EVENTS | VIEW CALENDAR
Special Report on Organ Models: All in
Special Report on Organ Models
Is model complexity helping or hindering adoption?
By Randall C Willis
My head is pounding, and my stomach is upset. Rising from my bed, I wait a moment for stability.
Gingerly making my way to the bathroom, I squint in the harsh light and slowly take in my reflection.
I have no idea what’s wrong, but the guy in the mirror doesn’t look too good. If only I could send him to the doctor for tests.
In a recent DDN podcast, Walter Kolch, co-founder and director of Systems Biology Ireland, talked about the application of computational modeling for accelerated target validation in drug discovery. As he concluded his thoughts, he took a moment to look into the future.
“If you think of medicine, the ideal world would be that you have a digital twin where you can simulate diseases, preventions, interventions on the computer first,” he speculated.
“In a very safe and efficient way, [we can] simulate what’s wrong and, before we actually go to the patient, interrogate the computer model, which gives us the best advice for how we should treat the patient,” he enthused.
Despite the best efforts of Kolch and others, the world is not there quite yet.
That said, early steps toward an in-vitro patient avatar were published 10 years ago in the pages of Science with the description of mechanically active “organ-on-a-chip” (OOC) microdevices.
In the decade since that paper, advances in culturing methods, microfluidics and cell analysis have led to an explosion of tissue and organ modeling systems, not all of which can be classified as OOCs.
“There has been a muddling of articles on organoids, OOCs, microphysiological systems, multiwells that have some flow and tiny bioreactors,” says Don Ingber, founding director of the Wyss Institute and scientific founder of Emulate. “There is a belief that it’s all the same thing, and it’s really not.”
As Ingber and colleagues explained in that 2010 paper, the OOC reconstituted “the tissue-tissue interfaces critical to organ function.” For University of Central Florida’s James Hickman, the key word there is function, looking beyond the multicellular architecture of 3D cell culture.
“What is the good of having the cardiac anatomy if you can’t measure conduction velocity, force and these kinds of very key elements?” he asks.
Alongside long-time collaborator Cornell University’s Mike Shuler, Hickman has spent much of his career working on various aspects of cell culture technology, both in terms of the cells and of the supporting mechanical and analytical resources. The pair co-founded Hesperos, the self-described Human-on-a-Chip company.
Thus, for Hickman and many others, OOCs are not simply 3D tissues grown in a multiwell plate.
“It’s important that people do know that there’s a difference between just growing a bunch of cells in the middle of a plate and trying to slosh the nutrients around them, and actually actively cycling the nutrients through the cell matrix, growing them on a very complex lattice and being able to feed them constantly,” says Jean-Pierre Joubert, product manager for CN Bio Innovations, which has developed the PhysioMimix platform.
Beyond nutrient and waste cycling, however, Joubert suggests that fluid flow is also critical to maintaining the sheer stresses that human tissues would normally experience.
“Cells actually need those cell stresses to grow and develop correctly,” he points out, “especially when you’re looking at things like barrier models.”
Ingber offers the example of a condition known as ileus, a bacterial overgrowth of the intestine that can lead to sepsis and death.
When a patient has surgery, he explains, they receive anaesthesia, which also serves to stop intestinal movement. Then, when the patient has been moved to recovery, medical practice dictates the patient quickly receive fluids and food to avoid ileus.
“The textbooks all say [ileus occurs] because fluid flow stops,” Ingber continues. “But in the chips, we kept fluid flow going but we stopped peristaltic motions, and it was the [lack of] mechanical motions that caused overgrowth.”
Beyond the mechanical stresses, the microfluidics also helps to reduce medium volumes, making the culture conditions more physiological and reducing resource use.
“One of the major drawbacks of standard cell culture is that the liquid-to-cell ratio is very high,” remarks Olivier Frey, head of technologies and platforms at InSphero. “You have around 1,000 or 10,000 times more liquid compared to the human body. With the miniaturization and the microfluidics, you keep these interactions between the liquid and the cells at the physiological scale.”
