Modeling planarian regeneration: a primer for reverse-engineering the worm
A central component of most planarian regeneration studies is the question of how cell type and tissue identity are specified in each location [105]. Although the ultimate goal is to understand how every planarian tissue and cell type is regenerated and maintained, here we concentrate on the most widely studied tissue types into which neoblasts differentiate [106]. Such studies are aided by the fact that the stem cell population can be selectively killed by irradiation, which prevents planarians from regenerating [30]. This method has been used to show that a single transplanted neoblast can rescue the regenerative capacity of irradiated planarians, as well as induce the production of gonads in asexual hosts [54]. Irradiation experiments have also been used to elucidate molecular identifiers of neoblasts. The piwi family of regulatory genes, known to be essential for maintaining stem cell populations by preventing cell differentiation, are expressed solely in neoblasts and are required for regeneration [107]–[109]. Biological markers that distinguish both early and late neoblast prodigy (descendants that will differentiate into tissue-specific cells) have also been identified [110].
GJC is also important for neoblast survival, as inhibition of the gap junction protein innexin-11 results in a loss of both piwi-positive cells and of the ability to regenerate [66]. Additionally, the phosphatase and tensin homolog (PTEN) protein (a known tumor suppressor) and its respective pathway have been shown to regulate neoblast activity in planarians. Disruption of planarian PTEN signaling results in both the hyperproliferation of neoblasts and the appearance of abnormal outgrowths reminiscent of tumors [111]. In mammals, loss of PTEN activates the target of rapamycin (TOR) pathway, resulting in tumor formation [112]. Pharmacological inhibition of TOR (with rapamycin) in planarians rescues loss of PTEN, preventing abnormal outgrowths and restoring regenerative ability [111]. Understanding the factors that link cell-level regulation of neoblast proliferation and differentiation to system-level needs such as polarity and patterning is a crucial next step for the field.
The planarian nervous system contains a full set of vertebrate neurotransmitter homologs, possessing among others dopaminergic, serotonergic, cholinergic, and GABAergic neurons [113]–[116]. Neural connectivity is mediated in part by the disheveled family of proteins, a Wnt signaling pathway member, most likely Wnt5 signaling [117]. Interestingly, several of the ML polarity regulators are also involved in the regulation of nervous tissue organization and when inhibited result in neural patterning defects in which nerves collapse towards the midline (slit inhibition [102]) or, conversely, displace laterally (Wnt5 inhibition [117]). The fibroblast growth factor (FGF) pathway, important in vertebrate neural formation and patterning, also participates in planarian brain regeneration. Loss of function of the gene nou-darake, a component of the FGF pathway specifically expressed in the head region, leads to the expansion of the brain through the body [118]. Finally, the netrin family of axon guidance proteins is also required in planarians for the regeneration and maintenance of neural patterning; when the netrin receptor is inhibited the overall organization of the nervous system is lost, disrupting the relationship between the brain and VNCs [119].
Graft transplantation experiments have historically been used to probe the functional identity of different regions and tissues in planaria and have provided important information about the types of signaling that occur between different tissues (Figure 2B). For example, when a small piece is cut out, flipped along the DV axis, and grafted back into its original location in the worm, a cup-shaped projection is formed on the boundary between the host and the graft; for anterior grafts this projection will develop a head-like morphology, while posterior projections will appear tail-like [21]. This suggests that the juxtaposition of dorsal and ventral tissues is a cue that signals the regenerative process. In contrast, if two dorsal halves are grafted together, no regeneration occurs [22]. Such results have driven many models in vertebrate systems in which growth is dictated by juxtaposition of regions with distinct “positional information values” [120]. Similarly, if the pharyngeal region is removed and the remaining head and tail fragments are then joined together, a new pharynx regenerates between them [121]; however, when two fragments from the same region of the worm are grafted together, no regeneration is observed [21]. Finally, if a small fragment of the head is grafted into the tail region, a new head and pharynx will appear with reversed polarity from the original AP axis; conversely, if a tail piece is grafted anterior to the pharynx, a second pharynx will form between the head and the transplanted tissue, again with reversed AP polarity [60], [61]. These results clearly indicate that, instead of adopting the identity of the host tissue, a grafted piece can signal the host to re-pattern and change tissue identities.
