Microscopic structure of the polymer-induced liquid precursor for calcium carbonate

Many biomineral crystals form complex non-equilibrium shapes, often via transient amorphous precursors. Also in vitro crystals can be grown with non-equilibrium morphologies, such as thin films or nanorods. In many cases this involves charged polymeric additives that form a polymer-induced liquid precursor (PILP). Here, we investigate the CaCO3 based PILP process with a variety of techniques including cryoTEM and NMR. The initial products are 30–50 nm amorphous calcium carbonate (ACC) nanoparticles with ~2 nm nanoparticulate texture. We show the polymers strongly interact with ACC in the early stages, and become excluded during crystallization, with no liquid–liquid phase separation detected during the process. Our results suggest that “PILP” is actually a polymer-driven assembly of ACC clusters, and that its liquid-like behavior at the macroscopic level is due to the small size and surface properties of the assemblies. We propose that a similar biopolymer-stabilized nanogranular phase may be active in biomineralization.

This paper brings us much closer to understanding the microscopic structure of the PILP phase. Using cryoTEM, liquid and solid-state NMR, as well as DLS, ATR-FTIR, and conventional microscopies, the authors provide a thorough and in-depth characterization of the early stages of PILP, and provide a unifying theory of its structure. This type of characterization has been lacking in the community, and I am grateful to the authors for providing us with such a clear picture of a very useful but poorly understood, until now, method for obtaining calcium carbonate structures with a range of nonequilibrium morphologies. The characterization is performed with attention to detail and the resulting conclusions regarding the formation mechanism are well-reasoned and supported by experimental data. For these reasons, I recommend publication in Nature Communications after minor revisions. Below are specific comments for the authors to consider when revising their manuscript.
Specific comments: 1) It would be helpful to have a clear and concise summary of the criteria for liquid vs. nanoparticulate phases for the non-expert reader who hasn't followed the years of PILP debate and discussion. Such a summary could go into the end of the introduction or beginning of the results section. Something along the lines of what characteristics in TEM, NMR, etc. would indicate liquid vs. nanoparticulate composite. The authors scatter these criteria throughout the manuscript, and I think it would be good to put them all into one paragraph early in the manuscript.
2) New data regarding the growth of calcite nanorods into track-etched membranes in the absence of additives is introduced in the Discussion section (Fig. S12). Please introduce all new results in the results section. I also suggest emphasizing that the reason the track-etch membrane results with and without additives are presented is because the ability to infiltrate pores and form non-equilibrium shapes has been attributed to the liquid-like nature of PILP (see comment #1 -if the track etch membrane experiments are motivated as one of the observations that needs explanation, it will make more sense why they are included.).
3) For both the ATR-FTIR and ss-NMR experiments, the authors need to provide more details regarding sample preparation. Are solid precipitates isolated by centrifugation? Are KBr pellets made? Was the solid packed into the NMR tubes? Such sample preparation details are critical to understanding the results. 4) On page 17, line 388, please provide a reference for the "traditional polyelectrolyte-enhanced wetting of surfaces." 5) On page 7, line 154, please specify in which figure the ds-DNA macromolecules can be seen. 6) In the Methods section, please provide an item number for the track-etch membranes used from VWR. These membranes are notoriously supplier and item number specific, especially as it relates to the surface treatments, which can vary. More information is required to repeat the experiments as described.
7) Units: a. Throughout the manuscript, lowercase l's are used for liter instead of uppercase L's. (e.g., g/l instead of g/L) b. Units are not provided for molecular weights. Please specify g/mol Gower procedure using 25 mg/L pAsp and 10 mM Ca 2+ . (see Text Ref. 20) This means similar film formation or nanorod growth in track etch membranes were observed in both system, and most of all, similar initial products (assemblies of ~2 nm sized ACC clusters) were formed as detected by cryoTEM. Since NMR spectroscopy only reflects the chemical change in short range (0.1 to 10 nm), we therefore feel comfortable to use the ds-DNA system for the NMR study of the PILP process.
Moreover, the ds-DNA system is the only system that can reasonably be used to monitor the PILP process by liquid state NMR. Typically the PILP process ( e.g. the original Gower procedure) is induced through the (NH 4 ) 2 CO 3 diffusion method, at a stable pH value (~10) with a low concentration of polymer (25 mg/L) and low supersaturation (10 mM Ca 2+ and up to ~5 mM CO 3 2-, Ref: Ihli J, Bots P, Kulak A, Benning LG, Meldrum FC. Elucidating mechanisms of diffusion-based calcium carbonate synthesis leads to controlled mesocrystal formation. Adv Funct Mater 23, 1965-1973 (2013).). In such a system, the only NMR measurement that would provide information about the reaction process could be 13 C liquid-state NMR. However, the 13 C content of natural or synthetic materials is only ~1%. Moreover, the gyromagnetic ratio of 13C (γ) is only 0.6 of that of 31 P. Since the measurement time for each reaction stage is also limited (<30 min), it is technically not possible for us to perform a reliable 13 C liquid-NMR measurement in a typical PILP system, even with 100% 13 C enrichment.
It is noteworthy that the system used by the Gower group for NMR study (see Text Ref. 33) is also not a typical PILP system, probably due to the difficulty we mentioned above. In their study a titration method was used (rather than the (NH 4 ) 2 CO 3 diffusion method), with a relatively high CO 3 2concentration (30 mM), moderate pH (~8.5) and 100% 13 C enrichment. This method generates relatively stable solutions that do not precipitate after 18 hours, and thus allows the 13 C liquid-NMR measurements. The extra peak observed in the 13 C liquid-NMR was attributed to the presence of PILP phase, and was considered as a direct evidence of the liquidity of PILP, although it was not shown that typical products of PILP processes (thin film or nanorods) could form in such a system.

