Perspective: The promise of multi-cellular engineered living systems

An important application of M-CELS could be in the development of biological robotics and actuation systems.110 Over the years, engineers have produced a variety of non-biological robots and machines with an immense impact on our lives and industrial production. However, they have fundamental limitations when compared to living systems, e.g., they cannot self-emerge, self-assemble, or self-heal. On the other hand, in nature, meso- and macro-scale functional organs and organisms (machines) emerge through complex interactions between individual cells and the extra cellular matrix. Organs, in turn, interact with one another through precise control algorithms that maintain homeostasis. However, the rules of the interaction between living biological components at various hierarchies and spatio-temporal scales remain elusive. With new methods of culturing and manipulating living cells, there is a growing interest in developing cell-based biological robots, many with engineered scaffolds. It is envisioned that such robots may have unprecedented capabilities, as they could reap the benefits of evolutionary pressures.

As an example, the development of biological actuators can form the building block of more complex emergent systems. Such engineered biological actuators consist of muscle cells (primary, cell lines or differentiated from stem cells or iPSCs), supporting cells such as fibroblasts, and an extracellular matrix (ECM) and may require a scaffold as a substrate. The cells interact with the ECM and remodel it (compact and align fibers of the ECM), approach, align111 and fuse with each other, undergo myogenesis, and emerge as muscle strips anchored to the scaffold. Unlike cells in other organoids, muscle cells interact over long distances using mechanical force. These complex interactions, both local and distant, involving diffusion and mechanical forces, result in an emergent muscle system that possibly maximizes force and motion and minimizes energy demand. Thus, the design of a muscle system involves solving an inverse engineering problem—given a prescribed form and function of the system, choose the appropriate scaffold and its biomaterial, ECM, and the muscle cell type and density.

Using the above principles, engineered muscle systems have been developed for in vivo applications such as cardiac tissue repair112 and in vitro applications such as pumping fluids at small scale113 and phenotypic assessment of neuromuscular disease.114 A variety of muscle systems have been explored to achieve multi-dimensional and complex motion and deformation from muscle contractions on 2D and 3D substrates.115 These engineered systems serve as a test bed for studying muscle force, their temporal dynamics, longevity, and overall performance in contrast to their in vivo counterparts, with the ultimate goal of providing muscle repair, analysis of muscle disease, and evaluation of drug efficacy. More interestingly, it is now possible to use human iPSCs to form patient specific self-organized muscle strips to develop disease models for prognosis, to interrogate disease mechanisms, and to study drug effects116 for individualized medicine. Recognizing that muscles are often activated by neurons for both voluntary and involuntary motions, recent work with M-CELS also involves muscles innervated by neurons, forming functional neuromuscular junctions.102

The force and deformation attributes of engineered muscle systems have recently been employed to generate locomotion of small structures. They include swimming in fluids mimicking flagella dynamics at small scale (low Reynolds number),117 coordinated flapping dynamics at larger scale (high Reynolds number),118–120 and walking using leg-like structures.115,121–124 These small robots move autonomously with cardiomyocytes as actuators, or they are stimulated to move by light or an electric field when optogenetic or regular skeletal muscle cells are used to actuate them. Through these elementary systems, we are learning to imitate different capabilities of natural organisms, including locomotion and active transport, which can lead to new applications.125 These include future in vivo applications in drug delivery and micro-surgery. However, major challenges need to be overcome, including the issues with biocompatibility, imaging and control of trajectories, viability, and their removal after the desired function. In the near term, they serve as test beds for studying emergent properties of biohybrid robots arising from clusters of heterotypic cells, ECM, and scaffolds. In the medium and long-term, such top down engineered biological robotics could consist of many more cell types to achieve specific functions and applications in the environment, energy, medicine, and others.

Success of engineering systems rests on the formulation of effective design principles that guide the creation of complex systems. Accordingly, achieving the M-CELS goals of creating complex synthetic multicellular structures with defined behaviors will require the establishment of experimentally verified rules and practices that guide the efficient and predictable formation of these systems. The key to success will be the integration of traditional engineering concepts, such as creating and characterizing reusable parts, establishing rules for the composition of such parts, appropriate layers of abstraction, and modular system design, along with design principles that take into account the unique properties and interactions of the biological substrate. In other engineering disciplines, one can achieve desired behavior by exerting fine-grain control over system components. In contrast, while precise spatiotemporal control over specific elements of a living system is becoming feasible, regulating all aspects at all times is not realistic. As such, a different approach must underlie our biological system design efforts.