There is also the question of the dynamic aspect of physiology and the response of a system to insult, whether from a toxin, drug or physical stress. The ability to sample a time-course and keep tissues alive longer is critical to doing the longer, more detailed experiments required to monitor biomarkers and metabolites.
Joubert agrees, offering the example of work CN Bio has done with non-alcoholic steatohepatitis (NASH) models where they wanted to monitor the impact of metabolites from one cell type on another.
“Then being able to apply a drug and see how long it takes for that drug to affect that interaction,” he continues, “whether you’re sampling the media to look for your secondary metabolites or biomarkers, or sampling physical cells and asking how the cell structure is starting to change or if the compound is keeping it alive longer.”
Increased model complexity isn’t always the right call, however.
“People say 2D is bad, 3D is better,” Hickman offers as an example. “They really don’t know what they’re talking about.”
“This is something where all OOC builders find their niche or decide at the beginning where is our system going to be used mostly,” says Frey. “What we are doing with the Akura Flow system is bridging the complexity with the high throughput.”
For InSphero, it is about using the same microtissues in a 384-well plate as in a much more complex flow system.
“With our approach, we can broaden the space where the system can be used in the drug discovery process with a continuity of the readouts and the tissue model over a longer process,” he comments.
This approach helps ensure that the results of early screens can translate to later screens.
Joubert concurs, suggesting you don’t want to reinvent the wheel, but rather find the synergies between data from broader, larger-scale studies and restricted, more refined experiments.
And as Joubert reminds us, more complex models come with their own challenges.
“Is it going to be so complex that it’s no longer useable or that the data is no longer consistent?” he asks. “That’s one of the really important things for us, the usability and then the robustness of the data.”
That usability question is also why companies like Hesperos went fee-for-service rather than rely on platform sales.
“We’re simply a service-based company because we figured it was going to be too complex to put into people’s labs and get them to figure out our very complex, multi-organ systems,” he says. “And if you look at some of the recent market analysis of things that people are doing, most of the places that were only going to be selling their systems are coming around to doing a fee-for-service model.”
Echoing his counterparts, Ingber suggests not all questions require complex solutions.
“You start with the simplest model you can, one cell type or a tissue-tissue interface, and if you mimic in-vivo functions, then you don’t need to add anything else,” he argues. “We have found in multiple studies that assumptions that you needed more complexity were wrong.”
He offers the example of a pulmonary edema model his group developed.
“One reviewer said our paper shouldn’t be published because we don’t have immune cells,” Ingber recounts. “Another reviewer said this is amazing; they just showed that you don’t need immune cells for pulmonary edema induction by the drug we were studying.”
It really comes down to the questions you’re asking and the scale of throughput you need.
“Organoids and transwells and some of these multi-chamber devices are fantastic as a front-end screen,” Ingber adds. “We even use them often to work out drug dosing, for example. But if you really want to recapitulate human responses, then the chips are where we feel it is most valuable.”
Ingber describes their models as living 3D cross-sections of the functional units of key organs. Because they are engineered to explore a specific function, they know the surface areas of the cells, they know exactly where that interface is, and they can examine it again and again over time—and then they can add variables to this fundamental unit.
“We could add immune cells at a specific point,” he says. “We could add virus. We could add microbiome. We could study interactions. We could collect out-flows in real-time. We could do multi-omics.”
It is also important, says Ingber, to remember that a lung chip is not a lung chip is not a lung chip.
“If your question is influenza infection, I may be looking at an airway chip,” he explains. “If it’s pulmonary edema, I’m looking at an alveolus chip. If it’s pulmonary fibrosis, I have to integrate fibroblasts into that chip. One thing that pharma needs to understand is that this is not a whole organ. These are models that need to be designed fit-for-purpose.”
Recapitulate the known
Although OOCs show a lot of promise to facilitate research into pathophysiology and drug development, the technology is still relatively new, so much of the literature revolves around efforts to recapitulate clinical results of animal and human studies.
