1. Comparison of High Resolution Whole Slide Imaging (WSI) vs
Conventional Fluorescence Microscopy
for Viewing and Analyzing
Multiplex Quantum Dot Immunostained (MQDS)
Microscopic Slides
No more lights off in the lab!
No more microscope!
Thank you for giving me this opportunity to present my project.
Today, I’m going to talk about approaches to viewing tissue specimens stained with multiple antigens. I will briefly introduce and compare the different approaches and what we need for next step – to incorporate this into the routine workflow of pathology.
The goal is to develop new tools to take advantage of the “digital revolution” in pathology by empowering immunohistochemistry and data analysis.
One Disclosure, before starting the talk, I would like to make it clear that , even though I will show equipments used for digital images, I am NOT representing or promoting any particular company or product.
Kumiko Isse, M.D., Ph.D.
Demetris Lab
Department of Pathology, Division of Transplantation
Thomas E. Starzl Transplantation Institute
University of Pittsburgh Medical Center

2. What is Quantum Dots (Qdots)?
Quantum dots are 10-20nm microspheres composed of a Cadmium Selenide core, which is encased by zinc sulfide, and then coated with polymers. The final hydrophilic exterior layer allows for conjugation of biological molecules, such as immunogulobulins and strept-avidin.
Qdots are similar in size to traditional fluorescence molecules and can be used in conventional immunohistostaining protocols.
The diameter of each Qdot determines the wavelength which is emitted by the exited molecules.

3. Traditional Fluorescence Markers vs Qdots (1)
This figure shows the absorption and emission spectra of conventional fluorescence markers.
Each fluorophore has a unique absorption spectrum and the emission spectrum peak is close to the absorption wavelength. This difference between the absorption and emission wavelengths is defined as Stoke’s shift. If the Stoke’s shift is small, excitation from other fluorochromes can bleed through and cause high background. Moreover, the emission spectras for traditional fluorophores are relatively wide and they often overlap each other.
Qdots are all excited by single UV light source. And each Qdot has a narrow, symmetric emission spectrum. This prevents overlap of colors. Furthermore, Qdots have large Stokes’ shifts, which helps to eliminate fluorescence bleed-through. Therefore, the images are very clear and crisp.

4. Traditional Fluorescence Markers vs Qdots (2)
Rhodamine
Rhodamine
3min
Qdot
Qdot 1H
This figure shows the comparison of a traditional fluorescence marker, rhodamine, to a Qdot.
Given the same excitation, the emission peak of the Qdot is 20 times higher and two-thirds less in spectral width than rhodamine. In addition, a Qdot is significantly more photostable. It remains at the plateau strength for hours (graph B). As many of you have experienced, traditional fluorescence emissions can bleach-out in minutes (graph C). The graph on the right shows that Qdots have a stronger signal than other conventional fluorochromes.
Adapted from 25 SEPTEMBER 1998 VOL 281 SCIENCE
Adapted from Invitrogen.com

5. Qdots No photo-bleaching Wide Stokes’ shift Narrow emission spectra
Permanent
Multiple staining
Expensive
Special filter
Therefore, Qdots are a very powerful tool. They are highly resistant to photo-bleaching. They also have a large or wide Stokes’ shift and narrow emission spectra resulting in clearer images. In addition, the stained slides are permanent and do not require slide storage in the dark or at 4C. Since the qdot emission is very strong, it is easy to separate individual colors on multiplexed stained tissues.
On the other hand, there are some issue to using Qdots. They are relatively expensive compared to other labels. And their use requires that specific filters be added to the imaging equipment.

6. Multiple Staining
Enable pathologists to contribute to the molecular revolution in medicine by merging traditional morphologic examination with multiple markers to precisely characterize specific cell types and investigate intra-cellular signaling pathways
Time consuming
Complicated protocol
The next question is “why do we need Multiple staining?”
Traditional histopathology needs to provide “more value” or more information from tissue examination.
As our reserch knowledge expand, we will become interested in more and more molecules. Especially in the immunology field, multiple markers are required to define one specific cell type. And if you want to see the specific cell’s function, multiple staining is necessary.
One drawback to multiple staining is that the staining protocol is time consuming and complicated.

