Appendix A. Autopoietic SCs Based on Gene Expression?
The production of proteins inside SCs can be achieved by employing the “PURE system” [194
], a reconstituted kit for TX-TL reactions with a known and minimal
36 proteins (20 amino acyl-tRNA synthetase, 10 translation factors, 4 energy-related enzymes, T7 RNA polymerase, methionine trans-formilase)
E. coli ribosomes (composed by 3 rRNAs and 55 ribosomal proteins)
tRNA mix from E. coli
(46 according to [196
small molecular weight (MW) compounds (amino acids, nucleotides, etc).
When compared to cell extracts, the PURE system leads to lower yields (as mentioned, ca. one third [74
]), but its major advantage refers to a Gedankenexperiment
with relevance for the construction of autopoietic or quasi-autopoietic SCs.
It is possible to imagine SCs that contain the PURE system and an artificial genome encoding for all macromolecular components of the PURE systems. This genome would be constituted by about 140 genes: a figure that is about 70% of the minimal genome, estimated as 204 genes [76
]). Note that the construction of such an artificial genome is not unrealistic after the well-known experiments of the Craig Venter team in 2010–2016 [197
Next, under the very permissive hypothesis that gene expression regulation and post-transcription/post-translation modifications are not strictly required, the result of the PURE system TX-TL reactions would be the entire set of components of the PURE system itself, provided that enough small MW components are available and by-products removed (not to poison the reactions). The resulting hypothetical SCs would still require macromolecular components to replicate DNA, and to produce lipids, but—at least in principle—should be able to produce all PURE system components. This minimalistic scenario should give, then a quasi- autopoietic SC where all internal components, but DNA and lipids, are self-produced. The entire mechanism would be made possible by a continuous supply of small molecules from the environment, and a release of waste chemicals, for example by small pore in the membrane (e.g., the α-haemolysin pore). Such a pore, however, would not allow the release of important information-carrying and function-carrying macromolecules in the environment [4
One of the several potential hurdles is the synthesis and assembly of functional ribosomes in situ, but encouraging results have been reported recently [199
]. The continuous production of internal components, without compensative degradation pathways (as requested by autopoiesis) excludes the possibility of a homeostatic state, favouring a growth phase. However, the resulting system could not undergo growth-and-division, because lipids are not synthesized. There are experimental indications that the machinery for lipid synthesis can be functional in vitro [28
], and thus such a “module” can be added to the hypothetical system shown above. In proper conditions, which till now have been demonstrated only for fatty acid vesicles [107
], and not for pure phospholipid vesicles, the spontaneous division of growing SCs could be obtained. Finally, DNA replication would be required. As mentioned, a recent report also indicates that it is possible to rely on machinery synthesized in situ for DNA replication (ca. 20 kbp) [182
The participation of the environment as a source of small MW nutrients and as a sink for byproducts recalls an important consideration, which in the autopoietic theory plays an important part. Living systems, as we know them, live in a specific environment. Their internal mechanisms not only depend from the available “building blocks” found in the environment (and from the energy extracted from the environment), but are adaptively and evolutively coupled
with environment composition, transformations, and fluctuations (in autopoietic terms: engaged in a “structural coupling”). The organism/environment duality, indeed, is just a result of our operation of “distinction”, which literally creates in our understanding—the observers’ understanding—the organism and the environment as two entities. However, they are not. The functional coupling of an organism with its environment and environmental fluctuations is inseparable, and is at the basis of biological—i.e., autopoietic—cognition. Such an additional aspect becomes fundamental when aiming at constructing artificial living
Appendix B. SC Complexity: A Still Unexplored Topic
Defining and measuring complexity in physical, chemical, and biological systems is a challenging task. As mentioned, there is no universal answer, but different definitions and measurements are best suitable for certain systems rather than to others [184
On the other hand, the exercise of focusing on SC complexity, even if not leading to a definitive answer, can be useful to better defining, considering, evaluating what is behind the enterprise of constructing SCs, exploring additional facets that can inspire further investigations.
Here we would like to briefly comment on a couple of possible approaches of SCs complexity, based on structure and organization. The approaches will involve operational strategies where informatics and computer science can be of great help. (Note
: the discussion has only pragmatic scopes: it is not intended as a contribution to the long-standing debate on the cell/computer or brain/computer analogies [210
. Information is needed to describe the SC structure, at different levels. For example, at the molecular level, to describe DNA genes, RNAs, enzymes, small MW molecules, information refers to the number and the types of components, and to their molecular structure (e.g., size, sequence, connectivity between atoms). At the level of reaction networks, the conversion of one chemical species into another can be seen as a set of sets of vertices connected by edges—a graph, whose complexity can be also determined. Here we will deal with the compositional/topological information associated to the SC level. The SC structure can be described operatively as the set of instruction needed to build a static model of it (e.g., by a computer). The amount of information
associated to the construction of the SC, and therefore the complexity associated to it, can be discussed in an analogous way to a proposed method for estimating the phenotypic complexity of organisms [215
]. In particular, the complexity of an object can be measured by the amount of information required to offset the Shannon entropy of the randomly constructed equivalent objects (i.e., those that could have been originated instead of the target one). Accordingly, the information value (in bits) results to be log2 N
, where N
is the number of different, but equivalent, randomly constructed objects (giving the same probability to all possible structure is equivalent to say that no previous knowledge is available; details can be found in [215
]). Some examples can clarify this approach.
