Perspective: The promise of multi-cellular engineered living systems
Recent technological breakthroughs in our ability to derive and differentiate induced pluripotent stem cells, organoid biology, organ-on-chip assays, and 3-D bioprinting have all contributed to a heightened interest in the design, assembly, and manufacture of living systems with a broad range of potential uses. This white paper summarizes the state of the emerging field of “multi-cellular engineered living systems,” which are composed of interacting cell populations. Recent accomplishments are described, focusing on current and potential applications, as well as barriers to future advances, and the outlook for longer term benefits and potential ethical issues that need to be considered.
Over the past several decades, bioengineers, biophysicists, and biologists have made steady progress toward the creation of “multi-cellular engineered living systems” (M-CELS). These systems are composed of living cells and tissues organized in a way that produces novel functionalities by design. For present purposes, we consider the subset of M-CELS composed of mammalian cells and used primarily for biomedical applications and exclude, for example, other potentially important applications such as those in plant systems, energy harvesting, or the microbiome. Defined in this way, it includes organ-on-chip or tissue chip systems being developed for drug screening or disease models1,2 with the potential to expedite drug discovery and provide important new insights into fundamental disease processes. It also encompasses implantable “hyper-organs,” ones that, for example, sense a biological signal and synthesize and secrete a biologic product in response. Also included are biological actuators or bio-robots that have applications in various fields. These M-CELS might be assembled in vitro from clusters of individually differentiated cells or co-differentiated within a single aggregate of pluripotent cells. An important distinguishing feature is that these systems are designed to possess a specific form and function by design to perform in ways that are not found in natural systems today and ultimately that they can be produced in quantity and in a sufficiently robust manner, thereby making them reliable and amenable to large-scale manufacture.
While we have a tremendous knowledge base to draw upon for the design and manufacture of M-CELS, derived from the study and design of non-biological engineered systems, much is not directly applicable to M-CELS. This is a consequence of at least two important features that distinguish M-CELS from abiotic systems: first, our lack of a fundamental understanding of their inherent complexity, and second, the central role played by emergence in M-CELS formation. In this context, we define emergence as a self-directed, multicellular response occurring as a result of collective interactions of individual cells between themselves and the extracellular environment at microscale which manifests itself by phenomena at macroscopic, system-level scale. Living systems, even at the level of a single cell, are remarkably complex. Cells employ a vast array of signaling pathways to govern their phenotype and behavior, and when used as the building blocks of multi-cellular systems, the complexity quickly becomes overwhelming. Notably, models that are capable of predicting the phenotype of even a single simple cell from its genotype are only now becoming available.3 When multiple cells and cell types interact, new phenomena and properties emerge which can only be attributable to their collective behavior and extend far beyond the capabilities of single cells. While these collective, emergent behaviors are in principle predictable, they are enormously complex and arise from biological reactions that are only partly understood. While there is little doubt that the transition from single-celled organisms to more complex multicellular ones was absolutely essential for the richness of form and function we see in living systems today, our ability to understand and predict cell population behaviors remains nascent.
In order to make meaningful progress in developing the methods and tools needed to create M-CELS, we must draw upon expertise from various disciplines. Certainly, various biological sub-disciplines—synthetic biology, developmental biology, systems biology, and stem cell biology—are essential as are engineering approaches reflected in biomaterials and tissue engineering. However, we must also look into basic engineering design and manufacturing and a variety of enabling technologies, in order to make meaningful progress. Relevant to this, the need for “convergence” was recognized and articulated in the NRC Report, “Convergence: Facilitating Transdisciplinary Integration of Life Sciences, Physical Sciences, Engineering and Beyond.”4 Convergence remains key to the development of M-CELS, across numerous sub-fields (Table I).
While steady progress in each of these disciplines fuels our ability to create M-CELS, several major recent advances are particularly enabling and noteworthy. First, the ability to reprogram adult, committed cells to generate induced pluripotent stem (iPS) cells5 has freed us from reliance on embryonic stem cells to form new organs or other systems. We can now differentiate induced human pluripotent stem cells (iPSCs) into a variety of cell types and begin to construct complex systems with a consistent genotype. Second, recent success in the growth of organ-like structures—organoids—from embryonic stem (ES) or iPS cells has shown that it is possible to co-differentiate cells into multiple cell types that begin to show the form, and in some cases, the function, of a real organ.6 Third, with the help from several new government programs worldwide, these technologies have been brought to bear on the growing field of tissue- or organ-on-a-chip development.