7. Tissue Staining, not Flow Cytometry
Panoramic overview of tissue at low magnification
Distribution, localization and cell-cell interactions visible
Can unlock decades of human biology/pathology from paraffin blocks
Connect to conventional morphology (H&E)
Immediate sample collection and triage
Complicated to analyze
Artifacts (wrinkles, bubbles, dust, scratches, etc.)
Another question is, why do we need to do multiple staining on the sections? Because flow cytometry is a well established method for detecting multiple markers. However, flow cytometry provides no “tissue context” for inflammatory and neoplastic diseases. For example, we are not able to see whether lymphocytes are destroying bile ducts and arteries. And since the final goal is to understand the biology in human disease and have the information from basic science to return to the bed side, we cannot always homogenize the tissue for flow cytometric examination.
In contrast, tissue staining enables the pathologist to observe the global distribution of target cells and cell to cell interactions. And, obtaining paraffin blocks is relatively easy, and you can unlock decades of biology/pathology from the sample. Moreover, combining multiplex staining with H&E allows you to connect protein expression with conventional morphology all in one section.
Some negatives to tissue staining are, one, sample collection. It could be a bottle neck. Two, is analysis. Because many cells that you are not interested in are in the same field. Therefore, the analysis is not simple. Lastly, there’s a possibility that you get artifacts during the sectioning and staining process.

8. Protocol for Multiplex Quantum Dot Immunostaining (MQDS)
Requires the best antigen retrieval for all antibodies that you are going to use
Antigen retrieval
Avidin Block
Biotin Block
Non Serum Protein Block
1st Primary Antibody
1st Biotinylated Secondary Antibody
1st Streptavidin Qdot
Extra Blocking for each segment
Before starting the panel staining, titration of antibodies will be done by immunohistochemistry to decide the staining order
Avidin Block
Biotin Block
Non Serum Protein Block
2nd Primary Antibody
2nd Biotinylated Secondary Antibody
2nd Streptavidin Qdot
Amplify Signals by 2-step immunohistochemical staining
I’m going to show the experimental example. Before that,
For a better understanding of multiplex Quantum Dot immunohistostaining, I am showing a schema of the staining process. For multiplex staining we use a sequential staining method. In the first step, if the samples are paraffin embedded sections, one must select a single antigen retrieval that is best for all antibodies in the panel. The best antigen retrieval and titer for all antibodies is determined by immunohistochemistry, HRP reaction on positive and negative control slides. One carefully observes the staining patterns and signal intensities to decide on a protocol. Only after this is a multiple panel staining attempted.
Since we use an avidin biotin method, extra-avidin biotin blocking is done before each antibody. This is essential to block the cross-reaction of Qdots.
The reason for doing a 3-step IH is to maximize the signal intensity in order to minimize exposure time and diminish capturing autofluorescence .
Avidin Block
Biotin Block
Non Serum Protein Block
3rd Primary Antibody
3rd Biotinylated Secondary Antibody
3rd Streptavidin Qdot
Repeat for additional stainings

9. CD3 Ab + Biotynilated anti-rabbit
Comparison of 2-Step and 3-Step IHC
2-Step
3-Step
CD3 Ab + Biotynilated anti-rabbit
+ streptavidin Qdot
CD3 Ab + Anti-rabbit Qdot
These pictures show the difference between 2-step and 3-step staining. Using the serial sections, the same CD3 antibody was incubated. The left shows staining with a Qdot conjugated secondary antibody, and the right panel shows staining with a biotinylated secondary antibody followed by a streptavidin Qdot.
In theory, biotin binds to four avidin molecules, so that it gives us a signal that is 4 times stronger.