Comparison with NMR data from Gower group
Indeed the precipitation method we used is different form the NMR work of Gower group (see Text Ref. 33), on which we should have been more clear in the manuscript. The aim of our discussion about the work of Gower et al. however is not to directly compare these two results and demonstrate they are the same, but to show that the NMR results of Gower et al. do not necessarily demonstrate a liquid-liquid phase separation, but could also be explained by the formation of ~2 nm clusters as we observed.
Revisions made to the manuscript: We have clarified the experimental differences between our work and the NMR work of Gower group, and emphasized the aim of the discussion in the revised manuscript: On Page 16, Line 370-374: Apparently, our results contrast with the NMR work of Bewernitz et al. (Text Ref. 33) who used a 13 C enriched titration system with high CO 3 2-/HCO 3 concentration, moderate pH (~8.5) and presence of pAsp 18 mg/L). An extra 13 C liquid NMR signal was detected in this system after Ca 2+ titration which they attributed to PILP, although it was not shown that typical products of PILP processes (thin film or nanorods) could form in such a system.
On Page 17, Line 379-382: Hence, our results are consistent with previous reports in spite of the differences in experimental conditions, and all data can be interpreted such that PILP is a dispersion of dynamic assemblies of ~2 nm sized ACC clusters that are cross-linked by charged polymers, without necessarily the presence of a liquid-liquid phase separation.
Reviewer comments (General 2): In the same line, the authors emphasize generality of their observations, but overlook differences from specific cases. For example, while Pasp and DNA cases appear similar, PAA and PAH don't, by judging the size or structures of aggregates (figure 2 vs figure s8) Author reply: We thank the reviewer for the comment, and realize that we have not been clear enough on this issue. First of all, the images shown in the original Fig. S8 d-i may cause confusion. These are not showing individual PILP nanoparticles, but a liquid-like complex of pAH and CO 3 2-, indicating the coacervation of these objects, and are not to be compared with the images of PILP from pAA or pAH. To generate more clarity we have made two separate supplementary figures ( Fig. S13 and S14) for these images.
Meanwhile, indeed the PILP nanoparticles induced by different polymers are different in sizes (from 30 to 200 nm) and shapes (e.g., the particles induced by pAH or ds-DNA seem to be more spherical than those formed with pAsp or pAA). In spite of these differences which may result from the different properties of the polymers (e.g., molar weight, charge density, length, etc.), however, all of the PILPs are colloidal nanoparticles assembled from ~2 nm sized subunits. So the PILPs are essentially have the same build-up at the nanoscale, and are colloidal nanoparticles instead of liquid droplets since they do not coalesce with each other, and the ~2 nm clusters inside them also do not coalesce. This is also what we want to highlight in this report. Still, we agree with the reviewer that the differences in the PILP nanoparticles induced by different polymers should not be overlooked. We have now addressed this in the manuscript.
Revisions made to the manuscript: 1. We have made two separate supplementary figures (Fig. S13 and S14) for the original Fig. S8.
2. We have modified the text accordingly, and described the difference of PILP induced by different polymers: On Page 7-8, Line 170-178: Similar assemblies of ~2 nm subunits were found in the early stages of the pAA or pAH (25 mg/L) induced PILP processes (see Supplementary Fig. S13), underlining the universality of our observations. It is noteworthy that the assemblies induced by different polymers are different in sizes (~200 nm sized assemblies could be observed for pAH) and shapes (the particles induced by pAH or ds-DNA seem to be more spherical than those formed with pAsp or pAA). These differences which may result from the different properties of the polymers (e.g., molar weight, charge density, length, etc.). Meanwhile, when pAH was used at a higher concentration (1000 mg/L), indeed droplet-like objects were observed as was originally reported by Cantaert et al (see Supplementary Fig. S14).
Reviewer comments (General 3): I looked up the original statement about PILP, and found out in many papers, the phase has already been described as "liquid-like" behavior. Therefore, the closing remark of this paper "we therefore suggest that the abbreviation PILP for polymer-induced liquid-like precursor" would not convey any strong message as the authors wish. I suggest to reconsider it.
Author reply: We agree with the reviewer on this suggestion.
Revisions made to the manuscript: We have revised the closing mark accordingly: On Page 19, Line 444-448: We therefore suggest to update the abbreviation of PILP into PINP for polymer-induced nanoparticulate precursor, highlighting its nanoparticulate microstructure. The presented model for PILP (PINP) not only aids to our understanding on how control over crystal morphology can be achieved in materials of technological relevance, it may also provide mechanistic insights into the formation processes of minerals in biological systems.
Reviewer comments (Specific 1): Assuming those aggregates (very much uniform sizes) are coming from 2 nm clusters, one would expect to see many individual 2 nm clusters (single/double/triple…) in surrounding media. I found this most strange. This makes me think whether the very early stage was missed. From schematic/mechanism of the process, should Ca/polyelectrolyte complexes should be the first products? Followed by formation of many individual 2 nm clusters ? The authors also mentioned on page16 "PILP consists of assemblies of 2 nm sized ACC clusters that are physically cross-linked by charged polymers, and no other product was detected before the formation of this phase in our experiment" Author reply:

Individual clusters or small assemblies
We thank the reviewer for this inspiring question. Indeed, as shown by our cryoTEM/tomographic results in Figure 2, no individual ~2 nm cluster could be observed within the surrounding media when the 30-50 nm sized assemblies of these clusters are formed. Also no individual cluster, or smaller sized assembly, was observed in earlier products. To further confirm whether the invisibility of individual clusters is due to a lack of contrast, we applied the early stage products (15 min for ds-DNA experiment and 90 min for pAsp experiment) on graphene oxide (GOx) coated cryoTEM grids. This recently developed method allows the formation of much thinner vitrified ice layers (~10 nm, See Text Ref. 43), which greatly improves the contrast of cryoTEM images and benefits the detection of ~1 nm sized clusters. Still no individual clusters or smaller assemblies were detected in these experiments as shown by Fig. R1, suggesting that their concentration is negligible in our experiments.
We can understand the extremely low concentration of individual clusters or smaller assemblies taking into account the high surface area ratio of the clusters and the instability of the ACC phase in water solution. As a result, in our experiments most of the individual clusters will immediately dissolve unless they are nucleated on, or bound with the charged polymers, and assemble into the >30 nm sized assemblies. Given this, an extremely low equilibrium concentration is expected for the individual clusters or smaller assemblies in our experiments. shows the zoom-in image, which shows densely packed ds-DNA molecules. The higher density of ds-DNA is due to the 60 seconds of waiting time applied during the vitrification, which allows the ds-DNA molecules to condense on the GOx layers. (b) and (e) are both taken with a defocus value of -1.5 μm. Scale bars: (a, d) 500 nm, insets of (a, d) 2 nm 2. The corresponding experimental method was added into the Methods part: On Page 21, Line 501-506: To study the early stage samples, 3 μL of reaction solution samples were applied on a GOx coated TEM grid (Text Ref. 43), and 20% (v/v) IPA in ultrapure water was used in humidifier. After 60 s of waiting, the grid was blotted for 3 s and vitrified. CryoTEM imaging was performed under ~3 μm defocus on a FEI-Titan TEM equipped with a field emission gun (FEG) and operating at 300 kV. For samples on GOx coated grids, imaging was done using a parallel beam with an illuminated area of 670 nm (nanoprobe), with a defocus value of -1.5 μm.
3. Fig. S10 is referred in the manuscript together with Fig 4. The explanation for the low concentration of individual clusters or assemblies is inserted into the Discussions part: On Page 6-7, Line 389-392: Without being stabilized by the polymers and assembled into the >30 nm sized assemblies, these ACC clusters will immediately dissolve due to their high surface area ratio and the instability of ACC phase in water solution, leading to an extremely low equilibrium concentration of the individual clusters or smaller assemblies in our experiments (see Supplementary Fig. S9-10).

Ca/polyelectrolyte complexes
And indeed as shown in Fig.6, the Ca 2+ /polyelectrolyte complex is expected be the first product, which actually should already form upon mixing the CaCl 2 and polymer solutions. The cryoTEM images from these complexes are however not expected to be very different from those of the free polymers, unless they aggregate and phase separate from the solution, such as is the case for the pAH/CO 3 2complex. This was not observed for the other charged polymers used in our experiments (ds-DNA, pAsp and pAA). The presence of these complexes could still be reflected by the binding between Ca 2+ and ds-DNA as shown in the supplementary Table S5 and S6.
Revisions made to the manuscript: We have supported our description of Ca 2+ /polymer complex formation using Table S5 and S6: On Page 18, Line 411-412: Taking negatively charged polymers as an example (Fig. 6a), the reaction starts with the formation of a Ca 2+ /polymer complex (stage 1, see also supplementary Table S5 and S6).
Reviewer comments (Specific 2): The concentrations of polyelectrolytes are too low to stabilise all the 2 nm clusters. Taking Pasp (Mw 10000) for an example, it is only 1-2 uM, which is extremely low to stabilize mineral phase (10 mM) used here ? Moreover as the authors claim, If the wet phenomenon were due to polyelectrolyte stabilzing the clusters, then demonstration of same phenomenon with the same polyelctrolyte stabilized (solid) colloids would be the strongest and direct evidence? PAH or PAA are widely used to stabilize colloidal metal/inorganic nanoparticles in similar sizes, like Au clusters. Authors could benefit from discussion further regarding "nanofluid" behavior : Generalized Route to Metal Nanoparticles with Liquid Behavior. J. Am. Chem. Soc. 128, 12074-12075 (2006).

Author reply:
The authors thank the reviewer for the questions, which greatly helped us to further improve the paper. As for the several points raised here:

ACC clusters stabilized by pAsp
Indeed the concentration of pAsp we used is only ~2.5 μM, which however means ~0.22 mM of aspartic acids (-COOgroups). According to a detailed compositional study performed by Gower et al. on pAspinduced PILP (Ref: Dai L, Douglas EP, Gower LB. Compositional analysis of a polymer-induced liquidprecursor (PILP) amorphous CaCO 3 phase. J Non-Cryst Solids 354, 1845-1854 (2008).), the Ca 2+ /COOratio is ~10:1 for PILP induced by 20 mg/L of pAsp. Thus the 2.5 μM of pAsp used in our experiments should be able stabilize PILP representing 2.2 mM of Ca 2+ . On the other hand, the 10 mM of Ca 2+ ions in the reaction solution are not expected to be all transformed into the PILP phase, at least not at the same time. The actual concentration of PILP in our experiments is difficult to be accurately measured. However, the PILP products we got after centrifugal separation and drying is only 2~3 mg for each experiment (25 mL of reaction solution), which correspond to only to ~1.2 mM of Ca 2+ in the reaction solution. Therefore, the 25 mg/L (or ~2.5 μM) of pAsp used in our experiments should be sufficient to stabilize all the ~2 nm ACC clusters in the PILP phase.
The high Ca 2+ /COOratio in pAsp-induced PILP could be explained by the fact that the polyelectrolytes are binding only with the surface Ca 2+ of ~2 nm sized ACC clusters, instead of all the Ca 2+ ions in PILP. According to a recent computational study, each ~2 nm sized ACC cluster consists of ~20 Ca 2+ /CO 3 2ion pairs (See Text Ref. 60). The 10:1 Ca 2+ /COOratio therefore suggests that each ACC cluster is bound with two COOgroups in pAsp-induced PILP. Interestingly, a similar Ca 2+ /-PO 4 -ratio (10:1) was detected for the ds-DNA induced PILP by the ICP-OES measurements (see the newly added supplementary Table S2), and it remains unclear why a much higher concentration of ds-DNA (2.5 g/L) is required to induce the PILP process. We have incorporated part of this discussion into the manuscript.
Revisions made to the manuscript: 2. One paragraph of discussion is added into the manuscript: On Page 16, Line 357-369: The fact that PILP is assembled from ~2 nm sized ACC clusters also explains why they could be stabilized by a relatively low concentration of polyelectrolytes e.g., 25 mg/L of pAsp, which contains only ~0.22 mM of COOgroups. Instead of binding with all the Ca 2+ ions, the polyelectrolytes only need to bind with the surface Ca 2+ of the ~2 nm sized ACC clusters in PILP. Each of these ACC clusters consists of ~20 Ca 2+ /CO 3 2ion pairs according to a recent computational study (Text Ref. 60). As a result, much more Ca 2+ could be stabilized by one COOgroup. Indeed, as shown by a previous compositional study of PILP by Dai et al., the Ca 2+ /COOratio is ~10:1 for PILP induced by 20 mg/L of pAsp, (Text Ref. 61) suggesting that each ACC cluster is bound with two COOgroups in pAspinduced PILP. As a result, 25 mg/L of pAsp could stabilize PILP representing ~2.2 mM of Ca 2+ , allowing a higher concentration of PILP formation. Interestingly, a similar Ca 2+ /-PO 4 -ratio (10:1) was detected for the ds-DNA induced PILP by the ICP-OES measurements (see supplementary Table S2), and it remains unclear why a much higher concentration of ds-DNA (2.5 g/L) is required to induce the PILP process.

