We used isophthalic acid as a monomer that can oligomerize at the expense of the condensing agent EDC17. These oligoanhydrides spontaneously deoligomerize through hydrolysis. We attached the nucleobase thymine to the isophthalic acid to form the thymine-labelled monomer T (Fig. 2a). Monomer T can dimerize to DynT2, which can further react to form other oligomers (DynTn; Fig. 2b and Extended Data Fig. 1). Alternatively, fuel-driven oligomerization can also occur when oligomers react with one another. We designed the six-atom repeat in the backbone of the oligoanhydride with a bond length of 7.14 Å such that it matches the six-atom length of the backbone of the repetition unit of RNA or DNA with a bond length of 7.65 Å (Fig. 2a, Supplementary Table 1 and Supplementary Notes). Our monomer T interacts with (dA)10 DNA templates with a dissociation constant KD of 91.9 ± 12.7 mM per binding site determined via an NMR titration, which is about threefold lower than the KD value we measured for deoxythymidine monophosphate (dTMP; Extended Data Fig. 2a,b)21. It is likely that T binds more strongly due to the additional stacking of the isophthalic acid backbone and adenosine units of the template.

We dissolved 25 mM T in 200 mM MES-buffered water at pH 6 with 10 mM pyridine and added 10 mM EDC. Using HPLC and mass spectrometry, we analysed the evolution of the library composition in the presence and absence of a DNA template comprising ten adenosine units ((dA)10; Fig. 2a and Supplementary Tables 2–4). Without the template, we found that the fuel was consumed entirely after 30 min (Supplementary Fig. 1b), resulting in the transient oligomers DynT2–DynT5 (Fig. 3a and Supplementary Table 3). It is noteworthy that we did not observe oligomers greater than pentamers, even with higher fuel or lower pyridine concentrations. The major component was anhydride DynT2. By contrast, DynT3 had a roughly 19-fold lower maximum concentration at 0.05 mM. The maximum concentrations of DynT4 and DynT5 were even lower at 0.006 mM and 0.004 mM, respectively (Fig. 3b–d and Supplementary Fig. 1a). The maximum concentration decays with the oligomer length for two reasons. First, the product of one reaction is the precursor for the next; for example, DynT2 is needed for oligomerization towards DynT3. Second, the hydrolysis of higher oligomers yields shorter oligomers. As soon as all fuel was depleted, all oligomers had decayed, that is, the library exists only out of equilibrium and at the expense of chemical energy (Fig. 3b–d and Supplementary Fig. 1a,b). After the cycle, we found some unwanted side product N-acylisourea (T*), but its concentration remained below 0.5 mM (Supplementary Fig. 1c and Supplementary Table 4).

a, HPLC chromatogram of the library with and without 8 mM template (dA)10 after 38 min in the reaction cycle. The yellow star denotes DynT3*, the unwanted N-acylisourea side product of DynT3. mAU, milli-absorbance units. b–d, The concentration (denoted by the square brackets) of oligomers DynT3 (b), DynT4 (c) and DynT5 (d) as a function of time with and without 8 mM template (dA)10. A kinetic model was used to fit the data. e, The kinetic model calculated the average oligomer length for the template and solution fractions in the presence of template (dA)10 (8 mM). f,g, Schematic representation of template-based selection for oligomerization (f) and deoligomerization (g). h, The gel of (dT)10 and DynT oligomers with (dA)35 in native gel electrophoresis. Dots on the gels are dust inclusions. The weaker but slow migrating band most probably originates from the self-association of (dA)35 at lower pH, high salt and high DNA concentrations30. Pyr, pyridine. In a–d,f–h, the pictograms for thymine oligomers are coloured red and the DNA template is coloured green. i–l, Selection factor S31 of the oligomers as a function of the concentration of the added template (dA)10 (i), MgCl2 concentration in the presence of template (dA)10 (j), template length for 2 mM (dA)n (n = 4, 7, 10, 35) and polyadenine RNA pA (k) and template type for 2 mM (dX)10 (X = A, T, C, G) (l). Dotted lines are used to guide the eye. The labels C2, TC, C3 and the others denote DynC2, DynTC, DynC3 and so on, respectively. aRNA was used instead of DNA.

