Identification of Metal Oxide Nanoparticles in Histological Samples by Enhanced Darkfield Microscopy and Hyperspectral Mapping

Animal protocols were approved by the Institutional Animal Care and Use Committee at the investigators’ collaborating institution, Stony Brook University. A list of specific materials and equipment used for this paper can be found in Table 1.

1. Sample Preparation of Tissues

1. Prepare animal tissues exposed to metal oxide nanoparticles for histological or immunohistochemical staining as per previously described methods^21-23.

2. Imaging

1. Initializing the Microscope
   NOTE: For this study, a research grade microscope was used, outfitted with a high-performance darkfield light source, motorized XY stage controller, 14-bit depth hyperspectral camera, and multiple objective lenses (10X air, 40X air, 100X oil immersion). The system used here has a 64 nm spatial resolution at 100X (oil immersion) magnification. The condenser used for this study has both Koehler and critical illumination features and focuses a highly collimated light at oblique angles on the sample.
   1. Plug in and turn on the light source, XY stage controller, optical camera (for darkfield imaging), hyperspectral camera, and computer system. Set the light source to 75% power; raise the stage to maximum height; connect the light guide to the condenser for enhanced darkfield imaging or to the collimator on the back of the microscope for reflected brightfield imaging.
      NOTE: By setting the light source to 75% power, there is sufficient, uniform illumination of all pixels within the entire field of view that enhances duller pixels while still allowing for contrast between dull and bright pixels.
   2. Raise the condenser to operating position. Apply 3-5 drops of type-A immersion oil on the condenser lens carefully, avoiding the formation of any bubbles. Wipe away the oil and reapply it, if bubbles should form.
   3. Position the slide on the stage. Slowly raise the condenser until the immersion oil makes contact with the slide. This will be noticeable through the rapidly brightening ring of illumination where the oil makes contact with the slide.
2. Put the 10X objective in place. Focus and align the condenser by examining through the oculars.
   1. Move the stage up and down via the coarse objective focus knob until brightness is maximized.
   2. Move the condenser focus up and down via the condenser adjustment knob until maximum brightness is found. Attempt to create the brightest central spot possible in the field of view.
   3. Adjust the condenser alignment knobs as necessary to center the bright spot, if needed.
   4. Use the fine objective focus knob to bring the bright spot into focus. When changing objectives to obtain a different magnification, readjust the focus plane. When utilizing the 100X objective, apply a drop of immersion oil over the coverslip to enhance the image and avoid damage to the objective lens.
3. Capturing Images
   1. Use stage controller to find a region of interest. This region will be determined by the needs of the experiment. A typical indicator will be high contrast with surrounding regions, as areas suffused with metal oxide nanoparticles often appear brighter than areas without them.
   2. Bring the region into focus using the fine objective focus knob, adjusting the condenser focus as necessary to equilibrate illumination. Attain a high-contrast, well-defined region within the field of view.
   3. Determine what images (for the optical camera) or datacubes (for the hyperspectral camera) will be captured, and in what sequence. Typically, optical images are obtained with a 10X air objective, 40X air objective, and 100X oil immersion objective and a corresponding HSI datacube with a 100X oil immersion objective.
      1. Open the optical imaging software. Click on “Settings” in the menu bar. Select the “Image Capture Button for capture event”. Select the stored image format (TIFF was used for this experiment); assign a filename; browse and select a stored image folder; keep default timelapse; click “OK”.
      2. Select the exposure settings that create the highest contrast image in the “Exposure” menu (for this study, a level of 0.0%, gain of 3.0 dB, and shutter of 35 ms were used).
