Industrialized, Artificial Intelligence-guided Laser Microdissection for Microscaled Proteomic Analysis of the Tumor Microenvironment

This protocol integrates AI-driven image analysis for tissue segmentation. With histology selective harvest by laser microdissection and brings us closer to reality of Pathomics. Predefining tissue ROI using AI, limits the dwell time of tissue slides at ambient temperature, reduces the manpower effort to perform these harvests and decreases operator variability.

This technique is broadly applicable to any disease or pathology-based research involving the collection or enrichment of specific cellular populations from tissue specimens. This method is straightforward for users with basic histopathology knowledge and LMD experience. The key is to make sure you cut clean calibrator for the cells and to train the classifier well.

To begin, ensure that the slide is completely dry before cutting the reference calibration fiducials. Open the Laser Microdissection software, and open the default calibration dot sld file under the import shapes option. Load the slide with the tissue facing down and the label side nearer to the operator.

Into the slide holder on the laser microdissection stage, select the auto focus before cut option. Using the laser microscope and default calibration dot sld file, cut calibration fiducials into the PEN membrane. For laser microdissection enriched collections only, open the image analysis software.

Select open images, and from the popup window, select the dot svs image file generated from scanning the slide. Navigate to the annotations tab, select and use the rectangle annotation tool to draw a box around the tissue. Select the box annotation and right click on the image.

Select the advanced dropdown menu, then click the partitioning option. Set the tile size and space between to 500 and 40 respectively, and select okay to generate the tiles. Select and delete the perimeter box annotation used to generate the tiles.

Select the layer actions drop down menu, then click export to save the tiled annotations as a dot annotation file. Place a saved copy of the Python dapa algorithm, developed to merge the AI classified annotation layers into the same folder as the tiled annotations file. Copy the name of the tiled annotation file.

Open the Python program using the idle integrated development environment and paste the name of the tiled annotation file between the quotations at the bottom of the program. Select the run dropdown menu, then click run module. Wait for a few file to be generated that will have all the tiled annotations merged under a single layer.

Open the image analysis software and navigate to the annotations tab. Select the layer actions drop down menu, then click delete all layers to remove all annotations from the image. Select the layer actions drop down menu, then click import local annotation file.

In the popup window, select the merged dot annotation file that was generated by the script. Ensure that all imported tiles are under the same annotation layer. From the classifier tab, follow the manufacturer's instructions to highlight representative areas of the tumor, stroma and blank glass slide background ROIs.

Before running the classifier, click advanced classifier options, select the desired annotation layer by checking the ROI box, or boxes on the annotations tab. Use the annotation layer option from the classifier actions menu to run the classifier. Once the classifier analysis is complete, navigate to the annotations tab, and select the annotation layer generated from the analysis.

Select the layer actions dropdown menu, then click delete all layers but current to remove all other annotation layers from the image. Next, select the layer actions drop down menu, then click export to save the annotations as a dot annotation file. Create a folder for the session or project and save the dot annotation file inside a sub folder.

Labeled with the unique identifier for the slide. Navigate to the annotations tab, select the layer actions drop down, then click delete all layers to remove all annotations from the image. Select the pen tool and draw a short line from each calibration fiducial.

Draw lines from the marks in the following order. Top left, top right, bottom right. Select the layer actions dropdown menu, then click export to save the tiled annotations as a do annotation file.

Add underscore calib to the file name and place the file in the sub folder that contains the coordinates for the tiled shapes. Copy the address for the main project. Open the XML import generating script malliator using the idle integrated development environment then paste the project folder address between the quotes at the bottom of the script.

Select the run dropdown menu, then click run module to execute the script. Load the marked membrane slide with the tissue facing down and the label side nearer to the operator into the slide holder on the laser microscope stage. Select import shapes from the file dropdown menu.

Select the dot XML, LMD import file generated for the slide. Select no in the popup window to avoid loading reference points from file, and no in the second popup window to avoid using any previously stored reference points for calibration. Follow the prompts from the laser micro dissection application and align the calibration cross to each of the three calibration fiducials on the slide.

Look for calibration fiducials that appear in the top left, top right, and bottom right of the slide image. In the image analysis software that will correspond to reference points of the inverted laser microdissection slide on the microscope stage. Switch between using the 5x subjective lens to locate, and the 63x subjective lens to align each calibration fiducial.

Select no on the popup window to avoid saving the reference points to file, and okay in the second popup window to confirm that the slide is inserted. Move the 5x subjective lens into position and select yes in the popup window to use the actual magnification. Once the imported shapes appear, focus the camera onto the tissue.

Highlight and select all the shapes in the shapes list window. Drag them into place using one or two annotations within the field of view as references and align the vertical z-axis for cutting with the laser. Load the tubes into universal tube holder designed for PCT microtubes.

Review the imported shapes and assign them to the appropriate tube position for collection. Press start cut to start the laser. Unload and close the sample tube with LMD tissues and put on dry ice.

After thermocycling and centrifuging, the LMD harvested tissue samples, remove and discard the microcaps from the PCT microtubes. Add trypsin, add a ratio of one microgram per 30 millimeters squared tissue. And insert a micro pestle into the micro tube using the microcap tool.

Transfer the microtubes into a barrel cycle cartridge and assemble the complete cartridge. Place the cartridge into the barrel cycle pressure chamber and secure the lid. Barrel cycle at 45, 000 PSI for 50 seconds and atmospheric pressure for 10 seconds at 50 degrees Celsius for 60 cycles.

For LC-MS MS analysis, load auto sampler vials in the appropriate positions into the liquid chromatography auto sampler. Close the auto sampler and analyze individual fractions with an appropriate gradient and mass spectrometry method. Unsupervised hierarchical clustering using the 100 most variable proteins.

Resulted in the segregation of the high-grade serous ovarian cancer and ovarian clear cell carcinoma histo types from the laser microdissection enriched and whole tumor samples. By contrast, the laser microdissection enriched stroma samples from both high-grade serous ovarian cancer and ovarian clear cell carcinoma. Clustered together and independently from the laser microdissection enriched tumor and whole tumor samples.

Among the 5, 971 quantified proteins, 215 were significantly altered between whole tumor collections from high-grade serous ovarian cancer and ovarian clear cell carcinoma specimens. Of the 76 signature proteins quantified by Hughes at all, 57 were co quantified in this dataset and were highly correlated Training the classifier is the most challenging part. Following the Indica Lab software instructions and refining the classifier reproduced most accurate results.

Everything else is standardized. The tissue harvested using this AI-driven LMD workflow is compatible with a variety of downstream analytical applications including mass spectrometry based proteomics described here or genomi, transcriptomic, or other omic analyses.