The most important point can be summarized by one equation:

Failure/lack of test approval = no therapeutic product approval

If this doesn’t make get your attention, nothing will. An approved drug can cost from $200M to $1.2B to develop (wide variability in costs is largely due to indication type, you can make your own calculations). Let’s recast the equation in terms of the cost of failure for the companion diagnostic and the cost of failure for the therapeutic. A good example is HER2 IHC, which brings in $50M in total global annual sales divided across a handful of chemistry and image analysis players. Let’s assume the next big companion diagnostic is almost this big — and the firm has only four other competitors. So $10M in lost sales per year, perhaps spread over 10 years, and perhaps $10M spent on development. Total cost of failure for the CDx — $100M.

Now let’s look at the failure cost of the therapeutic due to CDx failure, continuing with the HER2 example. Herceptin had total ten-year sales from 1998 to 2008 of $7.7B, and has climbed higher the last two years. Let’s assume only 10 year sales, meaning total cost of failure is lost development costs of $1B and lost ten-year sales of $7.7B.

$100M =? 7.7B                    Lost Sales Cost
$10M =? $1B                       Lost Development Costs

One can see the imbalance clearly between the diagnostics and therapeutics industry. Having worked in both diagnostics and therapeutics, the only thing the two industries have in common is biology.

Clearly every pharmaceutical executive will re-examine carefully all the risks involved in their companion diagnostics programs, as CDx competency becomes a determining factor in success or failure for pharma. Does their translational companion diagnostic team have experience launching diagnostic products? Do they have experience in regulatory filings of 510k and PMAs? Is there any part of the diagnostic technology that is not already well proven and widely used in clinical laboratory settings? Has their CDx internal pharma team or external CDx team actually worked in clinical laboratory settings?

Six types of diagnostics were given as potential CDx:

1) Identify patients likely to respond or not respond to a particular treatment
2) Identify subgroups of the larger population with poor prognosis who are likely to benefit from a particular therapeutic product
3) Identify patients likely to be at increased risk for serious adverse reactions as a result of treatment with a particular therapeutic product
4) Monitor response to treatment for the purpose of adjusting treatment (schedule, dose, etc) to achieve improved safety or efficacy
5) Individualize the dose of particular therapeutic product
6) Use as integral part of therapeutic clinical trials conducted to support market approval of a therapeutic product

The guidance provides advice for both therapeutic and diagnostic test labeling. If the therapeutic has been shown to be safe and effective only in a certain patient population, the Indications and Usage section must clearly define the patient population in whom the drug is approved. If the test is essential for monitoring (tox or efficacy), the Warnings and Precautions must specify the type of test. The therapeutic label will likely refer to the FDA approved test, but would not refer to a specific product test. This will allow a new equivalent test to be used without drug labeling changes. Conversely, the companion diagnostic labeling will include the intended use with the specific therapeutic product.

The guidance generally leans very strongly towards the use of a PMA rather than 510k. A new therapeutic indication by the diagnostic, for example would result in a new PMA being required. This would be a substantial and probably costly change in the industry, meaning for example that current diagnostics for HER2 with IHC or FISH would need new clearances when used on a new drug beyond Herceptin.

Regarding enforcement, another key change is that the FDA expects to require compliance with the device regulations regardless of whether the test is lab developed or distributed as a kit.

During product development (investigational use), the document states that the diagnostic can either follow an IDE or be part of the therapeutic IND filing. This would be a substantial change, as most INDs do not currently consider a companion diagnostic as part of the process.

The Guidance strongly recommends the Pre-IDE meeting, which given the complexity of these processes, will likely become standard process in the future.

ABSTRACT:

Histopathology peer review is a vital part of preclinical toxicology studies and has typically been conducted by on-site pathologist evaluation.  With emerging digital pathology technologies, successful remote peer review may now be achievable.  For many years pathologists have utilized photographic and digital images, both high magnification and whole slide tissue sections, for ad-hoc peer reviews and alternate opinions.  This case study documents a peer review process utilizing entirely digitized tissue sections and remote access of the images.  A previously peer-reviewed and closed study was used to study remote peer reviews processes.  Whole tissue sets from pre-determined animals in control and high dose groups and target tissues from other dose groups were scanned and housed on a dedicated server.  Individual animal data and a link to the images were provided to three veterinary pathologists, with extensive peer review experience, who accessed and evaluated tissue images from remote locations (other than the physical server location).  The evaluation was conducted to see if the peer review pathologist could identify previously diagnosed lesions and assess the primary pathologist’s evaluation.  Information was provided by the pathologists on the virtual and remote peer review process and experience.  This information will enable hardware and software vendors, and fellow veterinary pathologists to understand the pros and cons of remote peer review.

