Multiplexed ion beam imaging (MIBI) is often used to image tissue microarrays and tiled, contiguous tissue areas, but current software for setting up these experiments is cumbersome. The tile/SED/array interface is an intuitive, interactive graphical tool developed to dramatically simplify and accelerate MIBI run setup.
Multiplexed ion beam imaging (MIBI) is a next-generation mass spectrometry-based microscopy technique that generates 40+ plex images of protein expression in histologic tissues, enabling detailed dissection of cellular phenotypes and histoarchitectural organization. A key bottleneck in operation occurs when users select the physical locations on the tissue for imaging. As the scale and complexity of MIBI experiments have increased, the manufacturer-provided interface and third-party tools have become increasingly unwieldy for imaging large tissue microarrays and tiled tissue areas. Thus, a web-based, interactive, what-you-see-is-what-you-get (WYSIWYG) graphical interface layer – the tile/SED/array Interface (TSAI) – was developed for users to set imaging locations using familiar and intuitive mouse gestures such as drag-and-drop, click-and-drag, and polygon drawing. Written according to web standards already built into modern web browsers, it requires no installation of external programs, extensions, or compilers. Of interest to the hundreds of current MIBI users, this interface dramatically simplifies and accelerates the setup of large, complex MIBI runs.
Multiplexed ion beam imaging (MIBI) is a technique to image 40+ proteins simultaneously on histologic tissue sections at up to 250 nm resolution1,2,3. After a histologic tissue section is stained using antibodies tagged with isotopically pure elemental metals, the MIBI instrument performs secondary ion mass spectrometry to simultaneously quantify all the isotopes – and thus expression of all 40+ antigens – at individual spots on the tissue. Performed across grids of millions of spots, the resulting 40+ plex images of protein expression enable the delineation of cell boundaries and identification of specific cell types while preserving spatial context1,2,3,4. This technique has been used by hundreds of users at roughly 20 sites to study the cellular composition, metabolic profiles, and/or architecture of dozens of tissue types as part of examining the immune response to tumors, tissue inflammation caused by infectious agents, neuropathology of dementia, and immune tolerance in pregnancy5,6,7,8,9,10,11.
A key bottleneck in MIBI instrument operation is setting up fields of view (FOVs) – 200 x 200 µm2 to 800 x 800 µm2 areas of the tissue – for imaging. The MIBI images one FOV at a time, up to 800 x 800 µm2, thus imaging larger areas requires stitching multiple FOVs together. Imaging a tissue microarray (e.g., eight circular tissues in Figure 1A) involves placing multiple FOVs spaced apart. To set up FOVs, the manufacturer interface provides 1) an optical camera image of the slide with a crosshair that roughly corresponds to the specified imaging coordinate (Figure 1A) and 2) a secondary electron detector (SED) image that shows the exact area at the coordinate, reportedly accurate to within 0.1 µm (Figure 1B). First, the user roughly positions a single FOV using the optical image. Because the image resolution is only about 60 µm per pixel, if the placement is off by two pixels (2 pixels x 60 µm per pixel), a standard 400 µm FOV will be off by 30%. Thus, the user must use the SED image to fine-tune the position – a tedious sequence of a dozen steps involving multiple popup windows, typing coordinates into text boxes, slowly nudging the SED with directional control buttons, and often even writing down coordinates on paper (Supplementary Figure 1). This process must be repeated for each spot of a 100+ core tissue microarray (TMA). Some third-party tools can help with the initial rough positioning12. However, they still require some programming knowledge, and final positioning is still done through the dozen-step process. It is also highly troublesome to position grids of adjacent FOVs, which will be later stitched together into a tiled panoramic image.
Thus, the tile/SED/array Interface (TSAI) was developed with the goal of enabling users to rapidly position large numbers of FOVs using an intuitive, interactive graphical interface. TSAI consists of two main components: 1) A web-based graphical user interface (web UI) for rapidly placing TMA points and tissue tiles, and 2) Integrations into the MIBI user control interface for generating a tiled SED image and adjusting FOV positions. If only using the optical image, many FOVs can be roughly positioned and then quickly adjusted using the FOV navigation/adjustment tools (Figure 2, TSAI, left branch). However, if the SED tiling is performed, FOVs can be accurately positioned on the tiled SED image without needing further adjustments in SED mode (Figure 2, TSAI, right branch). Of general interest to hundreds of current MIBI users, these tools make tiling and TMA positioning very simple even for novices and reduce complex MIBI run setups from several hours to a few dozen minutes.
