1. Scale development in reptiles and avians
Scales are fundamentally different from the existing types of integument within humans, and as such, there is very little published literature on their morphogenesis in terms of biochemical pathways involved. There are no published studies on scale growth patterning mechanisms to date that we are aware of, after substantial literature review. As such, there is insufficient information at present to create scales from human cells using natural pathways. Therefore, we will opt for a rationally-designed engineering approach to build a scale from components we can find.
In order to do this, we must first list the main properties we want to engineer as an end goal for us to work towards. Scales come in many different types and forms depending on the species. Snakes, for example, have directional and partially overlapping scales akin to roof shingles, whereas iguanas have bump-like non-overlapping scales. This is accompanied by a complex set of underlying epidermal and dermal layers constituting scale microarchitecture, and has been a subject of our review, but will not be discussed in detail here due to the type of system we are proposing to engineer. It will be discussed further in a future update.
As such, in our efforts, we should try to recreate the scale morphology depicted below as closely as possible. While there are also many constituent layers and cell populations in true reptilian scales, these are less important than the overall morphology, so we will not discuss those intermediate layers here.
Figure 1: Simple illustration of scale characteristics
These properties are quite varied, and ideally we would be able to tune all properties individually. The main properties to focus on in scales are the following, based on the scale morphology:
- Scale height (thickness)
- Scale size
- Scale asymmetry/shape
- Overlap between neighbouring scales
- Spatial distribution and variation
Each of these can be tuned by different genetic networks constituting patterns that direct morphogenesis, some of which may be artificially inserted into the stem cell pool of patient skin. However, we are currently limited by technical constraints due to lack of known patterning mechanisms in the literature as described, and limitations on the degree of genetic alteration able to be imposed at once, so we have focused on essential morphological characteristics: development of scale height and the hinge regions (valleys) between scales. We have, at this moment, assigned lower priority to patterning mechanisms for other features such as overlap, to fully focus on the essentials in the near term.
It is likely that these morphological features are governed by mechanisms akin to Turing patterns in nature, although they have not been described in detail in the literature to our knowledge. At minimum, we may assume that such mechanisms can produce said features, but whether they do would require experimental confirmation. For our purposes of engineering, the idea that patterning can produce scales in theory is sufficient for now, and underlies our further work described here.
1.1 Scale growth can be split into pattern-forming and morphogenesis parts
Natural systems often tightly couple pattern formation and morphogenesis, such as how the gradient of Wnt ligands simultaneously establishes a pattern and directs cell proliferation and other behaviours (e.g. Li et al., 2017). However, for engineering, we want more discrete systems.
In order to engineer a proper scale-forming mechanism, we need to separate this into tangible components that are independent of each other as compartmentalised “black boxes”. For our purposes, we will split the growth process up in two sequential steps: the initial pattern formation, and then morphogenesis. In this model, pattern formation would tell each small group of cells how fast to grow (determining scale thickness), in what direction to grow (determining scale directionality and overlap), whether it is in a border region where tissue needs to be more flexible, what colour it should have (in the case of melanocytes), etc. In such an engineered system, after each cell knows what to do, all cells can be given a collective ‘start signal’ to start growing, kickstarting morphogenesis. Using the information they have gained during pattern formation, the cells will start executing their growth and differentiation program. In essence, the morphogenesis step uses the pattern generated in the first step as its input, giving us two separated mechanisms we can study and engineer.
In practice, a growing tissue continuously cycles between pattern formation and morphogenesis, possibly even at the same time, pipelining the two processes. Even so, considering these as two separate mechanisms will allow us to engineer the system more easily.
1.2 Methods of pattern generation
We first need a method to tell specific cells what exactly they need to do, depending on their location. As such, we need to first pattern the skin such that each cell knows whether it is part of the thick part of a scale, whether it is part of the gap between scales, what the orientation is of the scale, how thick the scale should be, etc. Again, we aim to use simplified methods engineered on a de novo basis, as this will be substantially faster and likely yield a viable result within the coming decades. To do this, we propose two separate mechanisms to approach this problem with: a ‘paintbrush’ method, and a Turing pattern method.