"Wetting ability" of PILP
Author reply: We apologize for being not accurate on the discussion about the wetting ability of PILP, which is actually the ability of PILP nanoparticles to attach to surfaces. For this to happen, the PILP nanoparticles must first be stabilized by the charged polymers, otherwise they will crystallize into large crystals, rather than attaching to surfaces and forming non-equilibrium morphologies such as thin films and rods. In that sense, the polyelectrolytes are essential for the "wetting phenomenon" since they clearly play a vital role in stabilizing the PILP particles However, one requirement for the "wetting" is the affinity between PILP particles and substrates. This is not determined only by the properties of the polymers on the surface of PILP particles, as we mistakenly stated in the original manuscript. Indeed, PILP induced by different polymers leads to thin film products with differences in thickness and surface roughness on the same substrate, as shown by our experimental results using pAsp, ds-DNA, pAA or pAH, as well as by observations in previous reports (Ref: Gower LB. Biomimetic model systems for investigating the amorphous precursor pathway and its role in biomineralization. Chem Rev 108, 4551-4627 (2008).). This suggests that the affinity of PILP nanoparticles to surfaces is at least partly regulated by polymers. However, the interaction between colloidal particles and the surfaces is a complicated collective effect of many forces including electrostatic interaction forces, van der Waals forces, entropic forces and steric forces (Ref: Belloni L. Colloidal interactions. J Phys: Condens Matter 12, R549 (2000).). As a result, the interaction should also be effected many other factors such as the size and chemical composition of the colloidal particles, the way polymers interact with the colloidal particles, as well as the ionic strength, pH value, etc. This has also been suggested in the PILP system by a previous report, showing that the "wetting" of PILP on different substrates could be drastically tuned with presence of 30 mM of Mg 2+ (see Text Ref.35). It was proposed that the Mg 2+ might influence the hydration level of PILP, but the detailed mechanism remains unknown.
Furthermore, PILP is not exactly the same with typical colloidal particles stabilized by surface polymers. They are actually 30-50 nm sized bicontinuous structures assembled by ~2 nm ACC clusters and polymers. These assemblies could grow larger and later on transform into crystals. Thus the interaction of PILP with different surfaces could be even more complex. As a result, the experiments of Au or other metal clusters stabilized by the same polymers would unlikely provide any clear evidence about why the PILP particles has a strong affinity to certain surfaces. To fully understand how the affinity of PILP to different surfaces is regulated by different polymers still require systematic experiments in the future, examining the PILP induced by polymers with different molar weight, charge density, diameter, length, etc., and how they interact with different surfaces, which is beyond our current discussions.
Revisions made to the manuscript: We have corrected our inaccurate statement that the surface property of PILP is only determined by the polymer: On Page 17, Line 399-400: The PILP NPs will completely cover surfaces that match with their ACC/polymer surfaces (or not at all when the surfaces are incompatible with the surface of PILP).

Reference for the liquid-like behavior
Author reply: We agree with the reviewer on this suggestion.
Revisions made to the manuscript: Reviewer comments (Specific 3): Line133 page 7 "did not coalesce to form continuous spherical objects" if I understood correctly, f and i in Figure S8 represent "liquid" nature of objects. However, even those doesn't look very spherical, even though I do agree with the authors the absence of internal subunits. Out of curiosity, "nanoscopic fluidity" has ever been directly observed by in-situ EM?
Author reply:

Liquid droplets visualized by cryoTEM
We appreciate the reviewer for pointing this out. The f and i in original Figure S8 indeed show liquid-like objects that are pAH/CO 3 2complex, as we mentioned in the reply to the General Comment 2. However, we realized that our previous description that liquid droplets will always be spherical under cryoTEM was not rigorous. In the ideal case without considering gravity and small local inhomogeneities, the equilibrium shape of liquid droplets dispersed within another liquid indeed should be spherical (Ref: Dubochet J, et al. Cryo-electron microscopy of vitrified specimens. Q Rev Biophys 21, 129-228 (1988)). This is in order to minimize their surface area and reduce the liquid-liquid interfacial energy. For the droplets with smaller size (higher surface area ratio) and/or high liquid-liquid interfacial energy, their shape will be strongly regulated by the interfacial energy and thus close to spherical ( Figure R2a). This has indeed been observed for oil nanodroplets dispersed in water, which are continuous and spherical objects under cryoTEM (Ref: Klang V, Matsko NB, Valenta C, Hofer F. Electron microscopy of nanoemulsions: an essential tool for characterisation and stability assessment. Micron 43, 85-103 (2012).). For droplets with bigger size (lower surface area ratio) and/or lower liquid-liquid interfacial energy, however, the shapes will be more easily disrupted by local fluctuation in density, temperature or ionic strength, etc. As a result, although still being continuous objects with smooth edges, their shape will become irregular ( Figure R2b). This was observed for the pAH/CO 3 2complex droplet in original Figure S8 f and I, which are micron-sized continuous objects with irregular shapes. The results suggest that the interfacial energy between the pAH/CO 3 2complex and the solution is relatively low. Also these droplets are relatively larger (~microns). As a result, the shape of these droplets are less controlled by the interfacial energy, and their shapes are more irregular.
For the PILP phase, the 30~50 nm particles are basically spherical, but they consist of ~2 nm subunits, and occasionally are hollow (Figure 2). Furthermore, the micron sized aggregates of these particles also clearly show the nanoparticulate features as well as the ~2 nm subunits (Figure 2c and f), which are quite different from the pAH/CO 3 2droplets (supplementary Figure S15). As a result, we conclude that the PILP is not a liquid.
It is noteworthy that the above mentioned features observed in cryoTEM (continuous, spherical or cloud like) can be associated with the liquid-like nature of an object, but do not prove it.  2111-2115 (2014).). In this work, 3-10 nm sized gold nanoparticles encapsulated by ~30 nm sized water droplets on a flat solid surface were observed by in-situ liquid phase TEM. The observed nanodroplets are continuous objects with a close to spherical shape, matching with the cryoTEM observations. During the observation, the droplets flows and its shape deforms, accompanied by the movement of the gold nanoparticles within. Unfortunately, very high electron dose rate is usually required for high resolution in-situ liquid phase TEM observations (for the above mentioned work the electron dose rate is 2000 to 5000 e/(Å 2 •s)). Thus for electron beam sensitive materials such as ACC or polymers, the resolution of insitu liquid phase TEM technique is still limited, and therefore it was not used for current study.(see Ref: 1. Nielsen MH, Aloni S, De Yoreo JJ. In situ TEM imaging of CaCO3 nucleation reveals coexistence of direct and indirect pathways. Science 345, 1158-1162 (2014). 2. Patterson JP, Proetto MT, Gianneschi NC. Soft nanomaterials analysed by in situ liquid TEM: Towards high resolution characterisation of nanoparticles in motion. Perspectives in Science 6, 106-112 (2015).), Figure R2. Scheme for the morphological difference of liquid droplets dispersed within another liquid. (a) For droplets with small size and/or high liquid-liquid interfacial energy, their shape will be close to spherical in order to minimize the surface area and reduce the interfacial energy. (b) For droplets with large size and/or low liquid-liquid interfacial energy, their shapes will be more easily disrupted by the local inhomogeneity, and thus are more irregular.
Revisions made to the manuscript: 1. We have added Figure R2 together with the discussions on the liquid droplet morphologies into the supplementary information as Figure S1.
2. We have revised our text, clarifying that the liquid droplets in cryoTEM are continuous objects with smooth edge, as also asked by the Comment 1 of Reviewer 3: The NPs subsequently grew in size and aggregated to form larger structures, but did not coalesce to form continuous objects with smooth edges as would be expected for liquid droplets (Fig. 2b). Reviewer comments (Specific 4): In Figure 1. The crystalline platelet is formed by dissolution of ACC films, it contradicts again with the suggested mechanism/schematic here and the original hypothesis of PILP, where solid state transformation is implied?
Author reply:

"solid-transformation" vs "ACC dissolution"
In the PILP systems the first crystalline CaCO 3 plate usually forms within the ACC thin films. This is also observed in our experiments and may be due to a solid state transformation process, or to a dissolutionreprecipitation process, currently there is no evidence for one or the other in PILP systems. This is why the term "solid-transformation" was not used in our manuscript. The next step is the growth of the plate through the dissolution of the amorphous film. This is a well know phenomenon in ACC-crystal transformation and is driven by the different solubilities of the two phases where calcite acts as a sink for the precipitation of Ca 2+ and CO 3 2ions.
Although this does not contradict with the formation of the first plate through a solid state transformation hypothesis, the latter cannot describe the entire transformation as the ACC phase is hydrated ( ) and the polymers within PILP will be excluded after crystallization, a larger volume of PILP is consumed to form the same molar amount of dehydrated crystalline polymorphs such as calcite or vaterite. This then lead to a dissolution of the ACC thin film around the growing crystalline platelets, which is promoted by the decreasing supersaturation of the solution (Fig. 4d) due to the relatively higher solubility of ACC comparing with calcite or vaterite. After complete crystallization (stage 6, see also Fig.  1b), the majority of polymers are excluded from the crystals, while some of them become occluded.
Reviewer comments (Specific 5): NMR: the pH is not constant during the reaction, changes in ionic strength and pH in the solution will heavily affect the interaction and conformation of DNA and might diminish or enhance such effects. Therefore it is not clear that how to decouple the effect of interaction to ACC from pH effects, unless data of DNA/Ca at varied pH support otherwise.
Author reply:

Liquid NMR measurements: the effect of pH and [NH 4 + ]
We agree with the suggestion of the reviewer, and have investigated the effect of pH value change to our 31 P liquid NMR measurements. Only a slight fluctuation of the chemical shift (~0.02 ppm) and peak width (~0.04 ppm) was observed when the pH value of our reaction solution (with 10 mM CaCl 2 and 2.5 g/L ds-DNA) was raised from 5.3 to 10.0 by 50 mM NaOH, as shown in Figure R3a. This very small fluctuation may in fact be due to the intrinsic experimental variability of the measurements. So the pH will not have a significant effect on the chemical shift of 31 P signal during the reaction.
Nevertheless, we should consider that the generation of NH 4 + ions due to the in-diffusion of NH 3 may influence the conformation of ds-DNA by changing the ionic strength. We investigated this by adding 5 to 100 mM of NH 4 Cl into our reaction solution ( Figure R3b). The addition of NH 4 Cl indeed first shifts the 31 P signal to high field reaching a shift of  =-0.06 ppm at [NH 4 + ] = 10 mM. When the concentration was further increased to 100 mM -which is close to the actual NH 4 + concentration in the reaction solution (Ref: Ihli J, Bots P, Kulak A, Benning LG, Meldrum FC. Elucidating mechanisms of diffusion-based calcium carbonate synthesis leads to controlled mesocrystal formation. Adv Funct Mater 23, 1965-1973 (2013).)the signal shifted back to 0.05 ppm down field of the initial solution.
We therefore attribute the small down field chemical shift observed after 200 min of the reaction (Fig. 4a  and b) of 0.07 ppm (compared to the neat ds-DNA solution) mainly to the increase of the ionic strength and the influence of NH 4 + on the conformation of ds-DNA.
Beyond this small shift, our results show a significant high field shift (=-0.14 ppm) and broadening (0.27 ppm) of the 31 P signal of ds-DNA upon mixing with CaCl 2 , and then a continuous down field shift (0.21 ppm) and sharpening (0.27 ppm) of the signal during the diffusion experiment. This trend is unlikely due only to the in-diffusion of NH 3 , and should still be mainly attributed to the release of Ca 2+ from binding with ds-DNA due to CaCO 3 formation. Revisions made to the manuscript: We have added Figure R3 and the corresponding discussions into the supplementary information as Figure S15, and modified the text at these two locations: On Page 12, Line 269-271: The increase of pH value was found to have no significant effect to the free [Ca 2+ ] and liquid 31 P NMR measurements, while slight decrease of zeta potential was detected due to deprotonation (see supplementary Fig. S15a-b, Table S5 and S6).
On Page 12, Line 321-323: After 200 min the peak showed a width similar to the neat ds-DNA solution, but with a lower chemical shift due to the increased ionic strength related to the introduction of NH 4 + and CO 3 2ions in the solution (see supplementary Fig. S15c-d).
Reviewer comments (Minor 1): Line 62 page 3 "with unprecedented detail the formation and transformation" I would not agree with this.
Author reply: We agree with the reviewer on this suggestion.
Revisions made to the manuscript: We have revised this sentence: On Page 3, Line 68: Here we present an in-depth investigation of the formation and transformation processes of PILP Reviewer comments (Minor 2): Line 100 Page 4 "~extending over centimetres" all figures shown only show micron-meter scale, I believe it would be typo.
Author reply: We are sorry for this mistake. The thin films are usually several millimeters large, but they are impossible to be several centimeters sized since the glass slide used as growth substrate is ~2 cm sized. The low magnification SEM image below ( Figure R4) shows a ~1 mm sized region of thin film. Revisions made to the manuscript: We have corrected the mistake by changing "centimeters" into "millimeters" on Page 4, Line 106.