Consider a target SC consisting of one water-soluble molecule A enclosed inside a liposome. The construction of the SC is conceptually made in two steps. A compartment is firstly created, originating a division of the e space in an inner and an outer space. Next, A is located inside the compartment. This corresponds to a dichotomic decision, but a random choice could have lead also to A located outside the compartment. There are N = 2 equivalent randomly generated objects that can be constructed, and thus the amount of information required to describe the building of the target SC is 1 bit. Similarly, the amount of information associated to a SC enclosing a water-soluble molecule A and a membrane-embedded molecule B that can stay in two possible orientations, inward and outward, is 2 bits (N = 22). If there are n molecules A and m molecules B, the number of randomly constructed SC configurations becomes 2n · 2m, and the amount of information correspondingly increases to n + m. Due to the its logarithmic nature, information is additive and the several contributions can be calculated separately as far as they refer to independent events.
As a second example, consider a multi-compartment SC (a vesosome) consisting of a large vesicle V containing an inner vesicle v. Again, 1 bit is required to describe the resulting structure. Note that the space is now divided in three sub-spaces (v, V-v, and the external space). Suppose that the target SC requires two molecules, A, and B to function, and that a specific configuration is needed. For example, A should be inside v, and B inside V-v. To compute the amount of information needed to describe/build the target SC the total number of possible combinations is needed. In this case N = 2 · 32 = 18 and the amount of information (log2 18) is 4.17 bits (5 bits).
. Following a similar operational approach, the complexity of SC organization, intended as its dynamical pattern, can be estimated from the complexity of the network of chemical reactions that occur in the SC, or from the complexity of the algorithm that describes how the SC operates from a logical viewpoint. In both cases, the network graph, or the algorithm should be able to generate a computer model of the physical SC. The literature about network complexity is huge, and there are several metrics available [186
]. Similarly, there are several measures of the algorithm complexity. When adapted to SCs, the length of the shortest program that produces the behavior of the target SC can be recognized as a definition analogous to the Kolmogorov–Solomonoff–Chaitin complexity, introduced in the 1960s for strings of symbols [216
]. Cyclomatic complexity is another metrics, based on the number of linearly independent paths in an algorithm [217
Whereas the network approach appears quite general, an important caveat about the algorithm approach refers to the implicit hypothesis that SCs work as “information processing machines” with a finite number of states/pathways. This is actually the case of the currently studied SCs (e.g., logical genetic circuits implanted inside SCs), whose functioning can be indeed simulated by an algorithm: their behavior is Turing computable. On the other hand, because the Turing computability of truly autopoietic systems has been questioned [112
], the algorithmic complexity should be cautiously considered.
reports, as an example, the case of SCs that produce proteins by a gene expression mechanism based on the T7 promoter and thermal activation (at low temperature, gene expression is essentially latent). When looked from the network viewpoint, the complexity of gene expression network is obvious, involving about hundred chemicals as starting chemical species, and several hundreds intermediates. However, when the information processing operation of the SC is considered, the system essentially responds to a temperature increase by activating (a very complicated) reaction. In this case, the algorithm that generates the SC pattern is a simple IF-THEN-ELSE instruction. This type of SC operates by just taking one decision, and the space of the possible different patterns includes just two alternative routes (its cyclomatic complexity is 2). SCs that rely on more complex genetic circuits will correspondingly have higher algorithmic complexity; for example, SCs enclosing genetic circuits functioning as an AND gate, producing a chemical species only if two “activators” are simultaneously present. An example can be the synthetic AND genetic gate reported by Noireaux [218
], which produces a reporter protein only when the σ54
transcription factor and the NtrC regulatory proteins are present. For SCs operating in this, or in similar, manner, algorithmic complexity can be used to rank SCs by their complexity, because the comparison is made within the same given frame of description.
However, some interesting investigations have shown SCs, whose behavior differs from a synthetic step that is an end in itself. Four of them have been visually represented in Figure 8
(a short description is given in the figure caption). The common trait of these SCs, that we intuitively see as more complex, is that their behavior is not realized by a mere complexification of the internal genetic circuit, but thanks to interactions
between different SC parts (or between the SC and the environment), or by a hierarchical
organization. Hierarchy and interactions are at the core of any organized system, and are the signatures of complexity; they give rise, ultimately, to emergent properties. A network analysis is probably able to identify these features of SC organization. It is then left as open question the problem of defining what is complexity in the SC context and how to measure it, which strategy is best suitable and in which context. As mentioned, this proposal is mainly intended as a stimulus for further investigations on the subject.
A final remark should be done about nomenclature. Here, the word “complexity” has been used according to its loosely defined meaning. Probably this is acceptable owing to the fact that the discussion is still at a very preliminary level. Actually, a distinction should be done, narrowly speaking, between a complicated system and a complex system, especially when dealing with definitions and quantifications.