1.2.1 The paintbrush method: easy to control/engineer, but less robust over time
The first simple mechanism approach we propose involves ‘painting on’ the pattern of scales manually, hence why we call it the ‘paintbrush method’. The ‘paint’ can be any signal, such as light (optogenetics, such as cryptochromes fused to transcription factors that are activated by a particular wavelength of light), a custom ligand for an inserted receptor, or something else (Hernandez-Candia & Tucker, 2020).
Naïve painting: population extinction causes pattern decay
Before we explain the full system, we will first explore a strawman system that doesn’t yet have the right properties. A naïve approach might be to use gene therapy to create two populations of epithelial stem cells: one population produces the thick parts of scales, and the other population produces the thin membrane that spans between neighbouring scales. One could imagine the latter being the default human skin thickness, with the thick parts of the scales constituting local skin thickening above and beyond what is normal for humans.
This approach sounds rather intuitive: simply inject each population of cells into hexagon shapes on a patient’s skin, akin to applying a tattoo, and hexagonal scales will form. The boundaries will be established by the lack of scale-forming epithelial stem cells, or even the injection of the special ‘boundary’ stem cells that only produce a thin membrane. This alternating pattern between scale-forming and boundary cells would then give rise to scales on skin.
However, this approach has several key problems, as we have hinted. The most important problem relates to pattern decay: without some way to maintain the boundaries between scales, over time, the boundaries will disappear, necessitating another solution which we have developed. In this first case, as epithelial stem cells divide, they can choose either to divide horizontally to maintain the stem cell population, or to divide vertically to differentiate into skin cells. Normal, random cell death or symmetric differentiation will remove a stem cell from the population. Overall, this has the effect that the amount of stem cells of a certain type will fluctuate. Smaller populations have a high chance to go extinct if the number of stem cells in the population ever fluctuates to zero. As such, the population of boundary cells, indicated in figure 2 as red cells, will over time disappear, causing scales to merge into each other over time.
Figure 2: Population extinction causes pattern decay
Some simple simulations have shown that decay occurs quite rapidly. The width of the scale boundaries wildly fluctuates over the span of weeks to months, and in only a couple of years, or even as little as a few months, entire scale boundaries could disappear causing scales to merge or deform.
To verify, we have performed a simple simulation of the stem cell populations in the epidermis. Each population is represented as N stem cells, each of which independently divides at each cell cycle. Each cell cycle is taken to be approximately 1.75 days. At each cell division step, 25% of the cells will divide into two differentiated cells, removing them from the stem cell population (Kuri & Rompolas, 2018). 50% of the cells will undergo asymmetric cell division, meaning one of the daughter cells will differentiate and one of the daughter cells will remain a stem cell. This effectively maintains the stem cell population, since a single stem cell will give a single stem cell after division. The remaining 25% of the cells will undergo symmetric cell division to produce two stem cells. Since each cell will do this maintenance at random, the population size will fluctuate over time. In some of the worst cases, this may lead to the complete extinction of a small population due to pure chance. We have plotted these cases in figure 3 and 4, for different population sizes. Although this is not guaranteed to occur for each population, this happens relatively often, and each scale boundary has a chance of decaying as a result.
Figure 3: A stochastic simulation of the amount of ‘scale boundary’ stem cells over time, starting with a population of 30 cells. Over time, the population size will fluctuate, until randomly populations may go extinct. y-axis: cell population size. x-axis: time in cell cycles. We assume one cell cycle is approximately 1.7 days (calculated from data in Kuri & Rompolas, 2018 and Oakley 2008).
Figure 4: A stochastic simulation of the amount of ‘scale boundary’ stem cells over time, starting with a population of 300 cells. Despite the much larger boundary size, the population still regularly goes extinct after a few years. y-axis: cell population size. x-axis: time in cell cycles. We assume one cell cycle is approximately 1.7 days.
This model does not take random cell death or cell location into account, which we expect to make the onset of pattern distortion or decay somewhat worse.
To prevent this problem, we need to ensure that the population of ‘scale-forming’ stem cells and ‘boundary-forming’ stem cells are maintained over long periods of time.