As we develop the enabling technologies for M-CELS, at least two approaches for fabricating M-CELS can be envisioned (Fig. 1), here viewed in the context of a simple muscle-actuated sphincter or pump that requires a vessel lumen for flow, muscle to contract or collapse the vessel locally, and neural control, possibly by the use of optogenetically modified cells and activation by light. Both approaches begin with a concept and detailed mapping of the various cell types needed to produce the end-product. In one approach, differentiated pluripotent or primary cells are seeded or plated in a specified spatial pattern in two or three-dimensional constructs to produce the elements of the system. These may include, for example, an endothelial monolayer, a pancreatic islet, a muscle strip, or a cluster of neurons. These elements are then arranged within a device or substrate in such a way that they interact with each other in a desired manner. We liken this approach to current “top-down” engineering design and assembly of complex systems from simpler components, all according to a master plan for the complete, assembled M-CELS. In contrast, one could fabricate a M-CELS beginning with a disordered collection of pluripotent cells that are subjected to a variety of guidance cues—chemical, mechanical, electrical, and genetic—either globally or locally, which induce the cells causing them to co-differentiate into multiple cell types and self-organize into a new multi-cellular system. We define this to be an “emergent engineering” approach. Figure 1 shows these two pathways and the potential for interactions or cross-over between them.
At this early stage in the development of M-CELS technologies, we recognize the need to look ahead and consider the eventual scientific, commercial, and ethical impacts of our approaches and actions. The purpose of this paper is to provide a review of recent accomplishments, assess current state-of-the-art in M-CELS, and point to future needs and challenges. We also seek to foster a wider conversation of these issues, initiated in a Workshop on Engineered Living Systems, held in Chicago, IL, in August 2016 and continued in August 2018.
The intricacy of the tissue/organ self-assembly represents the essence of what engineers strive to emulate as we design M-CELS. In organoid systems, we seek to replicate, control, and even deviate from normal developmental processes, so it is key that we understand these at a fundamental level. In addition to the initial emergence of form via the progression of genomic and biophysical processes, development and regeneration offer numerous examples of self-organization and dynamic pattern homeostasis even after initial anatomical organization is achieved.7
The correct stereotypical size and morphology of a developing organ result from the interplay between signals that regulate individual cell behavior and mechanical and biochemical feedback that emerges from interactions among many cells.8–12 A well-studied example of how the organ size is achieved via this interplay is the Drosophila imaginal wing disc, which over a period of 5 days grows from ∼50 to ∼50 000 cells and then stops. This growth is guided by both the spatially distributed expression of growth factors (e.g., Drosophila bone morphogenic protein (BMP) and Wnt homologues)13 and their resulting gradients and by mechanical forces which influence tissue growth.14,15 A model that has emerged from this system is that the organ size reflects an intricate balance between biochemical signals and tissue mechanics.16,17
Many kinds of embryos, especially mammals, can be split in half or joined together, still resulting in a perfectly normal animal,18,19 revealing regulative pattern control. After embryonic development, tissues can undergo various degrees of regeneration, supported by stem cells and their niches. One example is the mammalian intestine, where in humans, the small intestinal epithelial lining is renewed every 5 days. Other animals, such as salamanders, can regenerate whole limbs, spinal cords, jaws, eyes, hearts, and portions of their brain, after damage.20,21 An important aspect of regeneration is that, similar to embryonic growth, regeneration stops when a structure of the correct size and shape has formed.22 A better understanding of how the correct dimension feeds back to alter cell differentiation, migration, and proliferation in precisely the right way to bring the pattern closer to the appropriate target morphology would allow for an altered homeostatic size and tissue function.23 Taken together, exploiting innate pattern-restoration mechanisms would allow M-CELS to adapt to external perturbations and recover rapidly from damage. This could be applied, for example, to M-CELS that can respond to atrophy in a native tissue, growing and maintaining its complementary size.