Last August, for example, Laura Perin of Children’s Hospital Los Angeles and colleagues developed a glomerulus-on-a-chip (GOAC), seeding human podocytes and human glomerular endothelial cells onto MIMETAS Organoplates. They wanted to determine if their GOAC functioned the same as glomeruli and could reflect injury associated with chronic kidney diseases like membranous nephropathy (MN).
“To test the hypothesis that our GOAC can model a kidney injury state, we exposed the GOAC to puromycin aminonucleoside (PAN), a nephrotoxic agent that alters podocyte morphology and function,” the authors wrote. “When added to GOAC, PAN induced podocyte injury as documented by cytoskeleton rearrangement and loss of permselectivity for albumin at 60 min after stimuli.”
They then tested the response of GOAC to serum from patients with MN, a nephrotic syndrome characterized by anti-podocyte autoantibodies that induce proteinuria, and they found they could recapitulate the disorder in vitro. Likewise, they could reverse the in-vitro proteinuria with α-MSH, a hormone used clinically to treat MN.
The authors recognized that their model was still relatively simple and that to truly reflect kidney functions, other cells and sub-organ structures would need to be added.
“MIMETAS has already demonstrated the feasibility of producing functional proximal tubules using their technology, so the next step will be to combine the GOAC and the tubules to generate a functional nephron on a chip, where filtration and reabsorption can be evaluated at the same time,” they added.
Looking beyond solid tissues, Wyss Institute’s David Chou and colleagues, including Don Inger, recently described their efforts to model hematopoiesis with a bone marrow (BM) chip. The chip was constructed of two channels: a hematopoietic channel filled with human CD34+ cells and BMSCs in a fibrin gel, and a vascular channel lined with HUVECs.
Upon confirming the model supported myeloerythroid development, they examined its ability to mimic BM toxicity associated with approved chemotherapy 5-fluorouracil (5-FU) and an AstraZeneca clinical candidate AZD2811. In both cases, they were able to recapitulate the dose-dependent and infusion regimen-dependent effects seen in the clinic.
The researchers then modelled the BM dysfunction seen in Schwachman-Diamond syndrome (SDS) using CD34+ cells from SDS patients.
“After 2 weeks of culture, these SDS BM chips displayed both broad defects in haematopoiesis and cell-type-specific abnormalities,” the authors wrote.
“Interestingly, the impaired neutrophil maturation was further discernible as an aberrantly ‘blunted’ CD13/CD16 expression pattern,” they noted. “This has not been previously reported in the BM of patients with SDS, although mild morphological dysplasia in the neutrophil lineage has been observed.”
Thus, not only was their model able to recreate known effects, but it also identified previously unexplored pathological mechanisms.
“Compared with static suspension or 3D gel cultures, the BM chip was better able to recapitulate the toxicity responses of human BM to clinically relevant dose exposures of AZD2811 and 5-FU, and also showed an improved ability to recover after injury,” the authors concluded. “Thus, the BM chip could be used to assess both drug toxicity and recovery without requiring the usual animal studies, which has significant ethical implications, in addition to expediting the drug development process.”
Expanding on the unknown, Anjaparavanda Naren of Cincinnati Children’s Hospital Medical Center and colleagues described their efforts to explore cystic fibrosis-related diabetes (CFRD) using a pancreas-on-a-chip model. As Naren suggested in announcing the study, mouse models of CF did not faithfully recapitulate this particular manifestation of the lung disease.
The researchers separated pancreatic ductal epithelial cells (PDECs) and islets with a porous membrane, which successfully mimicked the cell-cell communications and fluid exchange normally seen in a pancreas. Then then disrupted the CFTR gene and saw immediate impairment of these functions, resulting in insulin deficiency and disease pathology.
“Pancreas-on-a-chip helps answer the fundamental question in CFRD: is loss of CFTR function in PDECs primary to CFRD development,” the authors wrote. “Based on our data, it is indeed the case.”