1. Loading of TSAI
2. Loading the MIBI slide and creating a template file
3. Optical image-stage motor coregistration
4. Tiled SED scan
5. Tissue microarray (TMA)
6. Area/polygon tile
7. FOV navigation and adjustment
8. JSON file generation and import
TSAI provides two methods for setting up FOVs (Figure 2). One uses only the optical image (Figure 2, TSAI, left branch), similar to other existing methods. The second method – generating a tiled SED image – is unique to TSAI (Figure 2, TSAI, right branch). TSAI draws FOVs accurately onto this image, eliminating the need to spend hours nudging FOVs into place in the manufacturer interface SED mode. However, the correction coefficients for SED tiling must be set properly, or else the resulting FOV positions may not be accurate.
SED tiling coefficients can be assessed by inspecting the tiled SED scan of a large tissue section, at least 1.5 cm in size (steps 4.1 to 4.11). The expected result of a proper setup is an image with minimal misalignments in the tissue between adjacent tiles (Figure 3A-B). If the SED imaging square is at an angle relative to the stage motor, or if there is step motor miscalibration (e.g., a command to move 400 µm moves 390 µm in reality), then the tiled SED scan will show significant misalignments, duplicated areas, and/or gaps (Figure 3C-D). Ideally, these motor inaccuracies would be corrected by the manufacturer such that all the coefficients could be set to 0. Otherwise, coarse manual corrections may be performed using patterned slide burns (steps 4.12.1 to 4.12.8, Supplementary Figure 6) before finer corrections using large, tiled SED scans (step 4.12.9).
To further check positioning accuracy, one may verify that the FOV squares drawn on the tiled SED image match those shown in the manufacturer interface SED mode after navigating to the FOV coordinate (Supplementary Figure 1A-B). One may also use the FOV navigation/adjustment tools (step 7) to quickly perform these checks for many FOVs.
Figure 1: Current MIBI user control interface. MIBI user control interface showing a representative slide with tissue sections of the brain (upper two tissues) and a small tissue microarray (TMA, lower eight circular tissues). (A) The optical camera image of the entire slide. The yellow crosshair (X) denotes the current location of the SED image. (B) The secondary electron detector (SED) image of the location marked by the yellow crosshair. Each side of the imaged square is 1400 µm. (C) The import fields of view (FOVs) feature. Please click here to view a larger version of this figure.
Figure 2: Schematic of Tile/SED/array interface (TSAI) workflow compared to standard FOV setup. First, the slide is loaded into the MIBI. Standard FOV setup involves positioning FOVs one at a time. TSAI proceeds with optical image-stage motor coregistration (step 3) and loading the optical image into TSAI. One branch of TSAI (left branch) uses only the optical slide image to roughly position FOVs (steps 5 and 6), followed by fine adjustment using FOV navigation/adjustment tools (step 7). The right branch generates a tiled SED scan (step 4). FOVs are accurately positioned on the tiled SED image (steps 5 and 6) without the need for further adjustment. After creating the .json file and importing it into the MIBI user control interface (step 8), the MIBI operation proceeds per protocol. The tiled SED scan shows the scale at the bottom right of the image. Please click here to view a larger version of this figure.
Figure 3. Representative tiled SED scans of tonsil sections. (A) A properly tiled SED image shows minimal misalignments between SED tiles. (B) The dotted area is enlarged. (C) Suboptimal setup results in large misalignments between SED tiles. (D) The dotted area is enlarged with arrows pointing to discontinuities. Scales are shown at the bottom right of each image. Please click here to view a larger version of this figure.
Figure 4. Interactive graphical interface for positioning tiles/FOVs on a tissue microarray (TMA). At the left, a tiled SED image of a TMA is loaded into the TSAI web UI, with the scale bar shown at the upper left. On the right is the tiles column containing the tile/FOV options and tools. As described in step 5, (A, B) users first set the pattern of FOVs to copy, (C-F) then set the dimensions and naming options. (G-J) After clicking and adjusting the corners and (K) building the tiles, (L) move the tiles into place or (M) remove tiles when there is no tissue. Please click here to view a larger version of this figure.
