**Amino Acids and Peptides as Versatile Ligands in the Synthesis of Antiproliferative Gold Complexes** †

**Tina P. Andrejevi´c 1, Biljana Đ. Gliši´c 1,\* and Miloš I. Djuran 2,\***


Received: 6 March 2020; Accepted: 24 March 2020; Published: 27 March 2020

**Abstract:** Gold complexes have been traditionally employed in medicine, and currently, some gold(I) complexes, such as auranofin, are clinically used in the treatment of rheumatoid arthritis. In the last decades, both gold(I) and gold(III) complexes with different types of ligands have gained considerable attention as potential antitumor agents, showing superior activity both in vitro and in vivo to some of the clinically used agents. The present review article summarizes the results achieved in the field of synthesis and evaluation of gold complexes with amino acids and peptides moieties for their cytotoxicity. The first section provides an overview of the gold(I) complexes with amino acids and peptides, which have shown antiproliferative activity, while the second part is focused on the activity of gold(III) complexes with these ligands. A systematic summary of the results achieved in the field of gold(I/III) complexes with amino acids and peptides could contribute to the future development of metal complexes with these biocompatible ligands as promising antitumor agents.

**Keywords:** gold complexes; amino acids; peptides; cytotoxicity

#### **1. Introduction**

Gold and its compounds have been used for the treatment of a wide range of diseases throughout the history of civilization [1]. The use of gold in modern medicine began with the discovery of the in vitro bacteriostatic properties of the gold(I) complex, K[Au(CN)2], by the German bacteriologist Robert Koch [2]. This gold(I) complex was found to be lethal to the microorganism, *Mycobacterium tuberculosis*, which is causative agent of tuberculosis [2]. After its initial use for tuberculosis with favorable results, serious toxic side-effects were observed for K[Au(CN)2] complex and the treatment was switched to the less toxic gold(I) thiolate complexes (AuSR). The mistaken belief that the *Mycobacterium tuberculosis* was also a causative agent of rheumatoid arthritis led Landé and Forestier to use gold(I) thiolate complexes for the treatment of this disease [3]. After thirty years of medicinal debate, in 1960, British Empire Rheumatism Council finally confirmed the beneficial effects of the gold(I) thiolate complexes against rheumatoid arthritis [3]. Since that time, these complexes have been widely used in the treatment of a variety of rheumatic diseases including psoriatic arthritis, juvenile arthritis, palindromic rheumatism and discoid lupus erythematosus [3]. Nowadays, chrysotherapy is an accepted part the modern medicine and refers to the use of gold-based formulations for the treatment of joint pain and inflammatory diseases [4].

Following the medicinal relevance of gold complexes for the treatment of rheumatoid arthritis, research has continued to uncover the potential of gold complexes as agents for the treatment of cancer [4–13], and various bacterial and fungal infections and tropical diseases, such as malaria, trypanosomiasis and leishmaniasis [14,15]. In some cases, gold complexes were found to be more

active than the clinically used agents, e.g., cisplatin for the cancer treatment [5]. Some gold complexes showed an outstanding in vitro cytotoxicity toward cisplatin-resistant tumor cell lines, which indicates the difference in the mode of action between them and platinum-based agents [5]. Indeed, it was found that the antitumor activity of cisplatin is based on its interaction with DNA, while the antiproliferative activity of gold complexes usually involves the inhibition of enzymes, especially those containing thiol groups, such as thioredoxin reductase (TrxR) [6].

Different classes of ligands have been used for the synthesis of biologically active gold complexes, including phosphines, *N*-heterocyclic carbenes, thiolates, polyamines, pyridine, bipyridine, terpyridine, phenanthroline, and their derivatives, macrocyclic ligands (cyclam), porphyrins and dithiocarbamates [4–15]. Besides them, amino acids and peptides represent two important classes of ligands which are also important as building blocks of proteins and enzymes and show a wide range of biological activities [16]. As constituents of proteins, amino acids and peptides can be considered as biocompatible ligands that can deliver Au(I) ion to its biological target or, as polydentate ligands, they can stabilize the Au(III) ion, preventing its reduction to Au(I) or/and Au(0) under physiological conditions. More importantly, metal complexes with this type of ligand can be more selective toward the abnormal cells in respect to the healthy ones, due to the fact that the abnormal cells overexpress amino acids receptors and need more nutrients [17].

The aim of this review is to present the findings obtained in the field of synthesis and evaluation of gold(I) and gold(III) complexes containing amino acids and peptides moieties for their antiproliferative potential.

#### **2. Gold(I) Complexes Containing Amino Acids and Peptides Moieties**

Considering the great importance of ferrocenyl group in drug design [18], two ferrocene bioconjugates, FcCO-TrpOMe and FcCO-ProNH2 (Fc = ferrocenyl, TrpOMe = methyl ester of tryptophan and ProNH2 = prolinamide) were reacted with an equimolar amount of [Au(acac)(PR3)] (acac = acetylacetonate, PR3 = PPh3, triphenylphosphine or PPh2Py, 2-pyridyldiphenylphosphine) to yield gold(I) complexes, [Au(FcCO-TrpOMe-*N*)(PR3)] (PPh3 (**1**) and PPh2Py (**2**)) and [Au(FcCO-ProNH2-*N*)(PR3)] (PPh3 (**3**) and PPh2Py (**4**)) (Figure 1) [19]. Similarly, the reaction of FcCO-MetOMe with [Au(CF3SO3)(PR3)] (MetOMe = methyl ester of methionine) led to the formation of [Au(FcCO-MetOMe-*S*)(PR3)]CF3SO3 (PPh3 (**5**) and PPh2Py (**6**)) complexes (Figure 1) [19]. The cytotoxicity of ferrocene bioconjugates and corresponding gold(I) complexes **1**–**6** was evaluated by MTT assay against two human tumor cell lines, HeLa (cervical cancer) and MCF-7 (breast cancer), and one murine cell line, N1E-115 (derived from mouse neuroblastoma C-1300) (Table 1). The evaluated gold(I) complexes **1**–**6** appeared to be cytotoxic against these three tumor cell lines, while the corresponding ferrocene bioconjugates used as ligands, FcCO-TrpOMe, FcCO-ProNH2 and FcCO-MetOMe, did not show antiproliferative activity (IC50 > 1000 μM, the IC50 value is defined as concentration required to inhibit tumor cell proliferation by 50% compared to the control cells). The IC50 values of the gold(I) complexes determined after 48 h are in range from 18 to 32 μM in HeLa cells, 15 to 52 μM in MCF-7 cells and < 10 to 54 μM in N1E-115 cells (Table 1). Among the complexes, gold(I) complex **5** with MetOMe and PPh3 in its structure was shown as the most effective against the HeLa cell line; while **4,** having ProNH2 and PPh2Py moieties, displayed the best activity in the murine cell line, although all gold(I) complexes were less cytotoxic than the reference drug doxorubicin, with IC50 values of approximately 1.5 μM [19]. For all compounds, the percentage of cell survival decreased with the increasing of exposure time, although the difference between 24 and 48 h was not found to be significant. Complex **4** induced the cell death through apoptosis and formation of reactive oxygen species (ROS) in tumor cells, while the gold(I) complexes did not act as DNA intercalators.

**Figure 1.** Gold(I) complexes with the ferrocene bioconjugates **1**–**6** showing cytotoxic activity [19].

**Table 1.** In vitro cytotoxic activity (IC50, 48 h, μM) of gold(I) complexes with the ferrocene bioconjugates **1**–**6** [19].


<sup>a</sup> HeLa = cervical cancer, MCF-7 = breast cancer and N1E-115 = mouse neuroblastoma C-1300.

The antiproliferative activity against different tumor cell lines was shown by gold(I) complexes obtained from the reaction of [Au(SPyCOOH)(PR3)] complex (SPyCOOH = nicotinic acid thiolate) by the functionalization of its carboxylic group with different amino acids, ester or amide derivatives of these amino acids or with peptide moieties [20,21]. Gold(I) complexes **7**–**24** of the general formula [Au(SPyCOR)(PPh3)], in which R = methyl ester of amino acid (**7**–**12**), amino acid (**13**–**18**) and amide derivative of the corresponding amino acid (**19**–**24**) have been structurally modified. This modification was performed by changing the type of phosphine ligand in [Au(SPyCOR)(PPh3)] complex (PPh2Py instead of PPh3; complex **25**) and the nature of the coupled amino acid (R) including its structural modification or peptide functionalization (**26**–**32**) and increasing the number of Au(I) ions per complex unit (**33**) (Figure 2a,b) [21]. The antiproliferative activity of these gold(I) complexes was evaluated against three different human tumor cell lines, A549 (lung carcinoma), Jurkat (T-cell leukemia) and MiaPaca2 (pancreatic carcinoma), as well as against non-tumor R69 (lymphoid cell line) and 293T (embryonic kidney fibroblasts), and these cells were exposed to different concentrations of each complex for 24 h (Table 2). As can be seen from this table, the complexes **7**–**24** were active against the investigated tumor cell lines at low micromolar range, with Jurkat cells being the most sensitive; their IC50 values fall in the range from 7.4 to 30.5 μM in A549, 8.2 to 27.2 μM in MiaPaca2 and 2.4 to 7.7 μM in Jurkat cells. Moreover, these complexes exhibited some selectivity for leukemia cells in respect to the non-tumor R69 cells, but this difference was not observed in the case of solid tumors. The cytotoxic activity of the ester complexes **7**–**12** was slightly higher than that of the corresponding precursor, [Au(SPyCOOH)(PPh3)], in all the tested tumor cell lines, with the exception of MiaPaca2 (Table 2). The complexes containing coupled amino acids **13**–**18** and amide derivatives of these amino acids **19**–**24** were, in general, less active than the ester analogues **7**–**12**, although the difference in the activity is not remarkable. The proline-containing complex **12** was found to induce changes in cell and nucleus morphology, loss of the mitochondrial membrane potential, production of ROS and to inhibit the thioredoxin reductase (TrX), an enzyme which acts as a target for biologically active gold(I/III) complexes [21]. Interestingly, gold(I) species **7**–**24** showed much higher antiproliferative activities in vitro in the used cell lines than the cisplatin, the well-known antitumor agent used in medicine for the treatment of various cancers (Table 2) [22].

**Figure 2.** (**a**) Gold(I) complexes **7**–**24** of the general formula [Au(SPyCOR)(PPh3)] in which nicotinic acid thiolate is coupled with methyl ester of amino acid (**7**–**12**), amino acid (**13**–**18**) and amide derivative of the corresponding amino acid (**19**–**24**), and (**b**) complexes **25**–**33** obtained by different structural modifications of [Au(SPyCOR)(PPh3)] complex showing cytotoxic activity [20,21].

The IC50 values of the structurally modified complexes **25**–**33** (Figure 2b) are also at low micromolar concentrations (4.1–33.5 μM in A549, 1.2–29.3 μM in MiaPaca2 and 0.9–36.5 μM in Jurkat cells), with the Jurkat cell line being, in most cases, more sensitive to these complexes than A549 and MiaPaca2 (Table 2). However, for the latter two cell lines, the complexes **25**–**33** are more active than cisplatin (IC50 = 105 and 71 μM in A549 and MiaPaca2, respectively), while, in the case of Jurkat cell line, the IC50 values for the gold(I) complexes and cisplatin are similar (IC50 = 7.4 μM). The change of the type of phosphine ligand coordinated to the Au(I) ion (PPh2Py instead of PPh3; complex **25**) resulted in the same or slightly greater cytotoxicity in Jurkat and R69 cell lines, respectively, while in the remaining cell lines, lower cytotoxicity was observed in respect to the analogue complex **7**, although the differences are not significant. The coupling of lysine ester afforded complex **26** having the good antiproliferative activity, while the removal of the protective Boc (Boc is *tert*-butoxycarbonyl) group in this complex significantly decreased the cytotoxicity of the complex **27**. This was ascribed to the presence of free amino group, which acts as a strong nucleophile and can react with the other biomolecules, preventing the Au(I) ion

to reach the target [21]. The gold(I) complexes having Gly-ProOMe dipeptide (**28**), tertiary amide (**29**), d-amino esters (**30** and **31**) and d-amino acids (**32**) in their structures have shown lower cytotoxicity than the corresponding analogues (Table 2). Among the investigated complexes, the best antiproliferative activity was demonstrated by the dinuclear gold(I) complex **33**, which is functionalized as ester and contains rigid proline as amino acid moiety (Figure 2b). The IC50 values of this complex are found to be in the low micromolar range and even in the submicromolar range in the Jurkat cell line and are also lower than those for cisplatin (Table 2).


**Table 2.** In vitro cytotoxic activity (IC50, 24 h, μM) of thiolate-gold(I) complexes of the general formula [Au(SPyCOR)(PPh3)] in which nicotinic acid thiolate is coupled with methyl ester of amino acid (**7**–**12**), amino acid (**13**–**18**) and amide derivative of the corresponding amino acid (**19**–**24**) and complexes

NT—Non tested; aA549 = lung carcinoma, MiaPaca2 = pancreatic carcinoma, Jurkat = T-cell leukemia, R69 = lymphoid cell line and 293T = embryonic kidney fibroblasts.

A remarkable cytotoxic activity against the same human tumor cell lines (A549, Jurkat and MiaPaca2) was observed for the gold(I) complexes with cysteine-containing dipeptides **34**–**44** (Figure 3a and Table 3) [23]. Starting from the gold(I) complexes **34**–**39** of the general formula [Au(Boc-Cys-XOMe-*S*)(PPh3)] (X = Gly, Ala, Val, Phe, Met and Pro; Boc = *tert*-butoxycarbonyl), different structural modifications of these complexes, such as changes in the phosphine ligand (PPh2Py instead of PPh3; **40**), introducing new amino protecting group (benzyloxycarbonyl (*Z*) instead of Boc; **41**), the use of non-proteinogenic rigid octahydroindole methyl ester (OicOMe; **42**) and increasing the number of Au(I) ions per complex unit (**43** and **44**), were performed in order to investigate the influence of these structural changes on the cytotoxicity of the gold(I) complexes with cysteine-containing dipeptides. As can be seen from Table 3, the IC50 values for the complexes **34**–**44** are in low micromolar range, from 1.5 to 15.6 μM in A549, 0.4 to 2.2 μM in Jurkat and 0.1 to 5.4 μM in MiaPaca2 cells, being much lower than the corresponding IC50 values for cisplatin (105, 7.4 and 71, respectively). The structural modifications leading to the formation of the complexes **40**–**42** (Figure 3a) resulted in almost similar activity against MiaPaca2 and A549 cells, while the introduction of an additional Au(PPh3) <sup>+</sup> moiety to the complex **34** led to the formation of complex **43** (Figure 3a), which is the most potent against MiaPaca2 cell line (Table 3). On the other hand, the coordination of two Au(PPh3) <sup>+</sup> fragments (complex **44**, Figure 3a) did not significantly improve the cytotoxicity of the complex [23].


**Table 3.** In vitro cytotoxic activity (IC50, 24 h, μM) of gold(I) complexes **34**–**51** in A549, Jurkat and MiaPaca2 cell lines [23–25].

**Figure 3.** (**a**) Gold(I) complexes with cysteine-containing dipeptides **34**–**44** [23] and (**b**) with non-proteinogenic 4-mercaptoproline amino acid **45**–**49** [24] showing antiproliferative activity.

*Chemistry* **2020**, *2*

Similar cytotoxic activity to the gold(I) complexes with cysteine-containing dipeptides **34**–**44** was manifested by the gold(I) complexes bearing non-proteinogenic 4-mercaptoproline amino acid (**45**–**49**, Figure 3b and Table 3) [24]. This amino acid is a hybrid of proline and homocysteine, and has the properties of both amino acids, nucleophilic character and reducing properties of the thiol group of homocysteine and rigid structure of proline. *N*-Protected 4-mercaptoproline ester was reacted with [AuCl(PR3)] to yield [Au(Boc-Pro(SH)OMe-*S*)(PR3)] complexes (PR3 = PPh3 (**45**) and PPh2Py (**46**)), while [Au(Boc-Pro(SH)OH-*S*)(PPh3)] complex (**47**) was obtained after the basic hydrolysis of the amino ester moiety in **45**. The latter complex could be further transformed to the [Au(Boc-Pro(SH)-GlyO*<sup>t</sup>* Bu-*S*)(PPh3)] complex (**48**), while the reaction of **45** with [Au(CF3SO3)(PPh3)] afforded the dinuclear [Au2(Boc-Pro(SH)OMe-*S*)(PPh3)2]CF3SO3 complex (**49**) (Figure 3b). As with the case of the abovementioned complexes, the antiproliferative activity of **45**–**49** is mainly due to the Au(I) center. The role of the phosphine ligand is to stabilize this metal ion and to enhance the lipophilicity, allowing the crossing through the membrane, while the thiolate takes part in the substitution reactions with the biomolecules and has an influence on the complex transport or biodistribution [24]. The gold(I) complexes **45, 46, 48** and **49** showed an excellent cytotoxic activity in the investigated human tumor cell lines (A549, Jurkat and MiaPaca2), with IC50 values being lower than 6.1 μM and, in some cases, in the nanomolar range (Table 3). Complex **45** is found to be 100, 23 and 10-fold more active than cisplatin in A549, MiaPaca2 and Jurkat cell lines, respectively, and approximately 2-fold more active than its analogue **46**, which contains PPh2Py ligand instead of PPh3. Similar to the abovementioned gold(I) complexes containing a thiolate ligand functionalized with several amino acids or peptide moieties [21], the formation of the complex with 4-mercaptoproline acid (**47**) decreased significantly the cytotoxic activity, probably as the consequence of higher lipophilicity of the ester group or higher reactivity of carboxylic group, preventing the complex to reach its target [24]. On the other hand, complex **48** with a dipeptide containing 4-mercaptoproline and dinuclear complex **49** have the antiproliferative activity at similar extent to the parent complex **45**.

Two gold(I) complexes **50** and **51** with *N,S*-heterocyclic carbenes derived from the peptides containing l-thiazolylalanine (Thz-Ala) showed good cytotoxic activity in vitro against the A549, MiaPaca2 and Jurkat cell lines (Figure 4a and Table 3) [25]. The carbene complex **50** with iodide was more efficient in all tested cell lines in respect to the thiolate-containing complex **51**, what can be the consequence of the higher lability of Au–I bond in respect to the Au–S bond and higher lipophilicity of **50**. As can be seen from Table 3, the IC50 values of **51** in A549 cell line was higher than 25 μM and approximately 25 μM for the remaining two cells, while **50** showed an excellent cytotoxicity against the A549 cell line with the IC50 value being in the submicromolar range.

The phenylalanine-*N*-heterocyclic carbene gold(I) complex **52** and its amino acid and dipeptide derivatives **53** and **54**, respectively, were evaluated for their in vitro cytotoxic potential against the human cell lines HeLa (human cervix carcinoma), HT-29 (human caucasian colon adenocarcinoma grade II) and HepG2 (human hepatocellular liver carcinoma) (Figure 4b and Table 4) [26]. These three complexes have shown moderate to good antiproliferative activity, with HeLa cells being the most sensitive. Among them, amino acid conjugate complex **53** exhibited the best activity, while the remaining two complexes showed a decrease in antitumor activity, which may be the consequence of differential uptake or different intracellular interactions [26].

**Figure 4.** (**a**) Gold(I) complexes **50** and **51** with *N,S*-heterocyclic carbenes derived from the peptides containing l-thiazolylalanine [25] and (**b**) *N*-heterocyclic carbene gold(I) complexes **52**–**54** [26] showing antiproliferative activity.

**Table 4.** In vitro cytotoxic activity (IC50, μM) of *N*-heterocyclic carbene gold(I) complexes **52**–**54** in HeLa, HT-29 and HepG2 cell lines [26].


#### **3. Gold(III) Complexes Containing Amino Acids and Peptides Moieties**

The abovementioned ferrocene bioconjugate, FcCO-MetOMe, was also used for the synthesis of gold(III) species, [Au(FcCO-MetOMe-*S*)(C6F5)3] (**55**; Figure 5), which was evaluated for in vitro cytotoxicity [19]. The IC50 values determined for this complex of 87 ± 2.0, 88 ± 2.2 and 31 ± 2.4 μM in HeLa, MCF-7 and N1E-115 cell lines, respectively, are higher than the corresponding values for the gold(I) complexes with the ferrocene bioconjugates and phosphine ligands (PPh3 and PPh2Py) (Table 1), indicating that the presence of a phosphine ligand is important for enhancement of cytotoxic potential of gold complexes [19].

A decrease in the antiproliferative activity after oxidation of Au(I) to Au(III) was also observed for the phenylalanine-*N*-heterocyclic carbene gold(III) complex **56** (Figure 5) in HT-29 cell line (IC50 = 125.8 ± 49.7 and 282.5 ± 41.8 μM determined by crystal violet and resazurin assays, respectively) [26].

**Figure 5.** Structural formulas of gold(III) complexes **55** [19] and **56** [26], showing a decrease in the antiproliferative potential in comparison to the analogue gold(I) species.