      3. Capture the image by clicking the “Image” button in the menu bar. Capture several low-magnification darkfield images in addition to those at high magnification by changing the microscope objectives, in order to provide context.
         NOTE: Capture optical images at the same magnification as any datacubes that will be captured with the hyperspectral camera, as these optical images tend to have a larger field of view and better aesthetic appearance for later visual inspection. It is essential that any datacubes that will later be spectrally analyzed use consistent magnification, as it is possible that changing the objective will alter the transmission spectra of the microscope optics, changing the captured spectra, and thus reducing the accuracy of the spectral angle mapper (SAM). If one datacube is compared to another datacube obtained with a different objective, the SAM function may not work.
   4. Select the hyperspectral camera imaging detector by redirecting the light source knob on the microscope to the hyperspectral camera. If the light is directed towards the hyperspectral camera, no image will be shown in the optical camera software. Minimize but do not close the optical imaging software window, as one may need it as specified in step 2.4.4.
4. Capturing Datacubes
   1. Open the hyperspectral imaging software for acquisition of HSI datacubes. Ensure the light guide is directed to the hyperspectral camera.
   2. Open “Hyperspectral Microscope” in the menu bar and select “HSI Microscope Controls”.
   3. Set the objective magnification and the save path. Make sure to name all images and files distinctively so no overwriting occurs. Change the area capture in settings by adjusting the field of view or number of lines (use 720 lines for this study), with a significant amount of additional time required for capturing larger areas. Finally, set the exposure time (0.25 sec for this study). Leave everything else to default, and click “Preview HSI” to view the image.
      Note: The intensity graph that appears shows the HSI datacube that will be recorded in the horizontal axis. The vertical axis represents the wavelengths of the spectra that will be captured in the HSI datacube. Placing the cursor at any position on the intensity graph corresponding to a spectral wavelength causes the intensities of all the points across the image, at that wavelength, to be shown.
   4. Focus based on this preview by adjusting the fine objective focus to sharpen the peaks in this image. If this is too difficult, the optical camera’s focal length is fairly similar, so load the optical imaging software and shift the slide bar to the optical camera, focus, and then return to the hyperspectral imaging software and return the slide bar to the hyperspectral camera.
   5. Adjust the intensity from this preview by adjusting source brightness, condenser focus, or by cancelling the preview to adjust the exposure time. The latter yields the best results, but increased exposure time can take longer to image. The aim is for the more significant peaks to be sufficiently large, but not exceeding the maximum intensity for the detector; here, the ideal range is between 1,000 and 16,000 units.
   6. Click “capture”. The microscope will ask to take a dark current image. Redirect the upper slide bar or aperture (carefully, so as not to disturb the alignment and focus of any of the other optics), and click “OK”. Restore the slide bar or aperture to the correct position and click “OK” again when prompt to image appears. Imaging can take up to 30 min, though times of about 5 min are more typical. Longer exposure times lead to longer imaging times. A progress indicator is present. Be careful not to physically disturb the scope before the progress indicator completes.
   7. Observe four new windows with the names: “Available bands list”, “#1zoom”, “#1scroll” and “#1 RGB band”. Maximize the “#1RGB band” window as this is the datacube for all future references.
   8. Right click the datacube and save it as a TIFF and click “OK”. If finished imaging; put away all samples; clean all oil-exposed surfaces with 70% isopropanol in water; raise stage to maximum height; lower condenser to minimum height and press into non-operational position; shut down or unplug the light source, stage controller, and cameras.