Figure 5. Interactive graphical interface for building a tile with FOVs covering a polygonal area. At the left, a tiled SED image of a tonsil is loaded into the TSAI web UI, with the scale bars shown at the upper left. At the right is the Tiles column containing the tile/FOV options and tools. As described in step 6, (A) users select the polygon tool, (B-C) then click to set the vertices, (D) and double click to close the polygon. (E) Expanding the menu shows additional options. (F) The FOV map can be edited by clicking the checkboxes. FOVs can be individually toggled on and off by clicking on the slide image, (G) toggled on by click and drag , (H) or erased by click and drag. Rows may be inserted (I) above or (J) columns inserted to the left. (K) The entire tile can be moved by click and drag. Please click here to view a larger version of this figure.
Supplementary Figure 1. Current workflow for setting up FOVs using the manufacturer's user control interface. Using the optical image panel at the left, the user roughly positions a single FOV. Then, they fine-tune its position through a sequence of 12-13 steps: 1. Type the FOV X coordinate into the stage position X coordinate text box. 2. Type the FOV Y coordinate into the stage position Y coordinate text box. 3. Click Move. 4. Click Jog Stage. 5. Click the Arrow buttons in the jog stage popup window to adjust the stage position until the SED image shows the desired area for imaging. 6. Optionally, use a screenshot tool to capture the SED image and crosshair overlay. 7. Write the new adjusted X coordinate on a piece of paper. 8. Write the new adjusted Y coordinate on a piece of paper. 9. Click on the Pencil/Paper icon to edit the FOV coordinate. 10. Type the coordinate from step 8 into the FOV center point X text box. 11. Type the coordinate from step 9 into the FOV center point Y text box. 12. Select the appropriate section from the drop-down menu. 13. Click Confirm. The position for that single FOV is now adjusted. This sequence is repeated for every FOV, with some slides exceeding 100 FOVs. Please click here to download this File.
Supplementary Figure 2. Setting up a slide, section, and panel in the MIBI experiment tracker, per usual MIBI operation protocol. (A-B) Before a slide can be loaded into the MIBI, users must use the Project Data/Slides tab to create a slide that contains a section. (C-D) Then, use the Resources/Panels tab to apply a panel to the section Please click here to download this File.
Supplementary Figure 3. Loading a slide, creating a template .json, and downloading the optical image in the MIBI user control interface. (A) As described in step 1, users perform a sample exchange to load the slide into the MIBI. (B) Then they create a template FOV, (C) save the. json and (D) save the optical image for import into the TSAI web UI. Each side of the SED image square is 1400 µm. Please click here to download this File.
Supplementary Figure 4. Performing optical image-stage motor coregistration. As described in step 2, users: 1. Open then optical coregistration menu; 2. Click the button to copy the code to the clipboard; 3. Open the MIBI user control interface JavaScript console; 4. Paste the code into the console and press Enter; and 5. Click on the resulting link. Please click here to download this File.
Supplementary Figure 5. Performing a tiled SED scan. A representative TMA slide is shown in the TSAI web UI. As described in step 3, (A-B) users first select corners of the rectangular area to scan. (C) Then click the button to copy the code to the clipboard. (D) After running the code in the JavaScript console and setting scan parameters (particularly gain and focus), they press Shift+T to begin the automatic scan. In (D), each side of the SED image square is 1639 µm. (E) After the tiled SED image has been generated, it is loaded into the TSAI web UI. (F) Slide display options such as zoom and brightness can then be set. In (F), the scale of the tiled SED image is shown at the upper left of the image. Please click here to download this File.
Supplementary Figure 6. Manual correction of SED misalignments. (A) A schematic of a five-square checkerboard burn, before (left) and after (right) adjustment of the correction coefficients as described in step 4.13. The x f(x) and y f(y) coefficients are corrections for the step motor being slightly miscalibrated in the x and y axis, respectively. The y f(x) coefficient is an angular correction where a movement in the x direction requires a correction in y. The x f(y) coefficient is an angular correction where a movement in the y direction requires a correction in x. (B) An actual five-square checkerboard burn, before (left) and after (right) adjustment of the coefficients. (C) An actual nine-square checkerboard burn, before (left) and after (right) adjustment of the correction coefficients. SED images shown are all 1300 µm on each side. Please click here to download this File.
Supplementary Figure 7. Generating the .json file and importing it into the MIBI user control interface. As described in step 7, (A) one should check the FOVs and estimated time before generating the .json file. (B) After selecting a grouping option and (C) a splitting option, (D) the user may change the order of the FOVs in the .json file(s). (E) By default, the Sequential .json orders FOVs just as they are ordered in the TSAI web UI – by tile, then row, then column. (F) The Random .json randomizes the FOVs within the groups specified in (B). (G) An image of the slide and tiles is useful for record-keeping and sharing with collaborators. (H) The .json is then loaded into the MIBI user control interface before starting the run. Please click here to download this File.