Source data

We devised a kinetic model to describe the time-dependent concentrations of all reactants in response to the fuel (see Methods and Extended Data Fig. 1). The model considers the fuel-driven activation and spontaneous hydrolysis of oligomers. Through least-squares fitting of the rate constants, the kinetic model predicted the experimental data of the evolution of fuel and oligomers well. Noteworthily, we observed that the oligomerization and deoligomerization were dependent on the oligomer length (Fig. 3b–d, Supplementary Fig. 1a,e–o and Supplementary Table 5).

We tested the response of the chemically fuelled library, now in the presence of a template. With the (dA)10 template, DynT2 was also the predominant oligomer early in the cycle (Fig. 3a). However, when all of the fuel had been consumed, the maximum concentrations of longer oligomers, for example, DynT3, DynT4 and DynT5, were higher by fivefold, 13-fold and sevenfold, respectively, compared to without the template (Fig. 3b–d and Supplementary Fig. 1b). Moreover, we found that DynT2–DynT5 are present for between one and two hours longer than in the experiments without the template. From these observations, we conclude that the template strongly influences oligomerization and deoligomerization (Fig. 3f,g, respectively). We hypothesize that the feedback mechanism of the template on oligomerization involves the pre-organization of oligomers on the template akin to the template-based ligation of DNA or RNA22,23,24,25,26,27. We also hypothesize that the template protects the oligomers from deoligomerization (vide infra). Interestingly, similar template-based protection has been reported for RNA, that is, double-stranded RNA is more stable than single-stranded RNA28, and has previously been explored as a mechanism to promote sequence copying29.

Mass spectrometry showed evidence of the complexes made from (dA)10 with monomer T and oligomers DynT2–DynT5 (Supplementary Table 6). Moreover, native gel electrophoresis showed a broadening and shifting towards higher molecular weights of the template band in the presence of DynT oligomers (Fig. 3h and Supplementary Table 7). For pure (dA)35 and mixtures that lack (dT)10, we observed a weaker but slow migrating band, most probably originating from the self-association of (dA)35 at lower pH, high salt and high DNA concentrations30. It is noteworthy that the bands of oligomer/template complexes do not shift under denaturing conditions, corroborating that our observations originate from hybridization (Extended Data Fig. 3a).

We extended our kinetic model to include the role of templation by adding reactions that can occur on the template. The model predicts the concentration of each library member in solution and on the template, using the dissociation constant KD values between the library members and the template (see Supplementary Notes, Fig. 3b–d and Supplementary Fig. 1a–d). For the monomer, we used the experimentally determined value of the dissociation constant. We assumed that the binding constant decreases with increasing oligomer length, in line with the literature (Supplementary Table 8)31. We used the same rate constant values for oligomerization and deoligomerization in the solution and on the template that we used before. However, the kinetic model underestimates the concentration and lifetime of the oligomers. The experimental data can only be predicted well if we assume no deoligomerization on the template (Fig. 3b–d and Supplementary Figs. 1a–d and 2). This first observation concludes that hybridization protects from hydrolysis, which aligns with our experimental observations that oligomers survive longer with the template (Fig. 3g).

Furthermore, the model predicts that library members in solution consumed 90.3% of the EDC, whereas library members on the template consumed only 9.7% of the EDC (Supplementary Fig. 3). Given the high monomer concentration in the solution, most of this 90.3% of the fuel was used to form DynT2. Specifically, in presence of the template, the average length in the solution was 1.02, consistent with the observation that monomers dominate the library (Fig. 3e and Extended Data Fig. 4a). By stark contrast, the average length on the template was 1.78, resulting from a high amount of dimers, trimers and tetramers (Fig. 3e and Extended Data Fig. 4b). In other words, oligomers accumulate on the template because they have a lower dissociation constant than monomers31. Therefore, ligations between oligomers occur frequently on the template (Fig. 3f, on template). Consequently, even though library members on the template consume far less EDC, on average, each ligation produces an oligomer of higher length than when in the solution.