In addition to set-points for the organ size, there is organismal-level homeostatic control. Planarian flatworms shrink and grow in allometric proportion, remodeling their entire body continuously to the available cell number due to eating or starvation.24 Tails transplanted onto the flanks of salamanders slowly remodel into limbs—a structure more appropriate to their new location.25 Additionally, tadpoles with experimentally rearranged Picasso-style faces still metamorphose into normal frogs, as individual organs migrate through non-endogenous paths as needed to build the correct frog face.26 The ability of an engineered organism to maintain its correct shape through similar mechanisms is another strategy to ensure robustness upon damage either during or after initial assembly, representing an important design challenge for a synthetic morphology.
A key issue for the future of M-CELS and regenerative medicine is to find the appropriate paradigm with which to enable direct modification of the shape for engineering and biomedical applications. Currently, the assembly of correctly shaped and patterned tissues is limited by our ability to manage their construction at a micro-level (e.g., cell type, cell position, and scaffold used)—a major barrier to progress in the repair of complex organs such as hands and eyes. These constructs are difficult to assemble directly but are routinely regenerated in vivo by some model species. Alternatively, the self-assembly of organoids leverages endogenous patterning cascades but comes at the cost of losing control over the final shape and pattern of the M-CELS. Ultimately, a solution lies between these extremes, as a kind of guided self-assembly that provides judicious tweaks to an otherwise self-directed endogenous cascade of morphogenesis,27,28 as further discussed in Sec. III. Advances in the field will depend on finding the minimal amount of input required for a system to progress towards a desired complex outcome. Promising approaches include activating master regulators of developmental programs (e.g., “build an eye here”) and learning to modulate biochemical and mechanical signals that trigger pattern formation above the cell level (e.g., organ size and spatial relationship between organs). Ever finer-scale reductionist analyses of the mechanisms of signaling at the cellular and sub-cellular levels, which are progressing at a very rapid pace, are complemented by theoretical and experimental approaches that quantitatively investigate the top-down, computational aspects of pattern regulation employed in tissue shape generation and maintenance.29–31
It is essential to understand and exploit the inherent processes that direct patterning of complex structures.32 Cells and tissues need to make many decisions about what to build, where, and when, in the control of growth and form. In our example of a muscle-actuated pump of Fig. 1, multiple cell types arranged in a specific spatial pattern are needed. The required structure bears similarities to the embryonic heart that develops in a remarkably robust manner in vertebrates,33 so we know that such signals can, in principle, be created and potentially deciphered. The logic of such processes is beginning to be understood, as inferred from studies of molecular regulatory networks,34 the encoding of positional information,35 and bioelectrical encoding of target states.36 Together, the biochemical and biophysical layers are being investigated with the tools of information theory, control theory, and computational cognitive science to develop interventions that target the biological systems' perception, memory, and decision-making about large-scale (morphological) properties.28,37 Moving beyond regulating just a few molecular pathways, we will be able to provide multiplexed inputs to tissues which alter the perception space of cells, leading to activation of desired morphogenetic and repair responses. It is becoming increasingly possible to leverage the ability of tissues to robustly implement specific modular patterning tasks via self-modeling and goal-directed remodeling activity.38–41
One of the greatest challenges in creating M-CELS is to regulate the spatio-temporal differentiation of cells within a developing construct such as an organoid. This control can be exerted in a variety of ways (Fig. 2). It has long been known that transcription factors, growth factors, morphogens, etc., play a key role in cell-cell or cell-matrix communication, and these exert their influence through activation of a multitude of intracellular signaling pathways. As mentioned in Sec. II, morphogens and their spatial gradients control many aspects of early development and determine much of the pattern formation that occurs as cells respond in diverse ways to various, and often multiple, local morphogen concentrations. In the case of organoid systems, morphogens are often used to push cell differentiation down to a specific path in order to produce a certain cell type of the desired tissue. In the muscle-actuated sphincter or pump, we require endothelial cells for the vessel, skeletal muscle for the pumping action, and motor neurons for controlled activation. However, achieving the needed level of control is clearly beyond current capabilities, as it is a major challenge to manipulate concentration gradients with the precision found in embryos. So, this method lacks the spatial specificity required to produce tissues with multiple cell types by design, as needed in the generation of a complex M-CELS such as the pump of Fig. 1.