“Surprisingly, the absolute amount of insulin was around 50 percent decreased during inhibition of CFTR channel function,” they added. “This finding tells us that CFTR channel function plays a critical role in maintaining endocrine function and may provide critical insight into the etiology of CFRD.”
But just as organs are more than an aggregate of different cell types, so too are humans more than a liver, a kidney or a heart.
Because the molecular functions of one organ significantly influence the functions of another, researchers are increasingly exploring the pathology of diseases or the impact of therapies in model systems that connect multiple organ chips.
Multi-organ systems offer researchers the opportunity to increase experimental complexity step-by-step, according to Frey, whereas in an animal model, this isn’t really possible.
“The major driving force there is to bring this more systemic view on how a compound interacts on the human body from the in-vivo space into a controllable in-vitro space,” he continues. “That allows you then to look at paracrine signaling between different organs.”
You can monitor if compounds become toxic when they are metabolized by the liver, he presses. Or it can be used in the pharmacokinetic/pharmacodynamic (PK/PD) space or the absorption, distribution, metabolism and excretion (ADME) space, where you want to see how the compounds are distributed between the different organs.
It can also help to build models of diseases that arise from multiple factors, such as NASH. Here, you might combine liver and pancreatic islet chips and then increase the circulation of free fatty acids through the system.
Earlier this year, Camilly Pires de Mello and colleagues at University of Central Florida and L’Oreal Research described their efforts to explore topical drug administration using a heart-liver OOC with a skin mimic versus systemic administration. Applying a variety of drugs known to disrupt cardiac and hepatic function, the researchers found they could mimic in-vivo results.
Importantly, the system was also able to demonstrate differential effects from acute and chronic drug exposure.
A similar use of heart-only and heart-liver chips from Christopher McAleer and colleagues at Hesperos and AstraZeneca examined the pharmacodynamics of a clinical candidate. In this case, however, beyond metabolism via the liver, the researchers noted that the cardiomyocytes themselves were converting the parent compound to its metabolite.
This helped explain why the researchers saw the PD effect despite minimal concentrations of the metabolite in either model.
And in June, says Hickman, Hesperos and Hoffmann-La Roche were able to model the innate immune system in a three-organ chip. As the study authors Sasserath and colleagues wrote: “Immune components have been implemented in single organ systems such as resident macrophages in liver, lung, and others, but have not been demonstrated in recirculating multi-organ systems.”
In this case, the researchers used THP-1 cells, a monocyte/macrophage cell line, as the immune component, recirculating freely through liver, heart and skeletal muscle chips. When cardiotoxic amiodarone was introduced to the system, the researchers noted THP-1 cell infiltration to the cardiac tissue chip, suggesting monocyte conversion to M2 macrophages. Yet when LPA and IFN-γ were added to the system, nonselective damage of all three organs led to conversion to M1 macrophages.
“And we didn’t publish this work,” adds Hickman, “but we’ve also shown that PBMCs and T cells are compatible with the system, so we could do adaptive [immunity], but we just haven’t gotten there yet.”
“Most people can’t get the immune cells recirculating in their systems because they have pumps and the pumping damages the cells and activates them,” he continues. “We have shown we can actually get them to go for 14 days in our system without activating them. And it’s only when we specifically put in cues—either damage the tissue or put in bacteria—that they activate and then start expressing all the cytokines.”
But even two or three organ systems are not the limit, as companies and funding agencies push the technical boundaries.
One such agency is DARPA, a major source of inspiration and funding for Ingber’s research.
“They give you impossible challenges, knowing that there is no way you’re going to do it,” he explains, “but along the way, there’s going to be amazing technology fallout.”
In Ingber’s case, the DARPA challenge was to develop 10 different organ chips that were fluidically connected and to automate everything from culturing to analysis to imaging. And then, use computational model on the data from that system to quantitatively predict drug PK/PD.
The result was the Interrogator platform.