We introduce the selection factor (S) to quantify templated oligomer production. For example, S31 is the factor between the oligomer concentration determined via HPLC with and without the template 31 min after adding the fuel. At low template concentrations, that is, 2 mM (dA)10 expressed in monomer units, the S31 value for DynT4 was roughly 28, whereas it was around five for DynT2. In other words, the template’s presence increased the concentration of DynT4 by more than 28-fold. Noteworthily, S20 and S38 showed the same trend, implying that the selection trend is generalizable in a time window (Extended Data Fig. 3b). The experiments contained T at 25 mM, so the template-to-library-member ratio was 0.08. Increasing the template concentration to 8 mM (dA)10 increased the S31 value of DynT2–DynT4 but not DynT5 (Fig. 3i and Extended Data Fig. 3c). The decrease in DynT5 is not entirely surprising, as the fifth base pair starts to turn in a B-DNA helix with a helical turn every 10.5 base pairs32. It seems that the oligomers cannot accommodate a twisted DNA angle, thus favouring shorter oligomers. Surprisingly, when we increased the template concentration further, for example, to 25 mM (dA)10, S31 decreased drastically for DynT3–DynT5 but not for DynT2—more template leads to shorter oligomers (Fig. 3i and Extended Data Fig. 5).

The positive feedback mechanism can explain these initially counterintuitive observations. The experiments are performed with substoichiometric amounts of fuel, that is, for each monomer, there are only 0.4 equivalents of fuel available. How this fuel is used will affect the oligomer length distributions, for example, one fuel molecule can link to monomers to form a dimer or two dimers to form a tetramer. The presence of the template affects this ‘efficiency’ of fuel usage. Without a template, the chance that two oligomers react to form a longer oligomer is small because the oligomer concentration is low. At a low template concentration, the library members compete for binding sites. This competition favours longer oligomers with higher binding constants to bind. Thus, the average oligomer length on the template is high, which favours the oligomerization of oligomers. At a higher template concentration, more binding sites are available, giving more space for monomers. Effectively, more template dilutes the oligomers on the template. This decreases the chance of two oligomers oligomerizing. In other words, too high a template concentration reduces the selective oligomerization of longer oligomers.

If the above mechanism is true, increasing the binding constant of the monomer should weaken the positive feedback of the template on the production of longer oligomers. Thus, we added the salt MgCl2, which is known to decrease the dissociation constant in native DNA33,34,35,36,37. In line with our expectation, the S31 value of DynT5 decreased gradually with increasing MgCl2 concentration, whereas more DynT2 was formed under these conditions (Fig. 3j and Extended Data Fig. 3d). Note that our control experiments without the template with and without MgCl2 show that MgCl2 does not affect the concentration of oligomers DynT2–DynT5. The individual kinetic profiles are shown in Extended Data Fig. 5.

On the basis of the above mechanism, we expect the template length to affect the length distribution of the dynamic library. To prove this, we tested DNA templates of different lengths, that is, (dA)4, (dA)7, (dA)35 and RNA-based polyadenine (pA; Fig. 3k and Extended Data Fig. 6). We analysed the changes in oligomer composition by monitoring S31. Interestingly, with template (dA)4, DynT5 was almost completely suppressed, whereas DynT4 was still formed. This seems to suggest that the template length controls the maximum length of the oligomer obtained (Fig. 3k and Extended Data Fig. 6a–d). Increasing the template length to (dA)7, (dA)10 and (dA)35 or switching to polyadenine RNA not only recovered the formation of DynT5 but increased the concentration of all oligomers DynT2–DynT5, and thus their selection factor values by factors of 3–5 for DynT2 and 16–35 for DynT3–DynT5 (Fig. 3k and Extended Data Fig. 6). It is of note that adding polyadenine RNA selectively increased the concentration of DynT5, by factors of 44 and 128.

