Special Report on Disease Modeling: So life-like

It’s your son’s big day, his birthday, and he is surrounded by friends, cake, balloons and a ton of wrapping paper. But he’s been bouncing off the wall waiting for his gift from you. With a big smile, you give him a beautifully wrapped box.

 

“I know how much you love airplanes,” you wink as he rips into the package like a hyena on carrion.

 

Desperately, he claws at the top of the box and reaches inside to withdraw…a single sheet of blank printer paper.

 

You beam with pride. He stares confused. His friends stare at their shoes.

 

“I couldn’t get you an actual plane,” you explain. “But if you fold this just thus and so, it’s a pretty good approximation.”

 

A decade later, the same boy struggles at his lab bench to develop a new drug compound, when suddenly another scientist runs into the lab, all excited and carrying a small case.

 

“You have it?” the boy smiles, his friend nodding like a hypercaffeinated bobblehead.

 

The boy rips off the cover and reaches inside to withdraw…a culture flask of pinkish cells.

 

“I couldn’t get you an actual prostate,” the friend explains. “But if you shake this just thus and so, it’s a pretty good approximation.”

 

 

It is undoubtedly true that the biggest expense in developing a new drug and getting it to market is accommodating the failure of a molecule to translate preclinical success to the clinical setting. For any number of reasons, something is often lost between the efficacy and safety of a compound in an animal or cell culture model of a disease and in patients who actually have the disease.

 

“We have found more ways to cure heart disease in mice than you can imagine,” says Brian Wamhoff, co-founder and vice president of research and development for HemoShear, a company working on more physiologically relevant in-vitro models of human disease.

 

Wamhoff’s comment echoes the sentiment expressed years ago by oncology specialist Judah Folkman, who suggested that medical research has become very good at curing cancer in mice.

 

“You can create models of fatty liver disease in a mouse,” says Wamhoff. “It looks like it; it smells like it. But how that mouse develops fatty liver disease is completely different than how a human does it.”

 

“So you develop a drug to treat fatty liver disease in a mouse with a target that may or may not exist in a human, and then you go into a human and wonder why didn’t this work or worse, why is it causing liver injury now?” he adds, giving voice to the frustration felt across the pharma industry.

 

As Wamhoff suggests, part of the problem may be that in many animal models, a disease is just that: a model. It gives all the outward appearance of being, let’s say, rheumatoid arthritis (RA). The joint inflammation may show the same pathophysiology as human RA, but the question becomes whether it is really the same condition at the molecular level, be that gene expression or metabolic pathway perturbation.

 

And even if the disease is the same, does the compound react with the rest of the model animal’s physiology as it does in a human? Is the animal more or less tolerant of the test compound? Or are there unforeseen off-target effects to which the animal is less prone or completely immune?

 

Highlighting the ubiquity of the frustration of insufficient animal models, Wamhoff points to comments made by Elias Zerhouni, former director of the U.S. National Institutes of Health (NIH) and current president of global research and development at Sanofi, in June 2013.

 

“We have moved away from studying human disease in humans,” Zerhouni lamented to the NIH’s Scientific Review Management Board meeting. “We all drank the Kool-Aid on that one, me included.”

 

“The problem is that it hasn’t worked, and it’s time we stopped dancing around the problem,” he continued, suggesting researchers have become too reliant on questionable animal data. “We need to refocus and adapt new methodologies for use in humans to understand disease biology in humans.”

 

And that shift away from studying human disease in humans has potentially been expensive.

 

“It takes about seven years to get into the clinic and anywhere between $50 million and $150 million depending on what you’re developing,” Wamhoff suggests. “If the target and the disease biology you’re starting with from day one are wrong, you lose seven years.”

 

Thus the interest in moving back to more human-based studies, and the opportunity for companies like HemoShear.

 

“What our partners are telling us now is that we want to start with more meaningful targets and more meaningful human disease biology,” Wamhoff continues. “It may still take seven years, but after those seven years, we’re going into the patient for the first time with more understanding of the human disease than we’ve ever had before.”

 

The recent research of Robert W. Davis and colleagues in the Inflammation and Host Response to Injury, Large-Scale Collaborative Research Program may point to molecular reasons why the translation of results from mouse to human may be so difficult.