8 August 2011 — Fenway, MA. After many years of anticipation, a killer application has finally been discovered for digital pathology, by researchers at the Fenway University in Boston, Massachussetts.“One would think that simply being able to work from anywhere, share anything, with a full audit trail of what was viewed, and the ability for the computer to assist in scoring would be enough to drive digital pathology adoption,” said Dr. Mark Lowell, Professor at Fenway University. “However, the industry has been waiting for a killer application to drive adoption, and we believe we have finally found it.”

“We studied pathologist patterns while at the microscope and while traveling to peer reviews, conferences, and tumor boards, and we were struck by an amazing pattern. Pathologists that regularly flew through Newark Airport, were far more likely to go digital in their work than those that did not fly through this airport. Basically, we modified an advanced pattern recognition software that was previously only used in over-training results in gene expression datasets, and applied this to pathologist commuting patterns. The trend was consistent everywhere in the United States, the more a pathologist has to travel through Newark Airport, the more likely he or she will stay home and read slides digitally.”

“These results fit well with other evidence we examined in our algorithm,” said Dr. Karlton Phisk, a co-author in the study. “First, a ranking of airports has Newark rated first for the most delays. Second, we noticed that pharmaceutical pathologists seem to be adopting digital slides faster than clinical anatomic pathologists, and we can attribute this directly to them having to fly more frequently through Newark for corporate pharma meetings. Third, the Cambridge area of Boston is adopting digital pathology faster than other parts of the United States. Clearly Boston pathologists hate having to travel through the New York area more than other pathologists would, given the historic rivalry between these two cities.”

“Saving even one trip through Newark is well worth the purchase of multiple scanners,” said Dr. Karl Yastemsky, a third author on the study. “Actually, avoiding New York City for any reason is worth spending a few additional seconds to view the images digitally versus with glass.”

The results are not without controversy, on both sides of the Atlantic. In the United States, Professor R. Ruffen of Yanqui University in New York City strongly disagreed with the study’s conclusions. “First, scientists in Cambridge will buy anything, and second, Newark is a beautiful airport.  You can see all of New York City multiple times while circling the airport on most flights. If we applied the logic used in this study, we would expect to see British pathologists also adopting whole slide imaging faster than their peers, because Heathrow Airport is one of the worst to fly through.”

The study’s original authors disagreed with Dr. Ruffen’s logic challenge comparing Heathrow to Newark.“Everyone knows that British pathologists will take every chance they can get to travel, in hopes of escaping bad food and bad weather, so the effect of Heathrow is a net neutral effect,” said Mr. Jon Riddeck, a up and coming star and graduate student at Fenway University, and the fourth author on the study. “Although to be fair, the British breakfasts are quite good, but probably not enough to keep pathologists from traveling abroad” he added.

At a recent pathology meeting, several English pathologists first apologized for their Heathrow Airport as well as their weather and their food, but then asked why they were included in this controversy, that seemed entirely American in nature and had nothing to do with them.

Submitted anonymously to avoid reprisals from anyone and everyone

The novel staining approach combines the multiplexing capabilities of fluorescence with the tissue context of brightfield whole slide scanning

FLAGSTAFF, Ariz. — March 17, 2011 — Flagship Biosciences announced today a novel whole slide imaging approach called Bridge Staining™, that allows a brightfield background to be aligned and overlaid on a fluorescent image on the same tissue section, allowing visualization of tissue morphologic features on the fluorescent image. The patent-pending technique is versatile, and can be used either with three to four color fluorochrome stained sections or autofluorescence or polarized scanning.

“While generally we prefer to work with brightfield pathology, but fluorescence presents useful applications in multiplexing of proteins in single tissue sections,” said Dr. Frank Voelker of Flagship Biosciences. “We utilize whole slide fluorescence scanning and analysis when two to four antibodies are required to be evaluated in a single section. However, fluorescence makes the determination of the local tissue context difficult for pathologists. This new approach allows immunofluorescence to be conducted as usual, but with the additional step of adding chromogen staining without interfering with fluorescence emissions. Protein expression can be quantified in tissue with fluorescence, with the advantage of using the chromagen staining to assist in pattern recognition of the regions of interest.”

The multimodal scanning is another in a long list of whole slide scanning and image analysis services offered by Flagship Biosciences, utilizing both whole slide brightfield and fluorescence scanning. The Bridge Staining™ approach is being refined in conjunction with a number of tissue-based companion diagnostics programs with Flagship’s pharmaceutical partners.

