Intercellular communication between artificial cells by allosteric amplification of a molecular signal

Multicellular organisms rely on intercellular communication to coordinate the behaviour of individual cells, which enables their differentiation and hierarchical organization. Various cell mimics have been developed to establish fundamental engineering principles for the construction of artificial cells displaying cell-like organization, behaviour and complexity. However, collective phenomena, although of great importance for a better understanding of life-like behaviour, are underexplored. Here, we construct collectives of giant vesicles that can communicate with each other through diffusing chemical signals that are recognized and processed by synthetic enzymatic cascades. Similar to biological cells, the Receiver vesicles can transduce a weak signal originating from Sender vesicles into a strong response by virtue of a signal amplification step, which facilitates the propagation of signals over long distances within the artificial cell consortia. This design advances the development of interconnected artificial cells that can exchange metabolic and positional information to coordinate their higher-order organization.


Bulk test of Response Cascade for Figure 2h,i.
The function of the Response Cascade was tested in bulk reactions, using 1.0 mM Na2HPO4. The experiment was performed in duplicate and the error bars represent the standard deviation.

Direct comparison of the activation of Receiver GUVs at different distances from an AMP or Na2HPO4 point source
These graphs display the distance-dependent activation of Receivers when a, 1 mM AMP was added as a point source, with 1 mM Na2HPO4 homogeneously spread throughout the sample, or b, 1 mM Na2HPO4 was added as a point source, with 1 mM AMP homogeneously spread throughout the sample. The point sources were located at position 1; the other positions are progressively further removed from the point source, with equal spacing between positions. The data were extracted from the images in Suppl. Fig. 7 and related images using Fiji's particle analysis plug-in. Figure 3d displays the same data, yet grouped differently to directly compare the allosteric activation and substrate activation at four selected positions. Error bars depict the SD from analyzing 40-200 GUVs per position.

Washing of GUVs removes unencapsulated proteins
Washing apyrase-loaded GUVs according to the centrifugation protocol described in the Methods section removed any unencapsulated components from the external solution. Apy = apyrase solution, OP = fresh external solution (=outer phase), SN = supernatant, M = protein marker, the molecular weights of which are indicated. Before the first washing step apyrase was present in the supernatant (SN 1); after the first washing step it was still present at low concentrations (SN 2); the second washing step removed all remaining apyrase that was not encapsulated in the GUVs. The purified GUVs were used for subsequent experiments. The reported molecular weight of apyrase is 49 kDa. 1

Activation of Receivers by Senders; individual Receiver GUVs
The upper panel displays the individual traces for the graph in

αHL insertion into GUV membranes
αHL self-inserted into GUV membranes and permeabilized most GUVs. The kinetics of protein insertion and the release of low-molecular weight contents were heterogeneous amongst the GUV population.
Top row: brightfield micrographs of GUVs; left) before αHL addition, right) after αHL insertion. Before αHL insertion, the sucrose/sorbitol gradient created a difference in refractive index between the lumen of the GUVs and the external solution. This was lost upon equilibration of the sucrose/sorbitol gradient through αHL pores. The GUV membrane composition was 35/35/30 (mol/mol) DOPC/POPC/cholesterol + 1.0% DSPE-PEG2000biotin.

αHL insertion into GUV membranes (2)
Calcein-release profiles for individual GUVs. Each trace represents an individual, calcein-filled GUV after addition of αHL. GUVs were randomly selected from Supplementary Figure

NADH calibration curve
Calibration curve for the bulk reactions. All NADH concentrations were measured in the appropriate buffer and measurements were performed in triplicate. Error bars represent the SD. Absorbance @ 340 nm (a.u.) [NADH] (μM) Apyrase was used to generate the AMP since it has a high turnover and, therefore, can produce AMP efficiently. It hydrolyzes the γ-phosphate of ATP, and the ß-phosphate of ADP to generate AMP and two equivalents of orthophosphate. The enzyme from Solanum tuberosum has two isozymes: one with an ATPase/ADPase ratio of ~10/1 and one with similar ATPase and ADPase activities. Since the ADP is a non-active intermediate whose build-up would be undesirable, we opted for the isozyme with an ATPase/ADPase ratio of ~1. It requires Ca 2+ for optimal function, which is why this was included in the solutions. 1,2 Since ATP is an important metabolite in the energy metabolism of cells and known to inhibit GPb as part of a negative feedback loop, we investigated its influence on the Response Cascade. Low millimolar concentrations of ATP did not inhibit the activation, but at >5 mM ATP inhibition was observed. Therefore, ATP concentrations below 5 mM were used throughout this study.