With the aim to obtain gold(III)-based peptidomimetics with anticancer properties that could target two peptide transporters, PEPT1 and PEPT2, which are upregulated in some tumor cells, Fregona et al. synthesized the complexes of the general formula [AuX2(dtc-Sar-AA-O*<sup>t</sup>* Bu)] (dtc = dithiocarbamate; AA = Gly, X = Br− (**57**)/Cl− (**58**); AA = Aib (2-aminoisobutyric acid), X = Br− (**59**)/Cl− (**60**); AA = L-Phe, X = Br− (**61**)/Cl− (**62**)) (Figure 6) [27]. The in vitro cytotoxicity of these complexes was evaluated toward the human androgen receptor-negative prostate cancer PC3 and DU145 cells, ovarian adenocarcinoma 2008 cells and the cisplatin-resistant C13 cell line, and Hodgkin's lymphoma L540 cells over 72 h, while cisplatin was used as a reference (Table 5). Among these complexes, **61** and **62** were less active, with the IC50 values higher than cisplatin (except C13 cell line), indicating that an aromatic or highly hydrophobic fragment attached to sarcosine decreases the cytotoxicity of the gold(III) complex. On the other hand, complexes **57**–**60** were generally more efficient than cisplatin, with Aib-containing complex **59** being the most active towards the investigated tumor cell lines (Table 5). Importantly, **59** showed 30-fold higher activity than cisplatin in growth inhibition of cisplatin-resistant ovarian adenocarcinoma C13 cells, excluding the occurrence of cross-resistance. Both the most active Aib-containing complexes **59** and **60** exerted their cytotoxic activity within the first 24 h, while the activity of the cisplatin significantly increased with the increasing of the exposure time [27]. The fact that the two most active complexes contain 2-aminoisobutyric acid is not surprising, since this amino acid is abundant in a class of peptide antibiotics, showing anticancer and antiviral properties [28,29]. Moreover, this amino acid plays a crucial role in the biological activity of peptide antibiotics, by forcing the peptide backbone to fold into helical arrangements and providing a capability to cross and/or perturb cell membranes [30]. Apoptosis was shown to be the major mechanism of cell death in the case of prostate cancer PC3 and DU145 cells and ovarian adenocarcinoma C13 cells, for the most active complexes **59** and **60**, while, in the case of the cisplatin-sensitive 2008 cells and the Hodgkin's lymphoma L540 cell line, the majority of dead cells underwent late apoptosis/necrosis over 24 h, after exposure to the these complexes [27]. On the other hand, the remaining gold(III) complexes and cisplatin were less effective in inducing apoptosis.

The same group of authors further synthesized the gold(III)-dithiocarbamato derivatives of oligopeptides, [AuX2(dtc-Sar-L-Ser(*<sup>t</sup>* Bu)-O*<sup>t</sup>* Bu))] (X = Br<sup>−</sup> (**63**)/Cl<sup>−</sup> (**64**)), [AuX2(dtc-AA-Aib2-O*<sup>t</sup>* Bu)] (AA = Sar (sarcosine, *N*-methylglycine), X = Br− (**65**)/Cl− (**66**), AA = D,L-Pro, X=Br− (**67**)/Cl− (**68**)), [AuX2(dtc-Sar-Aib3-O*<sup>t</sup>* Bu)] (X = Br<sup>−</sup> (**69**)/Cl<sup>−</sup> (**70**)), and [AuX2(dtc-Sar-Aib3-Gly-OEt)] (X = Br<sup>−</sup> (**71**)/Cl<sup>−</sup> (**72**)) (Figure 6) and evaluated their cytotoxic activity toward four different cell lines (PC3, 2008, C13 and L540; Table 5) [31].

The IC50 values of the complexes determined after the exposure of L540 cells to the complexes **63**–**72** are in the range 1.4–5.4 μM, being similar to the corresponding value for cisplatin of 2.5 μM. The gold(III) complexes **63**–**68** showed antiproliferative activity comparable to or lower than cisplatin on

prostate cancer and ovarian adenocarcinoma cells, while the tetra- and pentapeptide derivatives **69**–**72** appeared to be less effective. However, against the cisplatin-resistant C13 cell line, all these gold(III) complexes were much more active than cisplatin (Table 5). Among the complexes **63**–**72**, **68** containing proline and 2-aminoisobutiric acid turned out to be the most active toward all the investigated tumor cell lines, having the IC50 values comparable to the abovementioned complex **59** [27,31].


**Table 5.** In vitro cytotoxic activity (IC50, μM, 72 h) of the gold(III)-dithiocarbamato derivatives of oligopeptides **57**–**72** against different tumor cell lines [27,31].

NT—Non tested; <sup>a</sup> PC3 and DU145 cells = the human androgen receptor-negative prostate cancer cells, 2008 cells = ovarian adenocarcinoma and C13 = the parent cisplatin-resistant cell line, L540 = Hodgkin's lymphoma.

Seven gold(III) complexes with different L-histidine-containing dipeptides, [Au(Gly-L-His-*NA,NP,N3*)Cl]Cl.3H2O (**73**), [Au(Gly-L-His-*NA,NP,N3*)Cl]NO3 . 1.25H2O (**74**), [Au(L-Ala-L-His-*NA,NP,N3*)Cl]NO3 . 2.5H2O (**75**), [Au(L-Ala-L-His-*NA,NP,N3*)Cl][AuCl4] . H2O (**76**), [Au(L-Val-L-His-*NA,NP,N3*)Cl]Cl. 2H2O (**77**), [Au(L-Leu-L-His-*NA,NP,N3*)Cl]Cl (**78**) and [Au(L-Leu-L-His-*NA,NP,N3*)Cl][AuCl4] . H2O (**79**) were evaluated for in vitro cytotoxicity against different human tumor cell lines (Figure 7 and Table 6) [32–34]. Different spectroscopic techniques confirmed that tridentate coordination of the X-L-His dipeptides (X = Gly, L-Ala, L-Val and L-Leu) through the N3 imidazole nitrogen (*N3*), deprotonated nitrogen of the amide bond (*NP*) and to the nitrogen of the *N*-terminal amino group (*NA*) stabilized +3 oxidation state of gold, preventing its reduction to Au(I)/Au(0) under physiological conditions. Firstly, complex **73** was tested against the tumor cell line A2780 (human ovarian carcinoma), both sensitive (A2780/S) and resistant (A2780/R) to cisplatin [32]. This complex exhibited a remarkable antiproliferative activity against A2780/S cell line (IC50 = 5.2 ± 1.63 μM), being slightly less active than cisplatin (IC50 = 1.6 ± 0.58 μM). Importantly, the complex retains a significant cytotoxicity on the A2780/R cell line (IC50 = 8.5 ± 2.3 μM), having the resistance factor of only 1.6. The results of this study also showed that the Zn(II), Pd(II), Pt(II) and Co(II) complexes with the same dipeptide manifested only modest activity toward A2780/S and A2780/R cell lines, confirming that the presence of Au(III) ion was crucial for the cytotoxic effects [32].


**Figure 7.** Gold(III) complexes with L-histidine-containing dipeptides **73**–**79**, which were evaluated for the in vitro cytotoxic potential [32–34].

**Table 6.** In vitro cytotoxic activity (IC50, μM) of gold(III) complexes with L-histidine-containing dipeptides **73**–**79** against different cell lines [32–34].


<sup>a</sup> MRC-5 = human lung fibroblasts, MCF-7 = breast cancer, HT-29 = colon cancer, HL-60 = human promyelocytic leukemia, Raji = human Burkitt's lymphoma, HeLa = cervix cancer, and A549 = lung cancer; <sup>b</sup> The results are from three independent experiments, each performed in quadruplicate. SD were within 2%–5%.

In a continuation, the antiproliferative activity of **74** and **75** was evaluated against five human tumor cell lines, MCF-7 (breast cancer), HT-29 (colon cancer), HL-60 (human promyelocytic leukemia), Raji (human Burkitt's lymphoma) and one human normal cell line MRC-5 (human lung fibroblasts) [33]. These complexes are less cytotoxic than cisplatin, with the exception of **74** in the case of the HT-29 cell line (Table 6). The latter complex showed the activity against all tested human tumor cell lines, being non-toxic in the MRC-5 cell line, while the cytotoxicity of L-Ala-L-His-Au(III) complex **75** was not observed against the tested cell line.

In addition, the cytotoxicity of all seven gold(III) complexes with X-L-His dipeptides **73**–**79** was assessed against two human cancer, cervix (HeLa) and lung (A549), cell lines and compared to the activity against the MRC-5 cell line (Table 6) [34]. As can be seen, these complexes did not manifest great anticancer potential; however, their cytotoxicity towards normal cell line was low (IC50 > 100 μM). Moreover, the complexes **76** and **79** showed significant antiangiogenic activity in vivo in a zebrafish embryos model (Figure 8). Although these two complexes achieved comparable antiangiogenic effect to the clinically used auranofin and sunitinib malate at 30-fold higher concentration, the zebrafish embryos following the treatment with Au(III) complexes had no cardiovascular side effects in comparison to those upon treatment with auranofin and sunitinib malate. The binding of the gold(III) complexes to the active sites of both human and bacterial (*Escherichia coli*) thioredoxin reductases (TrxRs) was confirmed by molecular docking study, suggesting that the mechanism of biological action of these complexes can be associated with their interaction with TrxR active site [34].

**Figure 8.** The effect of gold(III) complexes **76** and **79**, sunitinib malate and auranofin on subintestinal vessels (SIVs), intersegmental vessels (ISVs) and dorsal longitudinal anastomotic vessels (DLAVs) development in zebrafish embryos. Reduced SIVs (arrowhead), disrupted DLAVs (asterisk), thinner or reduced ISVs (arrow), and pericardial edema (dashed arrow) are designated. The Figure was adapted from the Reference [34].

Additionally, the antimicrobial activity of the abovementioned gold(III) complexes with l-histidine-containing dipeptides **73**–**79** were evaluated against the Gram-positive (*Staphylococcus aureus*, *Listeria monocytogenes, Enterococcus faecalis* and *Enterococcus faecium*) and Gram-negative (*Acinetobacter baumannii*) bacteria and two strains of *Candida* (*C. albicans* and *C. parapsilosis*) [34]. In most cases, the minimal inhibitory concentration (MIC) values were between 200 and 400 μM, indicating their moderate to low activity. Nevertheless, the MIC values of **74** and **79** against Gram-positive *E. faecium* and Gram-negative *A. baumannii* were found to be 80 and 100 μM, respectively. Beside these results, a study related to the antimicrobial potential of a series of gold(III) complexes differing in the ligand structure, showed that two gold(III) complexes with L-histidine-containing dipeptides **74** and **75** exhibited relatively weak effects in comparison to the other studied complexes against the investigated bacterial strains [35].

#### **4. Conclusions**

The review article summarizes the results achieved in studies on the antiproliferative activity of gold(I) and gold(III) complexes containing amino acids and peptides as biocompatible ligands that can deliver a gold ion to its target. Except the biocompatibility of amino acids and peptides, the advantage of these ligands could be the fact that the tumor cells overexpress amino acids receptors, resulting in the selectivity of the obtained gold complexes toward the tumor cells in respect to the healthy ones. From the presented data, it could be seen that a large number of gold(I) and gold(III) complexes have been evaluated in vitro against different tumor cell lines, while the in vivo studies are rather scarce. In general, gold(I) complexes have shown higher cytotoxic activity than the gold(III) species, being, in some cases, superior in respect to the well-known anticancer agent, cisplatin. Except amino acids and peptides, the gold(I) complexes usually had phosphines and *N*-heterocyclic carbenes ancillary ligands, which also contributed to their biological properties.

Gold(III) complexes with L-histidine-containing dipeptides, in which *N*-terminal amino acid is L-alanine and L-leucine, showed selectivity in terms of cancer vs. normal cell lines and achieved antiangiogenic effects comparable to the known inhibitors of angiogenesis—auranofin and sunitinib malate—without toxic-side effects, in contrast to those following auranofin and sunitinib malate treatment. These findings make them good candidates for the further development of antiangiogenic drugs.

From this review, it can be concluded that the use of amino acids and peptides as ligands for the synthesis of biologically active gold complexes has merit for the development of novel therapeutic agents for the treatment of cancer, which is a major burden of disease worldwide.

**Funding:** This research has been financially supported by the Ministry of Education, Science and Technological Development of the Republic of Serbia (Agreement No. 451-03-68/2020-14) and by the Serbian Academy of Sciences and Arts under strategic projects program—grant agreement No. 01-2019-F65 and project of this institution No. F128. The authors wish to thank Dr. *Jasmina Nikodinovic-Runic* (Institute of Molecular Genetics and Genetic Engineering, University of Belgrade) for her valuable comments on the manuscript.

**Conflicts of Interest:** The authors declare no conflicts of interest. The funders had no role in the collection, analyses or interpretation of the data, in writing of the manuscript and in the decision to publish the manuscript.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Natural and Engineered Electron Transfer of Nitrogenase**

**Wenyu Gu <sup>1</sup> and Ross D. Milton 2,\***


Received: 8 April 2020; Accepted: 23 April 2020; Published: 27 April 2020

**Abstract:** As the only enzyme currently known to reduce dinitrogen (N2) to ammonia (NH3), nitrogenase is of significant interest for bio-inspired catalyst design and for new biotechnologies aiming to produce NH3 from N2. In order to reduce N2, nitrogenase must also hydrolyze at least 16 equivalents of adenosine triphosphate (MgATP), representing the consumption of a significant quantity of energy available to biological systems. Here, we review natural and engineered electron transfer pathways to nitrogenase, including strategies to redirect or redistribute electron flow in vivo towards NH3 production. Further, we also review strategies to artificially reduce nitrogenase in vitro, where MgATP hydrolysis is necessary for turnover, in addition to strategies that are capable of bypassing the requirement of MgATP hydrolysis to achieve MgATP-independent N2 reduction.

**Keywords:** nitrogenase; ammonia; metalloenzyme; electron transfer; ferredoxin; flavodoxin; Fe protein; MoFe protein

#### **1. Introduction to Nitrogenase**

Ammonia (NH3) is an important commodity for agricultural and chemical industries that is currently produced at over 150 million tons per year [1,2]. Currently, the majority of this NH3 is produced from molecular hydrogen (H2) and kinetically inert dinitrogen (N2, bond dissociation enthalpy of +945 kJ mol−1) by the Haber–Bosch process, which operates at a high temperature (~700 K) and a high pressure (~100 atm) in order to optimize NH3 production [3]. These conditions, in combination with the production of H2 (commonly by steam reforming of natural gas), result in the consumption of 1–2% global energy and the production of around 3% of global carbon dioxide (CO2) emissions. Due to ever-increasing concerns and awareness of climate change, there is significant interest in the development of new catalysts for N2 fixation. For instance, the development of a new (bio)catalytic system that operates under mild conditions could enable the decentralization of NH3 production as a key strategy for improved environmental sustainability.

Select bacteria and archaea are able to produce an enzyme, nitrogenase, which can fix N2 to NH3 under mild (physiological) conditions [4]. Thus, nitrogenase is of interest to new biotechnologies and new bio-inspired N2-fixing catalysts. Nitrogenase is a two-component metalloenzyme consisting of a reductase (iron or "Fe" protein) and a N2-reducing protein (MoFe protein), where the name "MoFe" refers to the metals employed in its catalytic cofactor. There are two alternative nitrogenases that are dependent on V (VFe) and Fe only (FeFe), although the Mo-nitrogenase system from the soil bacterium *Azotobacter vinelandii* will serve as the model to outline nitrogenase's mechanism (Figure 1) [5].

**Figure 1.** (**Left**) Representation of Mo-nitrogenase from *Azotobacter vinelandii*. The Fe protein is shown in gray and the MoFe protein is shown in cyan (NifD, α-subunit) and olive-green (NifK, ß-subunit). This representation was adapted from PDB:4WZA, which used non-hydrolyzable ATP analogues to form a tight Fe:MoFe protein complex. (**Right**) The FeS cofactors of Mo-dependent nitrogenase, where the coordinating residues are also shown. The homocitrate partner of the FeMo-co is shown on the left of the FeMo-co. Fe = rust, S = yellow, C = gray, N = blue, O = red, Mo = cyan, Mg = green.

The Fe protein of *A. vinelandii* is a homodimer of approximately 66 kDa in mass encoded by the *nifH* gene. Each dimer contains an adenosine triphosphate (MgATP) binding site, whereas a [4Fe-4S] iron-sulfur cluster is located between each monomer and coordinated by two cysteine (Cys) residues from each monomer [6]. The MoFe protein is a α2ß2 tetramer of approximately 240 kDa encoded by the *nifD* (α subunit) and *nifK* (ß subunit) genes. Each αß dimeric half contains a [8Fe-7S] "P" cluster that bridges both subunits (coordinated by Cys residues from each subunit) and a [7Fe-9S-C-Mo-homocitrate] FeMo-cofactor ("FeMo-co") contained within the α subunit [7,8]. The FeMo-co, the cofactor at which N2 is reduced, is coordinated by a Cys and a histidine (His) residue. Each αß half of the MoFe protein repeatedly transiently associates with a 1e−-reduced and ATP-bound Fe protein, during which a single electron is ultimately transferred from the [4Fe-4S]1<sup>+</sup> cluster to the FeMo-co *via* the P cluster [9]. The order of events that transpire during this transient association is debated, although electron transfer, 2MgATP hydrolysis (to adenosine diphosphate, 2MgADP), the release of two inorganic phosphate (Pi) equivalents, and Fe protein dissociation occur. Oxidized Fe protein is subsequently reduced by flavodoxin and/or ferredoxin in vivo or commonly dithionite (DT) in vitro, thereby restarting the "Fe protein cycle" [10,11]. The reduction potential (*E*o') of the [4Fe-4S]2+/1<sup>+</sup> couple of the Fe protein is approximately <sup>−</sup>0.30 V vs. the standard hydrogen electrode (SHE, see [12,13]), free of MgATP or MgADP [9,14]. Upon the association of MgATP, the *E*o' of [4Fe-4S]2+/1<sup>+</sup> is modulated to <sup>−</sup>0.43 V vs. SHE, which decreases further to <sup>−</sup>0.62 V when the Fe protein associates with the MoFe protein making it a more potent electron donor [14,15]. Perhaps the two most-relevant redox couples of the P cluster (PN/1<sup>+</sup> and P1+/2+) both have *E*o's of approximately <sup>−</sup>0.31 V vs. SHE. Further, the *<sup>E</sup>*o' of the as-isolated FeMo-co (MN) and its one-electron oxidation production (MOX) is approximately <sup>−</sup>0.04 V vs. SHE, while a second lower-potential redox couple (MRED) is also thought to be relevant to nitrogenase's catalytic cycle, although its *E*o´ has not been accurately determined [9].

Electron transfer from the Fe protein is currently thought to be coupled to the rate-limiting step of Pi release (25–27 s<sup>−</sup>1), suggesting electron transfer takes place with a rate constant of around 13 s−<sup>1</sup> [16]. Along with a study by Duval et al. in 2013, this suggests that electron transfer from the [4Fe-4S]1<sup>+</sup> cluster to the P cluster takes place prior to MgATP hydrolysis [17]. Finally, a deficit spending model has been proposed by which an electron is transferred from the P cluster to the FeMo-co, prior to the P cluster being reduced by the Fe protein's [4Fe-4S]1<sup>+</sup> cluster [18,19]. Nevertheless, the proposal that electron transfer from the Fe protein's [4Fe-4S] cluster takes place prior to the hydrolysis of MgATP suggests that artificial electron transfer to the P cluster (independent of the Fe protein) is

possible. Thus significant interest has developed concerning the MgATP-independent fixation of N2 by nitrogenase, given that the hydrolysis of each MgATP accounts for around <sup>−</sup>50 kJ mol−<sup>1</sup> in vivo [20].

The reduction of N2 to NH3 requires 6e−, although optimal N2 fixation by nitrogenase occurs after eight transient association events of the Fe protein (and the transfer of 8e−):

$$\rm{NH\_2 + 8e^- + 8H^+ + 16MgATP \to 2NH\_3 + H\_2 + 16MgADP + 16P\_i} \tag{1}$$

Notably, the fixation of each N2 also results in the evolution of at least one equivalent of H2. Lowe and Thorneley developed an early model to describe the observation of H2 formation, which has since developed into a model that highlights the pivotal nature of a 4e−-reduced FeMo-co known as the E4 state [8,10,21–23]. By this model, the resting FeMo-co in its E0 state accumulates individual electrons and protons during each Fe protein association event in order to reach the E4 state at which N2 binds and undergoes subsequent reduction. These electrons are stored as metal-hydrides (M-H) on the Fe centers. Prior to the binding of N2, the E2-4 states can unproductively evolve H2 by M-H protonation, thereby "dropping" by Ex <sup>−</sup> <sup>2</sup> states [21,22]. In contrast, the productive evolution of H2 is thought to occur in order to accommodate N2 binding and its partial reduction, by the reductive elimination (*re*) of H2 from the FeMo-co E4 state [24]. At this stage, H2 can also undergo oxidative addition (*oa*) to the FeMo-co and displace N2, which is also considered to be unfavorable [22]. Four additional Fe protein association events (transferring 4e− and hydrolyzing an additional 8MgATP) then lead to the production of 2NH3. This serves as a suitable model to explain the production of one equivalent of H2 for each N2 reduced [21,22]. Thus, H2 can be produced by unproductive or productive pathways, with increased H2 evolution and MgATP hydrolysis resulting from a combination of both. Questions therefore remain surrounding the reversibility or catalytic bias of the E4 state, given that the alternative nitrogenases also appear to follow the same *re* mechanism while appearing to be less-efficient at N2 fixation [5,25]. A "just-in-time" mechanism could serve as a useful model to justify the rate-limiting nature of electron transfer from the Fe protein (~13 s<sup>−</sup>1) such that unproductive H2 formation and the *oa* of H2 is minimized [23]. Finally, research has also questioned the suitability of the proposed *re*/*oa* model for N2 fixation by nitrogenase [26]. Density functional theory calculations were employed to calculate the partial charges of the FeMo-co during key turnover states, where a *re* mechanism was not supported and the possible involvement of hydrogen atoms (including hydrogen atom transfer, formation, and elimination steps) was highlighted.