Wobble base pairs, such as G•T, G•U or A•C base pairing, can lead to nucleobase mutations of the copied strands in DNA replication38,39. We tested if our synthetic system suffered from ‘wobble’ structures, too, by testing the oligomerization of T monomers in the presence of non-matching templates with the thymine ((dT)10), cytosine ((dC)10) and guanine ((dG)10) recognition motifs. We found only minimal changes in the selection factor, indicating that wobble base pairing does not play a large role in the case of DynTn (Fig. 3l and Extended Data Fig. 7).

Next, we focused on other monomers besides T. We fuelled libraries with U (uracil) or C (cytosine) isophthalic monomers in the presence of (dA)10 and (dG)10. Noteworthily, the purine bases (G and A) were poorly soluble when attached to isophthalic acid and thus were not used. Fuelling with U, a demethylated version of T, a diverse oligomer library emerges, comprising DynU2–DynU5 oligomers (Extended Data Fig. 3e and Supplementary Tables 9 and 10). In the presence of the ‘correct’ template (dA)10, the oligomer yield and lifetime of DynU2–DynU5 increased. By contrast, the ‘incorrect’ template (dG)10 brought about only a slightly increased yield and lifetime of DynU3–DynU5 (Fig. 4a–c and Supplementary Fig. 4a–l). When fuelling with C, a diverse oligomer library emerged, comprising DynC2, DynC3 and DynC4 (Fig. 4a–c, Extended Data Fig. 3f and Supplementary Tables 11 and 12). Again, using the correct template (dG)10, the oligomer yields increased. However, the C library suffered from wobble base pairing, as the presence of (dA)10 increased the oligomer yields, too (Fig. 4a–c and Supplementary Fig. 4m–v).

a, Molecular structures of monomers T, U and C. The reactive molecule sites, recognition motifs (for example, thymine) or complete molecular structures are colour-coded as follows: monomer acid (blue, left), thymine (red), uracil (orange) and cytosine (blue, right). b,c, Selection factor S31 for correct (b) and incorrect (c) templates (dG)10 and (dA)10 (at 2 mM) when fuelling with T, U or C. 2mer, dimer; 3mer, trimer; 4, tetramer; 5mer, pentamer. d, The chemical reaction system converts a chemical fuel (EDC) into waste (EDU) while building up mixed oligomers from monomers T and C. e, HPLC chromatogram at 31 min of the T/C oligomers without any template (bottom) and with 2 mM of template (dA)10 (top) or (dG)10 (middle). f,g, Selection factor S31 for the (dG)10 (f) and (dA)10 (g) template (at 2 mM). h,i, Sequence logo of the mixed T/C oligomers at 31 min in the reaction cycle with 2 mM of the templates (dG)10 (h) and (dA)10 (i). In b,c,f,g, the dotted lines are used to guide the eye. In d,f–i, the pictograms for the thymine oligomers are coloured red, the cytosine oligomers are coloured blue and the DNA template is coloured green with different shapes for (dA)10 (downtriangle) and (dG)10 (U shape). In e–g, the labels C2, TC, C3 and the others denote DynC2, DynTC, DynC3 and so on, respectively. aDynT5 was not observed without template (dA)10.

Source data

Given the selective oligomerization of hybridizing library members, we tested whether we could select T monomers from C monomers (Fig. 4d). We dissolved T (at 12.5 mM) and C (at 12.5 mM) in a similar buffer as described above and added a batch of 10 mM EDC. Without any template, we found eight oligomers: the three dimers DynT2, DynC2 and DynTC (DynCT and DynTC are the same molecules), three trimers (DynC3, DynT3 and DynT2C) and the two tetramers DynT4 and DynC4 (Fig. 4e and Supplementary Tables 3, 4 and 11–13). The dimers DynT2, DynC2 and DynTC were the major oligomers in the first minutes, of which the mixed dimer DynTC had the highest yield (Extended Data Fig. 8a–c), which is unsurprising because of its two formation pathways. The trimers and tetramers are in the same concentration range at around 0.01–0.02 mM, whereas pentamers were not observed (Extended Data Fig. 8d–j).