10. Nuance 420nm Pseudocolor Image Captured Image 720nm
Unmixed Grayscale Individual Images
720nm
Subtract AF by program
unmix
To capture multiple staining, special equipment is needed. This system is Nuance from CRI. It has a liquid crystal tunable filter on the top of the microscope, and captures the field at every 10nm wavelength from 420 to This pile of image data is called a cube and the program digitally unmixes the image into individual single spectra that you want.
This is the actual screen of the Nuance program. Left image is a captured image, bottom images are unmixed individual gray scale images, and the right panel is the final image reconstructed by pseudo colors. This program has a function to subtract autofluorescence so that the image is clearer than non-treated images.
420nm
720nm

11. Autofluorescence (AF)
in Different Tissue
Skeletal Muscle 2.36
Liver 1.84
Heart 1.51
Small Intestine 1.38
This slide shows autofluorescence patterns in different tissue types. The images were captured at the same exposure time, and components with high autofluorescence were captured. Examples include lipofuscin, platelets, red blood cells, and connective tissue. Although the intensity is different in each tissue, they all show the same spectral pattern. The liver shows a slightly different pattern than others, since it has some cholestasis.
Same AF pattern in different tissues
Frozen Liver 1.21
Colon 1.03
Lung 1.0

12. LCACD68CD3CD4CD8 + = + = LCA CD68 CD3 CD68CD3 CD4 CD4CD8 CD8
This is a staining example.
By following sequential 3 steps immunohisto staining and using a Nuance microscope, we were able to obtain up to 5 colors on one section. 4 out of 5 of those antibodies are mouse anti-human antibodies. As you see, macrophages and T cells, CD4 and CD8 cell are well separated.
CD3
CD4CD8 + =
CD4
CD8

13. Disadvantage of Traditional Fluorescent Microscopes
Physical problems:
Unpleasant usually isolated environment
Limited availability
Data problems:
Multiple layered image with lower opacity is unclear
Individual colors need to be saved separately =
Although the microscope has significant limitations
Physical problem: To capture images, one needs to be in a dark room the whole time staring into a microscope. Also, the special microscope usage is limited, and hard to find a time when it is available.
Data problem: Individual colors need to be saved for multiple staining. Also, layering multiple images and adjusting opacity with a program, like photoshop, made the image outcome unclear. In addition, changing the pseudo-color is not a simple procedure.
Mechanical problem: To maximize the utility of the microscope, you need to adjust settings. The microscope is not powerful for lower magnification. So, one tends to use this for higher magnification and not for global observation.
Mechanical problems:
Repeated training often needed
Precise adjustment of settings needed for good quality images
Suboptimal at low magnification

14. What is Whole Slide Imaging (WSI)?
Total 12 slides are scanned in one time
10-20min by 20x lens, min by 40x lens for biopsy size tissue
2.2GB with 80% compression of JPG
Using filters specific for Qdots
Inside of the scanner
Zeiss/3D Histech scanner
Qdot Filter
Then, what about whole slide image?
Our aim is “bringing multiplex Qdot staining into the whole slide image”.
This is a table-sized machine from Zeiss/3D Histech, now it is selling from CRi.
Inside the machine, there is a lens, we installed x40 objective lens to have better image quality. A maximum of 12 slides go inside the machine and can be scanned automatically. With this size of tissue, the data size is about 2GB as JPG format with 80% compression. This is our scanner. You can find two scanners in our lab, E1516 now. We modified them to have a camera changer to switch to bright field and the CCD camera for fluorescence. And we have 6 different Qdot filters set up.

15. Digital x100 Digital x10 Digital x20 x40
The idea of the whole slide image is shown here.
Capture one field with 40x lens, and move to the next field, to the next field, and stitch the image to build a digital 20x image, 10x image, up to whole tissue. Or, it can go the other way. The image can show you 100x image without oil.

16. 12bit vs 8bit
This is an example of our technical improvement. Using the same slide and same exposure time, the slide was scanned with 12 bit or 8bit camera. It is obvious that the 12 bit camera on the left with 4096 color depth produced a superior image compared to the 8 bit camera image on the right.