In summary, it is clear that electron transfer within the nitrogenase complex, and by extension, to the nitrogenase complex, is of high importance to biological N2 fixation as well as new biotechnologies and bio-inspired N2 fixation systems. Thus, this article reviews natural and engineered electron transfer to nitrogenase, in the contexts of both in vivo and in vitro catalysis. Specifically, we discuss the nature and delivery of electrons to nitrogenase in vivo. We also review approaches to transfer electrons to nitrogenase in vitro, either through the Fe protein or independent of the Fe-protein for MgATP-decoupled catalysis by the MoFe protein.

#### **2. In Vivo Electron Transfer to Nitrogenase**

#### *2.1. Electron Transfer from Ferredoxin and Flavodoxin*

In vivo reduction of Fe protein requires low-potential electrons provided by the electron-transferring proteins flavodoxin and ferredoxin [27,28]. In vitro studies of nitrogenase typically utilize dithionite (DT, *<sup>E</sup>*o' ≈ −0.66 V vs. SHE) or Ti(III) citrate (*E*o'Ti(III)/(IV) <sup>=</sup> <sup>−</sup>0.8 V vs. SHE) as electron donors due to their ease of preparation and use [29–31]. However, recent studies have reevaluated electron transfer to the Fe protein with flavodoxin, which is crucial for a mechanistic understanding of the Fe protein cycle and of nitrogenase activity in physiological conditions [16]. It is also important to consider physiological electron donors when developing biological systems to maximize electron transfer to nitrogenase.

Flavodoxin (Fld) is a monomeric electron-transfer protein carrying a non-covalently bound flavin mononucleotide (FMN) as a redox center. It consists of a central five-stranded parallel β-sheet flanked on either side by α-helices. Based on the presence of a ∼20-residue loop splitting the fifth β-strand, Flds involved in electron transfer to nitrogenase are classified as long-chain flavodoxins [32,33]. The cofactor FMN has three pertinent redox states: oxidized quinone (Ox), semiquinone (SQ), and hydroquinone (HQ), where FMNOx/SQ and FMNSQ/HQ couples are 1e<sup>−</sup> redox reactions. Although not overly abundant in aqueous solution, FMNSQ is stabilized when bound to flavodoxin; further, crossed-potentials of flavins can be stabilized in proteins and can support a process named flavin-based electron bifurcation (reviewed in [34–36]). *E*o's of Fld II from *A. vinelandii* range from <sup>−</sup>0.25 to <sup>−</sup>0.1 V (vs. SHE) for the FldOx/FldSQ couple and from <sup>−</sup>0.5 to <sup>−</sup>0.4 V (vs. SHE) for the FldSQ/FldHQ couple [32]. The latter is one of the lowest reported potentials in the flavodoxin family [37]. Because the *E*o' of the Fe protein is <sup>&</sup>lt;−0.3 V (vs. SHE), the FldHQ state is expected to be the functional electron donor [27,38]. In *A. vinelandii*, Fld (*E*o'SQ/HQ <sup>=</sup> <sup>−</sup>0.46 V vs. SHE) is encoded by the gene *nifF*.

Ferredoxin (Fdx) is an FeS cluster-containing protein that was first isolated from *Clostridium pasteurianum* [39]. Fdxs coordinate FeS clusters by Cys ligands and can be divided into different groups based on the number and types of FeS clusters [40]. Fdxs that participate in electron transport to Fe protein are found to have [2Fe-2S]-type [41], [4Fe-4Fe]/[3Fe-4S]-type, or 2[4Fe-4S]-type clusters [42–44]. The [2Fe-2S]-type is normally found in cyanobacteria with the corresponding gene named *fdxH,* while the other two are found in diverse diazotrophic groups and are usually designated as *fdxN*. The possible redox states and redox potentials of Fdxs are modulated by their peptide structure (and ultimately, their coordination spheres) and occur in the ranges of −0.24 to −0.46 V, −0.05 to −0.42 V, and <sup>−</sup>0.28 to <sup>−</sup>0.68 V, for the couples [2Fe-2S]2+/1<sup>+</sup>, [3Fe-4S]1+/0, and [4Fe-4S]2+/1+, respectively (vs. SHE) [45]. In comparison to Fld, Fdx is more sensitive to O2 due to the lability of their FeS clusters. Phylogenetic analysis indicates that Fld (NifF) is enriched in diazotrophs with aerobic or facultative anaerobic life styles, and is believed to be an adaptive strategy for the diversification of Nif-nitrogenase from anaerobic to aerobic taxa during evolution [46]. Although other Fdxs are found to be involved in the assembly of nitrogenase cofactors or for protecting it against oxygen [47,48], the following discussion focuses on Fdxs involved in electron transport.

Fld forms a tight complex with Fe protein with high affinity (Figure 2). Reported dissociation constants for the pairs from *Klebsiella pneumoniae* are 13 μM (MgATP-bound Fe protein) and 49 μM (MgADP-bound Fe protein) and from *Rhodobacter capsulatus* is 0.44 μM. [49,50] It is generally believed that in the Fe protein cycle, reduction of Fe protein and exchange of MgADPs for MgATPs takes place after its dissociation from MoFe protein, implying that the oxidized Fe protein cannot be reduced when bound to MoFe protein. Further, the Fe protein is believed to interact with Fld/Fdx using the same binding interface that is used with the MoFe protein [16,33,51,52]. This hypothesis is guided by docking models that predict electrostatic interactions between the positively charged surroundings of the [4Fe-4S] cluster of Fe protein and the negatively charged surroundings of FMN of NifF from *A. vinelandii*, possibly assisted by an eight amino acid loop (residues 64–71) on NifF [16,32,52] (Figure 2). The MgADP-bound state of the Fe protein has the most complementary docking interface with Fld compared with the MgATP-bound state. Experimentally, this hypothesis is supported by results from cross-linking studies and time-resolved limited proteolysis using NifF and Fe protein from *A. vinelandii* [16,52]. Similar modeling results were obtained for Fdx, although experimental support has not yet been reported [33,41].

Early studies indicated that the specific catalytic activity of MoFe protein [27,51] and ATP/e− efficiency of nitrogenase are higher when NifF is used as the reductant of Fe protein rather than DT. Second-order rate constants of Fe protein reduction are two to three orders of magnitude times higher when NifF is used [53] or included [16] as reductant compared to DT [16] or Ti (III) [53]. Both physiological reductants and chemical reductants reduce MgADP-bound Fe protein faster than its nucleotide-free form. Nevertheless, recent reexamination of the rate difference under pseudo-first order reaction conditions showed similar trends [16]. The diminished performance of DT might be partially due to the slow generation of radical anion SO2 •− (Kd 1.5 nM, rate constant of <sup>∼</sup>2 s<sup>−</sup>1) [16].

**Figure 2.** Docking models for reduction of the Fe protein by NifF and Fdx (FdI). (**a**) MgAMPPCP-Fe protein (top, PDB: 4WZB) and MgADP-Fe protein (bottom, PDB:1FP6) interacting with NifF (PDB:1YOB). (**b**) MgAMPPCP-Fe protein (PDB:4WZB) interacting with NifF (PDB:5K9B) and (**c**) FdI (PDB:6FDR). Hypothetical electron transfer distances between the [4Fe-4S] cluster of the Fe protein (NifH) and the FMN cofactor of NifF are shown. Reprinted (adapted) with permission from [9]. Copyright 2020 American Chemical Society. Republished with permission of American Soc for Biochemistry and Molecular Biology Copyright 2017, from [52]; permission conveyed through Copyright Clearance Center, Inc. Adapted with permission from [33]. Copyright 2017 John Wiley and Sons.

The findings above help revise kinetic models of the Fe protein cycle, as the Fe protein reduction rate has been used to estimate its dissociation from MoFe protein. When using DT, the complex dissociation rate was measured as <sup>∼</sup>6 s−<sup>1</sup> [11,51] and was believed to be the limiting step of the Fe protein cycle. Yet a high rate of 759 s−<sup>1</sup> was obtained when NifF was used, indicating that Pi release (25–27 s<sup>−</sup>1) is actually the rate-limiting step [16]. In line with this, the specific activity of nitrogenase was shown to increase by 50–170% with the presence of NifF as compared to DT alone [16,51].

Electron transfer efficiency to nitrogenase is expected to be improved when using physiological electron donors. While the 1e<sup>−</sup> reduced [4Fe-4S]1<sup>+</sup> of Fe protein is commonly believed to be the only physiologically relevant state [6], Watt and Reddy discovered that the [4Fe-4S]1<sup>+</sup> cluster can be further reduced by 1e<sup>−</sup> to a stable all-ferrous [4Fe-4S]0 state by methyl viologen (MV) [54]. An *<sup>E</sup>*o' of <sup>−</sup>0.46 V vs. SHE for the 1+/0 couple was reported [54]. Since, many studies have characterized the [4Fe-4S]0 state of Fe protein reduced by Ti(III) or Eu(II), Eu(II) complexes employed to study nitrogenase typically have *E*o's ranging from <sup>−</sup>0.6 to <sup>−</sup>1.1 V vs. SHE [53,55–61]. One study reported an *E*o' of <sup>−</sup>0.79 V for the [4Fe-4S]1+/<sup>0</sup> couple [62]. Lowery et al. (2006) showed MgADP or MgATP-bound Fe protein can be reduced by FldHQ state of NifF (*A. vinelandii*) from [4Fe-4S]<sup>1</sup> to [4Fe-4S]<sup>0</sup> independent of catalysis, which supports a *<sup>E</sup>*o' of <sup>−</sup>0.46 V of the [4Fe-4S]<sup>0</sup> state and the possibility of its physiological relevance [63]. The Fe protein's [4Fe-4S]<sup>0</sup> state could allow two electrons to be transferred from FldHQ to Fe protein per two ATP molecules hydrolyzed, reaching a 1:1 ATP:e− ratio. This was indeed observed in a few studies using either NifF or Ti(III) [53,63,64]. Yet in most studies the ATP:e− ratio remained at 2:1, even when NifF was used as the reductant [9,16]. In contrast, DT only reduces the [4Fe-4S] cluster to the [4Fe-4S]1<sup>+</sup> state, and thus transfers only one e<sup>−</sup> per transient association cycle.

These findings call for further investigation into nitrogenase's natural electron donors. Identifying Fld/Fdx that directly transfers electrons to nitrogenase and to what extent could be difficult due to their redundancy in both genome and function; it is common for a bacterial or archaeal genome to encode multiple Fld/Fdxs [65]. Due to the high energetic expense of N2 fixation, nitrogenase genes, including the ones encoding for electron transport, are often co-located and/or transcriptionally co-regulated [66,67]. Mutagenesis combined with nitrogenase activity assays in cell extracts can provide direct proof of a gene's function in electron transfer to nitrogenase. However, most diazotrophs have more than two Fld/Fdxs capable of direct electron-transfer to nitrogenase (Table 1). In rare cases such as in *K. pneumoniae*, a sole Fld NifF is the electron donor and diazotrophic growth is abolished in *nifF* deletion mutants. In comparison, *A. vinelandii* is still able to grow diazotrophically (though much is undermined) when the two electron transport components *nifF* and *fdxA* are deleted [68]. In phototrophic bacteria, Fld is commonly found to serve as an electron donor under iron depleted conditions, whereas Fdx is the main electron donor under iron replete conditions [69].


**Table 1.** Electron transport components required for nitrogenase in representative diazotrophs. Gene names are shown with protein names, if available.

<sup>1</sup> These genes are expressed by cells growing under iron depleted conditions. <sup>2</sup> These genes are expressed by cells growing under anaerobic conditions in the dark.

#### *2.2. Electron Transfer to Flavodoxin and Ferredoxin*

The electrons transferred to Fld/Fdx directly come from pyruvate, NAD(P)H, or hydrogen. Five major enzyme systems capable of reducing Fld/Fdx have been identified (Figure 3, Table 1) and are discussed below.

**Figure 3.** Schematic pathway of natural electron transfer to nitrogenase. (Fd: ferredoxin or flavodoxin).

Pyruvate-fld/fdx oxidoreductase (PFOR) catalyzes oxidization of pyruvate to acetyl-CoA and CO2 with reduction of Fld or Fdx by using thiamine pyrophosphate and FeS clusters as cofactors. The *E*o' for the pyruvate cleavage is −0.5 V (vs. SHE). PFOR involved in electron transport to nitrogenase is commonly found as part of *nif* operon designated as *nifJ*. The deletion of *nifJ* in *K. pneumoniae* was found to abolish diazotrophic growth [86–88].

Fdx-NADP<sup>+</sup> Reductase (FNR) catalyzes the reversible reaction of Fdx oxidation with NADP<sup>+</sup> reduction by a flavin adenine dinucleotide (FAD) cofactor. Fdxs can possess *E*o's more negative than NADPH/NADP<sup>+</sup> (−0.34 V vs. SHE), and the forward reaction (*i.e.*, Fdx oxidation) is energetically favorable. Yet, FNRs that favor the reverse reaction are found in *A. vinelandii* and cyanobacteria and are thought to support nitrogenase activity based on enzyme assay and gene coregulation [70,71,80].

Rhodobacter Nitrogen Fixation (Rnf) complex is a transmembrane ferredoxin—NAD<sup>+</sup> oxidoreductase, which catalyzes the NADH-dependent reduction of Fld/Fdx with the a depletion of electrochemical gradient [110]. The Rnf complex shares homology with Na<sup>+</sup> pumps and is distributed among diverse N2-fixing and non-nitrogen-fixing microbes [29,110]. The disruption of genes in the *rnf* operon of *R. capsulatus* resulted in a significant decrease in nitrogenase activity and abolished its ability to grow diazotrophically [94,95].

Bifurcating FixABCX couples the endergonic reduction of Fld to the exergonic reduction of coenzyme Q (*E*o' = +0.01 V vs. SHE) (then to respiratory chain) at the cost of NADH by using a flavin at the bifurcating site [81,82,91,104]. This membrane protein complex was identified to have three FAD moieties, one each in FixA, FixB, and FixC, and two [4Fe-4S] clusters in FixX [81]. Disrupting the Fix system in *Rhodopseudomonas palustris*, *Rhodospirillum rubrum*, and *Sinorhizobium meliloti* completely abolishes or significantly impairs their ability to grow under N2 fixing conditions [92,102,104]. A double mutant of Fix and RNF1 abolished diazotrophic growth in *A. vinelandii* [81].

Uptake hydrogenase (hup) is a heterodimer of the *hupA* and *hupB* (or *hupS* and *hupL*) gene products that has either a Ni-Fe or Fe-Fe active center [111]. It recaptures H2 produced during N2 fixation (Equation (1)) for use as electron donors cycled back to nitrogenase [96], and also seems to protect nitrogenase from oxygen, possibly by catalyzing oxyhydrogen reaction [112,113]. Overproduction of H2 from nitrogenase was achieved by deleting the uptake hydrogenase [72,114]. The *hup* genes are co-regulated with nitrogenase through complicated and diverse regulatory pathways [83,84,97,103], which was exploited to identify H2-overproducing variants of the Fe protein [115].

The diverse strategies to reduce Fld/Fdx by different microbes can be better understood if put into metabolic context (Table 1). Bioinformatic analysis indicates proteins that reduce Fld/Fdx with H2 or pyruvate are enriched in anaerobes, while those with NADH/NADPH are enriched in aerobes, facultative anaerobes, and anoxygenic phototrophs [116]. Aerobic, facultatively anaerobic, and anoxygenic phototrophic diazotrophs produce NADH/NADPH, which is not considered reducing enough to provide electrons for nitrogenase; thus, FixABCX and Rnf are acquired to generate low potential electrons [46]. Specifically, oxygenic phototrophs can energize electrons to low potential using photosystem I (PSI), but the low potential electrons are not available for nitrogenase, as the latter must be spatially (forming heterocysts) or temporally (under the control of circadian clocks) separated from PSI, in which case FNR is used [70]. Interestingly, some non-heterocystous cyanobacteria show nitrogenase activity only under light conditions, implying the existence of molecular mechanisms to protect nitrogenase against oxygen evolved by photosystems [113,117,118].

#### **3. Electron Transfer to Nitrogenase in Engineered Biological Systems**

Developing and improving biological systems for N2 fixation is greatly motivated by the need to design sustainable and economic solutions to nitrogen limitation of crop productivity. Decentralization of NH3 production is considered one possible strategy to improve environmental sustainability. Great efforts have been made towards engineering plant-associated microbes or plants themselves to heterologously express nitrogenases [4]. From the perspective of synthetic biology, genes required for functional nitrogenase can be grouped into modules of structural genes, genes involved in biosynthesis

and maturation of cofactors, and electron-transport components (ETC) [119]. ETC include Fld/Fdx and the reductase of Fld/Fdx. ETC can come from the native host to nitrogenase or be substituted by homologues from the expression host, if compatible (Table 2). In general, the former results in higher nitrogenase activity, while the latter gives the advantage of expressing fewer genes and simplifying genetic engineering [120,121]. It is important to bear in mind that there is limited crosstalk between Fld/Fdx and the Fe protein, and between Fld/Fdx and their reductases from different hosts. For example, Fdx from either *Clostridium pasteurianum* or *R. rubrum* were ineffective in coupling pyruvate oxidation to nitrogenase activity in the cell lysate of *K. pneumoniae* NifF mutant. NifF from *A. vinelandii* was only one-third as effective as NifF from *K. pneumoniae* at transferring electrons from PFOR to *Klebsiella* nitrogenases [88]. Further, in vivo activity could not be predicted by in vitro assays. For example, Fdxs from *R. capsulatus* and *S. meliloti* equally support in vitro acetylene (C2H2) reduction to ethylene (C2H4) by *R. capsulatus* nitrogenase, yet heterologous in vivo complementation by each other's Fdx was unsuccessful [122,123].


**Table 2.** Summary of heterologous expression of active nitrogenase in prokaryotic hosts.

<sup>1</sup> MoFe-type nitrogenase (*nif*) was expressed in these studies unless otherwise noted. <sup>2</sup> Structural genes of Fe-only nitrogenase were expressed in combination with accessory genes of MoFe-type nitrogenase. <sup>3</sup> Other combinations of *nif* genes and expression hosts were explored in this study but are not listed here.

By now, there has been greater success in expressing active nitrogenases in prokaryotic hosts (Table 2). Since Dixon and Postgate conjugated N2 fixation genes from *Klebsiella oxytoca* into *Escherichia coli* in 1972 [129], nitrogenases from various sources have been cloned and expressed in different hosts (Table 2)*. E. coli* has been used as a platform to identify minimal gene requirements for active nitrogenase [121,124] and to optimize expression [125,126,130]. It was found that *fldA* (homologue to *nifF*) and *ydbK* (homologue to *nifJ*) in *E. coli* can support nitrogenase activity, although activity is significantly improved when *nifFJ* from diazotrophs are employed [121,124]. Non-diazotrophic, oxygenic cyanobacteria have also been explored as expression hosts, as they serve as a simpler model of plant chloroplasts [113,118]. The O2 tolerance of nitrogenase was enhanced by co-expressing an uptake hydrogenase in *Synechocystis* sp. PCC 6803 [113]. Epiphytic and endophytic bacteria are desirable platforms for N2 fixation, as they can be directly applied. Challenges here remain in the fact that different plants have specific colonizers, and that most natural diazotrophic endophytes do not express nitrogenase under the desired conditions or to a desired level [131,132]. A recent study by Ryu et al. applied different strategies to a wide range of hosts and engineered inducible promoters in combination with nitrogenase genes to tackle both of those problems [128]. Overall, heterologous nitrogenases showed significantly lower activities compared to that in their native hosts with the exception of MoFe nitrogenase from *K. oxytoca* expressed in *E. coli* [125,126,128].

Expressing nitrogenases in eukaryotic systems—either plant or yeast and green algae—as simpler models, have limited success [4]. Separate Nif components have been expressed in these systems but the formation of a fully functional nitrogenase has not been achieved [4]. Both mitochondria and chloroplasts are candidate organelles to host the expression of nitrogenase. The former provides MgATP and a low O2 environment due to active respiration, while the latter produces abundant reduced Fdx and MgATP by photosynthesis with the byproduct of O2 [4,133]. Several lines of evidence suggest that chloroplasts might be a more suitable choice. While both locations have Fdx-FNR-type electron transport modules, the ones in chloroplasts function as part of the photosynthesis pathway and share more similarity to those in diazotrophic bacteria. In mitochondria, the Fdx-FNR homologue modules are called adrenodoxin and NADPH-dependent adrenodoxin oxidoreductase as part of biosynthesis pathway of biotin [134–136]. Yang et al. (2017) used *E. coli* as a chassis to study the compatibility between MoFe and FeFe nitrogenases with ETC modules from plant chloroplasts, root plastids, and mitochondria. They found that Fdx-FNR from chloroplasts and root plastids can support the activities of both types of nitrogenase, while the ETC module from mitochondria could not. A hybrid module of mitochondrial FNR and the cyanobacteria Fdx could support nitrogenase activities [137]. In addition, chloroplast genomes of some plants and algae encode a nitrogenase-like enzyme called dark-operative protochlorophyllide oxidoreductase (DPOR) that participates in biosynthesis of chlorophylls and is also O2 sensitive [138]. It was demonstrated that the Fe protein from *K. pneumoniae* nitrogenase can be expressed and functionally substitute for its homologue in DPOR in *Chlamydomonas reinhardtii,* suggesting that chloroplasts have the potential to provide a suitable environment for nitrogenase [139,140].

