Rising to the challenge: applying biofabrication approaches for better drug and chemical product development

The continuing growth in global population, together with rising calorie consumption, has resulted in a potentially unsustainable demand for food. It is the vision of agribusinesses to bring greater food security in an environmentally sustainable way to help meet this demand for yield gain. As such, their product portfolios include a broad range of plant protection products (PPPs; e.g., herbicide, insecticide, or fungicide) which require comprehensive mammalian toxicity evaluation, including carcinogenicity studies, before they can be registered in most regions of the world.

It is a requirement that all active substances be tested in at least two species (usually rat and mouse) for their potential to act as carcinogens. The liver is a common target organ in agrochemical carcinogenicity studies. Hepatocellular hyperplasia is a common injury response to chemical exposures in rodent toxicity studies and may result in unfavourable toxicity profiles in long term toxicity studies. Current approaches for examination of liver toxicity require samples taken from repeat dose in vivo studies, but in a conventional testing paradigm such studies are conducted only after other in vivo toxicity studies have been completed. Earlier identification of unfavourable areas of chemistry may reduce the number of animal studies conducted with compounds which are ultimately determined to be unsafe or unregisterable.

However, the utility of these studies has been called into question. Carcinogenicity studies run in rats are not necessarily predictive of the result of similar studies run in mice [15], and liver tumours in rodent models are not conclusive evidence that a compound will exert a carcinogenic effect in humans (as a result of rodent-specific tumorigenic mechanisms). To examine the human relevance of tumours observed in rodent carcinogenicity studies, in vitro and in vivo models are used to establish the adverse outcome pathway (AOP) underlying the observed effects [16]. Once established in rodents, the sequence of causative key events in the AOP can be interrogated across species to examine the relevance of rodent toxicities to humans. Currently the establishment and cross species interrogation of key events in AOPs is a complex, iterative undertaking requiring a number of animal studies.

Whilst current in vitro models are able to provide insights into the AOPs of liver toxicity, existing models do not accurately reproduce the architecture of liver tissue and often do not include the full range of cell types present in vivo (including hepatocytes, cholangiocytes, Kupffer cells, liver endothelial cells, and hepatic stellate cells). As a result, existing in vitro liver models are generally not suitable for histopathological analysis, limiting the potential for validation of these models against existing in vivo toxicity databases and their ability to predict the outcome of in vivo studies. Furthermore, the absence of architecture and use of limited cell types in existing models prevents accurate mimicry of tissue microenvironments, potentially reducing model predictivity.

Bioprinted organ models and tissues could potentially be used to both establish AOPs and as a tool to facilitate evaluation of the relevance of rodent tumours to humans, increasing the confidence in safety assessments and reducing the reliance on animal models. This could help guide early stage agrochemical research projects away from toxic areas of chemistry through early detection of hepatic toxicants and structure-activity relationship building. Rapid advances are being made in this area [17–19], but there still remain a number of critical considerations which need to be accounted for during future model development to maximise the potential of these tools in the context of safety assessment of PPPs—see table 1.

To be useful models for mode of action (MOA) investigation and as a predictive tool for compound development, the response of bioprinted constructs to hepatotoxins should be demonstrated to mirror the physiological, histological and biochemical responses observed in humans and rodents. The properties that a bioprinted liver model should exhibit in order to function as a predictive model are likely to require further investigation, however factors including metabolic capability (e.g. expression and activity of phase I and II enzymes), gene and protein expression (particularly of key nuclear receptors including the peroxisome proliferator activation receptors, constitutive androstane receptor, pregnane X receptor, retinoid X receptor, aryl hydrocarbon receptor, farnesoid X receptor, and glucocorticoid receptor) and cellular structure (open questions include the necessity of Kupffer cells in modulating responses to hepatotoxicants) are likely to prove important in model development.