17. Advantage of WSI Permanent data
Share the same slide with many people at once
Observe anytime, anywhere, portable.
No need to reserve microscope
Easy surveillance and analysis
Preservation of context and detailed morphological information
Saves space in your lab
Large data
Mechanical problems
Requires lot of adjustment
Cost
Over all, having a whole slide imager gives us several advantages.
It’s permanent data. And it is not under the microscope, so you can share the same slides with many people at once, anywhere, anytime through a network or portable device. Once you obtain the data, it’s yours, and even if you want to take extra pictures, you don’t need to reserve the microscope room or pull out your slides from the box. Since the image is already saved as data, the workflow from capturing the image to analysis is smooth. After scanning the slides, you don’t need the glass slides, which saves lab space.
However, as I have shown, the data files are quite large. Also, the system occasionally requires some maintenance, updating, and adjustment for each condition. In addition, cost can be an issue.

18. Digitally Preserving and Sharing the World’s Cultural Heritage
The utility of slide scanning for pathologists has parallels outside of pathology.
The other day, I was reading national geographic magazine, and found a very interesting story. It was about digital graphics of historical landmarks for backing up history.
This is the NPO group, CyArk’s home page. This Tomb was a UNESCO world heritage site. It was straw thatched building built in In Feburary 2009, CyArk group scanned the building and saved all kinds of images. Unfortunately, about one year later, in March 2010, it caught fire and was destroyed.
Our samples are not at the same level as UNESCO, but for us, our slides and data are valuable.

19. (3D HISTECH/ CRi Pannoramic Viewer)
WSI : HLADRCK19CD31
(3D HISTECH/ CRi Pannoramic Viewer)
Let’s go back to tissue imaging.
This is an example of the multiple staining using the viewer. The tissue is a human liver biopsy. This frozen section is stained for HLADR, CK19 which is a bile duct specific cytokeratin in the liver, and CD31 for endothelial marker.
You can observe whole tissue, and zoom in to the target area. And turn on and off the multiple colors by just clicking buttons. You can change the color on the screen, too.

20. Data Obtained From WSI
5 fields from liver, all portal tracts in the biopsy using
three different antibodies + DAPI = 4 colors
Total 19 cases
------over 400 images
This is an example of results from whole slide image.
These graphs show delta 1 and delta 2 TCR positive T cell ratio in the liver in the parenchyma or portal tracts. In liver lobules, 5 different areas were chosen. For portal tracts, all portal tracts in the liver biopsy were counted. Total 4 individual images for each color was required. And total case number was 19. So, If this analysis was done by microscope, over 400 images would have to be saved.
By using whole slide imaging, it is easy to obtain data from different anatomical locations.

21. Microscope vs WSI Microscope x4 Microscope x20 WSI Digital x4
Adapted from Zeiss “Microscopy From The Very Beginning”
Microscope x4
Microscope x20
Then, what else is different from microscope to whole slide image?
Here I am showing the advantage of using Whole slide image. The images are from a slide stained with CD31 and captured in x4 objective lens or x20 lens using a conventional microscope. Lower images are from whole slide image captured with x40 objective lens and digitally re-constructed to be equivalent to x4 and x20 images. In 20x, the signals are almost the same in microscope and WSI, but in the x4 microscope image the sinusoidal pattern is lost. The reason is that the higher objective lens (B) captures more protons than a lower objective lens (A) because of the physical distance from the tissue.
WSI Digital x4
WIS Digital x20
CD31

22. Disadvantage of WSI Unfocused Area Limited Abs numbers
because of hardware
 Multiple focus points
 Dark field condenser
Limited Abs numbers
because of AF and DAPI
 Digitally subtract AF
However, fluorescence whole slide image still has challenges.
For example, it is very hard to completely avoid having areas that are out of focus, especially around the edge of the tissue in the automated scanning. To avoid this problem the scanner should have multiple focal planes. Also, other manufactures use a dark field microscopy technique which identifies the tissue, thereby improving the focus.
The second problem is that we can barely use Q525, since it overlaps with AF and DAPI. Better scanned images require distinct DAPI staining, but it makes it impossible for Q525 to overcome the DAPI. This problem might be improved if whole slide image data can be applied to the program which digitally construct images.
Lastly, there can be shifting problems. There are two cells, and Q705 locates on nucleus, but the signal overlaps to the next cell. The reason could be due to a physical position shifting of the filter or it could be the fluorochrome spectra shift itself. This problem should be solved by mechanical adjustment, but it is a difficult task to adjust a few micrometers. Another option, which we are doing now, is to move the layer by using the software.
Shifting Problem
 Mechanical adjustment
 Layer adjustment in the software
DAPI
DAPI + Q705