Alternative strategies have been explored to increase nitrogenase activities using natural diazotrophs. Several studies exploited a MoFe nitrogenase variant (α-V70A, α-H195Q) [141] to catalyze the reduction of CO2 to methane as a way for in vivo production of biofuels in *R. palustris*. While ATP is supplied by cyclic photophosphorylation, it was found that the electron flow to nitrogenase can be enhanced by manipulating metabolic pathway or state, such as providing cells with organic alcohols, diverting electrons away from biomass synthesis by using nongrowing cells, or blocking the Calvin–Benson–Bassham cycle [142,143]. Further, the same group demonstrated that wild-type Fe-only nitrogenase in *R. palustris* reduces CO2 simultaneously with nitrogen fixation to yield CH4, NH3, and H2. Excitingly, this seems to be a universal feature for the Fe-only nitrogenases. The amount of CH4 produced was low but sufficient to support the growth of an obligate methanotroph in co-culture with oxygen added at intervals [143]. Further studies are needed to see whether the above strategies of metabolic engineering can be applied to enhance the activity of Fe-only nitrogenase and whether similar principles can be applied to engineer other diazotrophs.

Liu et al. demonstrated the use of H2 generated from catalytic water splitting driven by renewable energy to support diazotrophic growth of autotroph *Xanthobacter autotrophicus*. In that way the production of *X. autotrophicus* as biofertilizer or NH3 (when glutamine synthetase was inhibited) is efficiently connected to atmospheric nitrogen. The biomass produced was applied to radishes and significantly increased the yield of radish storage roots (Figure 4) [144]. Their system achieved a nitrogen reduction turnover numbers of <sup>∼</sup>9 <sup>×</sup> 109 bacterial cell−<sup>1</sup> and 2 <sup>×</sup> 106 nitrogenase−<sup>1</sup> and a turnover frequency of 1.9 <sup>×</sup> 104 <sup>s</sup>−<sup>1</sup> per bacterial cell, or <sup>∼</sup>4 s−1·nitrogenase<sup>−</sup>1.

**Figure 4.** (**Top**) Schematic of ammonia (NH3) production in a biohybrid system that produces hydrogen (H2) from renewable electrical energy and sunlight. Hydrogenases within *Xanthobacter autotrophicus* subsequently oxidize the H2 to ultimately supply electrons to nitrogenase for dinitrogen (N2) fixation. The generated NH3 is incorporated in biomass (pathway 1) or can diffuse extracellularly by inhibiting biomass formation (pathway 2). Red arrows represent carbon cycling and blue arrows represent nitrogen cycling; line widths represent the relative fluxes of these pathways. (**Bottom**) Enhanced radish growth upon biofertilization with *X. autotrophicus* grown from electricity/sunlight-sourced H2. Reproduced from [144] with permission.

#### **4. In Vitro Electron Transfer to Nitrogenase**

#### *4.1. Fe protein-Dependent Activity*

For in vitro nitrogenase activity, research has sought to artificially deliver electrons to the Fe protein. Subsequently, the reduced Fe protein transfer electrons to the MoFe protein in the presence of the MgATP that is required for the Fe protein cycle. As discussed above, the most commonly employed electron donor is DT, which supports the formation of the 1e−-reduced [4Fe-4S]1<sup>+</sup> state of the Fe protein. While DT supports nitrogenase catalysis, it can be undesirable in many ways: (i) it is considered single-use since its regeneration is complex, (ii) DT is not particularly stable in aqueous solutions, and (iii) the reduction potential of DT is dependent on its concentration and pH, which can complicate thermodynamic and kinetic interpretations of metalloenzyme properties [30].

In addition to DT, other electron mediators employed with the Fe protein include MV (and derivatives), Ti(III) citrate, Eu(II) complexes, and one of the presumed in vivo electron donors, Fld (NifF) [9]. While Ti(III) and Eu(II) complexes are useful due to their low reduction potentials, they are typically prepared as a stock of a single-use reductant. In the case of Ti(III), Seefeldt and Ensign reported in 1994 that Fe protein reduction could occur such that catalysis by the MoFe protein could be supported, where similar rates of H<sup>+</sup> and C2H2 reduction were observed in comparison to DT-driven

assays [145]. Further, the oxidation of Ti(III) to Ti(IV) could be followed spectrophotometrically. MV and NifF have also been employed as electron donors to the Fe protein (for subsequent catalysis); however, they are typically employed alongside a relatively higher concentration of DT as the reducing agent [9,16]. As mentioned above, Yang et al. demonstrated that the reduction of nucleotide-free Fe protein by DT increased around 10-fold when NifF was included, presumably as mediator for DT reduction of the Fe protein [16]. Additionally, it was observed that the presence of either MgADP or MgATP diminished DT-driven Fe protein reduction by ~100-fold although this was alleviated by the inclusion of NifF. H2, NADH, and KBH4 have also been employed to transfer electrons to MV prior to catalysis by nitrogenase (in cell-free extracts of *C. pasteurianum*), although in all of the reported cases the oxidation of MV by nitrogenase cannot be followed, since the oxidized MV is regenerated in these assays [146,147].

#### Electrochemical Methods

Recently, MV has been utilized as the sole reductant for Fe protein-dependent nitrogenase activity assays. In 2017, Milton et al. demonstrated the use of electrochemically reduced MV (in the 1e−-reduced MV•<sup>+</sup> state) as the sole electron donor to the Fe protein [148]. First, it was demonstrated that the oxidation of MV could be followed spectrophotometrically. This was employed to rapidly determine the optimal Fe:MoFe protein ratio for MV-dependent assays (~20:1), although it was also observed that increasing MV concentrations resulted in diminished nitrogenase activity [146,148]. The nature of this has been recently ascribed to the dimerization of MV [149]. In addition to spectrophotometric activity, an electrochemical method was also employed to reduce MV in situ for nitrogenase activity assays [148]. First, this paves the way for the utilization of renewable electrical energy for "bioelectrosynthetic" N2 reduction under ambient temperature and pressure (Figure 5).

**Figure 5.** (**A**) Bioelectrocatalytic dinitrogen (N2) fixation by Mo-dependent nitrogenase using methyl viologen (MV) as an electron mediator. (**B**) Cyclic voltammetric data (scan rate = 0.002 V s<sup>−</sup>1) for bioelectrocatalytic N2 fixation. The electrochemical cell contained MOPS buffer (100 mM, pH 7), a MgATP-regenerating mixture (ATP, creatine phosphate, creatine phosphokinase), 0.1 mg mL−<sup>1</sup> of MoFe protein, and 20<sup>×</sup> equivalents of the Fe protein. The addition of Mg2<sup>+</sup> permits MgATP hydrolysis by the Fe protein and initiates bioelectrocatalytic N2 fixation. Reprinted (adapted) with permission from [150]. Copyright 2019 American Chemical Society.

This method was capable of electrochemically driving N2 reduction to NH3 where 59% of the electrons delivered to nitrogenase were directed towards N2 fixation. This initial example reported NH3 production at a rate of 35 nmol NH3 min−<sup>1</sup> mg−<sup>1</sup> MoFe protein, corresponding to a turnover frequency (TOF) of 4.2 (single time point). In this section, TOF is calculated to be the number of moles of product formed per mole of MoFe protein (assuming, for consistency, a mass of 240 kDa for the MoFe protein) per minute. Second, the current recorded at the electrode corresponds to the rate and magnitude of electron consumption by nitrogenase (and thus, substrate reduction), which provides a method to study substrate reduction by nitrogenase in real time [151]. The subsequent optimization of this MV electrochemical approach resulted in an observed rate constant for electron flux through nitrogenase when fixing N2 approached the expected value of 13 s−<sup>1</sup> (reported as 14 s−1), where rate-limiting electron transfer at 13 s−<sup>1</sup> corresponds to an optimal TOF of 195 (accounting for *re* of H2). Further, MV (and derivates) can also serve as an efficient electron donor to other (metallo)enzymes, such as formate dehydrogenases and hydrogenases, either for H2/formate (HCOO−) oxidation or for H+/CO2 reduction [152]. Of particular interest is the ability of MV to mediate electrons or enzymatic NADH formation [153,154]. To this end, the Minteer group has extended the use of reduced MV in bioelectrosynthetic cells to reduce NADH. In this way, the NH3 formed by nitrogenase can be upgraded to further products of interest, such as chiral amines and chiral amino acids [155,156].

#### *4.2. Fe Protein-Independent Activity*

While nitrogenase fixes N2 under considerably milder conditions than the Haber–Bosch process (i.e., at ambient pressure and temperature), nitrogenase can still be considered to be energy-intensive due to hydrolysis of at least 16MgATP for the fixation of each equivalent N2. The hydrolysis of MgATP corresponds to around <sup>−</sup>50 kJ mol−<sup>1</sup> in vivo; thus, there is significant interest in decoupling the Fe and MoFe proteins [20]. Not only does this present the possibility of MgATP-independent N2 fixation, but catalytic rates could also be improved, given that the rate-limiting step of N2 fixation is associated with the Fe protein [16]. As outlined above, evidence to supporting the idea that electron transfer between the Fe and MoFe proteins occurs prior to MgATP hydrolysis suggests that artificial reduction of the MoFe protein independent of the Fe protein (and therefore MgATP independent) could be possible. Since 2010 (when two papers submitted within two weeks of each other reported Fe protein-independent MoFe protein catalysis), chemical, photochemical, and electrochemical techniques have sought to deliver electrons to the MoFe protein for substrate reduction; prominent examples are covered below.

#### 4.2.1. Chemical Methods for Electron Transfer

In 2010, Danyal et al. reported substrate reduction by the MoFe protein from *A. vinelandii*, for which low-potential Eu(II) was employed as the artificial reductant [157]. Eu(II) was prepared by bulk electrolytic reduction of a Eu2O3 (*E*o' = <sup>−</sup>0.36 V vs. SHE) solution, followed by the addition of a chelating polyaminocarboxylate ligand that lowered the potential of the pre-reduced Eu(II) center −0.88 or −1.14 V vs. SHE (depending on the ligand employed, reported at pH 8) [60,157]. Thus, electrochemistry was employed to prepare the reductant in batch mode, and this example can be treated as a "chemical" reduction approach. While the authors noted that N2 fixation was not observed, the 2e<sup>−</sup>-reduction of hydrazine (N2H4) to NH3 was observed; further, a ß-Y98H MoFe protein mutant was required for the formation of significant quantities of NH3 when compared to blank/control experiments. The ß-Y98 residue, located between the P cluster and FeMo-co, was previously identified as a residue that is important to electron transfer [158]. A TOF of 41 was reported for the system; further, N2H4 is an important substrate where its 2e<sup>−</sup>-reduction to NH3 may represent an intermediate step of N2 fixation by nitrogenase. In 2015 Eu(II) complexes were coupled with two additional MoFe mutants, ß-F99H and α-Y64H, which also revealed enhanced N2H4 reduction in comparison to the WT MoFe protein [159]. The α-Y64H mutation was found to improve to TOF to 72. It was also demonstrated that these MoFe protein mutants could reduce azide (N3 <sup>−</sup>) to NH3, albeit it with a reduced TOF of 2 (single time point). However, only the originally reported ß-Y98H MoFe protein was found to reduce

H<sup>+</sup> to H2 when coupled with a Eu(II) ethylenediaminetetraacetic acid (EDTA) complex. Of the Eu(II) complexes tested within this study, the Eu(II)–EDTA complex has the mildest reduction potential (−0.84 V vs. SHE). However, the Eu(II)–EDTA complex was selected, as minimal H2 is evolved when using this Eu(II) complex in the absence of MoFe proteins. Thus, one explanation for the inability of the ß-F99H and α-Y64H proteins to reduce H<sup>+</sup> could be that the *E*o' of Eu(II)–EDTA complex was not reducing enough. Further, H2 evolution was measured to be ~7 nmol H2 min−<sup>1</sup> mg−<sup>1</sup> MoFe (ß-Y98H) which corresponds to a TOF of <2 (single time point). In contrast, Fe protein-derived electron transfer with a rate constant of ~13 s−<sup>1</sup> loosely corresponds to the formation of ~1625 nmol H2 min−<sup>1</sup> mg−<sup>1</sup> MoFe and a TOF of 390 (H2).

In 2012, Lee et al. further demonstrated the ability of Eu(II) complexes to catalyze substrate reduction by nitrogenase [160]. However, substrate reduction was performed by variant MoFe proteins that lacked the FeMo-co and/or contained P cluster precursors (2[4Fe-4S] clusters). In *A. vinelandii*, the deletion of *nifB* leads to a MoFe protein that lacks the FeMo-co; instead, the deletion of *nifH* (Fe protein) leads to a MoFe protein that lacks the FeMo-co and has immature P-clusters (P\*, 2[4Fe-4S] pairs) [6]. Remarkably, it was demonstrated that MoFe protein lacking FeMo-co could catalyze the reduction of H<sup>+</sup>, C2H2, C2H4, N2H4, cyanide (CN−), carbon monoxide (CO), and CO2 to products including, H2, NH3, and alkanes/alkenes spanning C1–C7 [160]. TOFs for this system reached maxima of 0.02 (for CH4 production from CN−, single time point) and 0.17 (for NH3 production from CN−, single time point). Interestingly, the authors observed a ~2.5-fold increase in activity when using the Δ*nifH* MoFe protein vs. the Δ*nifB* MoFe protein; Jimenez-Vicente et al. recently demonstrated that the Fe protein of the V-nitrogenase system (encoded by *vnfH* and typically repressed in the presence of Mo) can substitute the NifH Fe protein in Δ*nifH* strains, which could explain the observed elevated activities [161]. In addition to the Mo-nitrogenase system, Eu(II) complexes have also been utilized to deliver electrons to the alternative V-nitrogenase system for MgATP-independent substrate reduction [162]. In 2015, Rebelein et al. demonstrated that Eu(II)-DTPA (*E*o' <sup>=</sup> <sup>−</sup>1.1 V vs. SHE, DTPA <sup>=</sup> diethylenetriaminepentaacetate) could function as an electron mediator to the VFe protein for CO2 reduction [163]. In addition to the formation of CO and CH4 formation (with a TOF for CH4 production reaching 0.4 <sup>×</sup> <sup>10</sup><sup>−</sup>3), C2–C4 hydrocarbons were also produced.

#### 4.2.2. Photochemical Methods for Electron Transfer

In 2010, another approach for MoFe protein reduction was reported by the Tezcan group, which employed a Ru-based photosensitizer to deliver electrons to the FeMo-co through the P cluster [164]. [Ru(bpy)2(phen)]2<sup>+</sup> possesses a long-lived photoexcited-state (\*RuII), which upon quenching, results in the generation of a reducing RuI species that is not expected to be too dissimilar to the related [Ru(bpy)3] <sup>+</sup> complex with an *E*o' of around <sup>−</sup>1.28 V vs. SHE [9,164,165]. The authors prepared a Cys-reactive Ru complex which was subsequently attached to a Cys mutation introduced in proximity to the P cluster (α-L158C). In this way, the Ru photosensitizer could be placed ~15 Å away from the P cluster (important for efficient electron transfer rates)—similar to the location of the Fe protein's [4Fe-4S] cluster during MoFe protein association [9]. Wild-type MoFe protein has a single solvent-exposed Cys residue; it was hypothesized that the Ru-attachment to this residue would not yield significant photo-excited electron transfer and MoFe protein catalysis. Indeed, significant substrate-reduction activity was observed following the introduction of the α-L158C mutation, where the authors first observed H<sup>+</sup> and C2H2 reduction at approximately 14 nmol H2 min−<sup>1</sup> mg−<sup>1</sup> MoFe protein and 16 nmol C2H4 min−<sup>1</sup> mg−<sup>1</sup> MoFe protein (both being 2e−-reductions), corresponding to TOFs of ~3.4 and ~3.8 respectively. In 2012, the authors expanded on this approach and demonstrated that photosensitized MoFe protein can also facilitate the 6e−-reduction of CN<sup>−</sup> to CH4 with a TOF of <0.1 [166]. However, this approach was not able to produce NH3 from N2 fixation at detectable quantities.

In 2016, a second photochemical approach for MoFe protein reduction was reported in which CdS nanorods were mixed with wild-type MoFe protein and N2 fixation to NH3 was observed [167]. In contrast to the low reduction potential afforded by the Ru-photosensitizer system, CdS nanorods

offer a milder *E*o' of <sup>−</sup>0.8 V vs. SHE, which can also be accessed upon illumination with visible light. Under N2, this system was found to produce NH3 at 315 nmol min−<sup>1</sup> mg−<sup>1</sup> MoFe protein which corresponds to ~63% of NH3 production by the Fe protein-driven, ATP-dependent activity of the MoFe protein and a TOF of 75. Recently, Harris et al. expanded on this approach by investigating electrostatic interactions between the MoFe protein and the surface of CdS nanorods modified by charged functional groups (Figure 6) [168]. In addition, the authors also prepared a flexible linker to enable the attachment of the MoFe to the CdS nanorods by a Cys residue introduced near to the P cluster (α-L158C). For the optimal configurations reported, maximal H2 evolution rates of approximately 1250 nmol H2 min−<sup>1</sup> mg−<sup>1</sup> MoFe protein were observed, which correspond to TOFs of approximately 300. In this example, N2 fixation to NH3 by the CdS nanorod biohybrid was not reported. In summary, precisely how the CdS biohybrid system bypasses the Fe protein and affords close-to-native TOFs is unknown. Yet, the ability to achieve N2 fixation is promising and highly attractive to future ATP-independent NH3 biotechnologies.

**Figure 6.** Harris et al. investigated the attachment of the MoFe protein to CdS nanorod surfaces: (**left**) surface-capping groups with different charges were employed, and (**right)** solvent-exposed Cys residues of the MoFe protein were used to covalently conjugate the MoFe protein to CdS nanorod surfaces. Reprinted (adapted) with permission from [168]. Copyright 2020 American Chemical Society.

#### 4.2.3. Electrochemical Methods for Electron Transfer

Since 2016, electrochemical approaches have gained significant interest for artificial MoFe protein reduction and catalysis. Milton et al. reported the immobilization of the MoFe protein within a polymer at an electrode surface [169]. In this approach, bis(cyclopentadienyl)cobalt(II) (cobaltocene1+/0, *E*o' = <sup>−</sup>0.96 V vs. SHE) was used as a low potential electron mediator to deliver electrode-derived electrons to the MoFe protein. In addition to H<sup>+</sup> reduction, N3 <sup>−</sup> and nitrite (NO2 −) reductions were also observed. As observed by Danyal et al., the use of the ß-Y98H MoFe protein mutant resulted in enhanced catalytic currents (consistent with enhanced catalysis): Faradaic efficiencies (FEs) of 35% and 101% and TOFs of 12 and 40 were reported for N3 <sup>−</sup> and NO2 − reduction (single time points). This method provided a method by which electron transfer to (and thus, catalysis of) the MoFe protein could be observed in real time. Hu et al. also adopted this approach to demonstrate that the MoFe and FeFe proteins could reduce CO2 to HCOO<sup>−</sup> with respective FEs of 9% and 32%. Khadka et al. also demonstrated that this approach could be employed to investigate the rate-limiting step of the MoFe when catalyzing H<sup>+</sup> reduction [170]. In this study, the cobaltocene concentration was titrated such that electron transfer to the MoFe protein was not limiting, and bioelectrocatalytic H<sup>+</sup> reduction to H2 was evaluated as a function of catalytic current. Following the addition of increasing D2O fractions to the electrochemical setup, a decrease in the catalytic current was observed, and this was attributed to a kinetic isotope effect. This was further exemplified by the use of a range of MoFe protein mutants and FeMo-co-transporting/storage proteins. Supporting density functional theory (DFT) calculations led to the conclusion that the protonation of a M-H at the FeMo-co is the rate-limiting

step of H2 evolution (within the MoFe protein component). In 2018, Cai et al. also employed this approach to investigate catalysis of the VFe protein for H<sup>+</sup> and CO2 reduction [171]. In addition to unsubstituted cobaltocene, mono-cobaltocene ([Co(Cp)(CpCOOH)], *E*o' = <sup>−</sup>0.79 V vs. SHE) and di-carboxy cobaltocene ([Co(CpCOOH)2], *<sup>E</sup>*o' <sup>=</sup> <sup>−</sup>0.65 V vs. SHE) derivatives were also investigated as electron mediators for the VFe protein. After the passage of 4 C of charge within an electrochemical cell that contained the VFe and protein and dicarboxy-cobaltocene, ~850 μmol H2 was detected μmol−<sup>1</sup> VFe protein (corrected to control experiments). Upon the addition of bicarbonate (HCO3 <sup>−</sup>) as the CO2 source, the passage of 4 C of charge led to the formation of ~20 nmol C2H4 μmol−<sup>1</sup> VFe and ~30 nmol propene (C3H6) μmol−<sup>1</sup> VFe protein (corrected to control experiments).