As Wyss Institute’s Richard Novak and colleagues described in April, Interrogator “employs liquid-handling robotics, custom software and an integrated mobile microscope for the automated culture, perfusion, medium addition, fluidic linking, sample collection and in-situ microscopy imaging of up to ten organ chips inside a standard tissue-culture incubator.”
Testing the platform on eight different OOCs, the researchers were able to maintain perfusion, viability, morphology and organ-specific function for three weeks in two separate experiments. They also verified medium perfusion by infusing inulin-FITC dye into the gut chip each week and daily monitoring tracer dye concentration throughout the organs.
Rather than use peristaltic pumps, however, the Interrogator relies on automated liquid transfer.
“We went to a robotic transfer of fluid with a liquid handler because we realized that there is a lot of dead-space in tubing that would have made it very difficult to do the analysis,” Ingber offers. “There’s absorption into tubing along the way, as well.”
More importantly, however, the robotics allowed them to remove one droplet for mass spectrometric analysis for drug levels and then transfer another droplet to the next chip.
“That turned out to be quite a useful innovation,” he says.
According to Sasserath and colleagues: “The modular design of both the Interrogator hardware and control software enables the rapid addition or removal of organ chips without disrupting the experiment, including—importantly—the perfusion of media.”
And, they continued, drug dosing could be modified to any mode of delivery, whether oral via the gut chip, intravenous via the vascular medium, topical via a skin module, or aerosol via lung.
Despite this modularity and flexibility, however, Ingber really doesn’t see the likelihood of anyone needing 10 different chips for an experiment.
Instead, he suspects that most work will be done with single OOCs or perhaps two, three or four connected OOCs.
“I don’t think it’ll ever make sense to do 10,” he says. “I really don’t.”
And yet, the ability to use such a multitude of organ systems does bring us a step closer to the human avatar, or at least to the animal models many hope to eliminate.
Not an animal
Given the low-throughput but high-content nature of OOCs, Ingber sees strong parallels with animal models.
Because these are human tissues, however, they can model responses that might be unethical to test in humans or that are undetectable in animals.
But this hasn’t slowed demand for animal data.
“No matter what we do, we get requests for mouse studies when we’re doing things that were never found in animal models,” Ingber complains. “But we know that when it comes to drug development, animal models are wrong 70 percent of the time. This happens again and again.”
Hickman echoes Ingber’s frustration, offering the example of gene therapy researchers, whom he suggests are married to their animal models.
“We write proposals for muscular dystrophy all the time because we believe that this is the perfect model,” says Hickman. “We can take the cells from the actual patients.”
“But they’ve got a dog model,” he continues. “And every time we try to put an application in at the NIH, there is at least one reviewer who kills it because ‘this isn’t necessary; it’s not significant because we’ve got a dog model that works.’”
But the dog models aren’t working, Hickman stresses, suggesting that the researchers can’t get anything approved because their results in dogs aren’t translating to humans in clinical trials.
Recognizing that the community is not quite ready to let go of their animal models, Joubert sees an opportunity for OOCs to at least inform next steps as researchers move from in-vitro screening to animals.
“What the OOC systems can do is to optimize things such as your drug dosing,” he offers as an example. “So, when you then have to move into your animal models, at least you’re able to optimize your systems before you move on.”
There are signs, however, that the research community is at least aware of its blinders, perhaps seeing them more as handcuffs. Ingber recounts requests from pharma companies to not only deliver a human model of liver toxicity, but also equivalent models in rat and dog.
“I was like no, we want to do human,” he laughs. “And they’re saying no, no, we have to use rat and dogs for the FDA.”
“So, we made rat, dog and human liver chips with primary hepatocytes from each species, primary liver sinusoidal endothelial cells, stellate cells, Kupffer cells,” he recounts, “and with drugs from Janssen and AstraZeneca, we could mimic species-specific liver toxicities.”
For his part, Hickman reflects on situations where animal models often do not exist or may not be possible: rare human diseases. Of 7,700 rare diseases, he says, there are only 400 research programs because of this challenge. This is where OOCs can be combined with stem cell and genome editing to develop rare disease models for therapeutic screening.