Although not perfect, due to the fact that pattern decay only becomes severe after a few months to years, one possible strategy would be to manually re-paint the pattern in the clinic periodically (say, annually). This presents a different set of technical challenges and risks, in that while patterning of naïve skin with an optogenetic light array is straightforward, doing the same for scaled skin requires extra steps. There are a number of methods we may use to address this, which would involve recalibration of the original instrumentation used. There are several techniques we are considering for this purpose, but for the sake of brevity, we will discuss them in a future update.
A more advanced and difficult-to-engineer approach is also being considered, where the pattern is maintained without requiring clinical intervention, by using engineered cell-to-cell communication. At this stage, such a method is hypothetical and would not be used in a “generation 1” system.
We will first discuss the ‘compromise’ strategy of clinical re-patterning, since it is technically the easiest to implement. This less-than-perfect, but viable strategy also provides the building blocks for the more advanced self-maintenance mechanism, allowing us to use it as an introduction for both methods.
“Smart paint” method: periodic clinical maintenance
For this method, the entire stem cell population in skin will be homogeneous, and contain a single circuit. Each cell listens to input, and switches to a ‘scale-forming’ state or ‘boundary’ state by flipping an internal bi-stable switch analogous to those that exist in nature (Verdugo et al., 2015). This way, the input signals can induce the right hexagonal pattern we need for scales, and when pattern decay occurs, we simply re-apply the pattern (e.g. an array of light) to reprogram cells that are at the wrong position. This mechanism, in a more advanced form, is actually present in many natural biological systems as well, as we will go into later.
Figure 5: the simplest possible Gene Regulatory Network (GRN) or Chemical Reaction Network (CRN) bi-stable switch. This switch forms a memory module for a cell. A and B are (macro)molecules that promote themselves, but repress each other. As such, when A is present, B is repressed and vice versa. Like a see-saw, this system is bi-stable, meaning it can switch back and forth by nudging the concentrations of A and B using the inputs. The concentrations of A and B can subsequently be read out by cellular circuits, such as by having A and B act as transcription factors, which forms the output of the system. For example, if A is present, this could mean that the cell is a scale-forming cell and must produce the thick layers of scales.
One way in which this can be done is depicted in figure 5. This system uses two inputs to switch a memory module called a bi-stable switch into a certain position. By applying input B, we can flip the cell into state ‘B’. By correctly programming the cell, this circuit could continuously listen to inputs to flip states, thereby allowing us to reprogram each cell that is mispatterned as often as necessary.
The ‘paint’, in this case, is the input signal of the circuit. It can be anything we choose, ranging from small diffusible morphogens to light waves that influence optogenetic receptors. We lean towards the use of optogenetic receptors due to their ease of programming and the ability to avoid using small molecules that themselves would require safety testing. All that being said, regardless of the ‘paint’ being used, we do need to insert this switching system into skin stem cells before they can switch based on morphogens or light waves. There are several ways to accomplish this using gene therapy, such as using a viral vector, virus-like nanoparticles, etc.
Constantly listening means constantly changing
One caveat that needs to be addressed is that if the circuit is constantly listening for input signals, especially if the signal is common (e.g. ambient light), the circuit might repattern scales in uncontrolled ways. As such, we need some mechanism to transiently enable this ‘listening’ behaviour, and then after clinical re-patterning is complete, remove the ability to change patterns again. We can do this by transiently expressing input receptors, such as through inducible expression systems activated by a small molecule (e.g. tetracycline, although many systems exist to select from for our range of use cases) that is administered before the re-patterning treatment begins. As an example, although not one we have committed to yet, we note that tetracycline is well-known in inducible expression systems and is used clinically as an antibiotic, where a one-time low dose for induction of expression is likely to be very well-tolerated.
Since any small molecule used for induction of expression would have a fairly short half-life, it will have cleared out of the system by the time (re-)patterning is complete, and the stem cell state becomes ‘fixed’. Now the patient will be able to expose their skin to many environments without risking accidental re-patterning.
The full circuit would therefore have to look something like what is presented in figure 6:
Figure 6: The proposed gene regulatory circuit that listens to and remembers external patterning information from the optogenetic or chemical ‘paintbrush’.