3. Creation of Reference Spectral Libraries

1. Selection of Reference Spectra
   1. Select a positive control that is known to contain the material of interest contained in the same matrix as the experimental samples, since HSI spectral profiles are matrix-dependent.
      1. For this study, use porcine skin injected with high doses of metal oxide nanoparticles as positive controls for comparison to the experimental porcine skin tissue from a topical exposure study. The spectral profiles of the metal oxide nanoparticles in suspension were generated and examined and found to be an inappropriate positive control for the histological experimental samples due to the different matrices.
   2. Obtain a datacube from the positive control as described in step 2.4., using the hyperspectral camera.
      1. Be particularly cautious when setting the intensity, as this is the primary metric by which the internal particle filter will identify the materials. Any intensities above the saturation point of the detector (here, 16,000 units) will result in invalid data (refer to step 2.4.5.).
   3. Right click in the image window, and left click “Z-profile spectrum”. A pop-up Spectral Profile window will appear. Left click onto pixels of interest on the datacube, particularly the brightest ones or those that can be confidently identified as representing the material of interest. Observe the Spectral Profile window showing the associated spectrum. Take note particularly of their lowest and highest value and which wavelength corresponds to it.
      1. When surveying the positive control sample, survey by clicking the regions of interest that are very bright relative to the surrounding tissue, especially those with readily identifiable particles. These particles are most likely to be the materials of interest, especially in the case of metals.
   4. Use the “Particle Filter” tool under the “Particle Analysis” menu to identify particles present in the datacube. In the new pop-up window, observe the “Spectral Max Must Exceed”. This will be determined by observations in step 3.1.3. Set this value so it is higher than background pixels, but lower than the materials of interest. The “Valid Data Max” is the maximum intensity (here, 16,000 units).
      1. Leave other parameters to default, but exclude objects based on size by adjusting the “Size Threshold” box. Save this data either at this point, or after running the analysis. For this experiment, use the following parameters: Spectral Max Must Exceed: 5,000; Valid Data Max: 16,000; Size Threshold (pixels): 400. Once all parameters have been set, click “OK”.
   5. Observe the resulting graph with the details of particles detected within the indicated intensity threshold; this data can be exported. If the characteristic maximum reflectance wavelength of the material of interest is known, select those particles that have a similar “Max WL” value; otherwise, click “select all”. Then click “Export”; “To Spectral Library”. Choose a file name then click “OK”.
2. Remove False-positive Spectra
   1. Select a sample that will serve as a negative control. Such a sample should have been prepared and treated in the same way as all experimental samples save that the absence of nanoparticles similar or identical to the material of interest is guaranteed.
      NOTE: It is important that the matrix of the negative control is the same as the matrix of the experimental samples, as HSI spectral profiles are matrix-dependent. For this study, we used porcine skin that was not exposed to metal oxide nanomaterials as negative controls.
   2. Obtain several datacubes from the negative control as described in step 2.4.7, using the hyperspectral camera. At least one is required, but more can be captured to increase selectivity (this is especially important in the event that some contaminant may be in the reference spectra).
   3. Use the hyperspectral imaging software to filter the spectral library collected in step 3.1 (positive control) against each datacube captured in step 3.2.2 (negative controls) serially, as instructed in the following steps:
      NOTE: Save the resulting, filtered spectral library to a separate file, and use as the reference spectral library for the material of interest.
      1. Click “Filter Spectral Library” under the Analysis menu, located on the main program toolbar. Click “Open”; “New File” and select the spectral library created in step 3.1. for the positive control as the Input file. Click “OK”.
      2. For external source, select “Image” under the Spectral Data box. Keep default settings under the Processing Parameters box. Select an Output Name For Filtered Library (this should be different than the original, or the raw spectral library collected will be lost). Click “OK”.
      3. Click “Open”; “New File” and select the first datacube captured in step 3.2.2 for the negative control, when prompted to select a source image.
         NOTE: The software will analyze the spectral library and remove each spectrum that matches any spectrum of the negative control datacube. Removing the background from the selection criteria thereby reduces potential for false-positives. A summary will be available when this is done (note that this summary is not saved automatically).
      4. If additional filtering is desired, repeat from step 3.2.3.1, except: step 3.2.3.2, select the last filtered spectral library created (resulting after step 3.2.3.3. completes); in step 3.2.3.3, select the next datacube from step 3.2.2.
         NOTE: Correcting and normalizing for the lamp spectrum may be necessary if samples scatter a considerable amount of light (e.g., carbon nanotubes) and/or if zero spectra remain when filtering a positive control spectral library against a negative control. These circumstances did not arise for our study and so this correction was not performed.