 

Publishing their results in PNAS in early 2013, the researchers examined gene-expression profiles in both humans and mouse models of trauma, burn and endotoxin assault. They found that the genomic responses across the different inflammatory stresses were highly similar within the human populations but that these patterns were not reproducible in the mouse models of the same stresses, suggesting a disconnect that could easily translate to different responses to potential treatments.

 

To some extent, they suggested, the differences could be explained in evolutionary terms, with different immune systems maturing from different environmental stressors.

 

“Relative to the human response, mice are highly resilient to inflammatory challenge,” the authors wrote. “For example, the lethal dose of endotoxin is 5 to 25 mg/kg for most strains of mice, whereas a dose that is 1,000,000-fold less (30 ng/kg) has been reported to cause shock in humans.”

 

By no means, however, are the researchers advocating for the elimination of mouse models, but rather for the application of more stringent model parameters.

 

“Because virtually every drug and drug candidate functions at the molecular level, one practical approach forward is to raise the bar by requiring molecular detail in the animal model studies indicating whether the model mimics or fails to mimic the molecular behavior of key genes, key pathways or the genome-wide level thought to be important for the relevant human disease,” they suggested.

 

“The quality of the animal model could then be determined by how well it reproduces the human disease on a molecular basis rather than simply phenotype.”

 

One of the complaints about this study, however, was that the researchers only examined one strain of mice, and several commentators suggested that the findings might be less clear if a broader range of test animals had been studied.

 

To some extent, this belief was borne out in a similar study by Michigan State University’s Daniel Hollern and Eran Andrechek, published earlier this year in Breast Cancer Research.

 

Using extensive databases of mouse mammary tumor samples used to model human breast cancers, the researchers compared gene expression and pathway activation patterns both within and across mouse models (e.g., Myc, Neu, p53, BRCA). Even within models, they found significant heterogeneity at both the gene expression level and the pathway activation level, with some genes or pathways elevated in sample subgroups within each model.

 

They then compared their mouse analysis with similar analyses in human breast cancer tissues and found a large number of mouse mammary tumor models had similar gene expression profiles to human breast cancers. Interestingly, however, they found that no single group of human breast cancer was modeled by a single mouse model at the pathway level.

 

Thus, the researchers concluded that mouse mammary tumors could be effective models of human breast cancer, but cautioned that “great care should be taken to appropriately choose the mouse model to use and that a genomic and histological characterization of tumors should be completed following experimentation.”

 

So what about cell culture models using human cells?

 

 

Cell culture models bring researchers closer to the organism of interest—in this case, humans—but even here problems can arise, because primary human cell cultures can be difficult to grow and maintain. Immortalized cells, meanwhile, are easier to grow but may be so significantly modified at the molecular level from their normal progenitors that the results of compound screening efforts may be suspect.

 

The advent of stem cell technologies, however, has opened a door to not only studying normal cells—healthy or diseased—but also studying the cells of individuals with the disease of interest.

 

“What you can do today is get or make a stem cell from that person, turn it into a liver and now technically in that dish you have a liver cell that has the same genetic mutations as the liver cell in that child,” Wamhoff says.

 

Furthermore, as highlighted time and again at the International Society for Stem Cell Research (ISSCR) conference back in June, gene editing tools such as CRISPR and TALENS mean that researchers can go into these disease-affected cell lines and “fix” the errant genes to create healthy controls that have essentially the same genetic makeup as the donating patient. This affords scientists the opportunity to then test for the impact on the compound on these cultures and clearly distinguish between disease-specific effects and potential off-target effects.

 

Such efforts may be particularly useful when examining conditions where cell biopsy can be difficult and/or dangerous, such as in neurological conditions.

 

The Salk Institute’s Fred Gage and colleagues presented some of their efforts to model autism spectrum disorder (ASD) by generating neural progenitor cells and mature neurons from affected and age/gender-matched control cell lines. While the results are quite preliminary, the researchers noted altered cell cycle and levels of excitatory and inhibitory markers of neural cells during early stages of cell differentiation, providing a possible window into autism pathology.

 

Also working in ASD, researchers at CHOC Children’s Research Institute described their efforts to build a repository of more than 200 cells lines from ASD patients and unaffected volunteers that can be differentiated into neurons and glia. The goal is to provide a resource to evaluate and compare data from different cells lines to better understand the causes and pathophysiology of ASD, whether to develop new therapeutics or better diagnostics.

 

But even here there may be a problem, for it seems that newly minted stem cells may possess all of the tools of their newfound trade, but that doesn’t make them identical to the cells they’re mimicking.