During everyday life, planarian cells are continuously replaced by the differentiating progeny of the neoblast population in a process known as morphostasis [4]. The mechanism is not well understood, although it is known that mitogen-activated protein kinase (MAPK) signaling is required for neoblasts to stop proliferation and undergo differentiation [78]. This homeostatic process of cell turnover, which constantly renews all cells without changing tissue size or proportion [55], [122], is common among organisms. However, planaria exhibit a unique remodeling ability that enables them to dynamically scale their body in relation to the available cell number. Upon starvation, planaria use their own cells for energy, reducing their body size in a phenomenon known as degrowth [3]; this is reversible, and growth resumes upon feeding (Figure 2C). The worm's total cell number changes linearly with respect to body length and time, while keeping the proportion of all cell types constant. This requires active remodeling of existing tissues to scale the structure in a process known as morphallaxis and is quantitatively described by an allometric equation [52], [123]. The mechanisms regulating this exquisite rearrangement are not well understood, but it is accomplished in part through coordinated cell death and neoblast proliferation [58], [124].
This brief overview of some of the questions that the planarian regeneration field has investigated, both historically and currently, highlights the increasing wealth of information that continues to be generated and must be integrated into any comprehensive model of planarian regeneration. Even within a narrow area (e.g., AP polarity) the field lacks a truly cohesive model that explains the system-level (patterning) data in terms of the behavior of lower-level (cell pathway) components.
Most modern planarian regeneration studies result in a gene regulatory network or protein interaction pathway. However, these networks largely do not constrain shape or shape regulation, and such pathways are compatible with a very broad range of morphologies. A few researchers have proposed algorithmic models, which precisely define the steps that cells (or tissues) must take to assemble or repair a given morphology. In specifying the information needed to make decisions, such models could be fleshed out in molecular terms to provide testable hypotheses about the mechanisms underlying information exchange, computation, and links to the ultimate execution of morphogenesis via cellular effectors. Due to space constraints, the basic features (assumptions and outputs) are given in Supplemental Text S1.
Many algorithmic models propose the existence of a coordinate system that facilitates the patterning of the organism. The positional information model [125] is based on diffusible substances that create concentration gradients; depending on the specific concentration of such substances at different locations, different cell programs are triggered [126]. The serial threshold model [20] combines positional information and cell migration to explain how planarians can restore AP polarity. Reaction-diffusion models [127] are based on diffusible substances that react with each other; they can generate most of the patterns found in biology [81] and explain regeneration of polarity in planaria [128]. Other algorithmic models proposed for planarian regeneration are based on bioelectrical signals [23], dorsal-ventral interactions [21], and the intercalation of missing regions [129].
However, none of the proposed models in the literature are able to explain more than one or two facets of planarian regeneration. There is a tremendous opportunity to integrate existing knowledge of pathways with the anatomical and physiological data to produce novel models of the worm as a system that senses damage [130], carries out the distributed processing needed to restore a target state, and then ceases growth. The major outstanding questions that a complete algorithmic model should be able to explain are summarized in Box 1, as a challenge to modelers from the computer programming, engineering, physics, mathematics, complex systems science, and artificial life communities.
Detection of missing tissues: What is the signaling mechanism that triggers the regeneration of only the exact missing regions, structures, and organs?
Growing boundaries delimitation: What signals trigger growth and cause remodeling to cease when regeneration is completed?
Plasticity of organism size: How are planarians able to remodel existing tissues to maintain the correct internal organ proportions as size changes due to available cell number?
Neoblast migration: What are the driving forces that permit neoblasts to migrate in the direction of the wound to initiate blastema formation and how are positional values encoded (and to what spatial resolution)?
Specification of cell types: How are different cell types generated from the same neoblast population, and how are anatomical regions of different size (head vs. eye vs. photoreceptor cell) specified spatially?
Organization of cells into organs: Once cells are committed to a given cell type, what is the process that orchestrates their development into specific organs?
Small vs. large fragments paradox: Experimentally suppressing essential mechanisms (such as anterior-posterior positional gradients) in large fragments (for instance middle-third trunk fragments), disturbs normal regenerative patterning; so why are small pieces cut from a worm often still able to regenerate normally, despite such loss-of-function treatments, when they have suffered relatively more damage?
Specification of target morphology: What is the mechanism (whether directly encoded or an emergent property of the remaining tissue) that specifies target morphology during regeneration?