23. FarSight__ Developed by Dr. Badri Roysam
I’ve been talking about staining, and image capturing. So, the final step is analysis.
Creating multiplex stained WSI can be a very powerful tool, but the colors often overwhelm the human mind. Therefore, computer-assisted slide interpretation is needed.
To achieve this last step, we have established a collaboration with Dr. Roysam who is the developer of free open source software, Farsight. His website is shown here.
Briefly, the program is based on nucleus segmentation. The software locates and defines each nucleus in the microscopic field of view and then finds the geographic center of the nucleus. Then, any signal showing a spatial relationship to the center of that particular cell is assigned to that individual cell. It does this by, dividing the nucleus into 8 segments and then calculates how many total pixels and how many segments are around that nucleus. In other words, this software is able to assign characteristics on an individual cell basis.

24. FarSight__Nucleus Editor
HLADRIL10TGFbDAPI
I’m going to show you how this FarSight program works.
Yellow is HLADR, Pink is IL10 and Cyan is TGFb.
After loading a set of multiple color images as a xml file and load the definition file, farsight starts nucleus segmentation. Then, the program stops. This is the time that you can edit the segmentation. You can merge, carefully observe the outcome of segmentation, and if you don’t like this cell, divide, delete, and add segment. Then, re-start the program. This is a signal identification step. Then, load the training set. This set has all the information such as nuclear volume, signal intensity, segment’s number etc, which you previously chosen as positive and negative examples for each signal in the representative images.
Then, run classification. HLADR is yellow staining, and you see the positive cells with magenta and negative cells with cyan plots. If you find the cells, like here, 205 should be positive for HLA, and 172 should be negative, you can go back to the training system, and type in 205 as positive and 172 as negative cells. And run the classification, you see the correction. Next color, IL-10 is magenta signal, after classification, you see these double layered circle. The inner circle represents IL-10 positivity. If you turn on IL-10 channel, you can see the magenda in the center. We can do this classification up to three colors. So, the same thing can be done in TGFb. You can see the triple layers, and to confirm that, turn off other markers, and see magenta or cyan in the center of the circle.
Furthermore, the program shows scatter plot, and by chosing the x and y axis, for example HLADR and TGFb total signal, this area suppose to be a double positive cells, you see as a yellow segmentation.
This is a table which has all the information of each cells, and you can export this table as an excel file. Also, you can save whole this segmentation and classification result as a project set.

25. Data Obtained From FarSight
HLADR expression and HLADR +TGFβ+/- cell numbers
This is an example of using farsight. You can count HLADR positive cell numbers, and also see the positivity of TGFb in HLADR cells.

26. Problem of FarSight or Human??
But there are still challenges. Good automated digital image analysis requires crisp, distinct, and consistant staining.
For example, we compared a Farsight analysis to human counting.
N shows the number of fields from each case. The graph shows % of delta 1 or 2 positive cells in total CD3 T cells. The green bar shows discrepancy between Farsight and human analysis. This occurred because delta 1 TCR signal in these particular slides was weak in contrast against the background. The red and yellow bars were similar because the stain was better.
The question is, then, “what is the truth?”. To answer this question, our plan is to do flow cytometry and staining from the same tissue and compare the outcome.
N=3
N=8
N=8
N=8
N=4
N=10
N=10
N=10
X40 magnification, unknown field size
x50 magnification, 230x350μm2
Vδ1+CD3+
Vδ2+CD3+
Vδ1+2+CD3+