 

“It turns out that they are very naïve,” Wamhoff explains. “They’re immature at best, almost fetal-like. So they’re lacking all of the properties that the adult cell in the disease state has.”

 

In a review published in Acta Pharmacologica Sinica, Harbin Medical University’s Xiao-hong Xu and GSK China’s Zhong Zhong concur, placing their focus on neurological diseases.

 

“Many neurodegenerative diseases are late-onset diseases, and their key phenotypes may not manifest easily within a short period of time in culture,” they wrote, further suggesting that many of these conditions also involve interactions between cell types and/or responses to environmental stressors.

 

“Therefore, it may be necessary to expose cells to the relevant biological, chemical or environmental stressors to reveal the underlying disease phenotypes when modeling late-onset, non-cell-autonomous and complex multifactorial diseases using iPSCs,” they concluded.

 

At ISSCR, Daniela Cornacchia and colleagues from Sloan-Kettering Institute for Cancer Research and Weill Cornell Medical College described their efforts to do just that, by looking for factors that could induce aging in iPSCs. Earlier efforts by the same group had shown that they could reinduce age markers erased during cellular reprogramming through ectopic expression of progerin, the mutant protein involved in the premature aging disease progeria. In the current studies, they set about to identify age-related transcriptional and epigenetic markers by comparing primary cells from young and old donors, as well as the iPSCs arising from those cells.

 

“Differential factors identified by our studies are employed to yield an improved ‘aging cocktail,’ aimed at testing our primary hypothesis that induced in-vitro aging allows the development of more faithful models of late-onset degenerative disorders including [Parkinson’s disease],” they wrote.

 

A potential challenge to this approach, however, is that you are introducing artificial factors to the cellular mix, albeit factors based on biological reasoning.

 

Wamhoff, in contrast, takes a more reductionist view, advocating the idea of going back to the original physiology.

 

“You need to take those cells and put them back in their physiological context,” he continues. “You need to find their neighbors and bring them back in. You need to restore blood flow, restore contraction.

 

“And when you do that, the really naïve rare disease liver cell you created from a stem cell can now become like an adult rare disease liver cell, and now you can go after a therapy.”

 

 

As noted in the feature "Life moves on" (July 2014 issue of DDNews), there has been significant movement toward the development of 3D cell cultures as a mechanism to gain some of the biologically critical information lost when cells are plated 2D.

 

In a review published in Stem Cells in 2013, Robert Hynds and Adam Giangreco of University College London noted that complex intracellular communication and organization networks normally found in tissues can be difficult to identify or may be absent in 2D cultures.

 

“This is because in vivo, cells exist within a complex network that provides important signalling and biomechanical components,” they wrote, echoing Wamhoff’s thoughts. “Overall, 3D cultures recapitulate in-vivo cell-cell and cell-matrix interactions more successfully than 2D plastic substrate cultures. Thus, 3D culture models allow for the emergence of more physiologically relevant cell phenotypes.”

 

Increasing awareness of this phenomenon has led to rapid growth in the market for cells, tools and reagents for 3D cell culture. Companies like InSphero and Scivax USA produce spheroid cell lines for a variety of tissue types, while other companies such as 3D Biomatrix, AMS Biotechnology, Essen BioScience and Thermo Fisher provide reagents, instruments and assays for labs to develop and test their own cell lines.

 

At the same time, suggested Anna Herland and colleagues from the Karolinska Institutet, despite spheroids providing some improvement in disease models over 2D cultures, there is still room for further improvement.

 

“Due to the self-assembling nature of spheroid cultures, they are difficult to control, and the cell microenvironment differs significantly depending on the distance to the spheroid surfaces,” they wrote in a paper published this year in Biomaterials.

 

They and other researchers took the idea of 3D culturing one step further by generating cultures that more closely resembled the organs they were trying to mimic, moving from a relatively homogeneous clump of cells to create a more clinically relevant microenvironment. With a nod to their morphology, these bodies were called organoids.

 

Earlier this year, Yan Li and colleagues at Emory University and Florida State University described various efforts to use iPSCs to develop organoids, publishing their thoughts in Organogenesis.

 

“Most organoids are formed through the process involving intrinsic tissue mechanics and the programmed internal interactions, known as self-organization,” they explained, breaking that process into steps involving relative cell positioning, control of cell status and morphogenesis.