The full paintbrush system
The full system is summarised in figure 7. In step 1, the novel gene circuit is embedded into the epithelial stem cells of the skin using gene therapy. In step 2, the transient listening circuit is induced by a small-molecule inducer, enabling the cells to switch states. In step 3 & 4, the scale pattern is applied to the skin, causing the cells to switch states appropriately. In step 5, the stem cells have switched to be either a scale-forming or boundary stem cell, and produce the appropriate cell layers comprising scales.
The downstream effects of scale development requires a deep examination of scale-forming pathways in either reptiles or birds, which is an ongoing research effort.
Figure 7: two-state pattern induction
1.2.2 The Turing pattern method: likely harder to engineer, but much more robust
As a parallel method in development, we are attempting to engineer an artificial or semi-artificial Turing pattern which resembles the paintbrush method as described above. However, instead of requiring repeated application of an exogenous pattern for maintenance, it would be self-stabilising, as occurs in nature. The mechanism of scale morphogenesis by Turing patterning is not fully understood, so an artificial pattern could either make use of a natural one if it is discovered, or else be engineered orthogonally on a de novo basis. Currently, a small amount is known about natural patterning, in that it may share components with the first stage of feather patterning (Li et al., 2017).
Such an engineered mechanism would essentially have two steps. Step 1 inserts a genetic control circuit into the relevant stem cell population, which will be dormant until provoked by a manually administered global signal that starts the patterning process. Step 2 then involves cells communicating in a complex network to generate an alternating pattern that induces scale development.
Simple Turing patterns have recently been engineered in both prokaryotic and eukaryotic cell colonies (Duran-Nebreda et al., 2020, Sekine et al., 2018). Even so, bioengineering of controlled, robust or complex Turing patterns in mammalian cell populations remains a serious challenge. One of the major issues is orthogonality, where it is important that the patterning network components do not have off-target effects on other pathways in the cell. Generating orthogonal pathways can take two forms; either it can involve rational protein engineering of custom receptor-ligand pairs, or it can use non-protein methods such as nucleic acids, which we view as more tunable based on unpublished proprietary designs. Our future work will involve assessing methods for construction of patterning mechanisms that are tuneable, robust, and effective in vivo for generating structures that are otherwise unnatural in humans, but this is necessarily a longer term goal which we feel will likely come after simpler methods are established as described previously.
Figure 8: The proposed system of scale maintenance in vivo. Scale boundaries are maintained through long-range inhibition and short-range activation. In nature, this leads to so-called ‘Turing patterns’, which is the process in which hair follicles, scales, and feather follicles are distributed on the skin during embryogenesis. In this system, we would mimic this process, albeit at a larger length scale to maintain the patterns.
1.2.3 The real system may use a combination of both methods
In reality, we will likely use a combination of both methods. For example, Turing patterns are relatively robust, and will tend to produce patterns with the right size and properties. However, without a good nudge, the formed pattern may not look exactly the way we want it to look. We can supply the Turing mechanism with an initial pattern using the paintbrush method to help it along, which in mathematical terms is known as an ‘initial condition’. This has the advantage of giving us a lot of control over the final position and size of scales, while also ensuring that the pattern remains stable over long periods of time.
1.2.4 The genetic network complexity is limited by gene vector size
Regardless of what we design, we are limited in the complexity of artificial gene networks that we can package into clinically viable viral vectors. This is because for most vectors used in gene therapy, 10 kb (kilobases) is the maximum size, with a hard limit of about 12 kb, including regulatory sequence elements. In practice, about 8 kb can be used for transgenes, and we would aim to include the necessary genetic modification in as few vectors as possible due to concerns about transduction efficiency. To achieve this, we may, for example, package multiple genes in the same multicistronic vector, separated by internal ribosome entry sites (IRES).
Very small proteins take as little as 0.6 kb, but some are large – for example, a receptor for Shh, called PTCH1 (Patched) in humans, is 4.3 kb. An average gene is around 1.5 kb, so on average, we could expect to include 5-6 genes, making for a very minimalistic network.