Details regarding the manufacture of bioprinted constructs, such as whether constructs are created in the presence or absence of a scaffold are unlikely to be a serious concern for industry, as the primary concern for these models will be around predictivity and validation rather than method of manufacture. How amenable the tissue is to histology is also an important consideration in being able to back translate responses to existing liver toxicity data and provide a degree of confidence in the ability of the model to predict human responses. Whilst precursor events are commonly observed in sub-acute and sub-chronic toxicity studies, most liver carcinogenicity observed with agrochemicals is the result of chronic exposure. As such the longer term stability of the model (up to 28 days) in culture is important to provide sufficient time course data to be able to predict chronic effects and support the development of appropriate AOPs.

The pharmaceutical industry is facing considerable challenges in the development of new safe and efficacious innovative medicines. Productivity has declined steadily over the last two decades, despite the investment in drug development doubling every nine years over the last 60 years. Only 10% of drug candidates entering the clinic achieve US Food and Drug Administration approval [20] and, according to some studies, it now costs approximately $2.6 billion to develop a drug [21]. A number of factors contribute to the high attrition rate and costs in drug development, including higher regulatory hurdles for safety and efficacy and commercial considerations including reimbursement strategies as well as strategic portfolio decisions driving project termination. But a substantial proportion of this failure can be attributed to the lack of safety or efficacy, suggesting the preclinical models used to assess these poorly translate to the clinic [22].

Respiratory disease remains an area of considerable unmet medical need with significant barriers to new drug development [23]. Many of the new respiratory drugs that have failed in the clinic have performed well in preclinical animal models, suggesting more predictive models and tools based on the latest technologies, including bioprinting, are required. But what these models should look like will clearly depend on the question being asked, and what particular respiratory disease you are studying. A 'one size fits all' model will not work, and therefore engagement and collaboration between the bioprinting community and end-users early in model development is essential for delivering bespoke models with utility.

Diseases such as asthma and chronic obstructive pulmonary disease are complex, heterogeneous diseases with diverse aetiologies. Not one single model is likely to replicate all of the necessary features to recreate these diseases in vitro, but there are some common features that are essential starting points. These include being able to model airway hypersensitivity and inflammatory cell recruitment. For example, bioprinting the bronchiolar epithelium should include not only the simple ciliated columnar cells but also secretory Clara cells, as well as neuroendocrine cells. Alveolar units should comprise Type I cells and Type II cells while pulmonary vascular tissues should include endothelial cells from different vascular structures, smooth muscle cells, and adventitial fibroblasts [24]. Importantly, any bioprinted lung mimetic should incorporate cells from the innate immune system or even adaptive immune systems if studying a particular pathogen to aid our understanding of the immune processes which drive respiratory disease. Given the chronicity of respiratory disease, it is important that bioprinted material can be used over a prolonged period of time and therefore demonstrate viability of weeks and preferably months, rather than days. As our ability to stratify patient phenotypes improves, we also require models that represent the heterogeneity of patient subpopulations. Such an approach is also likely to require mathematical modelling to determine the appropriate number of variants to create and study.

Understanding the basic anatomical and cellular components for bioprinting lung models suitable for drug discovery is the first step towards creating useful models for improved efficacy and safety testing without animals. Working collaboratively with the end-users will ensure that material is printed and models are built that address the gaps evident with current models. Related to this, by working in a partnership the bioengineering community will understand how and why the models are used and at what point in the drug discovery/development pipeline they will be deployed. Of particular value is the bioprinting of cells for use within phenotypic screens that elucidate new targets or drive a greater understanding of biological targets early in preclinical drug development. This has the benefit of enabling companies to identify early those compounds destined to fail and remove them from development before the more expensive animal testing stages—fail early; fail cheap is the commonly used mantra. For these early screens where thousands of potential therapeutics will be assessed to identify if they can alter the phenotype of a cell or tissue in a desired manner, the models need to be fairly simple to meet the demands for high-throughput and high-content systems and importantly cost-effective and highly reproducible. However, if bioprinted material is employed in the latter stages of discovery, for example prior to candidate drug nomination, a lower throughput, more complex system is preferred. In this scenario, the model should try to emulate physiological or pathophysiological responses that are not detectable in animal models or other complex in vitro model systems.