In the same way that chemical factors act to control the biochemical signaling pathways, cellular and extra-cellular mechanics can also exert control over the cell function during development.42 Force transmission through matrix, cells, and their constituent molecules, in fact, can be thought of as a parallel “mechanical signaling pathway,” one that communicates with the biochemical pathway via several channels, including changes in molecular confirmation. Thus, as a result of force transmission through the extracellular matrix (ECM), cell-cell junctions, the cytoskeleton, or the membrane, proteins might alter their binding affinities or expose cryptic binding sites, stretch-activated ion channels might open, and cell surface receptors might become activated.43,44 However, just as in the case of biochemical control, mechanical control remains a relatively blunt tool to use for co-differentiation of cells into a desired spatial pattern. Methods are being developed, however, drawing upon existing technologies such as optical or magnetic tweezers or AFM, which offer finer control. Just as cells secrete factors to control biochemical signaling, cells can also exert mechanical force internally or externally, as a means of regulating their interactions.
An important layer of the morphogenetic field that orchestrates cell behavior is electric. All cells, and not just excitable nerve and muscle cells, communicate electrically using the same ion channels, neurotransmitters, and electrical synapses (gap junctions) which evolution optimized into brains. Parameters such as resting membrane potential (Vmem) control proliferation, migration, cell shape, and differentiation45,46 and interact with chemical gradients, gene-regulatory networks, and tensile forces. More importantly, recent studies revealed that the dynamics of bioelectric networks process a kind of tissue memory that specifies high-level properties such as size control, organ identity, and axial polarity of the entire body plan.47 Recent developments of pharmacological and optogenetic techniques for specific manipulation of bioelectric circuits in non-neural, somatic tissues have shown the ability to trigger appendage regeneration, reprogram gut into complete eyes, alter the number of legs and heads, reprogram tumors into normal tissues, and overcome mutations in key patterning genes.48,49 Bioelectric properties control gene expression and chromatin state via a number of known transduction mechanisms; major opportunities for advances in this field focus on cracking the bioelectric code—the mapping of large-scale voltage gradient patterns onto organ-level outcomes.50,51 Given the degree of plasticity revealed by bioelectric controls of growth and form, advances in this field will be an invaluable aid in the top-down programming of the shape in bioengineering applications.
Genetic engineering of cells allows us “from the inside” to modify and extend the biological programs that underlie cellular behavior, through the design and implementation of synthetic gene regulatory networks.52 With the advent of the Cas9/CRISPR system,53 it is now possible to edit the genome and dynamically regulate specific genes. Synthetic biology has evolved from demonstrating a simple gene circuit in prokaryotes54,55 to large multi-input circuits in prokaryotes56 and eukaryotes.57 These circuits usually comprise genetically encoded sensors that detect levels of intracellular and extracellular biomarkers, a computational core that processes sensory information and makes decisions about which specific actions to take, and actuators that affect the cell state and the surrounding environment. Initial emphasis in synthetic biology focused on the creation of gene circuits that operate orthogonally to the cell, attempting to minimize bidirectional interactions and dependencies between the circuit and the host. In contrast, genetic circuits fashioned for M-CELS must at their core be conceptually and practically embedded within the host and tissue, responding to dynamically changing cellular and extracellular conditions and controlling the cellular milieu towards desired phenotypes.
Unlike many traditional approaches in engineering, M-CELS must develop the capability to cope with or exploit the heterogeneity of cell types, states, environmental conditions, and fluctuations in gene expression (or “noise”). Engineering noise of M-CELS will require addressing noise at multiple-scales throughout development. Predictive modeling and advanced tools to modulate the noise of genetic and regulatory circuits within the cell will be needed.58–61 Addressing these issues within an ensemble includes accounting for the coupling of fluctuations in cell-to-cell signaling and the microenvironment. This presents an opportunity to develop noise engineering approaches in both single-cell and multi-cellular systems. Engineering stochasticity using synthetic systems58,59 or exogenous drug targeting60 of gene circuits and regulatory motifs and networks can bias cell-fates and provide a role for stochastic noise engineering of growth and pattern development.62,63 Integration of engineering noise and stochasticity into the fundamental framework of bioengineering living systems can account for an iterative design of fluctuations in signaling pathways at both single- and multi-cellular levels.