The new digital pathology technique focuses on vascular access to the tumor tissue rather than the vessels themselves

FLAGSTAFF, Ariz. — March 1, 2011 — Flagship Biosciences announced today a vessel proximity analysis algorithm, which can quantitatively define the spatial relationships of tumor cells and vessels on a whole-tissue level in either histology or radiology imaging. The patent pending technique is a departure from the standard industry approach of quantifying the vasculature based on an endothelial stain in histology sections or a radiology tracer in three dimensional imaging.

Microvessel density (MVD) has been the industry standard approach for several decades for vascular analysis in histology tissue samples. It normally consists of using an endothelial stain to identify the vessel walls, and more recently can incorporate other secondary stains for muscle cells that are associated with these vessels. Once the vascular structures are identified, a microvessel density measurement is determined using a range of related approaches, either involving individual vessels or a composite area measurement of the overall vasculature. The techniques are widespread in research applications in many pharmaceutical companies, but have seen limited application in common clinical practice.

The challenge in using microvessel density is the wide ranging variability of the assessment in real-life clinical samples. The vasculature can range from single well defined vessels to a vast network of interconnected structures, making reproducible quantitation difficult. In addition, neovascularization driven by tumor expansion is different and more irregular than normal vascular architecture, and this irregularity is not captured by current microvessel analysis approaches.

Flagship Biosciences takes a fundamental new approach. The starting point is the tumor tissue itself, and measures the percentage of tumor tissue that is within an empirically set distance from vessels. This distance can be determined using histology markers for oxygenation (or lack thereof). The results provide a relative measurement of the oxygenation of tissue, rather than a measurement of the number of vessels.

Flagship’s pathologists and image analysis experts have participated in many different microvessel analysis projects. “The fundamental challenge in the current image analysis techniques that start with the vessels is the difficulty in reproducibly defining single vessels in a section of tissue,” said Trevor Johnson, Director of Image Analysis at Flagship Biosciences. “This often means that in real-world practice one must revert to a simple area measurement of tumor vasculature, which does not capture the distribution or geometry of the tumor architecture.”

The new technique depends on first discriminating tumor from adjacent normal and/or necrotic tissue using pattern recognition or new stain-assisted pattern recognition techniques like Feature Analysis of Consecutive Tissue Sections. The vessels are identified using techniques already common in the industry, and do not need to be individually segmented. The percentage of tissue within a given distance of these vessels is computed.

“We are working on studies with several pharmaceutical clients to determine the extent that this approach will better correlate with other measurements of a compound’s efficacy,” said Steven Potts, CEO of Flagship Biosciences. “In the rare case that the vasculature is evenly distributed in tissue, the analysis would perfectly correlate with microvessel density type measurement. However, we believe that this new approach may provide new endpoints that better match the fundamental question of neovascularization – how well oxygenated is the tissue?”

The technology approach is equally applicable to both two-dimensional histology sections and three-dimensional radiology imaging.

Below is representative images from five whole slide images of a protein that shows primary expression in the cytoplasm. Deliberately, we are using a difficult example, because of the wide variation in staining patterns from slide to slide. We will train on one slide, and then apply to all other slides with no adjusting of parameters between slides.

Example sections from five whole slide images in a diverse nuclear cytoplasm marker

Let’s walk through the steps coded in this example, using one of the five as the test set and then applying the result with no tuning of parameters to the other four slides. Implementation in C++ makes the most efficient, fastest running algorithm.

First, we will run color deconvolution (from Ruifrock and Johnson’s seminal paper in 2001 on color deconvolution).  Notice that there may be two approaches to separating the target from non-target tissue – either use the morphological differences in the nuclei in hemotoxylin, or the differences in staining in the DAB. We could use either, or a hybrid, but in this example we will use the latter, the differential staining pattern in the DAB stain.

Color deconvolution into hemotoxylin and DAB

Now we run a filter to smooth the DAB image slightly and then use Otsu’s method applied globally, looking for two classes, positive DAB stained target tissue and negative DAB stained non-target tissue. We use a number of statistics based thresholding methods, but Otsu’s (named after Nobuyuki Otsu) works best in examples like these where we cannot make any assumptions about the images in the experiment, especially the percentage of the image covered by target and non-target tissue.

Smoothing and thresholding of DAB using Otsu's method applied globally

The next step is find and filter out nuclei based on the DAB mask prepared above. This is shown below — there are a wide number of ways to find nuclei, this particular implementation does not involve any hardcoded thresholds, but rather is looking for objects of this approximate size and shape.