There are more exotic methods for gene transduction that can insert larger sequences, such as hydrodynamic gene therapy (HGD), sonoporation, lipid nanoparticles (LNPs), etc., but they are not currently in clinical use, so research and development would take longer. It is therefore important for us to create a barebones gene network for our purposes that can work with currently clinically usable minimalist vectors. We will keep the option of larger gene networks with the aforementioned future methods open as a possibility for use at a later date.
1.3 Methods of morphogenesis
1.3.1 Top-down approach: 3D bioprinting
The field of bioprinting and engineered tissues has introduced many new possibilities in the way we approach both development in vitro as well as integration in vivo of artificial integument (Weng et al, 2021, Kolesky et al, 2014). Artificial constructs may be constituted by direct application of suspended cells in the form of “bioinks” and/or fabrication of biocompatible scaffolds to be infused with cells in a bioreactor afterwards. Importantly, this approach sidesteps patterning by instead manually constructing the desired morphology.
Bioprinting and biofabrication serve two important purposes in our future research. The first would be to generate scaled skin for use in grafting, although applying grafts to wide areas of skin is very invasive and thus not ideal. More interestingly, the second purpose is to generate artificial skin for testing of the paintbrush method in vitro long before patient use.
1.3.2 Bottom-up approach: Cellular Growth & Asymmetric Cell Division
When scales grow, they grow outwards, in the so-called ‘apical’ direction. The amount of growth is probably modulated using an underlying pattern which controls the proliferation rate, although we are not aware of a known patterning mechanism for this yet. More complex patterns that are rationally designed may be able to introduce more complex scale shapes. However, for overlapping scales, which many forms require, just growing upwards is not enough to cause overlap, no matter how complex the growth pattern. As such, for overlapping scales, the growth direction must also be taken into account. There are two ways in which this directionality might be induced in nature. A generally accepted common method, present in many species, requires interfacing with the polarity axes of the epidermal stem cells. A second method might involve asymmetric patterning, causing the scales to fold and overlap like origami. These two mechanisms might even be intrinsically linked.
In either case, some way to control the direction of growth, whether perpendicular or parallel to the skin, or at some controlled angle, will be necessary to induce overlap in scales through simple growth or complex folding. As such, we will briefly examine the polarity axes within skin cells.
Figure 9: Overlapping scales with simple patterns.
1.3.3 Cell polarity axes in skin
There are two polarity axes of relevance present in these cells: the apical-basal (up-down) polarity axis and the planar (side-to-side) polarity axis. Normally, skin growth is aligned with the apical-basal polarity axis when the epithelial stem cells differentiate. During stem cell maintenance, cell division is aligned along the planar polarity axis (Muroyama & Lechler, 2013). These two processes must remain in balance for the skin to properly maintain itself.
It is possible that scale growth may be oriented at an angle in scales such that it does not grow perpendicular to the skin. Alternatively, there may be a balance between apical-basal and planar growth that yields a net angular direction of growth. The orientation of polarity, whether it be along the scale’s normals or aligned with an overall angled axis, should be assayed in reptilian skin by immunohistochemistry.
In hair follicles, the growth direction is dependent on the basal polarity axis (Chen & Chuong, 2013). However, simply defining the angle with respect to the vector pointing out of the skin is not enough, because the orientation that the hair/scale/feather grows in – the direction of the ‘grain’ of the hair, i.e. the direction in which you should pet a cat – must also be known. That is likely defined by the planar cell polarity of the actively growing cells, i.e. those differentiating from the epidermal cell pool near the base.
For overlapping scale development (snake scales, fish scales, etc), it is important that the growth direction is modulated with respect to the planar cell polarity axis as well as the apical-basal cell polarity axis. Cell division is oriented along the polarity axis by aligning the mitotic spindle with that axis, and this, in turn, is controlled by forces applied to the spindle through microtubules and accessory plasma membrane proteins. It thus becomes important to control orientation of polarity, as this is directly upstream of controlling cell growth direction.
1.4 Section Conclusion
Together, these mechanisms provide a clear research direction for creating scales from scratch. Although these theoretical systems are expected to work, our remaining challenge will be to identify physical targets that abide by the design’s parameters, and this forms the content of some of our current work.