A key question concerns the possible existence of a direct encoding of “target morphology”. It is commonly held that shape is an emergent property of cell interactions. However, recent data suggest that at least basic AP polarity in worms may be directly stored. A transient modulation of physiological events in a worm results in a bipolar 2-headed outcome; remarkably, although this change did not affect the DNA sequence, the patterning change persists upon multiple subsequent amputations in the absence of any other perturbation. The shape to which the animal regenerates upon damage (the target morphology) has been stably altered, suggesting the possibility that the large-scale axial anatomic plan may be encoded in physiological networks and thus directly modifiable by non-genetic experimental interventions. The concept is highlighted in the following hypothetical experiment, illustrated in Figure 4. Take one planarian from each of two species with clearly different morphologies: S. mediterranea with a rounded head, and P. felina with a hammerhead. In this experiment, the neoblasts are killed off (by irradiation) in half of the S. mediterranea worm. Subsequently, live neoblasts from the P. felina worm are transplanted into the irradiated worm. If this chimeric worm is now cut, forcing it to regenerate its head, which head shape will be regenerated (round or hammer, or perhaps a hybrid, or perhaps regeneration will never cease as each set of neoblasts works to remodel the head)? Without a theory of how the organism's final morphology determines stem cell-driven processes during planarian regeneration, it is impossible to predict the outcome of this experiment. Truly successful algorithmic models, that can be used predicatively to forward the regeneration field, should be able to answer these types of questions.
In addition to testable in silico models [131]–[134] (or indeed, hardware implementations based on swarm intelligence models [35]), there are several other areas in which information sciences can make a transformative impact on regeneration research. We suggest the urgent need for the development of a bioinformatics beyond sequence and regulatory networks—a bioinformatics of shape, including:
A formalization of patterning outcomes in model systems. We currently have no standard formal language in which outcomes such as 1-headed vs. 2-headed worm can be encoded for informatics approaches, and precise quantitative morphometrics need to be augmented by symbolic representations that focus on large-scale patterning changes.
Creation of databases where patterning phenotypes (and the associated manipulations that produced them) can be stored, queried, and mined. The field is currently limited to searching abstracts for keywords, and a new investigator in this field cannot easily do a search to find out if any existing experiment produced, for example, a worm with three eyes in a row.
A standardized formal modeling environment and representation format. Models arising from functional studies could be reported in the primary literature in a formal way that facilitated both their comprehension and implementation, to determine whether they indeed match the patterning behavior that was supposed to be explained by the genetic pathway being elucidated.
Artificial intelligence tools to augment human ingenuity in proposing testable models. We are currently inundated with functional data to the point that it is becoming increasingly difficult to come up with models that are consistent with available results. As this problem grows, human reasoning needs to be assisted by computational tools that can infer testable, algorithmic models from databases in which molecular-genetic perturbations are linked to their morphological outcomes.
Using the planarian regeneration data as a proof-of-principle test case, our lab is pursuing these directions. We hope the above primer on the flatworm system will motivate others with complementary expertise to join the molecular regeneration community in attempting to solve the puzzle of autonomous large-scale pattern formation and repair. Paradigms such as cellular automata [135]–[137], formal grammars [138]–[142], formal rules [143], [144], neural networks [145], and Boolean circuits [146], [147] are just some of the tools that remain to be brought to bear on this crucial problem at the center of developmental biology.
The planarian model organism represents an excellent opportunity for the application of the proposed model-building, formalization, and computational methods for automated model discovery. This approach can be also extended with little effort to other regenerative model organisms, such as hydra, axolotl, Xenopus, newt, zebrafish, and mammalian organs (e.g., liver, deer antlers, etc.). The incorporation of strategies from other fields, and the integration of truly interdisciplinary approaches with the now-accepted molecular genetics-bioinformatics-computational biology efforts will greatly facilitate fundamental insight into the general questions of how complex systems (organized on many scales of size) ascertain their shape, alter it to match functional needs, and recover from injury. With the ability to truly understand the generation and adaptive self-regulation of complex morphology will come exciting advances in evolutionary developmental biology, regenerative medicine, and synthetic biology. Moreover, the payoffs will extend far beyond biology, contributing significantly to cybernetics, computer science, dynamical control theory, robotics, and many areas of engineering that can benefit from understanding how living systems actually perform the remarkable tricks developed by millions of years of evolution.