Nuclei are identified and then filtered if they appear in non-target tissue

Now we want to identify cytoplasm and then determine the level of protein expression based on DAB in these cells. In this case the cell membranes are not differentially stained (or left unstained), so we approximating the cytoplasm two ways, first with a hybrid propagation / watershed approach and second with a defined distance approach (e.g. donut appearance).

A hybrid propagation / watershed approach to find cytoplasm from filtered nuclei

Notice that this method performs remarkably well, although there is some concern in the southeast corner, where there was not enough nuclei to effectively assign cytoplasm. To avoid this, let’s look at a second approach for defining cytoplasm, that defines cytoplasm as a circular distance from nuclei. This should generate similar results to the propagation/watershed hybrid, and in our in-house image analysis services we often will run both approaches to make sure that there is minimal differences between them.

A distance-based approach to finding cytoplasm

Finally, we apply the propagation / watershed hybrid to the other 5 whole slide images. The results from the same representative regions are shown below. We can then compute statistics with differential ratios between nuclei and cytoplasm across the target tissue on each slide. The two approaches should yield similar results.

Cell-based cytoplasm results using a propagation / watershed hybrid

The fifth annual retreat kicked off on Monday in sunnny Miami Florida in a packed room of 120 pathologists. Every year this conference provides the most in-depth analysis and discussion of best practices in immunohistochemistry. Below are some notes on the first day:

Conference started with a pretest of 20 questions, presenting various cases and diagnoses. For those attendees new to this conference, the level of questions certainly reflected the depth of knowledge presented on IHC at this conference.

Stephen Hewitt, NCI. Standardization of Preclinical Variables

First talk by Stephen Hewitt at NCI was given remotely.  “The H&E slide will likely remain the most valuable tool in diagnostic pathology for the next 100 years — the question is how to improve it.” Steve’s lab specializes in high-throughput quantitative pathology, and has been dealing with standardization in research settings for many years. The main topics he discussed were the pre-analytic factors. The talk followed an approach called “Fit-For-Purpose” –  matching process with goals. Some interesting points:

• Looked at the impact of various pre-analytic factors, as measured by RNA yield. RNA shows substantial degradation in first 24 hours, especially at temperatures higher than 4 degrees C.  The challenge of using IHC as an endpoint in measuring quantition is the variability of antigen retrieval, which while extremely flexible, makes quantitation with IHC difficult. Editorial comment: I’d like to see impacts on protein expression, rather than RNA, as the majority of the audience deals with protein in daily IHC practice. I appreciate that RNA is easier to measure than protein.
• Looked at three preanalytic factors:  fixation time, processor time, and fixative buffer.

1. Fixation time: Short fixation is more dangerous than overfixation, fixation not the only important parameter. 24hours seemed optimum from RNA data, 48 and 72 hours were poor (Don’t leave tissue in formalin over a weekend).
2. Fixative buffer: Use a phosphate buffer for RNA, commercial suppliers have different and proprietary formulations for the phosphate buffers. Pay attention to quality – it would help if vendors would expose their formulas for these buffers
3. Processor time: Would have expected processor time optimum to be short, commonly seen 5 to 45 minutes, most people assume shorter is better. THE LONGER YOU PROCESS THE MATERIAL, THE BETTER THE QUALITY! This rather surprising result was explained by this being a hydration process –  the more removal of water the better.

• Process time and fixative time far more important than archiving time. Degradation due to poor processing
• Humidity a bigger problem than temperature as a degradation effect. EFFECTS OF WATER IS SUBSTANTIALLY GREATER THAN TEMPERATURE! (both exogenous water and endogenous water has effects.
• Higher quality tissue processing with dryer tissues helps with all other effects
• FFPE Stability: length of fixation plays almost no role, tissue processing is key
• Exploring alternative fixatives: Many of the older fixatives that have beautiful histomorphology are highly acidic, and acidic characteristics generally bad.
• Differences between verification and validation: Europe and US use these terms differently. Verification is defined as “measuring what you intend to measure,” and validation is “determining the expected outcome”. Diagnostic versus prognostic validation (nice example of ER being prognostic in breast cancer, where PR is really diagnostic, as no differential therapeutic is given based on PR).
• Total Test: validation applied to the entire test. any element changes requires revaliation of whole test. Gives flexibility by focussing on end result.

RECOMMENDATIONS:

• Grossing: rapid from patient to pathology lab
• Fixation: 10% neutral buffered formalin, minimum 10 volumes, 24 hours fixation time, plus/minus 8 hours, alternative fixatives must be documented and justified.
• Tissue processing: longer is better, goal is 30 mins/station. Replace/replenish reagents weekly
• Store slides dry, the drier the better.