An additional approach to electrochemical MoFe protein reduction was reported in 2016, wherein Paengnakorn et al. reported the reduction H<sup>+</sup> by the MoFe protein also entrapped within a polymeric matrix [61]. A mixture of three Eu complexes was employed to mediate electron transfer to the MoFe protein, with *E*o's spanning <sup>−</sup>0.63 to <sup>−</sup>1.09 V vs. SHE (at pH 8). Excitingly, this approach was coupled with an attenuated total reflectance infrared (ATR-IR) spectroelectrochemical cell such that substrate binding at the FeMo-co could be interrogated as a function of potential. Interestingly, the binding of CO to the FeMo-co (which does not typically affect H<sup>+</sup> reduction) was found to commence at potentials of ~<−0.7 V vs. SHE. H2 formation was also confirmed by gas chromatography resulting in a TOF of 14 for the ß-F99H mutant MoFe protein; this mutant and the ß-Y98H mutant (TOF = 13) MoFe proteins showed elevated electrocatalytic currents over the wild-type MoFe protein (TOF = 7) (single time points).

In contrast to the use of electron mediators to deliver electrons to the MoFe protein, Hickey et al. reported on the ability to directly deliver electrons to the MoFe protein when immobilized at electrode surfaces [172,173]. This approach represents an exciting new approach by which the MoFe protein (and the VFe and FeFe proteins) can be explored in greater detail, as this permits the direct observation of the redox state and reduction potentials of enzymatic cofactors. This approach permitted NH3 production with TOFs of around 0.1–0.6, alongside a TOF of 2.6 for H2 formation (under Ar, single time point) [172]. Interestingly, this article also identified that the background production of H2 by the electrode (at low potentials) is inhibitory to the fixation of N2 by nitrogenase, which should be avoided in future electrochemical investigation of nitrogenase.

#### **5. Conclusions**

As discussed, nitrogenase is the only enzyme known to reduce N2 to NH3. Thus, many streams of research are underway to understand how nitrogenase fixes N2 and to engineer (semi)biological systems for NH3 production. The understanding of nitrogenase's N2 fixation mechanism could lead to the design of new bio-inspired catalysts with improved efficiencies, or lead to biotechnologies that exploit nitrogenase's reactivity to produce NH3 (such as directing renewable electricity towards MgATP-independent N2 fixation by the MoFe protein). The ability of engineered plants or symbiotic systems to produce their own NH3 also represents a significant milestone. As presented here, many pathways (both natural and engineered) exist, or are being developed, to deliver electrons to nitrogenase for N2 fixation. Nevertheless, ambiguity surrounds the possibility of the Fe protein undergoing the transfer of 2e− during each transient association with the MoFe protein, which has the potential to halve nitrogenase's MgATP hydrolysis requirement. Further, a range of natural and unnatural electron donors are being explored for electron delivery. In the case of natural electron donors, it is still necessary to improve the efficiencies of electron transfer systems in non-diazotrophic hosts to support higher nitrogenase activity. In artificial in vitro systems, it is unclear as to why certain systems support H<sup>+</sup> reduction by nitrogenase (representing necessary M-H formation at the FeMo-co) but N2 is not observed. For example, how are photochemical methods (such as CdS nanorods) able to support N2 fixation, while other photochemical methods (i.e., Ru-based photosensitizers) are not? Similarly, it remains unclear as to how mediated electron delivery to the MoFe protein (or VFe/FeFe proteins) cannot currently support N2 fixation (although H<sup>+</sup>, CO2, N3 <sup>−</sup>, and NO2 − reduction has been achieved), whereas direct electron transfer appears to facilitate NH3 production. Thus, further research into

nitrogenase is required in a multitude of areas; hence, multi-disciplinary research efforts surely present the best opportunity (and are clearly necessary) to advance nitrogenase research and understanding.

One method of improving environmental sustainability is the decentralization of key global-scale processes. To that end, nitrogenase-based technologies present opportunities for on-demand NH3 production. However, much development is still required to overcome some of the limitations surrounding the utilization of nitrogenase. For example, MgATP hydrolysis presents a significant energetic requirement, and the sensitivity of nitrogenase to O2 also inhibits deployment. Further, nitrogenase assembly and maturation is complex and limits in vitro utilization. Nevertheless, perhaps the most promising approach for nitrogenase deployment in new biotechnologies is presented by Liu et al., where photoelectrically generated H2 feeds a N2-fixing microbe for NH3 production and/or biomass accumulation [144].

**Author Contributions:** W.G. and R.D.M. conceived and wrote the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** R.D.M. acknowledges funding from the University of Geneva.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References and Note**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Appropriate Bu**ff**ers for Studying the Bioinorganic Chemistry of Silver(I)** †

#### **Lucille Babel \*, Soledad Bonnet-Gómez and Katharina M. Fromm \***

Chemistry Department, University of Fribourg, Chemin du Musée 9, 1700 Fribourg, Switzerland; sole82@bluewin.ch

**\*** Correspondence: lucille.simond@unifr.ch (L.B.); katharina.fromm@unifr.ch (K.M.F.); Tel.: +41-2630-087-32 (K.M.F.)

† Dedicated to the radical chemist Prof. Bernd Giese on behalf of his 80th birthday.

Received: 21 February 2020; Accepted: 18 March 2020; Published: 22 March 2020

**Abstract:** Silver(I) is being largely studied for its antimicrobial properties. In parallel to that growing interest, some researchers are investigating the effect of this ion on eukaryotes and the mechanism of silver resistance of certain bacteria. For these studies, and more generally in biology, it is necessary to work in buffer systems that are most suitable, i.e., that interact least with silver cations. Selected buffers such as 4-(2-hydroxyethyl)-1-piperazineethane sulfonic acid (HEPES) were therefore investigated for their use in the presence of silver nitrate. Potentiometric titrations allowed to determine stability constants for the formation of (Ag(Buffer)) complexes. The obtained values were adapted to extract the apparent binding constants at physiological pH. The percentage of metal ions bound to the buffer was calculated at this pH for given concentrations of buffer and silver to realize at which extent silver was interacting with the buffer. We found that in the micromolar range, HEPES buffer is sufficiently coordinating to silver to have a non-negligible effect on the thermodynamic parameters determined for an analyte. Morpholinic buffers were more suitable as they turned out to be weaker complexing agents. We thus recommend the use of MOPS for studies of physiological pH.

**Keywords:** silver; buffer; association constant; HEPES

#### **1. Introduction**

A well-known list of buffers was published between 1966 and 1980, called Good's buffers, for their use in biological systems [1]. This list contains essentially sterically hindered amines that aim to replace common buffers used in biology such as imidazole, sodium phosphate and sodium citrate. Indeed, these previously employed buffers are inadequate for certain experiments because of their reactivity towards small molecules (ATP), metal ions, or because of their toxicity for the cells [2–8]. For example, a phosphate buffer leads to precipitates with many cations and is known to inhibit or enhance certain reactions of a cellular system [2,3]. Imidazole is a very good complexing ligand for many metal cations and, due to its similar structure, could replace histidine residues in metal binding proteins [9–11]. Good's buffers on the contrary were believed to be largely inactive towards the cell metabolism and thus should not interact with any biological molecule and/or metal ions. Nevertheless, since this list was established, many studies have proved that most of these sterically hindered tertiary amine-based buffers are able to coordinate slightly some metal ions [12,13]. Therefore, binding constants determined for other ligands could be affected by the presence of these buffers, which are usually in large excess compared to the ligand to ensure a stable pH, hence it is a necessity to know these values. A correction can then be applied to the thermodynamic model to take into consideration the effect of the buffer. To limit the effect of this correction, careful consideration of the metal ions in solution and the concentration of the buffer is necessary prior to use. For example, complexation of copper(II) by buffers was thoroughly studied over recent years and it was shown that

Good's buffers coordinate the metal ion with variable but non-negligible affinities of 3 ≤ log KCu,L ≤ 5 [14–16]. However, most of the studies found in the literature concern divalent metal cations and little is known on monovalent ones [17,18]. Moreover, publications on the morpholinic and piperazinic family of buffers are sometimes concluding to contradictory results [12].

Our group is interested in the use of silver as an antimicrobial agent. Silver is used in in vitro studies to investigate e.g., the silver resistance mechanism of some bacteria or in studies investigating toxicity and/or antimicrobial properties of silver agents, yet appropriate buffers for this kind of studies are lacking in the literature. We have recently been studying peptide models inspired by the protein SilE, a protein of the silver efflux pump in Gram negative bacteria, which is able to bind a large amount of silver(I) [19,20]. In this case, phosphate buffer could not be used because of the immediate formation of the poorly soluble silver phosphate salt.

HEPES contains N-donors and is not innocent with respect to silver(I) as shown by a crystal structure of a HEPES-silver(I) complex [21]. Two nitrogen atoms from the piperazine moieties of HEPES molecules as well as two oxygen atoms from the alcohol and sulfonate functions coordinate the silver ion in a distorted tetrahedral geometry. However, the binding affinity was not quantified.

Herein, we determined the affinity of HEPES for silver ions in order to quantize the buffer effect. In comparison, we also studied the effect of other buffers that were expected to possess the least interaction with silver ions (Scheme 1) to find out which one would be ideal for studies with silver(I) in biological media.

**Scheme 1.** Structures of buffers investigated for their affinities with silver ions.

#### **2. Materials and Methods**

Silver nitrate AgNO3 was purchased from Carlo Erba reagents (RPE, Analytical 99+%). 4-(2-hydroxyethyl)-1-piperazineethane sulfonic acid (HEPES), 3-(*N*-morpholino)propanesulfonic acid (MOPS), tris(hydroxymethyl)aminomethane (Tris) (Roche), sodium nitrate NaNO3 and potassium hydrogen phthalate (KHP) (Merck) were purchased from Sigma-Aldrich. Piperazine-1,4-bis(2-ethanesulfonic acid) (PIPES) and 2-(*N*-morpholino)ethanesulfonic acid (MES) were purchased from Roth. Nitric acid was purchased from Fluka and NaOH pellets from Acros. HNO3 0.1 M stock solution in 0.1 M NaNO3 was standardized towards KHP (0.4 g) where the equivalence point is followed with the help of phenolphthalein indicator. NaOH 0.2 M stock solutions in 0.1 M NaNO3 were standardized with stock solution of HNO3 0.1 M and used within two weeks to avoid carbonate formation. Buffers and silver nitrate were dissolved at a concentration of 0.05 M in 0.1 M NaNO3. PIPES was insoluble in water, and NaOH had to be added up to a 0.069 M concentration (1.4 eq.).

Buffers were titrated manually in presence of 0.1 M NaNO3 at 296 K over the pH range of 2–11 (HNO3 was added to obtain the starting pH of 2) with NaOH 0.2 M as titrant. Changes in pH were monitored with a glass electrode (Primatrode with NTC Methrom, combined glass-Ag/AgCl electrode), calibrated daily with standard buffers at pH 4 and 7. Titrations were conducted in triplicates for each buffer at three different concentrations between 2 and 12 mM with a sample volume of 50 mL. Silver

nitrate was added at three different ratios from 0.2 to 1.0 equivalents compared to the buffer. The titration data were analyzed using the SUPERQUAD software according to equilibriums defined in Appendix A. Mass spectrometry was performed on an ESI-MS Bruker Esquire HCT in H2O/MeOH solution (0.8:0.2) on the positive and negative mode with each buffer adjusted at pH 7 and 0.5 equivalent of silver nitrate.

#### **3. Results**

Acid dissociation constants were first determined without silver (Table 1, Figures 1 and S1) [22–25]. In the presence of silver nitrate, titrations were stopped at pH 8.0 because silver hydroxide and silver oxide are known to precipitate above this pH. The titration curve for HEPES with silver was found to have a lower plateau compared to HEPES alone, likely due to the coordination of HEPES to silver ions (Figure 1).

**Table 1.** Acid dissociation constants pK*an* (*n*= number of protons dissociated, see Figure S6 to visualize equilibrium considered) and complexation constants β Ag,B 1,*<sup>m</sup>* (*m* = number of buffer molecules bound by one silver ion for the formation of the complex [Ag(B)*m*], see Figure S12 for proposed structures) obtained for the different buffers with potentiometric titrations and comparison with literature (L = HEPES, PIPES, MOPS, MES, Tris).


<sup>a</sup> Acid dissociation constants fitted on titration points between pH 2.0 and 11.5. <sup>b</sup> Stability constants fitted on titration points between pH 2.0 and 8.0.

**Figure 1.** (**A**) Titration curves obtained for HEPES (7 mM) without (circles, dashed line) and in presence of silver nitrate (diamonds, plain line, 0.7 eq., 5 mM). (**B**) Speciation diagram according to pH for the species involving HEPES buffer.

We supposed the formation of a complex with one silver ion per HEPES ligand, based on the crystal structure obtained by Bilinovich et al. in 2011 resolved as a 1D coordination polymer with alternating HEPES and silver ions (Scheme 2) [21]. As solid-state structures do not always reflect the speciation in solution, three different ratios of silver to HEPES were tested. The titration curve fitted well (a 1) to the formation of a 1:1 complex and gave a stability constant of log(*K*Ag,HEPES 1,1 ) = 2.36(2) (Table 1). A stability constant for the formation of a hypothetical complex [Ag(HEPESH)]<sup>+</sup> with protonated HEPES in acid medium could be excluded as the fitting immediately results in negative values when considering this equilibrium. Thus, the protonated complex [Ag(HEPESH)]<sup>+</sup> is unlikely to form in solution.

**Scheme 2.** Structure of HEPES in its neutral form and schematic representation of the crystal structure obtained in presence of silver(I).

Due to the non-negligible amount of silver bound to HEPES buffer, we investigated other buffers in the same way as well: another piperazine type buffer PIPES, and two morpholine type buffers MES and MOPS as well as Tris buffer (Figure S2–S5). Indeed, morpholinic and piperazinic families were selected to be the most innocent buffers because they contain bulky tertiary amines and a low number of other weakly coordinating groups (alcohols, sulfonates). Indeed, at physiological pH, these two families were considered to be suitable buffers due their weak complexation ability with other metals [12]. Tris buffer, which is widely used in biology, was expected to yield higher binding constants with silver ions due to the weak steric hindrance of the amine.

Stoichiometry of the complexes was proposed according to mass spectra and by testing various models for the determination of binding constants. Nevertheless, m/z signals in the positive and the negative modes for silver complexes were not observed for HEPES, PIPES, MES and MOPS buffers either because these are polymeric species or because the major species is neutral (Figures S7–S10). For the Tris complex, a 2:1 species was observed with two ligands around one metallic center [Ag(Tris)2] + (Figure S11). For the determination of silver binding constants, larger errors were obtained when considering [Ag(MES)2] <sup>−</sup> or [Ag2(PIPES)] (Table S1) and negative values were found for [Ag(MOPS)2] −, so we decided to give only one stability constant for the formation of the [Ag(L)] complexes, with L = PIPES, MOPS, MES (Table 1 and Figure S12).

#### **4. Discussion**

Acid dissociation constants for the buffers alone were in good agreement with data from the literature (Table 1), confirming the validity of our measurements [22–25]. For the titration experiments with silver ions, the stability constant obtained for HEPES log(*K*Ag,HEPES 1,1 ) = 2.36(2) was lower than the value obtained for the 1:1 complex of HEPES with copper(II) log(*K*Cu,HEPES 1,1 ) = 3.22(2) [15]. This trend is expected as copper(II) is usually presenting greater affinities with nitrogen ligands due to its higher charge density [26,27].

Given the relatively high value for a buffer considered to be innocent of log(*K*Ag,HEPES 1,1 ) = 2.36(2) for the silver-HEPES complex, we simulated how binding constants of an analyte binding silver would be affected by the presence of HEPES buffer (Table 2, Appendix B). The decrease on stability constants that would be measured without taking the silver-HEPES complex into account depend on the concentration of the analyte and the relative stoichiometry with the buffer. However, these effects are still quite weak on the logarithmic scale of the stability constants, except when working at high concentrations, i.e., using Nuclear Magnetic Resonance (NMR) spectroscopy to obtain the stability constants.

**Table 2.** Apparent binding constants log(*K*Ag,L app,1,1) corrected for the effect of buffer for various real values of binding constants log(*K*Ag,L 1,1 ) (L = peptide or analyte investigated for its complexation to silver, B = HEPES buffer at pH 7.4, 40 equivalents) and at different concentrations. Percentage of decrease is indicated in parenthesis.


In our previous study of SilE in presence of HEPES, the fact that HEPES binds to silver ions can be neglected. Binding constants of SilE model peptides with silver ions were indeed determined in presence of HEPES, but using a competitor with known binding affinity for silver ions. The competitor was then similarly affected by the buffer as the peptide ligand. The stability constant of the competitor was itself calculated in competition with imidazole (whose contribution to the thermodynamic equilibrium was taken into consideration).

Looking for evidence for the stoichiometry of the complexes formed with silver, published crystal structures were examined (Scheme 3) [28–30] but no structures were found for MOPS or Tris [31]. Triethanolamine buffer (TEOA) yields a [Ag(TEOA)2] <sup>+</sup> complex, and this, together with the linear [Ag(NH3)2] <sup>+</sup> complex, suggests the possible formation of a complex [Ag(Tris)2] <sup>+</sup> [32]. Interestingly, for the crystal structures of the silver-PIPES and silver-MES complexes, the silver(I) ions always has at least a coordination number of four (Table S2). The silver ion is typically maintained by two quite strong coordination bonds, preferentially with nitrogen atoms, and by two weaker secondary bonding interactions with oxygen atoms of sulfonate and alcohol groups. The silver-MES complex includes a benzimidazole ligand (Bz) together with the complexation of MES buffer. According to these structures, one could expect a 2:1 silver to buffer ratio for PIPES and a 1:2 ratio for MOPS, MES and Tris. The stoichiometry was confirmed by mass spectrometry for Tris buffer where the complex [Ag(Tris)2] + was clearly identified as the main species in solution (Figure S11). Indeed, only the model with a 1:1 silver/buffer complex was working while fitting potentiometric data. Possible second binding constants are likely too weak to be precisely determined (Figure S12).

**Scheme 3.** Crystal structures obtained for PIPES [28] and MES [29] buffers in presence of silver(I).

Unsurprisingly, the primary amine Tris is the strongest silver binder in this study and the stability constant obtained is comparable to other amine ligands such as ethanolamine [33–35]. This value is in line with other studies at different ionic strengths and temperatures that have been quantifying the interaction between Tris buffer and silver(I) [36,37], validating our approach. Morpholine type buffers were less coordinating than piperazine type buffers, as expected by previous results on unsubstituted morpholine [33] and piperazine [38] molecules. MOPS turned out to be clearly the least coordinating

buffer of the buffer series studied here (Figure S13). Compared to other metal ions, a lower first stability constant was obtained for silver (I) compared to the ones for copper(II), and similar to nickel(II) or cobalt(II) as found for amines in the literature [26,27,37,39].

To fully benefit from these results and apply them to the standard conditions of a titration (i.e., at constant pH, maintained with a buffer), stability constants were corrected to take into account the partial protonation of the buffer ligand (Figure 2A, Appendix C). The apparent binding constants are slightly decreased compared to the original values, especially when working at high concentrations. Please note that accurate determination of stability constants lower than log(*K*Ag,L app,1,1) = 3 will ultimately necessitate the use of higher concentrations for the analyte and so for the buffer in order to see the association process. At these concentrations, and according to the third line of Table 2, buffer complexation cannot be neglected. Only high stability constants (log(*K*Ag,L app,1,1) ≥ 4) can thus be determined when using buffers. Another way to see the effect of the buffer on metal ion interactions is to calculate the amount of silver(I) ions bound to the buffer (Figure 2B, Appendix C). According to this percentage, a high proportion of silver ions -more than 90% for Tris buffer- would be complexed by the buffer. Fortunately, when measuring high stability constants at low concentrations for an analyte, the fact that silver ions are not free but bound to the buffer does not affect much the formation of the silver-analyte complex.

**Figure 2.** (**A**) Conditional (or apparent) stability constant for the complexation of silver(I) to the buffer at physiological pH 7.4 (for molecules comprised in their buffer range). (**B**) Fraction of silver bound to the buffer (total concentration of buffer 20 mM and silver 1 mM) at pH 7.4.

One could also decide to work at a lower pH (so MES could be considered, but not Tris, Figure S14 and S15B) or to work at different concentrations of buffers (Figure S15A). In the buffer range of the molecules studied here, whatever the conditions, MOPS was always the most suitable buffer. MES, HEPES and PIPES had similar coordination strength regarding silver(I) ions. They can reasonably be used if taking into consideration partial complexation to the buffer for accurate determination of stability constants of ligand/silver complexes.