Other methods that hold promise for localized control in M-CELS are optogenetics,64 using light-activated ion channels, and magnetogenetics,65 locally activating ion channels, applying force, or precisely controlled heating. In these, either light or magnetic fields, often in combination with nanoparticles, are used to control local processes. For example, in the sphincter/pump of Fig. 2, activating spatially separated cells to differentiate into the three indicated cell types might be accomplished by localized stimulation. This might be accomplished by focusing light of different wavelengths at cells engineered to synthesize specific transcription factors through the use of optogenetics. The capability of such spatiotemporal activation, however, has yet to be demonstrated, and remains an obstacle to progress.
Thus, cells possess a variety of regulatory mechanisms, and each of these provides a means of regulating cell differentiation and function in the generation of M-CELS and, in particular, to exert control over the morphogenesis of organoid systems. To do so requires more than merely a collection of “tools” that can be used to manipulate the system; instead, it would need a systematic approach to gaining a more fundamental understanding of the principles that determine form and function. Moreover, such an understanding would then facilitate generation of a set of design principles, as described in Sec. VII. This represents one of the major challenges of the field going forward.
Several recent scientific advances have helped seed the nascent field of M-CELS, but none more so than those in organoid culture systems. Organoid, in this context, refers to an aggregate of multiple organ-specific cell types structured similarly to the in vivo organ and sharing its key functions. Research in organoids, built on work that started in the early 1900s,66 is now accelerating as it incorporates recent advances in stem cell biology. Numerous publications have identified culture conditions that support the differentiation of pluripotent stem cells (PSCs) and the differentiation and/or self-renewal of adult stem cells in a manner that produces organoids.67,68 PSCs have been differentiated into organoids representing a variety of organs or complex systems including but not limited to the liver, brain, small intestine, stomach, pancreas, lung, kidney, and esophagus.69–80
While there are large gaps in our understanding of the mechanisms of organoid formation, such as the rules that govern morphogenesis, regulate the size and shape, and lead to the emergence of adult stem and progenitor populations, researchers are able to successfully maintain organoids in culture for extended periods of time.67,77 The most common approach to developing an organoid culture system has been to follow nature's template, translating findings from developmental biology to direct differentiation in vitro. However, there are likely alternative paths that could produce similar results because in vitro systems are freed from evolutionary pressures and the dynamics of a complex organism. Discovering these “short circuits” to achieve emergent organoid features will likely be a vibrant area of research moving forward. A point of caution is whether or not the resulting organoid from such an approach will be sufficiently similar in form and function to the organ and whether or not it will suffice for its engineered purpose.81,82
Despite the advances made in the last six years in a variety of tissue types, many challenges remain to further development of organoid systems. Many of these challenges stem from a lack of fundamental understanding associated with the biochemical, physical, and mechanical drivers of differentiation within a multicellular environment which give rise to aggregates with the form and function of organs. For example, how do asymmetrical structures emerge? What structural components are critical for the emergence of the desired function? And why is there such variability in how PSCs respond to morphogenic cues? Some of the variability is due to stochasticity or underlying biological noise (see also Sec. III). Engineering approaches may ultimately leverage this stochasticity to drive differentiation or “canalization” in directions that otherwise would be unfavorable under naturally occurring conditions. The same features that give rise to cellular heterogeneity, however, also result in inconsistency with organoid differentiations regardless of the tissue type, which is a barrier to wider scale adoption of organoid systems in industrial screening settings.
Before organoids can readily be incorporated into M-CELS or used at an industrial scale, there are practical implementation issues to address. First, at each stage of organoid formation, the ability to scale in batches for suitable use in a high-throughput screen (e.g., >10 000 small molecule compounds) is limited. Second, existing protocols do not address a fundamental issue of how heterotypic tissues may have varied nutritional requirements and media needs; currently, a given organoid will be supported by a common medium and culture conditions. Third, organoids can be cultured long-term, but like other PSC derivatives, they tend to arrest developmentally before reaching a mature, adult-like stage.83 Furthermore, the current repertoire of organoids does not contain all organ-specific cell types or contain mature cell types from other systems, like the circulatory or nervous systems. Finally, most of these organoid systems still rely on undefined, highly variable components such as Matrigel and serum, which restricts our ability to control and fine-tune the emergence of organoids. These factors contribute to the variable rates of success in creating an organoid and our inability to alter the resulting structure, by design, to produce new functionalities.