However, while traditional engineering approaches may have limitations, if they are pursued with regard to the biological substrate, it is important to note that M-CELS efforts should be inspired by, but not limited to, natural designs. Design principles that have emerged from the evolution of biological systems are instructive regarding what works and what does not work in a biological context. However, evolution is incremental and produces intermediate designs that are both functional and competitive within their environmental context. In contrast, M-CELS and their designs are not constrained in the same manner. Both top-down and emergent engineered systems may favor configurations that are easier to understand, intentionally support and simplify future enhancements and modifications, and arrive at solutions that leapfrog existing designs.

Several fundamental properties of living systems introduce challenges to engineering design. Importantly, the high-level behavior of living systems emerges from the properties and interactions between their constituent low-level components.126 To design M-CELS effectively, we must consider the vast range of scales in which they operate (from the subcellular to the organismal) and the need to integrate different control modalities. In biology, as in other areas, the system function is tightly interwoven with the structure. The structure/function relationship manifests itself at different scales of living systems, and it is also impacted by the numerous ways in which cellular systems exert control over themselves and their surroundings. In the efforts to engineer living systems, we have at our disposal a large set of proven methods for manipulating biological behavior. It can be modulated “from the outside” via micro and macroscale methods described in Sec. III.

To fully realize the potential of M-CELS, these control methods can serve as composable functions. These functions should have inputs and outputs and perform some transformation on the living systems. The functions operate in space and time and at defined scales. While they always perform operations physically, their impact can be both physical and regulatory/information-processing. Functions have specific energy usage, operate at different modalities, and may require coordination among multiple cells and time scales. While the concept of a function is relatively straightforward, implementing a design framework that effectively integrates different modes of control modalities as discussed in Sec. III remains a challenge. 3D multicellular simulation tools such as Morpheus127 and others87,128 integrate gene regulation, cell signaling, and biomechanics and may serve as a useful basis for M-CELS design tools. However, missing are effective abstractions that support M-CELS 3D organ and multicellular machine design objectives, akin to gene circuit design tools.129–132

New scientific directions inevitably require new technologies. In the context of M-CELS, we have already encountered technological limitations for sensing, modeling, and reengineering of the form and function of cell-based systems. Technologies are needed for the formation and characterization of organoids and cellular clusters, 4D imaging of these biological systems, 3D dimensional patterning and fabrication, flexible means of providing bioelectric and metabolic stimuli during guided self-assembly, and computational approaches capable of predicting the behavior of cell populations of multiple cell types whose form and function emerge over time.

Enabling technologies can be categorized in terms of where they fit within the life cycle of a M-CELS. First, the systems need to be assembled or arranged into a configuration that facilitates emergence and self-assembly. Technologies such as 3D printing, biomaterial scaffold design, and advanced microfluidics have already demonstrated their benefits and potential but need further development. Microfluidics has the capability of culturing cells in 3D with multiple communicating cell types that can be arranged through 3D printing so as to facilitate the essential interactions needed to generate the necessary cell-cell and cell-matrix interactions and realization of autonomous operation of these systems in a wide range of environments and ambient conditions.

Methods are also needed to interrogate the developing systems and to assess their final outcome—the end product of the biomanufacturing process. Imaging will be key, but we are now facing limitations in terms of our ability to visualize the structures deep inside a M-CELS, especially in a label-free manner in live cells over time. Methods such as CLARITY133 and 3DISCO134 and their variations have already proven useful for imaging large multicellular constructs, but these, too, have limitations and provide little opportunity to assess the cell functionality of internal structures.

Another consideration is the need to assess system stability over time. Unlike abiotic systems, M-CELS have the tendency to continue to change with time. Some aspects of change—adaptability and self-repair—have benefits, but others such as loss of differentiation and progressive cell death are clearly detrimental to function. It will therefore be necessary to incorporate methods to continually assess the stability of the system so as to ensure its performance. Functional measures such as the secretion rate of an essential chemical or force generation are examples.