Reviewer comments (Minor 3):
Using unified units, hour or mins, will help readers Author reply: We thank the reviewer for the kind suggestion. However, we used different time units on purpose. For the in-situ measurements (CryoTEM, liquid state NMR, pH/[Ca 2+ ] curve, DLS/zeta potential measurement, etc.), we used "mins" since these measurements reflect the real time status of the reaction solutions. For the characterization on dry samples, the samples have to be taken out of the reactions solution, thoroughly washed by ethanol, and then dried in room temperature for overnight before measurements. Several minutes will be required before the reaction process is fully quenched by de-hydration, thus the time scale could no longer be as accurate as the in-situ measurements, and "hours" were used for these results. For this reason we prefer to keep the units unchanged.
Author reply: We appreciate the reviewer for pointing this out. ACC was reported to show only one broad peak around 1085 cm -1 in Raman measurements (Ref: Addadi L, Raz S, Weiner S. Taking advantage of disorder: amorphous calcium carbonate and its roles in biomineralization. Adv Mater 15, 959-970 (2003).). Our Raman spectra measured on the ACC thin films, however, show 3 humps around 500-600, 800 and 1085 cm -1 . Indeed 3 similar humps are also detected from the glass substrates we used ( Figure  R5). For the ACC thin film samples, however, the hump around 1085 cm -1 is significantly enhanced compared with the glass substrate, still indicating the presence of ACC in the thin films. Figure R5. Raman spectra taken on the products grown on a glass substrate with 25 mg/L of pAsp or 2.5 g/L of ds-DNA. The spectrum of the glass substrate is also shown for comparison. The glass substrate shows 3 humps around 500-600, 800 and 1085 cm -1 , respectively. For the thin films grown with pAsp for 5 h or with ds-DNA for 2 or 5 h, the hump at 1085 cm -1 increases significantly, suggesting the formation of ACC. (SI Ref. 3) The rhombic platelet (platelet 1) grown with pAsp and the platelet grown with ds-DNA found at 24 h showed peaks corresponding to the CO 3 2ν 1 symmetric stretch mode (1087 cm ) and external mode (300 cm -1 ) of vaterite. (SI Ref. 4,5) Revisions made to the manuscript: 1. We have added the signal of glass substrates to supplementary Fig. S3 and S12f.
2. The following literature is added into the supplementary reference list as SI Ref. 3: Addadi L, Raz S, Weiner S. Taking advantage of disorder: amorphous calcium carbonate and its roles in biomineralization. Adv Mater 15, 959-970 (2003).

Reviewer comments (Minor 5):
Supplementary table 5 : measurement at actual reaction pH (-9-10) would be more informative. Author reply: We appreciate the reviewer for the suggestion. The original supplementary Table S5 is remeasured at pH=10.0 (Table R1), which shows only small deviations (~0.1 mM) to the previous results. This suggests the binding between Ca 2+ and ds-DNA is relatively strong and thus not affected by the protonation/deprotonation of the phosphate groups. The zeta potentials in original supplementary Table  S4 are also re-measured at pH=10.0 (Table R2), which are slightly lower than the values measured at pH=5.3 due to the deprotonation at higher pH. ] measurement of 10 mM CaCl 2 solution containing different concentrations of ds-DNA. The phosphate group concentration was also estimated from the molecular weight (~225000) and number the number of base pairs (~300) of the ds-DNA. The measurements were performed at pH=5.3 and 10.0, respectively. It shows that the free Ca 2+ in the solution decreased with increased ds-DNA concentration. With 2.5 g/L ds-DNA, only 7.25 mM of free Ca 2+ was detected in the solution with pH=5.3, indicating 2.75 mM of Ca 2+ was bound to ds-DNA. The free [Ca 2+ ] detected at pH=10.0 only deviate slightly with the values detected at pH=5.3, suggesting the binding between Ca 2+ and ds-DNA is relatively strong and thus not significantly affected by the protonation/deprotonation of the phosphate groups.  Table R2. Zeta potential measurement of 2.5 g/L ds-DNA solutions mixed with different concentrations of Ca 2+ . The measurements were performed at pH=5.3 and 10.0, respectively. The zeta potential of pure ds-DNA solution was highly negative due to the negatively charged phosphate groups. The zeta potential significantly increased with increased Ca 2+ concentration, indicating the binding between Ca 2+ and ds-DNA. The zeta potentials measured at pH=10.0 are slightly lower than those measured at pH=5.3 due to the deprotonation at higher pH. Revisions made to the manuscript: We have added the Table R1 and R2 into the supplementary information as Table S6 and S5, and described the effect of pH value to free [Ca 2+ ] and zeta potential measurements in the text: On Page 12, Line 269-272: The increase of pH value was found to have no significant effect to the free [Ca 2+ ] and liquid 31 P NMR measurements, while slight decrease of zeta potential was detected due to deprotonation (see supplementary Fig. S15a-b, Table S5 and S6). This suggests that the interaction between ds-DNA and Ca 2+ is relatively strong and not affected by the deprotonation of ds-DNA.