Using stem cells derived from patient tissues, he explains, Hesperos is developing phenotypic models of different diseases and comparing them to non-disease phenotypes.
“Then, we treat it with drugs and see if we can bring it back to the wildtype,” he says.
Frey describes this as the “You-on-a-Chip” approach.
Beyond journal reviewers, however, how are regulatory agencies viewing these advances?
Perhaps one of the greatest barriers to uptake of any evolving technology has nothing to do with the technology itself, but rather with the acceptance of data by regulatory agencies.
“People are always willing to try something new but until you actually have that stamp from a regulatory authority, you won’t be able to move too far forward on it,” says Joubert. “It’s important to engage with those regulatory authorities—here’s our system, would you like to try it out?”
As Hickman adds, this need for regulatory acceptance was why he and his colleagues started the engagement process almost a decade ago.
“I was on the board of directors of an organization called AIME—American Institute for Medical and Biological Engineering—and we organized six workshops with NIH, FDA, EMA on looking at validation and qualification of these systems for use in the regulatory process,” he recounts. “FDA was not very interested initially, but by the sixth workshop, they were actively promoting how to do this.”
Hickman suggests that the FDA is quite willing to look at efficacy data arising from OOCs, but that toxicity data is going to be a bigger challenge.
Still, he says, data from Hesperos platforms are starting to be used in IND applications.
“I have certainly been at meetings with representatives of the FDA and the equivalent in Europe where they are telling pharma and biotech, 'we would love to see that done,'” Ingber adds.
In February, Ingber and colleagues from across the industry and academe published the output of a three-day workshop—the t4 Workshop Report—focused on microphysiological system (MPS) applications in drug discovery that included discussion about acceptance by regulatory agencies.
“Despite qualification and validation, a frequent argument justifying the lack of industrial use of MPS-based methods and tests in safety assessment is that regulatory agencies need to formally communicate that MPS-based methods and tests are accepted or indicate what data are needed to obtain regulatory acceptance,” the authors wrote. “However, this is only partly true, since regulators, in turn, have noted that applicants have so far submitted very little or no MPS-based assay data to the FDA or EMA. Therefore, experience with and confidence in these data cannot be gained by regulators.”
They noted similarities with gene expression data 15 years ago, when regulatory agencies encouraged companies to submit exploratory data sets under “safe harbor.” They acknowledged that the field is still quite nascent and that it will take some time for early-adopters and agencies to truly understand the limits and opportunities of the technology.
That said, the authors were pleased to report the active participation of the FDA and EMA not only in this particular working group, but also in other initiatives and collaborations. Still, they offered thoughts on how regulators could facilitate the process.
“Firstly, regulators could begin working to establish performance criteria for MPS that outline what regulators need to see in order to accept MPS in lieu of traditional testing,” the authors suggested. “Secondly and more broadly, regulators can work to establish a clear pathway for the evaluation of new methods that includes communication, such as guidance how to translate a scientific method into a valid assay which will be useful within a specific context of use.”
Although it is currently possible to qualify a platform on a case-by-case basis, they argued, these applications may be considered as drug-specific and learnings kept confidential. To address this concern, they called for two-way communications and guidance where end-users feed information to the regulatory agencies and vice-versa.
For Joubert, that two-way flow of information also needs to occur between OOC developers and the end-user community.
“I think one of the key challenges is just getting the message out there,” he remarks. “There are not enough people letting the wider community know that the technology exists.”
“The publication rates are increasing at a dramatic rate,” he presses. “They need to know it’s a technology and a field that’s taking off really quickly.”
As Frey puts it: “You need some push of technology to understand what can we replicate, and on the other side, you need the application or the draw of the market or from the questions that arise that define what are the parameters that need to be reproduced in such an organ system.”
Joubert is also open to conversations with developers in related fields to evaluate opportunities to integrate different systems.
For a technology with such transformative potential, maybe this is a first step to those conversations.