Response Cascade
The Response Cascade was built using three important enzymes from the glycogen metabolism and the pentose phosphate pathway. Glycogen phosphorylase was selected because its unphosphorylated form, glycogen phosphorylase b, can be allosterically activated by AMP, and would as such provide an inducible enzymatic reaction that leads to high levels of signal amplification. It hydrolyzes a terminal α-1,4-glycosidic bond using orthophosphate and produces glucose-1-phosphate and a shorter glycogen chain. The phosphorylated isoform of GP, called GPa, is constitutively active and insensitive to AMP. The presence of low levels of GPa in the commercial GPb preparation accounts for the weak NADH output that can be observed in the absence of AMP. Activation of GPb by addition of AMP, however, induces a far stronger NADH output, clearly providing a strong on/off modus operandi for the Response Cascade.
To generate a physiologically relevant output from the Response Cascade, we selected PGM and G6PDH to convert the product of GPb, glucose-1-phosphate (G1P), into NADH. Additionally, this provided a convenient read-out for the activation of GPb, because NADH is fluorescent with an excitation maximum at 340 nm and an emission maximum around 460 nm. PGM needs divalent cations for its activity (especially Mg 2+ ) as well as α-D-glucose-1,6-bisphosphate (as a phosphate donor for the serine residue in the active site). The equilibrium between G1P and G6P that is catalyzed by PGM has a ΔG of -7.5 kJ mol -1 , 3 therefore favoring the conversion into G6P.
The concentrations of all three enzymes were optimized as shown in Suppl. Figure 1; furthermore, the concentrations of the other components were optimized as well. Subsequently, the kinetics and amplitude of the Response Cascade could be tuned as outlined in the main text. Due to the presence of a small amount of constitutively active GPa in the commercial GPb preparation, some background activity in the absence of AMP is always observed. Therefore, a trade-off has to be made: if a fast reaction (to get an NADH burst) is required, high concentrations of GPb (3.0 U ml -1 vs. 0.5 U ml -1 ) and Na2HPO4 are beneficial. If, however, the background reaction should be suppressed (for instance because of the time needed for AMP to diffuse to the Receiver GUVs), low concentrations of GPb and Na2HPO4 are preferable (Suppl. Figure 3).

Supplementary Note 2. Activating the Response Cascade in GUVs
To generate the Receivers, 200 mM sorbitol was used in the outer phase solution instead of the widely used glucose, as glucose was shown to have an inhibiting effect on the Response Cascade at > 50 mM; most likely this is caused by a negative product feedback loop in the glucose metabolism.
Since AMP is anionic, it cannot readily permeate phospholipid membranes. Therefore, α-hemolysin was added to the GUVs to allow the influx and efflux of small metabolites like AMP through the 1.4 nm pores. 4 However, since only macromolecules like enzymes and glycogen could be retained inside the GUVs, all other metabolites and ions, like NAD + and CaCl2, were added to the external solution as well to prevent drainage of the Response Cascade from the Receivers.

Supplementary Note 3. Artificial cell heterogeneity
As discussed in the main text, throughout our experiments we observed a marked heterogeneity in the response of the Receivers to AMP (Fig. 3a, 4a, 6d). Although the population of Receivers as a whole successfully detected the presence of AMP, interesting differences in lag time, kinetics and yield of NADH production were observed. The many variables involved in the activation of the Receivers confound the establishment of a clear causal relationship between input parameters -such as GUV size, permeability and loading -and the response of individual Receivers. A clear correlation with the size of the GUVs was not observed (Supplementary Figures 6, 12). Moreover, the size of the GUVs and the enzyme concentrations used here implicate that each GUV contained 10 6 -10 7 enzymes (see table  below). Therefore, stochastic fluctuations in solute partitioning during GUV formation are not considered to be likely causes of the observed variability in the response of the artificial cells. Other non-stochastic effects, however, may play an important role in solute distribution. 5 Average GUV diameter = 50 μm Average GUV volume = 65 pL Most likely, differential permeability of the GUVs has a more significant influence on the activation kinetics and NADH output of individual GUVs. As αHL inserts into the phospholipid bilayers spontaneously, controlling the degree of insertion per vesicle is difficult, which causes divergent inand efflux rates of small polar molecules (Supplementary Figure 14). Moreover, several studies have pointed out the existence of subpopulations of GUVs with significantly more permeable membranes -even in the absence of membrane pores. 6,7 Finally, although GUVs produced by the droplet transfer technique are generally unilamellar, a fraction of vesicles can be expected to be multilamellar. This would also have strong influence on the permeability of an individual GUV.
A more detailed analysis of the different factors that govern the heterogeneity of the response to AMP would establish the important functional consequences of artificial cell heterogeneity, yet requires a full quantification of all relevant concentrations, permeability coefficients, etc. for individual vesicles. Such further quantitative studies -although outside the scope of the current study -will likely provide important insights for the implementation of cell-like communication networks.

Supplementary Methods
Concentrations of key reagents for each experiment.