 

Through these processes, the cells adapt within the context of one another to more closely resemble the morphology and various cell types of the target organ, whether liver, intestine, heart, lung or pancreas, among others.

 

The goal, according to Hynds and Giangreco, is to then use these more complex systems much as we currently use 2D cell culture models of human disease.

 

“Multiwell plate-based organoid assays would then be channeled into compound toxicity and efficacy screening systems such as gene expression microarray, protein mass-spectrometry and multiplex ELISA platforms,” they suggested. “High-throughput and high-content analysis would be achieved using automated cell manipulation and readout systems.”

 

But even organoids have their limitations. Beyond a certain size, diffusion becomes a limiting factor for any test because of a lack of circulation, whether of nutrients or test compounds.

 

To some extent, this issue can be moderated through improvements in bioreactor technology, but as Wamhoff indicated earlier, there is more to life than simply having the right cell combinations and being able to feed them.

 

 

“It was not just a matter of bringing two cells together in a laboratory, because people had done that before and that wasn’t working,” he says. “There is something else missing.”

 

That something else was physiology and the physical forces that act on those cells within the human body.

 

“In a blood vessel, the cell that lines the blood vessel wall senses blood flow as soon as the heart starts to beat in development,” he continues. “The cell responds to blood flow and that blood flow dictates the function of that cell. And that cell talks to its neighbor and dictates the function of it.”

 

Thus, to create a more accurate model of human health or human disease, it is critical to reintroduce the dynamics of physiology back into the cell culture system, and while technically challenging, this has been done in a number of ways.

 

At the Wyss Institute at Harvard, for example, Founding Director Don Ingber and colleagues have taken a microfluidics approach to essentially create organs-on-chips. Organ-appropriate cells line the channels where they can be exposed to each other and to fluids or gases. But just as importantly, the chips—about the size of a memory stick—have been designed to allow physical processes such as flexing to be incorporated.

 

In a video, Ingber introduces the lung-on-a-chip model: “It has human airway cells from the air sac on a membrane that’s porous. On the other side of the membrane are human capillary blood vessel cells. There’s air on one side. There’s flowing medium with human blood cells in it like blood on the capillary side. And the whole thing stretches and relaxes, just like our lung does when we breathe.”

 

The breathing action is the result of changes in air pressure in two channels that line the main physiological channel. As the vacuum increases in these passages, it stretches the tissue, which then relaxes as the vacuum is diminished.

 

“We mimic various types of physiological responses to drugs, toxins or various types of materials that we encounter on a daily basis,” adds Technology Development Fellow Dan Huh.

 

In proof-of-concept experiments, the researchers were able to introduce bacteria to the airway and watch as white cells in the blood stream responded by moving through the membrane and attacking the bacteria. They monitored cell migration using high-content imaging.

 

The group was also able to mimic IL-2-induced edema that can occur in cancer patients receiving the cytokine. At IL-2 levels commonly given to cancer patients, small amounts of fluid translated from the blood stream side to the airway when the system was static. When the system mimicked lung expansion and contraction, however, the fluid completely filled the airway chamber and blood clots were noted in the airway.

 

Since their first publication in 2010, the organization has developed more than 10 organ models, including chips for liver, gut, kidney and bone marrow, and in late July, they announced the launch of the company Organs-on-Chips to commercialize the technology.

 

Aside from the chips themselves, however, the group has also developed an instrument to automate the various chips and fluidically link the organs-on-chips together to better mimic whole-body physiology, human-body-on-chips.

 

HemoShear took a somewhat different approach (see the sidebar article below titled “How’d they do that?”).

 

“We set out to create, first, a healthy human blood vessel in a laboratory,” Wamhoff explains. “And we did that by superimposing on the vascular system human physiological parameters that were deduced from a human high-resolution MRI.”

 

To do this, they co-cultured endothelial cells and smooth muscle cells in a 75-mm Transwell plate and then added a cone and plate drive to simulation hemodynamics, as well as in-flow and out-flow tubing to move culture fluids across both cellular surfaces.

 

As he goes on to explain, this work couldn’t have been done 15 to 20 years ago, as the technology to understand how the mechanical forces were somehow sensed and then recreate them on the bench really didn’t exist until the early 2000s, when molecular physiologist Wamhoff and company co-founder and biomechanical engineer Brett Blackman first met at the University of Virginia.

 

The combination of technologies was a total game-changer to Wamhoff.