Reviewer comments (Conclusion):
This is thorough studies with many beautiful data, particularly the cryo -TEM images and videos are striking, I am sure that crystallization/biomineralization community could benefit from. Thefore I would encourage publication if the authors address the comments and questions, hoping it would strengthen the paper. Author reply: We thank the reviewer again for the encouragement and fruitful discussions. We have modified the manuscript accordingly and believe its clarity has been significantly improved. We hope the manuscript now fits the standard for publication on Nature Communications.

Reviewer 2:
Reviewer comments (General): The authors studied the PILP process using advanced analytical techniques such as CryoTEM, Raman, and NMR techniques. The study highlight the mechanism behind mineral formation process, which may be relevant to biomineral formation or synthesis of novel biomimetic materials. The work is important. However, some of the results and discussion may be more convincing after more analysis and review of prior literatures. Author reply: We thank the reviewer for the positive comments, and have addressed the specific comments point-by-point as follows.
Reviewer comments 1: LIne 85-95: Figure 1 e-h and supplementary video 1 describes the formation of thin calcite crystals at the later stage. However, the reviewr highly doubt the dissolution of ACC film in regions over 200 um diameters resulted in the formation of the calcite crystal. It is highly possible that ACC film still there and polymer is completely consumed at the late stage, the remaining Ca (which may be released from dissolution ACC phase as well) will react with CO3 to form Calcite crystals. The observed "dissolution" may be just an artifact of focusing of objective lens on the crystals. If calcite crystal formed at the expense of dissolution of ACC phase, one should observe the disappearance of ACC film close to the crystal first, then extend the disappearance to outside. In summary, the reviewer doubts the observed disappearance of ACC from outside to inside. Author reply: We agree with the reviewer's viewpoint and this is indeed a misunderstanding. We intended to describe the process as outlined by the reviewer. Indeed the dissolution of the ACC thin film should start around the crystals first, and extend outside to the film edge, as shown also in the supplementary Movie S1. The focusing of objective lens is not changing here, which could be reflected by the stable contrast of the crystalline edges.
Revisions made to the manuscript: We have described the process more clearly in the figure caption of Figure 1: On Page 4, Line 97-98: A crystalline platelet nucleated within the film in (f), and grew in (g) and (h). The film near the growth front of the crystalline platelet started to dissolve in (g). The dissolution extended outside to the film edge and the film was fully dissolved in (h).

Reviewer comments 2:
Line 246-251: The authors talked about the release of Ca from Ca-DNA complex, which is correct. However, when Ca is released from Ca-DNA, there is already CaCO3 phase formation. The observed Zpotential change measurement is based on particles. AT this stage, there are two types of particles that contribute to the Z-potential: ACC particles and maybe macromolecular DNA particles. Thus the change of Zeta-potential does not confirm the release of Ds-DNA. Author reply: We thank the reviewer for pointing this out. Indeed the decrease of Z-potential could also be an effect of Ca 2+ /CO 3 binding. The release of ds-DNA could still be confirmed by our EDS, FTIR, NMR and ICP-OES measurements.
Revisions made to the manuscript: We have corrected the text accordingly: On Page 12, Line 268: During this process also the zeta potential decreased to -24.2 mV due to the partial release of ds-DNA and/or binding between Ca 2+ and CO 3 2- (Fig. 4c).
On Page 14, Line 318: During crystallization a continued decrease of the zeta potential to -34.3 mV was observed, in agreement with the release of the ds-DNA/formation of CaCO 3 during the process (Fig. 4c).
Reviewer comments 3: Line 110-111: Crystal growth at air-liquid interface and on glass slides substrates are different matters. Has Gower mentioned that 2 um sized particles in the air/liquid interface formed the film on the glass slide? The film formation refers only to the CaCO3 on glass substrates. Author reply: We again humbly feel here is a misunderstanding. As written in our manuscript, during the in-situ OM observation of the thin film formation process, the 2 μm sized particles were observed at the glass/solution interface, and there was no air/liquid interface involved in these experiments (please also check supplementary Fig. S4 for details). In Gower's report on PILP,(Text Ref. 20) similar 2 μm sized particles/droplets were observed on the surface of air bubbles in the solution, which might be the air/liquid interface mentioned by the reviewer. However, later on these particles were also observed near the CaCO 3 thin film formed on the glass substrate. Based on this observation, it has been proposed by Gower et al. that the CaCO 3 thin film was formed by the coalescence of PILP "droplets".
Revisions made to the manuscript: We have referred to Gower's work again after this sentence, making it clear that it was proposed by Gower et al. that the thin film was formed by coalescence of these 2 μm sized particles/droplets: On Page 5, Line 121-122: However it was never observed as proposed by Gower et al. (Text Ref. 20) that these particles deposited and subsequently coalesced with the film.
Reviewer comments 4: Line 317, Figure 6. The reviewer disagree with picture depicted in stage 6. Can the author proves there are free "excluded polymer" in the solution. During mineral formation, the polymer-Ca complex even release Ca to form CaCO3, the polymer stills bounds to mineral surface and polymers imbedded in the mineral phase (the polymer-poor region forms crystal, the polymer-rich region remains ACC phase). The close association of polyaspartic acid with CaCO3 mineral during and after the PILP process has been demonstrated by Gower's group by using the FITC-labeled Polyaspartic acid. Please confirm the existence of free polymer as the author demonstrated in Stage 6. There is a mechanism figure in the below reference which may be cited and discussed for comparison purpose. "Dai, L. J.; Cheng, X. G.; Gower, L. Certainly some of the polymers could be trapped within the crystals, in some cases forming "transition bars", as mentioned by the reviewer. However, the transition bars were not observed in our experiments, as well as most studies of PILP as far as we know, suggesting this is not a common phenomenon in the PILP systems. However, one thing that was missing in the Stage 6 of Figure 6 is that the polymers should preferably attach to the surfaces of crystals, from where they were excluded.
Revisions made to the manuscript: We have modified the State 6 of Figure 6, showing that the polymer preferably attach to the surface of crystals, and modified the text accordingly: On Page 18, Line 423-425: After complete crystallization (stage 6, see also Fig. 1b), the majority of polymers are excluded from the bulk crystals to the crystal surfaces or back to the solution, while some of them become occluded.
Reviewer comments 5: Line 443, EDS used to quantify the Ca:P:Cl ratio was not an accurate method and the quantification is only accurate on smooth surface and. A better method such as dissolving the material and measure by other technique including ICP-AES should be used Author reply:

ICP-OES measurements
We agree with the reviewer on this suggestion. Inductively coupled plasma optical emission spectrometry (ICP-OES) is performed to more accurately measure the elemental ratio of the samples.  1536-1541 (2007).), in order to fully digest the ds-DNA molecules in the samples for a more accurate phosphorous measurement. Mass fractions of four elements (Ca, P, Mg and Na) are measured for each sample (Table  R3). The results show similar trend of chemical composition change in different samples as indicated by the EDS measurements (Table R4), thus do not affect the discussions in the original manuscript.
Unfortunately, the sensitive spectral lines of chloride have wavelengths of around 135 nm, which fall in the vacuum ultraviolet region (<190 nm) and are difficult to be detected by conventional ICP-OES due to the low transmission of the vacuum UV light in atmospheric oxygen or water. As a result, we are unable to accurately measure the chloride fraction in the ds-DNA/Ca 2+ complex using our ICP-OES spectrometer.
Considering the similarity of the EDS and ICP-OES measurements and the thorough cleaning procedure that we used for the ds-DNA/Ca 2+ complex (5 rounds of centrifuge inside 70% ethanol), we think the actual chloride ratio inside the complex should not be very high and thus will not significantly affect the solid-state NMR measurement results. Ca (mg of sample g -1 ) P (mg of sample g -1 ) P (mg of sample g -1 ) Mg (mg of sample g -1 ) Mg (mg of sample g -1 ) Na (mg of sample g -1 )  The atom ratios of different elements in the ds-DNA, the ds-DNA/Ca 2+ complex and CaCO 3 products grown with presence of ds-DNA and slow stirring (100 rpm) measured by EDS and ICP-OES a , respectively. In the Ca containing samples the ratio of Ca was set as 1, while in ds-DNA the ratio of P was set as 1 for the sake of comparison.  On Page 2, Line 37-39: The formation of these non-equilibrium morphologies have been attributed to the liquid-like nature of PILP, being able to wet the solid substrates, or to be capillarity absorbed into the nanopores. (Text Ref. 20,23) Reviewer comments 3: For both the ATR-FTIR and ss-NMR experiments, the authors need to provide more details regarding sample preparation. Are solid precipitates isolated by centrifugation? Are KBr pellets made? Was the solid packed into the NMR tubes? Such sample preparation details are critical to understanding the results.

Ca
Author reply: We thank the reviewer for the kind suggestion.
Revisions made to the manuscript: We have added the information to the Methods part: On Page 22, Line 553-556: Solid state 31 P and 13 C NMR spectra were recorded on a 9.4 T solid-state Varian NMR system (VNMRS) using a Varian 3.2 mm T3-HXY MAS probe, configured in double-resonance mode for 1 H-31 P, respectively 1 H-13 C. The samples (except ds-DNA powder, which was measured as it is) were isolated by centrifugation, dried at room temperature, and directly packed into the rotor without further treatment.
On Page 23, Line 572-574: Attenuated total reflection Fourier transform infrared (ATR-FTIR) spectra of the ds-DNA powder or centrifuge-separated and room temperature dried precipitates were obtained directly on the samples using a Varian FT-IR 3100 Spectrometer with Golden Gate ATR accessory Reviewer comments 4: On page 17, line 388, please provide a reference for the "traditional polyelectrolyte-enhanced wetting of surfaces." Author reply: We thank the reviewer for pointing this out. As discussed following the specific comment 2 of reviewer 1, the ability of PILP to attach to different surfaces should not be only determined by the polyelectrolytes. Furthermore, it is misleading to use the term "wetting" again since PILP is not a liquid phase and thus could not really wet surfaces, but only attach to them.
Revisions made to the manuscript: We have corrected the statement into "the attachment of polyelectrolyte-stabilized colloidal particles to surfaces" on Page 18, Line 432, and has added the following literature as Ref. 66: Böker A, He J, Emrick T, Russell TP. Self-assembly of nanoparticles at interfaces. Soft Matter 3, 1231-1248 (2007).
Reviewer comments 5: On page 7, line 154, please specify in which figure the ds-DNA macromolecules can be seen. Author reply: We thank the reviewer for the suggestion.
Revisions made to the manuscript: We have modified the text accordingly: On Page 7, Line 166, In this case however, also the macromolecules (ds-DNA, highlighted by yellow arrows in Fig. 2d) could be observed, intertwined with the ~2 nm subunits and protruding into the solution.
Reviewer comments 6: In the Methods section, please provide an item number for the track-etch membranes used from VWR. These membranes are notoriously supplier and item number specific, especially as it relates to the surface treatments, which can vary. More information is required to repeat the experiments as described.
Author reply: We thank the reviewer for the suggestion.
Revisions made to the manuscript: We have added the information to the Methods part: On Page 20, Line 471-472, 10 μm thick poly-carbon track etch membranes with 50 nm (item No. 515-2026, supplier No. 110603) and200 nm (item No. 515-2029, supplier number: 110609) sized pores were ordered from VWR.

Reviewer comments 7:
Units: a. Throughout the manuscript, lowercase l's are used for liter instead of uppercase L's. (e.g., g/L instead of g/L) b. Units are not provided for molecular weights. Please specify g/mol Author reply: We thank the reviewer for pointing this out.