In addition to experimental work, computational approaches will be required at every step and closely coupled with experimental data, to analyze, predict, and probe the system's operation. Synthetic biology at the single-cell level has had many successes,56,57 but it has also had to accept that progress has been slower than once hoped.135 One of the many challenges is the non-intuitive behavior of molecular circuits: complex topologies, non-linear relationships, and feedback loops often make the dynamics of a circuit impossible to predict without the aid of computer simulations—these unexpected behaviors may be considered as examples of the emergence, discussed above. Much of this work still revolves around trying to understand the dynamics of simple circuits that are already constructed rather than trying to design de novo circuits, but this ambitious challenge of engineering new circuits is also gaining ground.34,130–132

Switching from single-celled circuits to multicellular systems brings a level of complexity for computer modeling. Feedback loops now exist at more scales than the purely molecular. Although the active dynamics of cells are known to be largely controlled by their molecular states, it is now equally clear that macroscopic events directly feed back to control molecular events. For example, tissue growth may push a group of morphogen-secreting cells away from their target cells, thus changing their expression profiles as a direct consequence. If the behavior of feedback loops in single-celled gene circuits is hard to predict, the complexity of these multi-cellular and multi-scale feedbacks is dramatically more challenging. That is, the degree of complex and subtle emergence is even higher than for molecular circuits alone. Understanding and predicting them will not be possible without good computer models, so a clear expectation for the future is that multi-scale numerical simulations must become a key goal for M-CELS.

A wide variety of modeling formalisms exist to tackle multi-cellular systems. At one end of the spectrum are continuum approaches such as Finite Element Modeling (FEM), which approximate a tissue as a continuous material, and they are optimal for questions involving physical mechanics.136 At the other end of the spectrum, Cellular Potts Models (CPMs) employ many discrete variables on a lattice of points to represent each cell of the tissue.137 They can therefore capture the irregular shapes of individual cells and have been used to simulate various developmental processes,138 as well as tumor growth and vasculogenesis.139 However, for general-purpose modeling of engineered tissues, their computational cost and limitations in representing the mechanical integrity of large-scale structures (e.g., macroscopic bone or cartilage elements) make them unlikely to become the primary formalism of choice for M-CELS. In between these two extremes are a collection of techniques which together may be termed off-lattice methods: agent-based models are often defined in terms of logic-based rules, or “state-charts,” and have been particularly successful for cases where the primary scientific question is about the control of differentiation through discrete cell states or fates.140 Vertex models in contrast have been used mostly to explore the physical mechanics of cellular interactions—how different types of cell divisions affect the packing and cell shape.141 Since this formalism explicitly represents the boundaries of cells, it is only really optimal for epithelial tissues having a neat polygonal packing arrangement. For volumetric 3D tissues such as mesenchyme, particle-based or cell-centroid models are preferable. They cope well with arbitrary 3D arrangements of cells, without the computational overhead of maintaining mesh integrity.

At the same time, they are ideal for simulating the molecular aspect of the system—cell autonomous reactions can be calculated on each particle, and diffusion can be well-approximated by exchange of molecular concentrations between adjacent cells. Impressive results have recently been demonstrated for early zebrafish development and gastrulation.142,143

Despite exciting improvements in these modeling approaches, we should be cautious about predicting the use of computer modeling as a tool for “rational” design of multicellular M-CELS. Thus far, the design principles found to be useful tend to exist at specific scales. For example, at the lowest scale of small molecular circuits, we may use the design concepts of Boolean logic or dynamical systems theory (attractors in phase portraits, etc.).34,56 At the intermediate scale of multi-cellular pattern formation, we can employ the unifying concepts of positional information or self-organized patterning.144,145 While at the scale of macroscopic tissues, interactions between active tissue growth and the material response properties of the tissue can also lead to design principles based on physical mechanics (the control of stress distributions, buckling, etc.).146 However, the interactions between design principles at different scales have rarely been explored.147 Since a key feature of engineered tissues and organs is their multi-scale nature and we hope to control dynamic macroscopic processes through “molecular programs,” it will probably take a significant phase of further basic research before computer modeling aids us in “forward” rational design of these systems. Indeed, for the foreseeable future, computer modeling research will probably focus on obtaining satisfactory models of existing multicellular systems, and this will largely depend on “reverse-engineering” from large quantitative datasets. Organoids may prove a particularly tractable system to work on in the short term, due to their both relative simplicity (compared to organs or embryos) and visual accessibility for digitizing with time-lapse imaging.