 

“When you take a cell and you put it in a dish, you can squirt a drug on it and get the response that you think you’re looking for,” he explains. “It turns out that the concentration of that drug is usually so high that you can never achieve that concentration in a human. So how do you make a decision off of that?”

 

“Once we let the cells talk to each other and gave them the physiological forces back, they now started responding to drugs at in-vivo concentrations.”

 

But as with drug discovery, success on the bench does not always translate into broader commercial success. Thus, it was critical for HemoShear to validate both their vascular and liver models.

 

As Wamhoff explains, potential pharma partners weren’t about to sign on to collaborate with HemoShear if they couldn’t validate their system, because any IND filings arising from the research would slam up against FDA questions.

 

“It took us well over five years and a lot of drugs and burning a lot of cash to validate it,” he says.

 

The company has screened more than 200 drug compounds, most of them FDA-approved therapeutics, to validate that they can reproduce the known in-vivo effect at clinical concentrations.

 

“We can show efficacy, safety or harm, and we’ve had a pretty good track record,” he adds.

 

As a more recent show of success, the company announced in October the successful completion of the first phase of a project with the National Cancer Insitute to recreate the cancer tumor microenvironment.

 

“We had tumor vasculature, the tumor cells and the stromal support cells, and the hypothesis would be that if you bring all of that together in the right physiological context, you’d see drug responses at clinical therapeutic concentrations,” explains Wamhoff.

 

As proof-of-concept, the company created a non-small cell lung cancer tumor platform and then probed their construct with cisplatin, a drug commonly used to treat various cancers. Unlike what had been seen in mouse or other in-vitro models, HemoShear was able to demonstrate that they could effectively reduce tumor growth at clinically relevant IC₅₀ levels.

 

The next step in the agreement is to generate models for other tumor types and then validate those models against other FDA-approved drugs and combination therapies.

 

Aside from simply filling in for traditional preclinical mouse models and tissue culture, Wamhoff also sees opportunities for the HemoShear system in areas such as drug repositioning (scanning other disease models), identification of off-target effects and potentially testing drugs in broader patient populations.

 

The last opportunity may become significant, as many drugs are approved based on data from focused patient populations, e.g., predominantly Caucasian adults aged 25 to 55. But what about the impact of the same drugs in the aged, children or people of different ethnic backgrounds or environmental exposures?

 

The company has already performed experiments where they examined differences in the gastrointestinal bleeding profiles of Caucasian women 70 years or older vs. Caucasian women less than 70 years old to understand the underlying cause of the bleeding.

 

But for all of their technical achievements, these new disease models run second to animal and traditional cell culture models in one significant way: throughput.

 

Wamhoff is the first to admit that the HemoShear platform is low- to medium-throughput, although he balances that against the type of information arising from his system and the company’s arsenal of upward of 120 simultaneous experiments being more than enough for a company HemoShear’s size.

 

That being said, he acknowledges that the pharma industry has built billion-dollar infrastructures around high-throughput screening, infrastructures that these companies are not going to mothball simply because new technologies have arrived.

 

Thus, he suggests, HemoShear and the nanofluidic device companies are going to have to evolve their systems to figure out how to plug into the high-throughput world.

 

Wamhoff calls that the biggest challenge in the near future.

 

 

To test the impact of hemodynamic flow on hepatocyte morphology, function and metabolic activity, researchers at HemoShear cultured rat hepatocytes for two weeks in either a nonflow collagen sandwich or in a perfused Transwell device that simulated controlled hemodynamics.

 

They then monitored the hepatocytes using a combination of methods, including confocal microscopy and transmission electron microscopy, a cytochrome P450 activity assay and an MTT assay for toxicity in the presence of dexamethasone.

 

They also examined the impact of hemodynamic flow by looking at the expression of various metabolic genes at both the RNA expression level using RT-PCR and protein level using western blotting.

 

In all cases, the morphology, function and metabolism of the hepatocytes more closely resembled that found in vivo in the cells cultured under hemodynamic flow conditions.

 

“Taken together, these results highlight the importance of interfacing in-vitro biology with in-vivo physiological parameters,” the authors concluded. “Specifically, the retention of in-vivo-like hepatocyte phenotype and metabolic function coupled with drug responses at more physiological concentrations requires the restoration of in-vivo physiological transport parameters in vitro.”

(Dash A., et al. Am J Physiol Cell Physiol. 2013;304:C1053-C1063.)