In conclusion, between pH 6.5 and 7.9, MOPS would be recommended for the studies necessitating the use of silver(I) as it was the less coordinating buffer.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2624-8549/2/1/193\T1\ textendash202/s1. Figure S1: Titration curves obtained for HEPES without and in presence of silver nitrate in solution, Figure S2: Titration curves obtained for PIPES without and in presence of silver nitrate in solution, Figure S3: Titration curves obtained for MOPS without and in presence of silver nitrate in solution, Figure S4: Titration curves obtained for MES without and in presence of silver nitrate in solution, Figure S5: Titration curves obtained for Tris without and in presence of silver nitrate in solution, Figure S6: Acid dissociation equilibriums considered in the present study for the different buffers, Figure S7: Mass spectra in positive and negative mode for HEPES buffer with silver nitrate, Figure S8: Mass spectra in positive and negative mode for PIPES buffer with silver nitrate, Figure S9: Mass spectra in positive and negative mode for MOPS buffer with silver nitrate, Figure S10: Mass spectra in positive and negative mode for MES buffer with silver nitrate, Figure S11: Mass spectra in positive mode for Tris buffer with silver nitrate, Figure S12: Proposed structures of complexes formed with silver. This stoichiometry was retained for determination of stability constants, Figure S13: Logarithm of stability constants

*Chemistry* **2020**, *2*

for the first complexation of silver(I) on ligands B (B= buffer studied in this paper), Figure S14: Logarithm of conditional (or apparent) stability constants for the first complexation of silver(I) on buffers at a fixed pH value pH = 6.7, Figure S15: Fraction of silver bound to the buffer, Table S1: Stability constants obtained when considering other equilibrium than the one for the formation of [Ag(L)] (complex [Ag2(PIPES)] or [Ag(MES)2] −, Table S2: Bond distances (Ag-donor atom), average bond valences (νAg,N1X2 and νAg,O3-5) and total atom valence (VAg) in the molecular structures of [Ag*x*(Buffer)*m*].

**Author Contributions:** Conceptualization, L.B.; methodology, L.B.; validation, L.B.; and K.M.F.; formal analysis, L.B.; investigation, L.B. and S.B.-G.; resources, L.B.; data curation, L.B.; writing—original draft preparation, L.B.; writing—review and editing, L.B. and K.M.F.; visualization, L.B.; supervision, K.M.F.; funding acquisition, K.M.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Swiss National Science Foundation and BNF Universität Bern program.

**Acknowledgments:** I would like to acknowledge Jihane Hankache for mass spectra and Aurélien Crochet for search of crystallographic structures.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **Appendix A. Thermodynamic Equilibrium Used to Fit Potentiometric Titrations**

$$\text{BufferH}\_{n} \rightleftharpoons \text{BufferH}\_{n-1} + \text{H}^{+} \quad K\_{\text{an}} = \frac{[\text{BHI}\_{n-1}][\text{H}^{+}]}{[\text{BHI}\_{n}]} \tag{A1}$$

For acid dissociation constant *Kan*, there are one to three constants depending on the sum of amine group (one) and the number of sulfonates groups present in the buffer molecule.

A stability constant was then fitted with the fully deprotonated buffer according to Equation (A2).

$$\text{Buffer} \rightleftharpoons \left[ \text{Ag} (\text{Buffer})\_m \right] + \text{H}^+ \quad \beta\_{1,m}^{\text{Ag},\text{B}} = \frac{\left[ \text{AgB}\_m \right]}{\left[ \text{B} \right]^m \left[ \text{Ag} \right]} \tag{A2}$$

For all buffers, *m* = 1 except in the case of Tris buffer where there are two constants for *m* = 1 and *m* = 2. For conversion between cumulative constants and stepwise constants (as usually found in literature):

$$
\boldsymbol{\beta}\_{1,1}^{\text{Ag,B}} = \boldsymbol{K}\_{1,1}^{\text{Ag,B}} \text{and } \boldsymbol{K}\_{1,2}^{\text{Ag,B}} = \boldsymbol{\beta}\_{1,2}^{\text{Ag,B}} / \boldsymbol{\beta}\_{1,1}^{\text{Ag,B}} \tag{A3}
$$

The presence of a complex [Ag(BufferH)] was tested for the fitting of titration curves for all buffers but could not lead to any reliable results (constants were systematically negative). Thus, we consider that this complex was unlikely to be formed in solution.

#### **Appendix B. Calculation of Apparent Binding Constants of a Ligand Binding Silver**

$$\text{Ag}^+ + \text{L} \rightleftharpoons \text{[AgL]} \quad K\_{1,1}^{\text{Ag},\text{L}} = \frac{[\text{AgL}]}{[\text{L}][\text{Ag}]} \tag{A4}$$

We define an apparent binding constant which will be the one obtained if not considering the buffer-silver complexation:

$$\mathcal{K}\_{\text{app},1,1}^{\text{Ag.L}} = \frac{[\text{AgL}]}{[\text{L}] \left( [\text{Ag}]\_{\text{tot}} - [\text{AgL}] \right)} = \frac{[\text{AgL}]}{([\text{L}]\_{\text{tot}} - [\text{AgL}]) \left( [\text{Ag}]\_{\text{tot}} - [\text{AgL}] \right)} \tag{A5}$$

The mass balance equation for the total concentration of silver is expressed in Equation (A6):

$$\left[\text{Ag}\right]\_{\text{tot}} = \left[\text{Ag}\right] + \left[\text{AgL}\right] + \left[\text{Ag}\left(\text{HEPES}\right)\right] \tag{A6}$$

Concentration of silver complexes can be expressed according to the binding constants:

$$\left[\text{Ag(HEPES)}\right] = \frac{K\_{1,1}^{\text{Ag,B}}}{1 + \sum\_{i=1}^{n} \beta\_{\text{an}} \left(10^{-p\text{fl}}\right)^{n}} \cdot \frac{[\text{Ag}][\text{HEPES}]\_{\text{tot}}}{1 + \frac{K\_{1,1}^{\text{Ag,B}}}{1 + \sum\_{i=1}^{n} \beta\_{\text{an}} \left(10^{-p\text{fl}}\right)^{n}}} \tag{A7}$$
 
$$[\text{AgL}] = K\_{1,1}^{\text{AgL}} \cdot \frac{[\text{Ag}][\text{L}]\_{\text{tot}}}{1 + K\_{1,1}^{\text{AgL}} \cdot [\text{Ag}]} \tag{A8}$$

Introducing Equations (A7) and (A8) in Equation (A6), we obtain an expression of total silver concentration as a function of silver free concentration [Ag].

$$\text{[Ag]}\_{\text{tot}} = \text{[Ag]} + \frac{K\_{1,1}^{\text{Ag,B}}}{1 + \sum\_{i=1}^{n} \beta\_{\text{th}} (10^{-pH})^{u}} \cdot \frac{[\text{Ag}][\text{HPOES}]\_{\text{tot}}}{1 + \frac{K\_{1,1}^{\text{Ag,B}}}{1 + \sum\_{i=1}^{n} \beta\_{\text{th}} (10^{-pH})^{u}} \cdot [\text{Ag}]} + K\_{1,1}^{\text{Ag,L}} \cdot [\text{Ag}] \quad (\text{A9})$$

This concentration is optimized to minimize the difference between the actual concentration [Ag]tot and the one calculated by Equation (A9). Once the concentration of free silver [Ag] at hand, the apparent binding constant can be calculated from Equation (A8) and reintroducing in Equation (A5).

#### **Appendix C. Calculation of Conditional Stability Constants at a Certain pH and Calculation of Percentage of Metal Bound to the Bu**ff**er** θ**<sup>B</sup>**

Conditional stability constants are defined as the apparent binding constants of the complex between silver(I) and the buffer B at a certain pH value. Thus, we considered that a certain part of the buffer is not coordinating silver(I) as it is protonated but it is still considered in the equilibrium as shown in Equation (A10):

$$K\_{1,1}^{\text{cond,pH est}} = \frac{[\text{AgB}]}{[\text{Ag}] \left( [\text{B}] + \sum\_{i=1}^{n} [\text{BH}\_{n}] \right)} = \frac{K\_{1,1}^{\text{Ag},\text{B}}}{1 + \sum\_{i=1}^{n} \beta\_{\text{th}} \left( 10^{-pH} \right)^{n}} \tag{A10}$$

For the calculation of the concentration of species and the percentage of metal bound to the buffer, we first established the mass balance equations:

$$\text{[B]}\_{\text{tot}} = \text{[B]} + \sum\_{i=1}^{m} m \text{[AgB}\_{\text{m}}] + \sum\_{i=1}^{n} [\text{H}\_{i}\text{B}] \tag{A11}$$

$$\text{[B]}\_{\text{tot}} = \text{[B]} + \sum\_{i=1}^{m} m \text{[AgB}\_{m}\text{]} + \sum\_{i=1}^{n} [\text{H}\_{i}\text{B}] \tag{A12}$$

Then we rearrange Equation (A6) and silver total concentration to express the concentration of free silver:

$$\mathbb{E}\left[\text{Ag}\right] = \frac{\left|\text{Ag}\right|\_{\text{tot}}}{1 + \sum\_{i=1}^{m} \beta\_{1,\text{m}}^{\text{Ag},\text{B}} \left[\text{B}\right]^{m}} \tag{A13}$$

Free concentration of silver was then reintroduced in Equation (A11):

$$\mathbf{[B]}\_{\text{tot}} = \mathbf{[B]} \left( \mathbf{1} + \sum\_{i=1}^{m} \frac{m \cdot \beta\_{1,m}^{\text{Ag,B}} \left[ \text{Ag} \right]\_{\text{tot}} \left[ \mathbf{B} \right]^{m-1}}{\mathbf{1} + \sum\_{i=1}^{m} \beta\_{1,m}^{\text{Ag,B}} \left[ \mathbf{B} \right]^{m}} + \sum\_{i=1}^{n} \beta\_{1m} \left[ \mathbf{H} \right]^{n} \right) \tag{A14}$$

*Chemistry* **2020**, *2*

We assumed a certain value for the concentration of the free ligand [B] to obtain a value of [B]tot, calc. with Equation (A14) at a certain pH value. Difference between the calculated total ligand concentration and the one set in the experiment was minimized by tuning the value of free ligand [B].

Once the parameter of free ligand/buffer [B] has been optimized, one could calculate the concentration of complexed species [AgB*m*] with concentration of free metal being determined with Equation (A12):

$$\left[\text{AgB}\_{m}\right] = \beta\_{1,m}^{\text{Ag,B}} [\text{Ag}] [\text{B}]^{m} \tag{A15}$$

The percentage of metal bound to the buffer is then calculated according to Equation (A16):

$$\boldsymbol{\Theta}\_{\rm B} = \frac{\sum\_{i=1}^{m} \text{[AgB}\_{m}\text{]}}{[\text{Ag}]\_{\text{tot}}} = \frac{\sum\_{i=1}^{m} \beta\_{1,m}^{\text{Ag},\text{B}} [\text{Ag}][\text{B}]^{m}}{[\text{Ag}]\_{\text{tot}}} \tag{A16}$$

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Giant Polymer Compartments for Confined Reactions**

#### **Elena C. dos Santos, Alessandro Angelini, Dimitri Hürlimann, Wolfgang Meier** *∗* **and Cornelia G. Palivan** *∗*

Department of Chemistry, University of Basel, Mattenstrasse 24a, BPR 1096, 4002 Basel, Switzerland; e.dossantos@unibas.ch (E.C.d.S.); alessandro.angelini@unibas.ch (A.A.); dimitri.huerlimann@unibas.ch (D.H.)

**\*** Correspondence: wolfgang.meier@unibas.ch (W.M.); cornelia.palivan@unibas.ch (C.G.P.);

Tel.: +41-61-207-38-02 (W.M.); +41-61-207-38-39 (C.G.P.)

Received: 25 April 2020; Accepted: 8 May 2020; Published: 12 May 2020

**Abstract:** In nature, various specific reactions only occur in spatially controlled environments. Cell compartment and subcompartments act as the support required to preserve the bio-specificity and functionality of the biological content, by affording absolute segregation. Inspired by this natural perfect behavior, bottom-up approaches are on focus to develop artificial cell-like structures, crucial for understanding relevant bioprocesses and interactions or to produce tailored solutions in the field of therapeutics and diagnostics. In this review, we discuss the benefits of constructing polymer-based single and multicompartments (capsules and giant unilamellar vesicles (GUVs)), equipped with biomolecules as to mimic cells. In this respect, we outline key examples of how such structures have been designed from scratch, namely, starting from the application-oriented selection and synthesis of the amphiphilic block copolymer. We then present the state-of-the-art techniques for assembling the supramolecular structure while permitting the encapsulation of active compounds and the incorporation of peptides/membrane proteins, essential to support in situ reactions, e.g., to replicate intracellular signaling cascades. Finally, we briefly discuss important features that these compartments offer and how they could be applied to engineer the next generation of microreactors, therapeutic solutions, and cell models.

**Keywords:** artificial cells; biomimicry; polymer GUVs; polymer capsules; single compartments; multicompartments

#### **1. Introduction**

Compartmentalization produces a remarkably efficient organization of membranes and biomolecules that is essential to cope with the complexity of metabolic reactions in cells, and whose stability and functions are vital [1]. Inspired by natural biocompartments, significant efforts have been made to produce compartments that mimic cells and organelles, either in terms of their membrane properties or of the reactivity of encapsulated biomolecules [2]. Micrometer-sized vesicles, namely giant unilamellar vesicles, GUVs for short, are preferably used in this context, since their size and architecture can mimic cells, such as to extract information regarding reactions in a bio-relevant confined space. In addition, they allow for real time visualization of the membrane structure (providing information regarding membrane fluidity and integrity), and of biochemical reactions and enzymatic crowding effects that occur within a controlled and simplified environment, yet still preserving defined characteristics of cells. Lipid based compartments are straightforward systems for mimicking a cell/organelle membrane, nevertheless, their mechanical instability and the presence of membrane defects are limiting factors. One elegant way of introducing robustness to compartments, and at the same time of expanding new membrane properties, is the use of compartments made of copolymers. With the progress in polymer chemistry, numerous amphiphilic block copolymers have been synthesized with a variety of compositions, block ratios

and functions [3]. Due to a greater chemical versatility compared to lipids, block copolymers increase the opportunities to achieve desired self-assembled morphologies made of membranes with tailored properties and excellent biocompatibility. The architecture of such compartments—whether they are micro- or nano-sized—offers three different regions: the inner cavity for encapsulating hydrophilic molecules, the membrane for insertion of hydrophobic molecules, and the external membrane surface, for attachment of specific molecules [4,5] and eventually, for immobilization onto external functional surfaces [6,7]. From a topological point of view, nano-sized compartments, such as small layer-by-layer (LbL) capsules [8], polymersomes [9] or liposomes [10], can be designed as to mimic organelles, the cellular subcompartments. They have been used as nano-scale catalytic compartments, serving as to produce various desired compounds, as artificial peroxisomes [11], acting in tandem to support cascade reactions or as subcompartments inside GUVs to allow development of multicompartment systems [12]. When nanocompartments are encapsulated inside polymer giants in combination with active compounds, they are able to communicate among them to allow reactions, similarly to intracellular organization [13]. Permeability of membranes (either in a single or in multicompartments) favors molecular transport (enzyme substrates and products) and can be achieved in various ways, resulting from the chemical nature of the copolymer [14], by insertion of peptides [15,16] and membrane proteins [17,18]. A schematic of the most common compositions that polymer single and multicompartments can attain is presented in Figure 1.

**Figure 1.** Schematic representation of the different types of polymer compartments (polymer giant unilamellar vesicles (GUVs) and layer-by-layer (LbL) capsules), showing their diversity in terms of size, arrangements and the different types of biomolecules, including their possible locations within the assemblies.

This review presents micrometer-sized compartments either as single or as multicompartment reaction space, as powerful tools for biomimicry, lowering the degree of complexity to enable studies on targeted processes. We first introduce the synthesis of amphiphilic block copolymers through the various known polymerization techniques as building blocks of such compartments and indicate the conditions and properties required to support in situ reactions. Different methods for the preparation of these vesicular structures and their combination with active compounds (e.g., enzymes and peptides/membrane proteins) are presented together with the crucial points ensuring an efficient compartmentalization for desired applications. Permeabilization methods will not be described in this review, as they were already discussed extensively elsewhere [19,20]. We rather explore the biomimetic approach of these compartments equipped with peptides/membrane proteins to render them permeable for molecular flow and containing active compounds/subcompartments. Reactions inside confined spaces at microscale allow studying and better understanding of natural mechanisms of such reactions and their role inside them to support applications in various domains, as sensing, therapeutics and catalysis.

#### **2. Polymers as Building Blocks of Micrometer-Sized Compartments**

The progress in polymer chemistry gave access to a variety of polymers with tailored properties and excellent biocompatibility, thus serving to select specific components, where the precise role of each leads to a well-controlled system. Two or more chemically different polymeric domains, covalently bound together are defined as block copolymers. More specifically, amphiphilic block copolymers are composed of both hydrophilic and hydrophobic blocks, often named as diblocks (AB), triblocks (ABA or ABC) or multiblocks (ABCBA, ABCD, etc.). According to the required properties, amphiphilic block copolymers are built/designed by combining specific types of hydrophilic and hydrophobic blocks.

#### *2.1. Amphiphilic Block Copolymers as Building Blocks for Generation of GUVs*

In order to prepare amphiphilic block copolymers, controlled polymerization techniques are commonly used: Atom transfer radical polymerization (ATRP) [21], reversible addition fragmentation chain transfer (RAFT) [22,23], ionic polymerization and combinations thereof [3,24]. Typically, sequential chain extension can be used, in which a first block is polymerized using the aforementioned techniques, forming the so-called macro-initiator. Immediate addition of a second monomer leads to chain-extension, yielding a diblock copolymer. This approach allows the adjustment of each block length by terminating the corresponding chain extension according to the desired degree of polymerization. Tri- or multiblock copolymers can be obtained analogously either by sequential chain extension (asymmetric ABA, ABC, ABCD, etc.) or by a bifunctional initiator (symmetric ABA, ABCBA, etc.). Monomer conversion as well as the living character of the polymer chain are fundamental parameters to be considered among each chain extension. In ionic polymerizations, high monomer conversions can easily be reached by maintaining narrow polydispersity [25]. For example, poly(ethylene oxide)-*block*-polybutadiene (PEO-*b*-PBD) has been successfully synthesized by anionic polymerization in a two steps sequential monomer addition [26–28]. Prior to the addition of the second monomer, modifications are required as for poly(ethylene oxide)-*block*-poly(ethyl ethylene) (PEO-*b*-PEE), in which a PBD precursor is first hydrogenated to yield the PEE macroinitiator. Subsequently, ethylene oxide is polymerized to obtain the diblock copolymer [29]. In another study, poly(acrylic acid)-*block*-polybutadiene (PAA-*b*-PBD) was prepared by sequentional addition of butadiene and *tert*-butylacrylate, followed by hydrolysis to its acid form [28]. The combination of different polymerization techniques is another possibility. Namely, the preparation of poly(dimethyl sulfoxide) (PDMS) by anionic polymerization was followed by activation and cationic ring-opening polymerization of 2-methyl-2-oxazoline (MOXA) monomers to obtain poly(dimethyl sulfoxide)-*block*-poly(2-methyl-2-oxazoline) (PDMS-*b*-PMOXA) diblock copolymers (Scheme 1) [30].

**Scheme 1.** Synthesis route of PDMS-*b*-PMOXA combining both types of ionic polymerizations. Reprinted with permission from [30]. Copyright c 2014 American Chemical Society.

Ionic polymerization techniques are limited due to their high sensitivity to impurities. Hence, the solvent choice is important and the preparation of each reactant has to be handled very carefully to reach the desired purity grade. Controlled radical polymerization techniques (CRP), such as ATRP or RAFT, have recently been developed and have provided interesting and more versatile alternatives for the production of block copolymers. Both techniques require an initiator and the polymerization is governed by an equilibrium between an active species and a dormant one. The latter is constantly re-initiated in order to form the active species responsible for propagation through the addition of monomers. With these techniques, it is usually recommended not to exceed monomer conversions of 90%, above which the probability of termination is higher, risking to form dead chains unable to continue chain-extension [22]. As an example, poly(ethylene oxide)-*block*-poly(4"-acryloyloxybutyl 2,5-bis(4'-butyloxybenzoyloxy)benzoate) (PEO-*b*-PA444) was obtained from PEO modified to a macroinitiator for ATRP on which PA444 has been polymerized [31]. By using RAFT, a diblock copolymer poly(pentafluorophenyl acrylate)-*block*-poly(*n*-butyl acrylate) (PFPA-*b*-P*n*BA) was firstly prepared using the appropriate chain transfer agent (CTA), followed by modification to yield the amphiphilic glycopolymer PN*β*GluEAM-*b*-PBA [32]. More recently, polymerization induced self-assembly (PISA) allowed the preparation of poly(ethylene oxide)-*block*-poly(2-hydroxypropyl methacrylate) (PEO-*b*-PHPMA). In a suitable solvent, a solution of monomer feeds the growing chain on the PEO macroinitiator, producing an amphiphile that gradually self-assembles into structures, while polymerization is ongoing and leading to turbidity in the medium (Figure 2) [33].

**Figure 2.** (**A**) RAFT polymerization of 2-hydroxypropyl methacrylate controlled by a PEO macroinitiator (**B**) Reaction mixture throughout the PISA polymerization process. Adapted with permission from [33]. Copyright c 2017 Springer Nature.