The biomanufacture of M-CELS poses significant challenges to existing bioprocesses.148 M-CELS present unique needs based on their inherent complexity and reliance on emergent behavior. The overarching need—developing effective methods to reliably and robustly produce M-CELS at a desirable scale—highlights a number of challenges and opportunities.

The development of M-CELS relies upon the integrated formation of cellular structures. Such structures can be achieved through either self- or guided-assembly of multiple cell types into functional units (see Sec. I). As we consider methods for manufacturing these systems, it will be important to understand whether such complexity can be reduced or segmented in a modular fashion to create M-CELS subunits, followed by assembly into desired M-CELS in a controllable or bio-foundry manner. It is expected that efforts to better understand the phenomena associated with emergence will help elucidate where such processes may be simplified or deconstructed and where they cannot. The goal of such efforts is to enable the creation of a “bio-assembly line” to achieve industrial scale production of desired M-CELS.

Biofabrication of subunits and their further assembly into M-CELS requires manufacturing technologies across multiple disciplines (Table I) and technologies (Table III). Simple 2D and 3D cell culture methods can be augmented or replaced by newer technologies such as 3D bioprinting,149–153 and automated bioreactor systems are key to enabling the generation of compartmentalized M-CELS structures in an organized manner (Fig. 3). The continuing development of effective 3D bioprinting technologies may enable the effective implementation of this approach in M-CELS biomanufacture. Although the resolution that is currently achievable with 3D bioprinting may not yet permit direct implementation on the length scales (10's of microns) required for M-CELS formation, they can help, at least, spatially position cells in a manner that will facilitate further self-assembly and may prove to be effective tools for automating and standardizing key steps in M-CELS biomanufacturing.

Technologies that enable both scale up—relying on novel culture platforms with increased capacity (e.g., bioreactors) and scale out—increasing the number of culture systems in use—will be of paramount importance to M-CELS biomanufacturing (Fig. 3).154 Improved control of M-CELS biomanufacturing poses unique challenges based on the features that differentiate M-CELS from traditional cell based systems—cellular complexity and emergent behavior. Since cell-based processes are subject to inherent biological variability, the push towards more defined and controllable components for use as cellular matrices and media is imperative in minimizing “external” heterogeneity of M-CELS biomanufacturing. These variabilities highlight the challenge for quality control (QC). Traditional cell-based products such as primary cells or stem cells are developed with a mindset towards maximizing homogeneity of the cell type(s) associated with these systems. As we begin to tackle the challenges of M-CELS biomanufacturing, the heterogeneity of the cellular component is not just a possibility but a necessity. Methods to understand and control this heterogeneity are therefore essential in the successful biomanufacturing of M-CELS.

Furthermore, the biomanufacturing of M-CELS will necessitate the development of effective and appropriate assays for evaluating the M-CELS identity and function. These assays will include the characterization of cell sources and biomaterials using many of the technologies described in Sec. VIII. Ideally, these should be noninvasive or minimally-destructive to enable live and real time monitoring of developing M-CELS. As our ability to discern the mechanistic underpinnings of M-CELS development evolves, we also anticipate an increasing demand for the use of molecular and analytical analysis in the selection of input materials. We already see the consequences of the considerable variability in these fundamental components of a M-CELS—variability in organoid formation from the same starting population155 and disparity in directed differentiation of various pluripotent stem cells.156,157 These examples highlight differences that can occur under ostensibly identical extracellular environments. We anticipate that technologies such as targeted sequencing, micro RNA analysis, epigenetic fingerprinting, and proteomics will help elucidate and establish appropriate specifications for these cells, but much work is needed before adequate procedures can be put in place.

Finally, the biomanufacturing of M-CELS demands the development of reliable tissue preservation technologies. While laboratory-scale M-CELS can be made and used on-demand,148 industry-scale production of these systems includes release and distribution of M-CELS through supply-chains that are dependent upon reliable preservation. Cryo- or other preservation technologies of M-CELS pre-products or sub-assemblies will streamline and simplify the manufacturing process. The ability to preserve completed M-CELS could enable building “finished-goods” inventories and shipping to end-users.