Many groups are working to address these issues both by furthering our understanding of the principles governing organoid emergence and also developing new tools to aid in these studies and provide new methods of manipulating the differentiation cultures. For example, the development of on-chip culturing devices provides more sophistication in geometric arraying for imaging and recovery for later biochemical analysis. Tunable hydrogels will enable detailed analysis of the role of the matrix,84 and biosensors for metabolic studies will be valuable tools to find common medium and culture conditions.85 Microfluidic devices can be designed to support analysis methods such as smFISH and immunofluorescence.86 Novel tools for multicellular image digitization into network properties87 have allowed for extraction and quantification of emergence, but more computational methods need to be developed in order to identify and predict desired features of organoid development, especially if large industrial-scale manufacturing of these systems is needed for the next generation of drug development.
Building on the success of microfluidic technologies and iPS and stem cell engineering and driven by the widely recognized need for transformative changes in the process of drug discovery and development, a variety of new assays have been advanced which enable certain aspects of the organ function to be replicated with in vitro models. While the “organ-on-chip” (OoC) technologies are still at a relatively early stage in development, nascent versions of cardiac muscle,88,89 liver,1,90 brain,91–93 lung,94 skin,95 placenta,96 and various other tissues have been reported (Table II). Similar systems have been created for the purpose of modeling and gaining new insights into fundamental disease processes such as cancer97 and Alzheimer's disease.98 Given the increasing availability and reduced cost of generating iPS cells, the prospect of developing patient specific assays to screen for optimal personalized therapies is on the horizon.
Practical challenges include phenotypic instability, low throughput associated with system complexity, material-drug incompatibilities of commonly used device materials such as PDMS, and biomaterial inconsistencies and limitations. A fundamental question for OoC technology is, will we be able to create microscale constructs that adequately recapitulate the macroscopic organs? There is a danger, even if we construct OoC using human cells with sufficiently mature phenotypes, that the resulting system will have the physiology more reminiscent of a small rodent rather than the humans from whom the cells were originally derived. Two major scaling issues arise in OoC design and construction: (i) maintaining absolute values of physiological parameters and (ii) maintaining relative sizes between different types of cells, tissues, and organs.
One example of changes in absolute physiological parameters due to scaling involves cellular level metabolic rates.104 As the size of an animal decreases, the basal cellular metabolic rate increases. Importantly, this does not occur because of cell-intrinsic differences in the metabolic rate. Instead, the basal metabolic rate of cells is dictated by the rate of nutrients and oxygen delivery. Thus, in the typical, nutrient-, and oxygen-plentiful culture environments, human cells and even faithfully constructed minimal functional tissue units would metabolize at mouse-like rates rather than human-like rates. While there are potential solutions such as limiting oxygen availability,105 effective implementation in OoC systems is still a challenge.
Even if one were to achieve human-like physiological parameters for each organ module, another challenge arises when connecting multiple OoCs together with a common medium that might be circulated by our muscle-actuated pump example. If we shrink all linear dimensions proportionally, the multiple OoC system will not reflect human proportions in terms of function. For example, if a lung is isometrically miniaturized by a factor of 100, the surface area available for gas exchange falls nearly 10 000-fold. The same change in linear dimension for muscle would reduce oxygen consumption rates by a factor of 106. Some potential solutions include “functional scaling” of organs where miniaturization factors would be assigned to different organs based on whether the organ function depends more on its surface area or its volume.104,106 While these and other potential solutions have been proposed,107–109 effective implementation and validation in OoC systems remain a significant challenge.
While the long-term goal may be the construction of an OoC that reproduces all aspects of human physiology, the near-term opportunity may be in fit-for-purpose OoCs that test for specific aspects of efficacy or toxicity. Such efforts will require close collaboration with target end-users to develop OoCs with “just enough complexity” to provide a valuable physiologic readout while also allowing for a robust and high throughput assay. The systems will also have to be extensively validated against existing gold-standard animal models using a large panel of reference chemicals. A helpful resource with various examples of validating non-animal technologies is The European Union Reference Laboratory for Alternative to Animal Testing (eurl-ecvam.jrc.ec.europa.eu). Finally, the continuing difficulties in predicting adverse immune responses and the emergence of a range of successful immunotherapies suggest a need and opportunity for OoCs that replicate tissue-level aspects of immunity.