27. Combitnation of H&E and MQDS
H&E  Qdot multiple staining
H&E after Qdot multiple staining
Fluorescence signal
Even though multiplex staining is very powerful, there is still a need to return to findings on H&E stained sections. It is ideal to compare expression patterns of antigens or signaling pathways to the conventional H&E staining and routine morphologic examination. The conventional method has been to prepare serial sections to see the “same lesion”. This method is close, but you can never be sure at a single cell level. The question, then, is “Can we simultaneously view both multiplex staining side-by-side with H&E stain?” The answer is “Yes, it is possible if we use whole slide image”.
These images are from one field captured in brightfield and fluorescence signal. You see the strong signal in elastic fibers. In fact, Eosin has fluorescence.

28. Eosin emission spectrum on the top of Qdots
Eosin has wide spectrum
( Eosin)
Eosin is strong Acid
Hematoxylin is strong Base
pH Ranges for Qdot® Nanocrystals
This is a screen shot of Nuance system showing the eosin spectrum on the tissue. One can see eosin emission overlap Qdot signals. Moreover, since eosin is very low acid and hematoxylin is very high base, all antigen-antibody binding comes off from the tissue and either Qdot won’t be stable. pH
Recommendations
>9
Not recommmended- Qdot® nanocrystals start to self-aggregate/clump. (Qdot® nanocrystals are not degraded by basic pH. )
>6 to <9
Qdot® nanocrystals most optimal stability in this pH range.
>5 to <6
Marginal stability is shown in this range
<4
Not recommended- The polymer will dissociate; exposed core/shell will start to dissociate.

29. WSI
MQDS H&E
CD31CD34aSMA
So, using Whole slide scanner, one can observe both fluorescence and H&E in the same section at the same time, comparing protein expression and morphology side by side. This is an example of Qdot and H&E after captured Qdot staining. One disadvantage is there is some damage of the tissue, but it is minor.

30. Microscope WSI Conventional Method Easy to use, portable
Stronger signal at lower magnification
Combination of FL and Bright Field
Direct connection to automated image analysis
Bottleneck because of limited resource (location and time)
Microscope often difficult to use
Not portable
Limited triage to automated image analysis
Looking at a tree and not the forest
There are many factors that we can improve. So far, microscope is easier to install and more widely understood than WSI. But, one tends to look at a tree and not the forest under high magnification to obtain a strong signal intensity. And, if you go to a lower magnification, you loose some signal.
The scanner is easy to observe on the computer screen, and the image is captured by x40 lens so that even small signals can be observed, and that helps you survey a whole tissue better and give you a better idea of the target cells’ distribution and localization. The weak points of whole slide images are 1) unfocus area in the tissue. However this problem is specific for the scanner that we are using. The other problem is autofluorescence. It is not a critical issue so far, although, it would be nice if the program would subtract autofluorescence since AF pattern is symmetric in different tissue.
Lastly, these techniques are still developing, so we are still trying to find ways to improve some of these problems.
Unfocused areas
Autofluorescence
Adjustments needed

31. Future Plan
Analyze whole slide image
Automated whole slide image analysis using a selected region of interest (ROI)
Better performance
Hardware – computer, scanner, filter
Software – imaging, analysis
In the future, the aim is 1) automated whole slide image analysis using a selected region of interest. There is still a gap between multiple color stained whole slide image and the analysis program.
2) is to have a better performance in both imaging system and analyzing system. Those are beyond my ability, so I am looking forward to seeing other developer’s wonderful work.

32. Thank you! MIDI and system Consultant Andrew Lesniak
Rensselaer Polytechnic Institute
Roysam Lab
Dr. Badri Roysam
Kedar Grama
Demetris Lab
Dr. Demetris
John Lunz III
Susan Specht
Yoshiaki Mizuguchi
Natasha Corbitt
Enrico Pegolo
RHS Lab
Lisa Chedwick
Lori Perez
Trevor Benyack
Eleck Walton
Traci Ondik
ISH Lab
Kathy Cieply