. Since we are dealing with network-like structures, a gaussian filter (smoothing) might not be enough. A better filter in this case is a ridge operator. These operators take one or more values called sigmas used as scales of the filter. For this specific dataset, we write 2, 3 in the Sigmas used to enhance network-like structures parameter.

Tip

Selecting the right sigmas for the ridge operator might require some trial and error. Some values that you can test are 1, 2, 1, 2, and 2, 3. The more sigmas you use, the longer the computation time. Test these values by both visually inspecting the result of the filter and the result of the segmentation (using the compute compute buttons).

For the Ref. channel gaussian filter sigma parameter instead, we leave this one at 0.75 to remove some background noise.

Now we need to choose whether to use the ‘Thresholding’ or ‘BioImage.IO model’ for the Ref. channel segmentation method. Since we know that ‘Thresholding’ works well in this case we will use that, but feel free to experiment with any of the models available at the BioImage Model Zoo.

Next, to choose the optimal Ref. channel threshold function we click on the compute button beside the Ref. channel segmentation method and we should be able to appreciate that thresholding_yen, threshold_otsu, and threshold_isodata all do a pretty good job at segmenting the mitochondrial network. We will proceed with thresholding_yen.

Finally, we can choose whether to Save reference channel segmentation masks and Save pre-processed reference channel image.

Spots channel

We are almost done, since this is the last section that we will setup.

For the Spots segmentation method we know that ‘SpotMAX AI’ works well in this case, but feel free to experiment with ‘Thresholding’ (which is much faster than the neural networks) and with any of the models available at the BioImage Model Zoo.

Note

If this is the first time you are using the ‘SpotMAX AI’ method, SpotMAX will need to install some libraries. Keep an eye on the terminal during this time and check that installation is successful.

After selecting ‘SpotMAX AI’ you will need to configure the parameters of the model. To do so, click on the cog cog button beside the parameter. If you want more information about the parameters of the AI see this section SpotMAX AI parameters. For this dataset, we know that the following parameters work well:

  • Model type: 2D

  • Preprocess across experiment: False

  • Preprocess across timepoints: False

  • Gaussian filter sigma: 0.0

  • Remove hot pixels: False

  • Config yaml filepath: SpotMAX_v2/spotmax/nnet/config.yaml

  • PhysicalSizeX: 0.0672498 (same as in the metadata_mtdna_yeast_tutorial)

  • Resolution multiplier yx: 1.0

Next, we can ignore Spot detection threshold function because we are using the ‘SpotMAX AI’ method.

Now, as we said at the beginning, we want to keep only those spots that are significantly higher than the reference channel in the same place. To do so, we can take advantage of a couple of features to filter valid spots. One way to compare the two signals is to perform a Welch’s t-test and keep only those spots whose p-value of the test is below a certain threshold. Therefore at the Features and thresholds for filtering true spots we click on the Set features or view the selected ones button and we expand the Statistical test (vs. ref. ch.) group. We then select the p-value (t-test) feature and we click Ok. Next, we select a maximum bound for the p-value of 0.025. Since this would result in keeping those spots whose mean intensity is both significantly and lower then the reference channel, to keep only those that have higher mean intensity we click on the plus Add feature button and we add the t-statistic from the Statistical test (vs. ref. ch.) group. Finally, we set a minimum bound on the t-statistic of 0, meaning that we will keep only those spots whose mean intensity is higher than the reference channel.

Here is how the window to select the features should look like:

../_images/features_and_thresholds_window.png

Window used to select features and thresholds for filtering true spots.

We now activate Optimise detection for high spot density, and we do not activate Compute spots size (fit gaussian peak(s)).

Finally, we can choose whether to Save spots segmentation masks and Save pre-processed spots image.

Note

Since we do not activate Compute spots size (fit gaussian peak(s)) we do not need to worry about the paramters in the SpotFIT section. Also, we can leave the Configuration section deactivated and we will get asked about it when we run the analysis

Running the analysis

Ok, we are finally ready to run the analysis!

To do so simply click on the cog_play Run analysis... button on the top right of the tab.

SpotMAX will now allow use to save the parameters to an INI configuration file and we choose ‘Yes’. This way we can load them back into the GUI any time we want by clicking on the load-folder Load parameters from previous analysis... button on the top-left of the tab.

Next, SpotMAX will ask us whether we want to select the measurements to save and we say ‘No, save all the measurements’.

Then we choose a filename for the parameters file and the folder where to save them. We will get a dialogue confirming that parameters where saved with the path where they have been saved. We click ‘Ok’ and we get a reminder that the analysis will now run in the terminal and we should keep an eye on that.

We click on ‘Ok, run now!’ and we move our attention to the terminal. In the terminal we will get asked some last questions about parameters that we did not selected and we simply confirm that we want to use the default ones.

The analysis will now run and the output files will be saved in the same folder of the dataset in a new folder called SpotMAX_output. For details about the output files see this section Output files.

Closing remarks

At the end of the analysis you can go back to the GUI and visualize and inspect the results using the tools in the Inspect and edit results.

That’s it! I hope you found this tutorial useful and you can let us know if you found mistakes or any other feedback on our GitHub page or by sending us an email at .

Until next time!