The consumer goods industry spans a number of different product types, including foods, drinks, homecare and personal care products. Since 2013, legislation in the European Union has banned the marketing of cosmetics and personal care products that contain ingredients that have been tested on animals. For some toxicity endpoints (e.g. skin corrosion, eye irritation and androgen receptor activity), Organisation of Economic Cooperation and Development test guidelines exist to allow data generation without using animals. However, for other important toxicity endpoints, no guidelines currently exist. The marketing ban has the potential to stifle innovation in the sector and so there has been a concerted effort towards the development and application of new in vitro approaches by the personal care products industry. Bringing greater focus on human relevant, in vitro MOA safety (and efficacy) assessment and embracing Safety Science in the 21st Century framework approaches [25, 26].

The majority of homecare and personal care products interact with the skin either intentionally or as a consequence of use in other capacities, and so assessing the possible effects of these chemicals at both the skin surface and underlying structures is an essential part of consumer product development. In vitro cell culture of human cells plays an important role in both the identification of new targets and leads and in assuring their safety. Numerous 2D and 3D model systems exist for this purpose, but their utility can sometimes be limited because of their (i) cellular composition (primarily keratinocytes and fibroblasts), (ii) lack of appendages and other macrostructures (e.g. hair follicles, sebaceous glands, vascularisation) and (iii) inability to represent other cell types in the skin (e.g. immune cells, adipocytes, dendritic cells, melanocytes etc). In this context, current 3D systems have only limited value in enabling the identification of new targets (e.g. in a hair biology/skin ageing context where the presence of macrostructures and vascularisation are key determinants of in vitro to in vivo functional equivalence); neither do they represent sufficient human biology to advance our mechanistic understanding of adverse events (e.g. skin sensitisation, where immune function is often poorly represented in the available model systems).

Bioprinting offers the ability to automatically fabricate robust and reproducible 3D skin models with improved human relevant functionality and which possess critical attributes of the target tissues they represent [27, 28]. These tissues can be configured to mimic the cell density of the target tissue and to be composed solely of the appropriate cells and extracellular matrix they produce. Spatially defined deposition of cell types also enables the design and fabrication of tissues that recapitulate key architectural features of the target tissue in vivo. This spatial patterning and cell alignment is crucial to the normal functioning of most tissues, enabling cell–cell communication to be more effectively emulated. More highly complex skin models containing blood vessels and nerve fibres are potentially feasible, allowing for questions of safety and efficacy to be answered in a more in vivo-like context—see box 1.

It is unlikely that it will be possible to develop a single model that is capable of representing the entirety of human relevant biology, and that bespoke assays may be required to address, for example, effects at the level of the hair follicle [29], or effects exploring aging biomarkers, or immune responses in the skin. It is clear that certain solutions/models are further off than others in terms of their development and application. Recapitulating functional structures in the skin (e.g. sebaceous glands and hair follicles) is on a longer term horizon than say the introduction of immune competent cells into models, although this too is no trivial endeavour. Maintenance of immune cells in skin models is a key challenge as they are often short-lived and are 'recruited' to the site of function.

Being able to measure the exposure of chemicals via the dermal route is very important in any skin model. Current approaches rely on the use of ex vivo skin samples which lack functional vasculature, and so penetration is measured as a function of transit through the skin. Building in more human relevant functionality, future bioprinted skin models should incorporate tissue printing around pseudo-blood vessels to support more relevant in vivo-like exposure studies. The availability and access to ex vivo skin is often limited, therefore future bioprinting of 3D skin samples could potentially address this need and bring greater consistency/standardisation concerning the tissue supplied. Other safety endpoints of importance that should be considered in the development of bioprinted skin models include standard measures of cytotoxicity, genotoxicity and allergy. Immunotoxicology is of particular interest because of the deficiencies acknowledged in currently available 3D models regarding the absence of immune function and critical cell types (for example dendritic cells, mast cells, T-cells). 3D bioprinted models able to incorporate and retain these cell types within the tissue construct would represent a significant step forward in consumer product development.