4. Image Analysis

1. Spectral Angle Mapping
   1. Open the “Spectral Angle Mapper” (SAM) from the Spectral menu, Mapping Methods submenu, using the hyperspectral imaging software once all the experimental datacubes have been acquired following step 2.4. The spectral angle mapper compares spectra by geometrically analyzing the changes in intensity as a function of wavelength (i.e., it compares two spectra by normalizing their intensity and comparing the angles required to trace the graph of each spectra)²⁴.
   2. With the datacube open, select the name of the experimental datacube in the pop-up window and click “OK”. If no file names are listed, click “open”; “new file”, choose the experimental datacube, then click “OK”.
      1. Observe a new pop-up window called Endmember Collection:SAM. Click “Import” in the menu bar, then select “from Spectral Library file”. A pop-up window named Spectral Library Input File will appear. Open the Spectral Library of reference that was created in step 3.2.3. and click “OK”. Observe a new pop-up window named “Input Spectral Library”.
      2. Click “Select All Items”; click “OK” right click on “Color” and select “Apply default colors to all.” Click “Select all,” followed by “Apply” and then choose the output file name and click “OK”. The Spectral Angle Mapper will take a few seconds to analyze and save the data.
   3. Open the datacube in the hyperspectral imaging software. Select “Classification” in the Overlay menu of the image window, then navigate to the file name and click “OK”. The user can now overlay any spectrum from the library to see where they map to in the image.
   4. Select the “Merge Classes” option from the Option menu in the “Interactive class tool” opened through the Overlay menu in the image window, if a unified color scheme is desired, such as for easier analysis by other software (as described in step 4.2.).
   5. Highlight all the classifications to combine (typically, everything except “unclassified” in the “classes to merge into base” list), and select a single spectrum from the “base class” list, then click “OK”. That color will now represent all selected spectra.
   6. Click on the color selected and all matching spectra will be shown in that color.
   7. To obtain a dichromatic image that will show the matching spectra over a black background, cIick on the “unclassified” colors box, which is black by default. Observe a dichromatic image. This step can be reversed by clicking again on the “unclassified” colors box.
   8. Right click to save the image as a TIFF, including any overlays currently active.
      NOTE: For future imaging processing, a dichromatic image will be used.
   9. Select “Class Distribution” from the Option menu in the “Interactive Class Tool” window to acquire the mapping statistics. This data represents how many pixels were unmapped (unclassified) and how many were identified as the material of interest. Note that the number of pixels does not correspond with the number of particles mapped. This information is not automatically recorded.
2. Particle Analysis in Hyperspectral Mapped Images
   1. Open the images in NIH ImageJ software.
   2. Use the Image menu, Adjust submenu, “Threshold” function. Select parameters that differentiate particles of interest from all other materials. Use the following threshold parameters for this study: default threshold method, color red, color space HSB, checked dark background checkbox, checked pass checkboxes for hue, saturation and brightness.
      NOTE: This may vary significantly from case to case. When analyzing a mapped datacube, this can be simplified by unifying all relevant classes to a single color and overlaying that color with all unclassified colors before generating the image (steps 4.1.4 to 4.1.8).
   3. Use the Analyze menu, “Analyze particles” function, which will retrieve information for area, mean, minimum and maximum values. For this study, use the parameters here: size 0-infinity, circularity 0-1, show nothing, checked display results checkbox.
   4. For statistical analysis, export this data to another software program enabling statistical comparison with the number and size of particles located in similar control samples. We suggest using a spreadsheet for analysis of ImageJ data. Data displayed in the results table after step 4.2.4 should be copied to a spreadsheet. The MAX function can be used on the first (untitled) column to determine the number of particles, while the AVERAGE function can be used on the Area column to determine mean size of a particle. In the future, experimental validation will be pursued to investigate this function in determining mean size against an established standard for sizing (e.g., TEM).