Although, the possibility of combining synthetic approaches broadens the library of accessible polymers, chemists still need to work hard on the quantitative attachment of the re-initation site for the next polymerization, which is highly recommended to prevent purification difficulties. To circumvent this problem, two or more homopolymers can be connected together using coupling reactions such as Diels-Alder, copper-catalyzed azide-alkyne cycloaddition (CuAAC) or thiol-ene click chemistry, thus offering an increased number of possibilities. To illustrate, PAA-*b*-PBD has been prepared by combining poly(*tert*-butyl acrylate) (P*t*BA) and PBD homopolymers, both synthesized beforehand. A hydrolysis step leads to the final diblock [34]. Poly(dimethylsiloxane)-*block*-poly(ethylene oxide) (PDMS-*b*-PEO) diblock copolymers were synthesized using ring-opening polymerization of hexamethylcyclotrisiloxane to obtain PDMS-N3 and further coupling with PEO-Alkyne chains via click chemistry [35]. However, some reactive conditions can require high temperatures or metal catalysts, which might not be suitable for biomedical applications [36,37]. Moreover, complete end-group functionalization and equimolar ratios of both homopolymers are required, preventing the challenging removal of unreacted homopolymers. Increasing the number of blocks introduces more challenges, especially in re-initiation, purification and finding suitable solvent for all the blocks.

The self-assembly of amphiphilic block copolymers in solution leads to the formation of many different assemblies including spherical, cylindrical, gyroidal and lamellar structures [38]. These assemblies are directly influenced by intrinsic molecular parameters of the amphiphilic block copolymers and the conditions in which the self-assembly process takes place (concentration of the copolymer, presence of solvents, temperature, etc). In this respect, the hydrophilic to the total mass ratio (*f*) calculated as the ratio of the molar mass of the hydrophilic block to the total molar mass of the copolymer is an important parameter, which governs the resulting supramolecular assembly. Vesicular structures are typically obtained for *f* values ranging from 0.20 to 0.40. Another molecular parameter influencing the self-assembly into different assemblies is the packing parameter (*p* = *v*/*a*0*lc*; *v* = volume of the hydrophobic part, *a*<sup>0</sup> = contact area of the head group, *lc* = length of the hydrophobic part) that describes the degree of curvature from the membrane. For low packing parameter values (0 < *p* < 0.5), the curvature gradually decreases from high to medium, resulting in the formation of spherical or cylindrical micelles, respectively. For higher values (0.5 < *p* < 1), the curvature of the membrane is considerably low, which is more favorable for vesicular structures. The dispersity, D, of the copolymer is affecting the size distribution of the formed vesicles: a narrow dispersity typically leads to uniform-sized polymersomes, whilst on the opposite, a more polydisperse population of vesicles is obtained [39–41].

#### *2.2. Polymers as Building Blocks for Generation of Polymer Capsules*

There are a few works that produced polymer capsules via methods originating from the LbL deposition, e.g., single-step polymer adsorption, surface polymerization and ultrasonic assembly [42]. However the vast majority have employed purely the LbL assembly technique [43], where different polymer segments are alternately deposited and adsorbed. These layers are typically formed by homopolymers. The wide range of polymers provides capsules with a variety of walls, as a result of adjusting important parameters, such as composition, permeability and surface functionality of the capsules [44]. Nevertheless, such polymers must have functional groups capable of providing electrostatic interactions or hydrogen bonds. For electrostatic interactions, polyelectrolytes having anionic or cationic groups in their side chains are used, poly(styrene sulfonate (PSS) or poly(allylamine hydrochloride) (PAH), respectively [45,46]. In the case of polymers forming hydrogen bonds, the side chains are composed of functional groups called "hydrogen-bond receptors", which have at least one lone pair (carbonyl, ether, hydroxyl, amino, imino, and nitrile groups), like polyvinylpyrrolidone (PVP), or "hydrogen-bond donors" represented by the presence of a hydrogen atom covalently bound to a more electronegative atom (hydroxyl, amino, and imino groups), like poly(methacrylic acid) (PMAA). These interactions are fundamental for the formation and maintenance of the layers during the LbL deposition.

#### **3. Technologies for Engineering Polymer Single and Multicompartments in Combination with Biomolecules**

Important features of supramolecular assemblies, designed as functional single or multicompartments to accommodate active compounds, are highly dependent on the preparation methods. Aiming at obtaining the desired structures with optimized characteristics as, size and size distribution, membrane composition and specific functionalities, biomolecular content inside cavities, etc., appropriate procedures need to be selected [47].

#### *3.1. Polymer GUVs*

A wide selection of methods to generate polymer vesicular structures are available; ranging from the fairly established bulk techniques, as electroformation and film rehydration, to more automated and high-throughput ones as microfluidcs, currently still underused in the domain of cell mimicry.

#### 3.1.1. Bulk Techniques

#### **Electroformation**

Electroformation, the most common method to obtain GUVs, involves the swelling of the amphiphilic polymer film in the presence of an electric field. The dry copolymer film is deposited on conductive indium tin oxide (ITO) coated glass slides and is subjected to an alternating sin-wave electric current while it is rehydrated in aqueous solution. The former contains the desired biomolecules to be encapsulated or incorporated inside the GUVs core or membrane, respectively. The electric field induces a periodic electroosmotic movement of the water in between the individual bilayer lamellae in the film, causing the vesicle detachment from the substrate surface [47] as represented in Figure 3B. This method was successfully employed in many different occasions [14,16,48,49]. In particular, Itel et al. [14] formed giants consisting of diblocks (PMOXA-*b*-PDMS) or triblocks (PMOXA-*b*-PDMS-*b*-PMOXA), yielding membranes with thicknesses ranging from 5–30 nm, which can represent 2–10 times that of the phospholipids. Albeit this feature can contribute to a hydrophobic mismatch between membrane thickness and the size of the proteins of more than 5 times, (PMOXA-*b*-PDMS) offered enough flexibility and fluidity to facilitate the membrane protein insertion [50]. To enable reactions, Lomora et al. [51] produced GUVs of different poly(2-methyloxazoline)-*b*-poly(dimethylsiloxane)-*b*-poly(2-methyloxazoline) (PMOXA*x*-PDMS*y*-PMOXA*x*) triblock copolymers. These were equipped with a peptide (Gramicidine, gA) for inducing a selective monovalent ion permeability. Another example was the formation of GUVs with PEO-12 dimethicone, in which permeability was induced by the addition of calcimycin, an ionophore that enabled the transport of Ca2<sup>+</sup> selectively, serving for the in situ mineralization of calcium carbonate [52]. Nevertheless, this method is not recommended for charged amphiphilic copolymers due to electrostatic interactions, which might affect the self-assembly process [40].

#### **Film-Rehydration**

A more suitable technique to circumvent the problem of electrostatic interactions is the direct rehydration of a thin polymer film to form the GUVs. For example, this method succeeded in forming GUVs made of a mixture of PMOXA5-*b*-PDMS58-*b*-PMOXA5 and the negatively charged PDMS65-*b*-heparin copolymers as a mimic for heparan sulfate, known to be exposed on the plasma membrane of most cell types [12]. In the film rehydration method, the block copolymers are first dissolved in an appropriate organic solvent, followed by evaporation either with a stream of nitrogen or by applying vacuum in a rotary evaporator. Rehydration takes place by pouring aqueous solution to the dried film, resulting in the detachment of the GUVs from the substrate surface, (Figure 3A). In general, the desired hydrophilic biomolecular content is encapsulated into the GUVs cavity by mixing it to the rehydration buffer solution. Whereas, as shown by Belluati et al. [15], the hydrophobic ion channels can be inserted in different steps of the hydration processes, e.g.: (i) blended and co-dried

with the copolymer film, (ii) added to the rehydration buffer or (iii) added to the pre-formed GUVs suspension (*ex post*). Moreover, aiming at obtaining functional compartments and to follow a reaction in situ, Garni et al. [18] simultaneously encapsulated a model enzyme horseradish peroxidase (HRP) inside the polymer GUVs and inserted a channel porin, Outer membrane protein F double mutant (OmpF-M), by adding these biomolecules to the rehydration buffer during the formation process. Self-assembly process of GUVs by film-rehydration and electroformation does not produce GUVs with homogeneous sizes, instead a mixture of GUVs in the size range of 1 to 40 μm is formed. In case smaller sizes and narrower size distribution are required, the polymer giants suspension can be subjected to additional processes [53], as freeze-thawing, sonication or extrusion through a polycarbonate membrane [16,51]. Dialysis and size exclusion chromatography are alternative steps to obtaining a relatively monodisperse population.

#### **Solvent Switch/Exchange**

With the solvent switch method, the supramolecular assembly is induced by adding water drop-wise into a dissolved and molecularly dispersed polymer organic solution, thus, gradually exchanging the organic solvent with water. The turbid solution that is formed is immediately quenched by being poured slowly into an excess of water under continuous stirring. Finally, the organic solvent is removed from the solution via dialysis [54]; an important step especially when envisaging biomimetic applications [55]. However, due to the possible denaturation and degradation caused by traces of organic solvents, such a method may be incompatible with sensitive molecules, limiting their use in biomedical applications [40]. As it has been demonstrated by Daubian et al. [56], depending on the chemical nature of the amphiphile, the solvent switch method may perform faster than the film-rehydration, especially when many metastable phases of the block copolymer can be formed, leading mostly to less aggregates. [57]. With this technique, GUVs are assembled via nucleation and growth of unimers [58,59]. Due to the solvent exchange, the great number of unimers formed deplete the unimers in solution reaching rapidly phase equilibrium, and thus are prevented to grow to larger sizes. GUVs produced hence are the smallest (≈1 μm), and can be tuned to form polymersomes on the nanoscale [57].

#### 3.1.2. Microfluidics

#### **Double Emulsion Method**

Microfluidic techniques allow for the production of defined polymer stabilized water-oil-water (w/o/w) double emulsions, which are used as templates for generating GUVs. Double emulsion formation proceeds when the inner aqueous phase, containing the biological solution is enveloped by the organic phase, typically consisting of the diblock copolymer dissolved in a volatile and water-immiscible solvent, which breaks up into double emulsions, due to shear caused by the external aqueous phase (Figure 3B) [60–62]. These GUVs have narrow size distributions, with mean sizes ranging from 10–100 μm, which are highly dependent on the microdevice channel sizes and junctions (where generally droplets are formed) [63,64]. To form GUVs from double emulsions, the amphiphile chains are brought together by evaporating the volatile solvent, forming the bilayer membrane. While the complete removal of the organic phase might not be trivial and implies a limitation, this method allows for efficient encapsulation of large amounts of water soluble biomolecules [64]. Thus, its employment is vastly recommended when high-efficiency encapsulation is required, e.g., for loading enzymes and pore-forming proteins within GUVs for mimicking cells. Despite essential contribution on the development of such compartments has been made, there exists only one example where biological machinery (i.e., an aqueous mixture containing *E. coli* ribosomal extract and MreB DNA plasmid) was encapsulated into semi-permeable poly(ethylene oxide)-*block*-poly(lactic acid) (PEO-*b*-PLA) GUVs for carrying out protein expression [65].

*Chemistry* **2020**, *2*

**Figure 3.** Engineering strategies for constructing polymeric single and multi-compartments, capsules and polymer-based giant unilamellar vesicles (GUVs). (**A**) Mechanism of polymer GUV detachment from the substrate surface by the film-rehydration method. Adapted with permission from Thamboo et al. [12]. Copyright c 2019 Wiley-VCH. (**B**) Double emulsion droplets formed in a microfluidic capillary device, which serve as templates for producing GUVs. Adapted with permission from do Nascimento et al. [64]. Copyright c 2016 American Chemical Society. (**C**) Polymer microcapsules produced via layer-by-layer (LbL) deposition onto hard colloidal sacrificial templates. Mechanism using Silica particles, adapted with permission from Yan et al. [66]. Copyright c 2012 Wiley-VCH. Mechanism using CaCO3 particles, adapted with permission from Postma et al. [67]. Copyright c 2009 American Chemical Society. (**D**) Polymer microcapsules produced via double emulsion technique, followed by UV polymerization. Adapted with permission from Xie et al. [68]. Copyright c 2017 American Chemical Society.

#### *3.2. Polymer Capsules*

Differently from GUVs, polymer capsules require the employment of either a soft or a hard template. They have operated as delivery vehicles, since they also allow for the selective diffusion of reagents/reaction products; yet their use as microreactors for mimicking cells has been limited.

#### **Layer-by-layer Microcapsules**

Fabrication of polymer microcapsules involves multiple synthetic steps and compositional complexity for the particular application. The LbL technique requires the use of a colloidal particle as a sacrificial template, which plays a pivotal role, since it determines the capsule size and shape, and most importantly the biomolecular encapsulation method. Soft sacrificial templates have been employed, including the commercial ones: poly(methyl methacrylate) (PMMA) and polystyrene (PS), however they do not allow for the pre-loading of the active components, hampering the microcapsules application for therapeutics, due to low reproducibility of the diffusion process involved in the post-loading method [69]. Instead, when employing hard sacrificial templates, e.g., calcium carbonate [66] or silica [67], encapsulation of enzymes and sensitive dyes is reached via their concurrent precipitation with the template, ensuring a high loading efficiency [70,71]. Decomposition of these templates is then induced for the creation of the inner cavity loaded with the specific biomolecule (Figure 3C). With respect to the outer shell, two polymers interacting by electrostatic forces or hydrogen bonding are deposited alternately on the template before it is dissolved to obtain the hollow sphere. Typically, PVP and PMAA which interact via hydrogen bonding at pH values below the pKa of PMAA are used for this technique. Using PMAA, the stability of these capsules can be extended to physiologically relevant pH by crosslinking. The resulting pure PMAA hydrogel capsules are biodegradable, nontoxic, semipermeable and thus well suited for biomedical applications. More recent studies replace the labour intense LbL assembly of PMAA/PVP capsules by polydopamine shells that are deposited on the template in a single step [72].

#### **Double Emulsion Templated Microcapsules**

Opposite to the conventional fabrication methods, where multiple laborious synthetic steps must be satisfied, microfluidics offers an alternative technique for a rapid, with low polydispersity and highly reproducible production of polymer microcapsules. To this aim, double emulsion droplets, formed following the same procedure aforementioned, serve as non-sacrificial templates [73]. For generating polymer microcapsules, flowing droplets are subjected to UV irradiation and thus continuously and rapidly polymerize (Figure 3D). Here, the oil phase contains a photocurable polymer and a photoinitiatior dissolved in a water miscible organic solvent [69]. For biomedical applications, poly(ethylene glycol) diacrylate (PEGDA) microcapsules of around 15 μm were produced and allowed for the diffusion of molecules as large as heparin labeled with Fluorescein isothiocyanate (FITC) (*MW* ≈ 10 kD) [68]. These results demonstrate the biosensing ability and the promising versatility of microfluidics for the preparation of microreactors.

#### *3.3. Building Multicompartments*

Multicompartments are considered as an advance towards functional models for eukaryotic cells and their cellular organelles, which are able to perform multiple, chemically incompatible, enzymatic reactions simultaneously by separating them in subcompartments. Multicompartment vesicles were pioneered when the so called vesosomes (liposomes encapsulated inside larger liposomes) were first developed [74]. This process was promptly transferred to synthetic polymeric assemblies, such as polymeric vesicles or LbL capsules, resulting in all conceivable combinations. Multicompartments consist mainly of bigger outer compartments that can be loaded with different kinds of subcompartments, as subsequently detailed.

#### 3.3.1. Loading Polymeric GUVS with Subcompartments

The encapsulation of subcompartments as, small polymersomes, micelles or liposomes, but also nanoparticles, is usually attained during the polymer GUV self-assembly. Each subcompartment can be previously equipped with biomolecules and/or the biomolecules can be encapsulated together with the mixture of empty subcompartments. For example, GUVs of polystyrene-*b*-poly(L-isocyanoalanine (2-thiophen-3-yl-ethyl) amide) (PS-*b*-PIAT) were prepared by the solvent switch method, using as aqueous phase, a mixture of the cyanine-5 conjugated immunoglobulin G proteins (Cy5-IgG) and a suspension of smaller polymersomes made of PMOXA-*b*-PDMS-*b*-PMOXA, previously generated by film rehydration and equipped with green fluorescent protein (GFP) [75]. Co-localized red and green fluorescence emission measurements were used to compute that only 45% of the supramolecular assemblies resulted in multicompartments. Marguet et al. [76] also demonstrated the generation of polymer multicompartments based on the emulsion-centrifugation method. The inner polymersomes were formed by nanoprecipitation of poly(trimethylene carbonate)-*b*-poly(L-glutamic acid) (PTMC-*b*-PGA), and subsequently loaded in GUVs made of polybutadiene-*b*-poly(ethylene oxide) (PBD-*b*-PEO) by emulsion–centrifugation. By using such technique, yet for formation of giant liposomes, the loading efficiency reached up to 98% [49]. Regardless of the method used, the obtained structures will always consist of a combination of single and multicompartments. Double emulsion microfluidics has also been used to form multicompartments made of PEO-*b*-PLA diblock-copolymers for both the inner and the outer membranes [77]. Despite promised control over the number of the inner polymersomes by solely adjusting the flow rates, no loading efficiency was reported.

#### 3.3.2. Layer-by-Layer Multicompartments

LbL multicompartments are constructed with either one smaller LbL capsule as single subcompartment (shell-in-shell structure) [78] or thousands of subcompartments that are deposited onto the template during the preparation of the micron-sized outer capsule. In this regard, the subcompartments may comprise small LbL capsules, polymersomes [9] or liposomes [10], with the former being used for the majority of the LbL multicompartments. The LbL deposition offers the control over the spatial positioning of the subcompartments. Depending on the polymers used for the precursor or separation layer, they either stay attached to the inner walls of the LbL capsule or become "free-floating" after template removal [79]. If only one hemisphere of the template is exposed to the subunit deposition, Janus type multicompartments can also be prepared by the LbL approach [80]. As for single compartments, it is possible to encapsulate the biological content inside the subunits or the lumen of the main compartment, in addition, it can be also found within or outside of the membranes. Replacement of the liposomal subcompartments with polymersomes offers the possibility to address challenges, as prolonged stability of the subcompartments to sustain activity of the encapsulated enzyme. However, examples for polymersome subunits in LbL capsules remain scarce [9].

#### **4. Vesicular Compartments for In Situ Reactions**

Biomimicry offers strategies for the creation of vesicular compartments with incorporated peptides/membrane proteins and encapsulated active compounds providing an approach for various applications. Polymeric compartments with encapsulated cargo have been employed in imaging, sensing, therapeutics, as artificial cells, etc. So far, such compartments were almost solely assembled by film rehydration, electroformation, and LbL.

#### *4.1. Reactions inside Single Compartments*

#### 4.1.1. GUVs

Reconstruction of biological structures and processes can be achieved with a bottom-up approach using GUVs. Encapsulation of enzymes inside the cavities of GUVs is an emerging way to fabricate artificial environments that mimic the complexity of cells by introducing similar functionalities. The resulting GUVs serve as platforms to visualize biological processes in real time, contributing to our understanding of human cells, which in turn promotes new developments of biomedical applications [81,82]. Since the permeability of polymeric GUVs is essential for in situ reactions, one biomimicry approach is to equip them with peptides/membrane proteins to allow molecular transport through the membrane. Up to now, there are only few examples of polymeric GUVs with incorporated membrane proteins/peptides and they are primarily based on PMOXA-*b*-PDMS-*b*-PMOXA triblock copolymer membranes. For example, the permeability of GUVs, to selectively transport Ca2<sup>+</sup> ions, was attained by inserting several ionophores: calcimycin [52], Lasalocid A, and *N,N*-dicyclohexyl-*N',N"*-dioctadecyl-3-oxapentane-1,5-diamide [83]. Furthermore, Gramicidine (gA) allowed Na<sup>+</sup> and K<sup>+</sup> ions to specifically pass the membranes of GUVs [16]. The hydrophobic mismatch of pore length and membrane represented a barrier to membranes thicker than 12.1 nm, whereas thinner membranes facilitated successful gA insertion. The bee venom melittin was inserted into various PMOXA-*b*-PDMS-*b*-PMOXA membranes. The insertion process and the resulting functionality of the peptide have been related to the membrane curvature [15]. Besides, the

membrane protein OmpF was successfully reconstituted in membranes of GUVs allowing an enzymatic reaction inside the cavity, which was monitored in real time with a confocal microscope (Figure 4) [18].

**Figure 4.** Reaction inside single polymer GUVs. (**A**) Schematic representation of a polymeric GUV equipped with the membrane protein OmpF. Substrates and products diffuse through the membrane, thus enabling an enzymatic reaction. (**B**) Fluorescence micrographs of a single GUV recorded at several time points after addition of the substrates showing the difference of GUVs with and without reconstituted OmpF. Adapted with permission from Garni et al. [18]. Copyright c 2018 American Chemical Society.

#### 4.1.2. Layer-by-Layer Microcapsules

LbL capsules with an encapsulated enzyme offer various possibilities in sensing and imaging [84]. The preparation of (PSS/PAH)4/PSS shell structures, the co-encapsulation of urease and the pH sensitive fluorophore enabled the quantification of urea on a single capsule level [70]. Continuing with this approach Kazakova et al. [71] managed to encapsulate lactate oxidase, peroxidase, or glucose oxidase, with respective sensitive dyes to detect lactate, oxygen, and glucose levels [71]. In another (PSS/PAH)4 system the detection of oxaloacetic acid with NADH as cofactor was possible. Thus, the efficacy of an enzyme fluorophore coupled system was demonstrated (Figure 5) [85]. Magnetic polydopamine capsules enhanced the activity after reusing and the long-term stability of the encapsulated Candida Rugosa Lipase compared to the free enzyme [86]. Reuse is a key factor for potential application in industry. Moreover, further attempts are required to validate the performance of these systems in vitro and in vivo. Another application of LbL capsules is therapeutics: e.g., microcapsules with encapsulated L-Asparaginase in poly-L-arginine and dextran sulfate layers were tested in vitro on leukemic cell lines resulting in a decreased proliferation [87]. LbL enables convenient encapsulation of enzymes in one single particle. However, their semi-permeable membrane allowing unspecific transport of small molecules is rather a deficiency, that needs to be overcome for future applications.