Supplementary Coding File 1: .zip file containing TSAI code. GitHub repository maintained at https://github.com/ag-tsai/mibi_tsai. Please click here to download this File.
Multiplexed ion beam imaging (MIBI) is a powerful technique for dissecting detailed cellular phenotypes and tissue histoarchitecture5,6,7,8,9,10,11. Computational efforts around MIBI have largely focused on processing the data after imaging, but little has been done to improve the instrument software's usability for common applications such as large TMAs and tiled tissue areas. As the scale and complexity of MIBI experiments have increased, the manufacturer-provided instrument interface and third-party tools have become increasingly unwieldy for these tasks12. Thus, the tile/SED/array interface (TSAI) was developed to enable users to rapidly position large numbers of FOVs using an intuitive, interactive graphical interface.
TSAI provides two methods for setting up FOVs. One method builds FOVs using only the optical image, while the other generates a tiled SED image upon which FOVs are built. The optical-only method operates within the same paradigm as the manufacturer's user interface – rough positioning using the optical image followed by precise positioning under the SED view. TSAI just accelerates the process with an interactive graphical layer (steps 5 and 6) and keyboard enhancements (step 7).
The tiled SED method (step 4) breaks the paradigm by providing precise positioning up front, obviating the need to finely adjust each FOV one-by-one using the SED view. Furthermore, users can perform light microscopy on an hematoxylin and eosin-stained serial section (adjacent slice of tissue), identify structures of interest, and more easily pinpoint their positions on the tiled SED image for MIBI imaging.
The critical factor in the tiled SED method is setting the correction coefficients to eliminate misalignments or, ideally, having the manufacturer calibrate the instrument such that all the coefficients can be set to 0. Large misalignments in tiled SED scans indicate that the coefficients need to be adjusted (Figure 3). Using patterned slide burns (steps 4.12.1 to 4.12.8, Supplementary Figure 6), one can quickly troubleshoot and perform coarse corrections before fine-tuning with subsequent tiled SED scans. This process for determining the coefficients typically takes several hours but only needs to be redone if the instrument hardware is modified. If the process is too troublesome to perform within an acceptable amount of time, one may default back to the optical-only method for setting up FOVs.
With regards to the code itself, TSAI has been tested on software versions 1.8 and 1.7 (specifically 1.8.2.1 and 1.7.0-0f60ffbc) and is immediately ready for use by hundreds of current MIBI users. As it stores settings in cookies, TSAI must reside on a website. Otherwise, it is written using HyperText Markup Language 5, JavaScript, and Cascading Style Sheets – existing standards already built into modern web browsers13,14,15. Thus, it requires no installation of plug-ins, extensions, or external programs or compilers such as Matlab, Python, R, Juypter, or Docker. TSAI leverages very helpful features of the MIBI user control interface – a way to import/export run parameters from a JSON file (Figure 1C) and being built within a web browser. Since web browser coding is platform-independent, with extensive graphics source code libraries and populous communities of programmers, users can integrate their own code much more easily than they could for a traditional, proprietary, compiled, closed-source Windows or Unix interface. Thus, instrument manufacturers may wish to embrace browser-based interfaces more often, empowering users to extend them beyond what the manufacturer originally envisioned and thereby spur further innovation.
The authors have nothing to disclose.
H. Piyadasa was supported by the Canadian Institutes of Health Research (CIHR) Fellowship (MFE-176490). B. Oberlton was supported by the National Science Foundation (NSF) Fellowship (2020298220). A. Tsai was supported by a Damon Runyon Cancer Research Foundation (DRCRF) Fellowship (DRG-118-16), the Stanford Department of Pathology, the Annelies Gramberg Fund, and NIH 1U54HL165445-01. Additional acknowledgments go to Dr. Avery Lam, Dr. Davide Franchina, and Mako Goldston for helping to test and debug the program.
MIBI computer | Ionpath | ||
MIBIcontrol (software) | Ionpath | ||
MIBIscope | Ionpath | Multiplexed Ion Beam Imaging (MIBI) microscope | |
MIBIslide | Ionpath | 567001 | Conductive slide for MIBI |
Tile/SED/Array Interface (TSAI) (software) | https://github.com/ag-tsai/mibi_tsai/ |
.