With template (dG)10 at 2 mM (expressed as the monomer concentration), the library composition remained similar for all oligomers apart from DynC3 and DynC4, whose concentrations increased from fourfold to 20-fold (Fig. 4e,f and Extended Data Fig. 8a–j). Interestingly, with template (dA)10 at 2 mM (expressed as monomer concentration), all of the oligomer concentrations increased. In particular, the concentrations of DynT2C, DynT4 and DynT5 increased drastically. As DynT5 was not observed without a template, its S31 value would be greater than 30 if we assumed concentrations below the detection limit of the liquid chromatography–mass spectrometry (LC–MS) measurements. Noteworthily, the concentrations of DynC2 and DynC5 increased roughly five and 20-fold, respectively, upon templating with (dA)10 (Fig. 4e,g and Extended Data Fig. 8m–x). We hypothesize that selection suffers from wobble base pairing, especially for A•C, which we have already demonstrated above. The sequence logos in Fig. 4h,i describe the relative frequency of monomers T and C within the oligomer groups (dimers, trimers, tetramers and pentamers) with the templates (dG)10 and (dA)10. In the presence of the templates, the dimers comprise roughly 55% T and 45% C. In the presence of template (dG)10, monomer T is less favoured (5%) in trimers and tetramers, whereas in the presence of template (dA)10, C is less favoured in tetramers (28%) and pentamers (0%). Taken together, the selective oligomerization of hybridizable monomers and their protection on the template increases the incorporation of T or C monomers into longer oligomers, respectively, for the correct template. However, selection for C suffers particularly from A•C wobble base pairing.

We wondered whether the selective binding of certain oligomers could be used to purify pools of monomers. The hypothesis is that a mixture of monomers becomes purer after a templated oligomerization–deoligomerization cycle, driven by different binding affinities and subsequent phase separation of the template–oligomer complexes21,31,40,41. Mechanisms here could also shine new light on purification mechanisms in a prebiotic setting—the prebiotic formation of the first nucleosides was unlikely to yield pure A, C, T and U, and purification mechanisms such as template-assisted ligation and selective precipitation probably played a role42,43,44,45,46.

As a selection mechanism, we used the ability of pA (RNA) to form complex coacervates by mixing pA as the polyanion at 0.86 mM (expressed as monomer concentration) and the cationic peptide Ac-F(RG)3N-NH2 (at 5 mM) at pH 6 (Fig. 5a and Extended Data Fig. 9a,b; phenylalanine (F), arginine (R), glycine (G) and asparagine (N)). After applying 50 mM fuel, we centrifuged the reaction solution, removed the supernatant and resuspended the pellet by adding buffer. We homogenized the pellet via sonication and then reinitiated the cycle. The cycle was performed three times in total (Fig. 5b). After each cycle, we analysed the composition of the resuspended pellet via HPLC (Fig. 5c). After the first cycle, the droplet consisted of 55% T and 45% Me (that is, the monomer with no nucleobase recognition motif). After the second cycle, the library contained almost exclusively T. The third cycle did not change the composition further. We hypothesize that the inefficiency of the first cycle results from both monomers partitioning in the droplets non-selectively (log P is 1.51 ± 0.15 for T and 0.99 ± 0.15 for Me). Thus, after centrifugation, there is no selection after the first cycle. In the second cycle, the concentration of monomers is much lower because many monomers have been washed away in the first cycle, leading to more selective, templated uptake.

a, Coacervate droplets (green droplets) made from Ac-F(RG)3N-NH2 (shown as red spheres with a charge of +3; the cationic side groups of the molecular structure are highlighted in red) and pA (green pictogram). b, Schematic of the experimental set-up to extract T from a mixed monomer pool. c,d,f, Composition of the coacervate phase after each cycle for the mixed T/Me pool (c), mixed T/C pool (d) and mixed T/C/Me-C/Me-T/Me/3-pyridyl IPA pool (f). Error bars depict the standard deviation of the mean (n = 18). e, Molecular structures of Me, T, C, Me-C, Me-T, 3-pyridyl IPA. The reactive molecule sites and recognition motifs (for example, thymine) are colour-coded as follows: monomer acid (blue, left), thymine (red), cytosine (blue, right), methylated cytosine (purple), methylated thymine (pink) and 3-pyridyl IPA (green). The respective monomer pictograms are coloured the same as before. The DNA template is coloured green. In c,d,f, the dotted lines are used to guide the eye.