“This is an exciting approach to generating an ongoing dialog between members in the area of biosurfaces,” said Dr. Lawrance Salvati, President of the Society. “It will allow discussion of topics throughout the year and expose members to new technologies and techniques in the biomaterials area, without the time and cost of meeting in person.”

The first Beneath the Surface seminar will be held on January 27th at 1:00 p.m. EST. The speaker will be Dr. Steven Potts, on ”New imaging approaches for quantitative analysis of the tissue-biomaterial interface.”

The February speaker will be Dr. Melissa M. Reynolds, Boettcher Investigator and Assistant Professor in the School of Biomedical Engineering, Colorado State University, speaking on, “Synthetic materials that maintain homeostasis”.

The March speaker will be Dr. Robert Kellar, President of Development Engineering Sciences and Faculty at Northern Arizona University, speaking on “Cardiac repair using regenerative medicine”.

Registration is on-line. While the main audience is Society members, the general public can register at no cost for a limited time for these first three science webinars

With the increased adoption of both brightfield and whole slide fluorescence scanning, the accessibility of skin samples seems ripe for digital pathology applications. But there are good reasons for going slow in this clinic area:

1) Questions of dermatology diagnostic equivalency of glass versus image. While the whole slide imaging technology keeps improving, Jonhan Ho et al (University of Pittsburg, 2008) rightly mentions concerns with the whole slide image in clinical dermatology usage. A recent publication by Bjarne Nielsen et al (Aarhus University in Denmark, 2010), is positive on skin tumor virtual microscopy provided pathologists have completed a period of digital pathology training. The good news overall is scanning technology keeps getting better, especially in reducing poorly scanned areas that are unacceptable in a clinical practice.

2) Complexity of dermatopathology. The skin may be the only place where the surface really is the most complex. There are an estimated 1500 different rashes and skin tumors, including variants, making dermatology and dermatopathology among the most complex specialties of medicine. How exactly does one approach an equivalency study design with this much disease complexity? It would challenge even a Dr. Holger Lange to device an adequate regulatory digital pathology study design for dermatology.

3) Complexity of image analysis approaches. One needs to first be able to have the computer identify layers of the derm, prior to looking at feature-level analysis (e.g. cell and object counting, or geometry measurements in a given layer). Layer analysis requires a white-box or rules-based programming that combines geographic knowledge (“Which layer am I in?”) with textual feature recognition (“Hey computer, this layer’s texture looks like this…”). In this sense it is similar to layer-based ophthalmology image analysis. It takes substantial time to write these layer-based detection algorithms and one has to constantly verify that the assumptions made in the algorithm match the tissue being analyzed (e.g as a superficial example, an algorithm looking to identify five stratum layers may work in thick skin epidermis, but will fall apart when working in thin skin with a missing stratum lucidum and only three or four of the five layers).

4) Economics of clinical dermatopathology. Most dermatology practices do not send out many of their dermatapathology cases, and cannot afford the hundreds of thousands of dollars for digital pathology scanners and software, which would not change how they would practice their discipline (at least not yet).

Despite these limits in clinical dermatology digital pathology adoption, we are excited about the possibilities that whole slide imaging brings to dermatology research and pharmaceutical clinical trials. Quantitative data on both efficacy and toxicology is key to all stages of pharmaceutical product development, and skin is no exception. To be successful, the dermatology trained pathologist must work closely with an image analysis expert. Particularly in skin, the specific image analysis design must be discussed beforehand with a pathologist, and a pathologist needs to review and sign off on each result. This is true whether the study approach is simple, like increased collagen or dermal thickness measurements, or complex, like multiplexed IHC melanoma proliferation studies.

Interestingly enough, thanks to the efforts of some remarkable pathologist pioneers in dermatopathology, the dermatology field has historically been described in algorithmic terms. Dr. Ackerman’s book in 1978 is the classic work, and written in a rules-based approach. There have been multiple new editions, but the original 1978 textbook, if you can find one, costs thousands of dollars, and is worth far more in the contribution he provided to this field.

As a digital pathology CRO, we do a lot of work in the area of skin samples, whether the sample is for implanted device material evaluation, cosmetics studies, or dermatopathology product development. This seems an opportunity for comparative pathology approaches, and the opportunity to participate in dialogs between veterinary and MD pathologists in developing dermatology image analysis applications is truly a privilege. However, finding MD dermatopathologists who have the interest, time, and training to be involved in dermatology product development is not easy.

Of all the various organs where digital pathology will have a major impact, the complexity of dermatopathology is perhaps the most humbling to image analysis experts. We are just scratching the surface.