A fundamental question for researchers working in synthetic biology, emergent behavior, and living systems is whether they are “creating life.” Some researchers in synthetic biology believe microorganisms are “just” machines, and that the creation of new such machines shouldn't be considered differently.149 Others may argue that researchers need to be cognizant of the degree to which projects might be perceived as creating life and what special obligations might exist in this creative process. The language of “creating life” raises fundamental questions, including religious issues, for some observers who ask whether there should be natural limits beyond which we should not be trespassing. While many of these issues have been raised in the context of synthetic biology, they grow in importance when, for example, we become able to produce a functioning brain organoid that can collect, process, and act based on information gained from the environment. In addition, the concept of pain and sentience150,151 would need to be addressed in the development of M-CELS that could be perceived as replication of living entities. As many of the future M-CELS could by pass the natural developmental process, the use of the 14 day rule150,152 employed to provide an ethical framework around embryo research would have to be modified.153 We are also cognizant that some communities, in various parts of the world, may be more sensitive to these concerns than others.

Living systems' research that replicates essential components of existing life forms may also raise questions regarding whether there should be limits to scientific efforts that “re-create” natural systems. Foundational to this discussion is the complexity of and current lack of full understanding of the underlying biology and its emergence. Limited knowledge on biological emergence makes it difficult to determine or project what the consequences might be of current research. Indeed, the concept of emergence in complex systems raises the prospect of unforeseen outcomes in the M-CELS that are created, with potentially negative consequences. Efforts to understand and incorporate system repair and healing pathways should also be assessed parallel to efforts to control or limit potential harmful outcomes. Designing “kill-switches” to halt unforeseen harm provides one possible pathway but one that is potentially at odds with efforts to support system self-repair.

A consequence-based analysis would include an assessment of the potential for harm and dual use—in which the technology itself or resulting living systems may have unforeseen, harmful applications with public health and security implications (e.g., bioterrorism). This analysis would also require discussion of possible safeguards that could be explored throughout the research process.

Medical applications of M-CELS, particularly in connection with the creation of non-natural organs (e.g., ones that might synthesize and secrete therapeutic factors for chronic illness) or tissues with enhanced capabilities (e.g., a muscle “patch” containing cells that out-perform natural muscle and thereby enhance athletic performance or an implant that enhances mental acuity) raise new issues not previously considered. At what stage do we limit the use of such performance enhancements? Would they only be accessible to a select few? Numerous scenarios can readily be envisioned which raise important ethical questions for society.

The asserted goal of work in M-CELS is to support the greater good of society, identifying constructive, efficient, new pathways for solving functional, real-world needs. As research in M-CELS advances, acknowledging both “precautionary” and “proactionary” risk management approaches/principles will be essential. The precautionary principle advises that when there is a potential that harm may occur, researchers should use extreme caution, stopping research until the potential for harm can be assessed.154 The proactionary principle, on the other hand, views research as positive for society and assumes that research is beneficial unless there is evidence to the contrary.155 An intermediate ethical position may provide a reasonable commitment to both accountability and responsibility. Collaboration between researchers would provide a system for checks and balances, with sharing of information between labs, while acknowledging the tension such sharing would have with pathways for preserving research/publication rights. M-CELS researchers, in close consultation with bioethicists, should develop a shared ethics framework and consider creation of a shared governance structure.

It is unclear to what degree researchers and trainees who work on M-CELS are considering or fostering discussion of these questions surrounding “creating” or “re-creating life” or controlling emergent behavior in their daily work. Those involved in research may benefit from an intentional structure for identifying underlying, competing values and ethical principles in their work and providing a forum for discussion, especially for trainees, about how scientific values and personal values or beliefs coincide. The development of such a frame work and an ethics code of conduct for M-CELS research is an imperative task for the research community to develop. Being able to articulate the ethics underlying one's work is so important in communicating within the scientific community, with funders and policy makers and with the public at large.

Following on recent advances in understanding single cell behavior156 and in developing simple, proof-of-concept biological machines,108,112 organoids,6 and organ-on-chip technologies,84 efforts are underway to develop the scientific and engineering principles that will ultimately enable the development of M-CELSs. The approaches, however, are widely divergent and often lack a sound basis due to the absence of a fundamental understanding of aspects unique to M-CELSs—complexity, the central role of emergence, and intrinsic variability in the starting components—and fail to take advantage of their extraordinary capabilities—self-assembly, growth, self-repair, adaptation, learning, etc.