Source data

Next, we attempted purification between the pyrimidine nucleobases (T and C in Fig. 5d). Both monomers had similar partition coefficients of 1.51 ± 0.15 and 1.33 ± 0.16, respectively. After the first cycle, equal amounts of T and C were present in the coacervate phase. After the second cycle, the coacervate phase comprised roughly 60% T and 40% C (Fig. 5d). In other words, the selection was not particularly efficient, probably because of A•C wobble base pairing. We expanded the monomer pool further to six members, that is, T, C, Me, methylated T (Me-T), methylated C (Me-C) and a non-natural base of 3-pyridyl isophthalic acid (3-pyridyl IPA, Fig. 5e). The library consisted of 25% each of T and C and 12.5% each of Me, Me-T, Me-C and 3-pyridyl IPA. We used coacervate droplets formed with the template (dA)17(dG)13 (or 0.34 mM adenine and 0.26 mM guanine when expressed in monomer concentration) and 2.5 mM polycation Ac-F(RG)3N-NH2 at pH 6 (Extended Data Fig. 9c). After the first cycle, the library composition had hardly changed. By contrast, the second cycle decreased the concentrations of the methylated and non-natural nucleobases, whereas the concentrations of C and T were increased slightly to 37% and 39%, respectively. The third cycle did not change the library composition further (Fig. 5f). Finally, we carried out similar experiments using hydrogels that comprise pA instead of coacervate droplets. These hydrogels showed similar purification abilities (see Supplementary Notes and Supplementary Fig. 5). The ability to preferentially ligate monomers offers a simple mechanism for purifying libraries of monomers via phase separation and centrifugation.

We demonstrated how templation affects the library composition. Next, we tested whether the opposite can also be true, that is, can the library affect its medium? Such reciprocal coupling between sequences (such as DNA or RNA) and their environment (a protocell or the cell) brings us closer to exploring Darwinian evolution in our system. For example, the library’s presence could increase the likelihood of a droplet producing offspring or surviving47,48,49,50,51.

We measured the fluorescence recovery after photobleaching (FRAP) of complex coacervate droplets made of pA, the polycation peptide Ac-F(RG)3N-NH2 and monomers T or C before and after adding fuel. Before adding the fuel, FRAP experiments on the polyanion pA showed that micrometre-sized droplets recovered their fluorescence within minutes, from which we calculated the diffusion constant D of 6.53 × 10−5 µm2 s−1. By contrast, after adding the fuel, the diffusion constant decreased roughly threefold in the presence of T oligomers, whereas it decreased roughly 1.5-fold in the presence of C oligomers (Extended Data Fig. 9d–i). Thus, the internal viscosity increases drastically upon fuel-driven oligomerization, probably related to triple helices with pA, which have a larger persistence length and act as cross-linkers52,53,54,55,56,57. For the mismatching nucleobase oligomers C, the effect seems to be smaller.

Coacervate-based droplets are known to fuse rapidly, which disadvantages a protocell as it loses its identity. The decrease in viscosity prompted us to explore if template-based copying could decrease fusion between protocells. For the protocells, we used the polyanions pDexS and pA. We mixed those with the cationic peptide Ac-F(RG)3N-NH2 and monomer T. We obtained multiphase complex coacervate droplets comprising a pDexS core and pA shell (Fig. 6a), which fused from roughly 130 multiphase droplets in a selected area to 19 droplets within 90 min (Fig. 6b,c). Their pA shell fused first, after which the pDexS core fused. The same experiment with fuel resulted in similar multiphase droplets in which the pA shell still fused. However, the increased viscosity resulted in a drastic decrease in the fusion of the pDexS cores, which led to large pA droplets with multiple pDexS cores inside them (Fig. 6c,d). By contrast, the same experiment with C instead of T did not decrease the fusion of the pDexS cores (Extended Data Fig. 10a). Mixing C and T in a 1:1 ratio, however, recovered the protection mechanism