#### *4.2. Reactions inside Multicompartmentalized Structures*

#### 4.2.1. GUV Multicompartments

In biological cells, evolution has developed the system of subcompartmentalization (cellular organelles) within individual cells in order to allow specific reactions to take place in a spatially defined manner. This is an efficient solution, as many reactions (e.g., protein lysis, electron transport) require very specific conditions (e.g., low pH, proton gradient) to occur. Careful application of biomimicry principles allows integral cell mimics as combining nano- and microstructures with biomolecules. In this respect, biomolecule equipped polymersomes have previously been shown to be functional as artificial organelles both in vitro and in vivo where they supported the natural cellular metabolism, and have even been shown to function "on demand" in a life-like manner [11,88]. Artificial cell mimics have been designed by constructing synthetic multicompartmentalized systems. The most significant

examples were found when using a larger polymer-based GUV loaded either with smaller nano-sized liposomes or smaller nano-sized polymersomes [89]. Synthetic GUVs based on (PBD46-*b*-PEO30) loaded with hydrophilic dyes, liposomes (DPPC), and polymersomes (PBD23-*b*-PEO14) allowed fast, selectively triggered release due to a light-induced increase in osmotic pressure, resulting in rupture of the GUVs [90]. Examples of polymeric GUVs acting as artificial cell-mimics are still scarce, but more complex multicompartmentalized GUVs exist. Namely, where enzymes and membrane protein equipped polymersomes coexisted in the GUVs inner cavity [12]. By applying the principle of multicompartmentalization, an artificial cell mimic is created with subcompartmentalized polymersomes acting as artificial organelles. This allows cascade reactions to occur successively due to the segregation of enzymes in different subcompartments. The proximity among subcompartments provided by larger GUVs, facilitates the diffusion of reagents and reaction products, while confining the enzymes to their individual subcompartments. PBD-*b*-PEO polymer GUVs can mimic structural and functional eukaryotic cells by encapsulating enzyme-filled intrinsically semi-permeable PS-*b*-PIAT polymer nanoreactors together with free enzymes and substrates to fulfill a three-enzyme cascade reaction inside the multicompartmentalized structures [13]. Although this study represents an important step towards artificial cells, it only reports the fluorescent product of the reaction, without providing detailed information about localization of the enzymes. In addition, such examples lack the complexity of cells because they are mainly developed with only few functional elements and by using buffer medium. For the creation of an artificial cell mimic by multicompartmentalization, there are requirements still not fulfilled. The selective permeabilization of every membrane of the involved compartments, which allows for a higher control of the diffusion of substrates and products across the membranes and a more complex medium mimicking the cytoplasm represent advancements not yet provided. Systems addressing this question are multicompartmentalized GUVs with stimuli-responsive and non-responsive subcompartments (Figure 6). With an external signaling molecule passively diffusing through the GUV's membrane, inducing the disassembly of the stimuli-responsive nanoparticle and the release of the entrapped cargo (peptides or enzyme substrates). These molecules allowed a selective ion flux through the GUV's membrane or an enzymatic reaction inside the GUV [12].

**Figure 5.** Reaction inside single polymer microcapsules. (**A**) Schematic illustration of an oxaloacetic acid (OAA) or nicotinamide adenine dinucleotide (NADH) sensing microcapsule. The encapsulated pH-sensitive fluorescent dye ( seminaphtharhodafluor (SNARF-1)-dextran) responds to a decrease in local proton concentration caused by the enzymatic reaction. (**B**) Reaction kinetics demonstrating NADH as the limiting factor. The first dose of substrate is added at (\*) and then added gradually, after the plateau was reached, from (\*\*)-(\*\*\*\*). The corresponding micrographs on the right hand side represent the reaction of one capsule (red (R), yellow (Y), transmission (TM), and the false-colored ratio I*r*/I*y* (R/G)). Adapted from Harimech et al. [85] under the terms of CC BY 3.0.

More and more experimental successes in combination with vesicle engineering techniques are leading into a new era of complexity in artificial cells. These have attracted increasing attention as substituents for living cells. Polymer GUVs and polymersomes offer an ideal platform to engineer cell mimics, allowing reactions to take place in compartmentalized spaces. Meanwhile, they remain stable for longer periods compared to their lipid-based counterparts. With the beforehand mentioned functionalization with membrane proteins and peptides, a life-like functionalization of membranes is approved.

**Figure 6.** Multicompartmentalized GUV with reduction sensitive and ion channel recruiting modular subcompartments (**A**–**D**). Dithiothreitol was used as a triggering signal (red arrow). Reprinted With permission from Thamboo et al. [12]. Copyright c 2019 Wiley-VCH.

#### 4.2.2. Layer-by-Layer Polymer Capsules as Multicompartments

Multicompartmentalization allows for separation in preparation and modularity in formulation to increase also functionality in theranostics [91,92]. LbL assembled capsules prepared of polyvinylpyrrolidone (PVP) and thiolated poly(methacrylic acid) (PMAA) containing smaller crosslinked capsules showed possibilities in catalysis and drug delivery [8]. PVP and tannic acid (TA) LbL capsules filled with POEGMA26-*b*-PDPA50 polymersomes loaded with pDNA could release the cargo in response to pH changes [9]. Microfluidics have been improving the development of such attractive microreactor systems with increased complexity and modularity. Encapsulation of glucose oxidase (GOx) conjugated on quantum dots or on gold nanorods (NR) into PEO-based microreactors acting as glucose biosensor, while amine functionalized NRs were employed as heparin sensors. GOx oxidized glucose to gluconic acid and H2O2 leading to fluorescence quenching by the quantum dots providing a sensitivity in the range of glucose sensors for diabetes diagnostics [68]. However, most capsules only work in defined pH-ranges and do not resist enormous local pH-changes, which would be necessary for in vivo applications [84].

#### **5. Conclusions and Outlook**

Biological systems as cells are highly compartmentalized across several length scales. Their precise features, as biomolecule compartmentalization through attachment to membranes and cytoskeleton scaffolds, lateral organization on membrane rafts, as well as compartmentalization in membrane-bound or protein-based organelles, are current subject of study. Understanding their underlying mechanisms opens exciting avenues in many application fields, notably in material science, biotechnology and medicine. As we have seen, efforts at achieving this are accelerating and synthetic approaches to mimic increasingly intricate biological structures are being developed.

This review demonstrates the relevance of polymer-based systems with special focus on polymer GUVs and capsules, for addressing the challenge of eukaryotic cell biomimicry. We only reviewed examples of supramolecular structures, whose membrane is equipped with peptides and membrane proteins since they genuinely represent bioinspired catalytic compartments where the membranes are mimicking biomembranes.

The progress in polymer chemistry gave access to a variety of polymers with tailored properties as biodegradability and non-toxicity that leads to enhanced structural properties and with the significant fluidity, necessary to cope with the insertion of peptides and membrane proteins. Due to their large variety of chemical composition and functionalization, a greater versatility for further improvement of their properties concerning the desired application can be achieved, regarding stability, loading efficiency, intervesicular interaction, and selective permeability for specific substrates. Besides the polymer, the selection of an appropriate preparation method for engineering the supramolecular structures in single and multicompartments to conform biochemical reactions, is detrimental. Although electroformation and film rehydration are the most commonly used techniques, both of them do not produce homogeneous sizes of GUVs. Alternatively, microfluidics became a serious candidate, showing great potential to generate polymer GUVs with narrow size distribution and controlled biomolecular content at high-throughput.

GUVs loaded with enzymes inside their cavities and equipped with peptides/membrane proteins acting as "gates" for the diffusion of molecules through the synthetic membrane, constitute complex reaction spaces. Careful application of biomimicry principles allows integral cell mimics as combining nano- and microstructures with biomolecules. The inner compartments (the organelle mimics) that are loaded in the outer one (the cell membrane mimic) do not necessarily need to encapsulate the same content, where the various contents can even be incompatible or act synergistically. As a result, combinatory drug delivery becomes possible in one single vector. Such systems serve multifold purposes. Besides the aforementioned design of an artificial cell, allowing systematic studies of biological phenomena in simplified environments, they are explored as (compartmentalized) microreactors, where segregation of the catalytic steps in separated compartments allows distinct chemical environment (e.g., different pH, redox states, presence of cofactors, etc.) to couple enzymatic reaction steps that would otherwise inhibit one another. Because of the GUV cell-size, real time imaging of the fluorescence activity of model enzymes can be monitored and used for enhancing diagnosis capabilities. Lastly, polymer capsules have been aimed for use in therapeutic applications, by encapsulating hydrophilic molecules in the aqueous core for enzyme therapy and controlled drug release. In addition, such structures can improve the therapeutic index of drugs by influencing drug absorption and metabolism, and can extent drug half-life as well as reduce toxicity. Although an effort at outlining a roadmap for the field had been made, there will surely be many new developments that will take this research area to unforeseen directions. From a material perspective, the advanced control in polymer synthesis and self-assembly that is available nowadays would certainly bring a real breakthrough in this cell biomimicry field. Hence, it is expected a growing interest in such biomimetic approaches to soon offer many new opportunities in drug delivery, cell-sized reactors, biosensors and imaging for therapeutics, which will offer better communication and interaction with living systems.

**Author Contributions:** E.C.d.S., A.A. and D.H. carried out literature research and wrote the manuscript. W.M. and and C.G.P. provided additional guidance and assisted in finalizing the manuscript. All authors reviewed the final version of the manuscript and approved it for publication.

**Funding:** The authors acknowledge financial support from the Swiss National Science Foundation, NCCR-MSE, and University of Basel.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **List of Abbreviation**


*Chemistry* **2020**, *2*


#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Combining the Sensitivity of LAMP and Simplicity of Primer Extension via a DNA-Modified Nucleotide**

#### **Moritz Welter and Andreas Marx \***

Department of Chemistry & Konstanz Research School Chemical Biology, University of Konstanz, 78457 Konstanz, Germany; moritz.welter@uni-konstanz.de

**\*** Correspondence: andreas.marx@uni-konstanz.de; Tel.: +49-7531-88-5139

Received: 16 April 2020; Accepted: 11 May 2020; Published: 14 May 2020

**Abstract:** LAMP is an approach for isothermal nucleic acids diagnostics with increasing importance but suffers from the need of tedious systems design and optimization for every new target. Here, we describe an approach for its simplification based on a single nucleoside-5 -O-triphosphate (dNTP) that is covalently modified with a DNA strand. We found that the DNA-modified dNTP is a substrate for DNA polymerases in versatile primer extension reactions despite its size and that the incorporated DNA indeed serves as a target for selective LAMP analysis.

**Keywords:** modified nucleotide; DNA polymerase; LAMP; primer extension

#### **1. Introduction**

Despite the widespread use of PCR-based amplification, the drawback of this technology, however, is its need for temperature cycling. Many attempts have been made to develop isothermal amplification methods that do not require heating of the double-stranded nucleic acid for the separation of templates [1]. These methods include strand-displacement amplification (SDA) [2], rolling circle amplification (RCA) [3], and helicase-dependent amplification (HDA) [4]. Another important method for nucleic acids diagnostics is loop mediated isothermal amplification (LAMP) [5,6]. LAMP relies on auto-cycling strand displacement DNA synthesis that is performed by a DNA polymerase with high strand displacement activity and a set of two specially designed inner and two outer primers (for details of the method, see Supplementary Materials Figure S1 in the ESI).

The reaction can be monitored by e.g., the addition of dyes used for nucleic acid staining [7] or turbidity analysis of precipitating magnesium phosphate [8]. Furthermore, the use of low-buffered reaction mixtures allows amplification monitoring with pH sensitive indicator dyes by the naked eye, as during the DNA polymerase reaction a proton is released for each nucleotide incorporation [9]. LAMP assays have also been adapted for many applications e.g., to cover genotyping and RNA detection [10–13]. However, in order to work as intended, the primers required for the amplification have to be carefully designed to meet particular requirements in regards to their melting temperature, spacing, and concentrations [5]. Thus, the design of suitable LAMP primers has to be optimized for every target that can be tedious, even when done with specific design software. While setting up the LAMP reaction, further problems are described such as the amplification of non-template controls, which drastically impedes the reliability of LAMP assays [14,15].

#### **2. Materials and Methods**

#### *General*

All reagents and solvents were obtained from Sigma-Aldrich (Darmstadt, Germany) and used without further purification. All synthetic reactions were performed under an inert atmosphere. Flash chromatography was performed using Merck silica gel G60 (230–400 mesh, Darmstadt, Germany)

and Merck precoated plates (silica gel 60 F254) were used for TLC. Anion-exchange chromatography was performed on an Äkta Purifier (GE Healthcare, Chicago, IL, USA) with a DEAE Sephadex™ A-25 (GE Healthcare Bio-Sciences, (GE Healthcare, Chicago, IL, USA) column using a linear gradient (0.1 M–1.0 M) of triethylammonium bicarbonate buffer (TEAB, pH 7.5). Reversed phase high pressure liquid chromatography (RP-HPLC, Shimadzu, Kyoto, Japan) for the purification of compounds was performed using a Shimadzu system having LC8a pumps and a Dynamax UV-1 detector (RP-HPLC, Shimadzu, Kyoto, Japan). A VP 250/16 NUCLEODUR C18 HTec, 5 μm (Macherey-Nagel, D) column and a gradient of acetonitrile in 50 mM TEAA buffer were used. All compounds purified by RP-HPLC were obtained as their triethylammonium salts after repeated freeze-drying. NMR spectra were recorded on Bruker Avance III 400 (1H: 400 MHz, 13C: 101 MHz, 31P: 162 MHz, Billerica, MA; USA) spectrometer. The solvent signals were used as references and the chemical shifts converted to the TMS scale and are given in ppm (δ). HR-ESI-MS spectra were recorded on a Bruker Daltronics microTOF II. KlenTaq DNA polymerase was expressed and purified as described before. [16] T4 polynucleotide kinase (PNK) was purchased from New England BioLabs (Ipswich, MA; USA). [γ-32P] ATP was purchased from Hartmann Analytics (Braunschweig, Deutschland) and natural dNTPs from Thermo Scientific (Waltham, MA, USA).

**Synthesis of nucleoside triphosphate 2.** To a solution of 18.9 μmol (10.3 mg) 5-(aminopentynyl) -2 -deoxyuridinetriphosphate tetrabutylammonium salt (**1**) [16] in 1 mL DMF, 94.5 μmol (5 eq., 9.5 mg) of NEt3 were added. In parallel, 37.8 μmol (2 eq, 11.2 mg) of 16- azidohexadecanoic acid, 94.5 μmol (5 eq, 9.5 mg) NEt3 and 37.8 μmol HATU (2 eq., 14.4 mg) were dissolved in 1 mL DMF and stirred for 30 min. Both mixtures were then combined and stirred at room temperature for additional 12 h. The solvent was removed under reduced pressure and the residual oil was subjected to C18-RP-HPLC (95% 50 mM triethylammonium acetate (TEAA) buffer to 100% MeCN). 2 was obtained in 58% yield as determined by Nanodrop ND1000 spectrometer with <sup>ε</sup> (290 nm) = 13,300 M−1·cm<sup>−</sup>1. The compound was diluted in MilliQ water and kept as a 10 mM stock solution at −20 ◦C.

Analysis of **2**: 1H NMR (400 MHz, Methanol-d4) δ 8.01 (s, 1H, H-C(6)), 6.26 (t, <sup>3</sup>*J* = 6.8 Hz, 1H, H-C(1 )), 4.63–4.57 (m, 1H, H-C(3 )), 4.34–4.26 (m, 1H, H-C(5 a)), 4.23–4.17 (m, 1H, H-C(5 b)), 4.11–4.06 (m, 1H, H-C(4 )), 3.29 (t, <sup>3</sup>*<sup>J</sup>* <sup>=</sup> 6.9, 2H, H-C(L16)-), 3.22 (q, <sup>3</sup>*<sup>J</sup>* <sup>=</sup> 7.1, 17H, <sup>−</sup>CH2CH2CH2NH-, Et3N), 2.46 (t, <sup>3</sup>*J* = 6.9 Hz, 2H, H-C(L2)), 2.32–2.25 (m, 2H, H-C(2 )), 2.1 (t, <sup>3</sup>*J* = 7.5, 2H, <sup>−</sup>CH2CH2CH2NH-), 2.20 (t, <sup>3</sup>*J* = 7.6 Hz, 2H, H-C(L2)), 1.79 (p, <sup>3</sup>*J* = 6.8 Hz, 2H, <sup>−</sup>CH2CH2CH2NH-), 1.60 (p, <sup>3</sup>*J* = 6.8 Hz, 4H, H-C(L3+15)), 1.45–1.27 (m, 51H, H-C(4-14), Et3N). 31P NMR (243 MHz, Methanol-d4): <sup>δ</sup> <sup>=</sup> <sup>−</sup>10.44 (d, <sup>2</sup>*<sup>J</sup>* <sup>=</sup> 20.5 Hz), <sup>−</sup>11.33 (d, <sup>2</sup>*<sup>J</sup>* <sup>=</sup> 21.3 Hz), <sup>−</sup>23.72 (t, <sup>2</sup>*<sup>J</sup>* <sup>=</sup> 21.3 Hz). HR-ESI-MS (*m*/*z*): [M <sup>−</sup> H]<sup>−</sup> <sup>=</sup> calcd: 827.2552; found: 827.2562.

**Preparation of dT15LAMPTP.** The split LAMP target sequence was ligated using T4 DNA ligase and a splint oligonucleotide. 1 nmol of LAMP\_TARGET\_A and LAMP\_TARGET\_B (10 μM, Biomers.net) were mixed with 2 nmol (20 μM) of the splint oligonucleotide in a total volume of 98 μL of 1× T4 ligase buffer provided by the manufacturer (NEB). The mixture was heated to 95 ◦C for 2 min and slowly cooled down to 25 ◦C. Subsequently, 2 μL of T4 Ligase (800 U) were added and the reaction was incubated at 16 ◦C overnight. The mixture was then diluted to 200 μL with MilliQ water and subjected to 95 ◦C for 5 min. Ion-exchange HPLC was performed at 85 ◦C column temperature using 100 μL of the solution on an analytical HPLC system with a semi-preparative Thermo Scientific™ Dionex™ DNAPac™ PA100 column and a gradient from IEX-HPLC buffer A (25 mM Tris-HCl, pH 8) to IEX-HPLC buffer B (25 mM Tris-HCl, 0.5 M sodium perchlorate, pH 8). Peaks demonstrating an absorbance at λ = 260 nm were collected and pooled in Amicon 4 centrifugal filters. After repeated washing with MilliQ water, the ligated LAMP target was transferred to a 1.5 mL reaction tube and absorbance was measured by NanoDrop ND-1000 spectrometry at 260 nm with ε (403 nm) = 1,752,400 M−<sup>1</sup> cm−1. To conjugate the 5 -DBCO labeled LAMP target with compound **2**, the above generated oligonucleotide was incubated with 10 eq of the nucleotide in 1× PBS (pH 7.4) overnight. IEX-HPLC and Amicon purification were repeated to yield **dT15LAMPTP**.

**Primer extension (PEx) in solution with dT15LAMPTP.** To 1× polymerase buffer (50 mM Tris-HCl, 16 mM ammonium sulfate, 2.5 mM magnesium chloride, 0.1% Tween 20, pH 9.2), 150 nM 5 - 32P-labeled primer and 200 nM template were added. The mixture was annealed at 95 ◦C for 5 min. Subsequently, DNA polymerase was added (100 nM KlenTaq DNA polymerase) and the reaction was started by addition of the dNTP (1 μM final concentration). Time points were collected by quenching 2 μL of the reaction mixture with 10 μL stopping solution (80% *v*/*v* formamide, 20 mM EDTA, 0.025% *w*/*v* bromophenol blue, 0.025% *w*/*v* xylene cyanol). Denaturing polyacrylamide gels (9%) were prepared by polymerization of a solution of urea (8.3 M) and bisacrylamide/acrylamide (9%) in TBE buffer using ammonium peroxodisulfate (APS, 0.08%) and *N,N,N ,N* -tetramethylethylene-diamine (TEMED, 0.04%). Immediately after addition of APS and TEMED, the solution was filled in a sequencing gel chamber (Bio-Rad) and left for polymerization for at least 45 min. After addition of TBE buffer (1×) to the electrophoresis unit, gels were pre-warmed by electrophoresis at 100 W for 30 min and samples were added and separated during electrophoresis (100 W) for approximately 1.5 h. The gel was transferred to Whatman filter paper, dried at 80 ◦C in vacuo using a gel dryer, and exposed to an imager screen. Readout was performed with a molecular imager FX.

**LAMP assay in solution.** To avoid contaminations, all LAMP reactions were pipetted with Biosphere filter tips. The LAMP target used for the positive controls was pipetted with a second pipette set. Initial LAMP reactions were performed according to the conditions reported by Tanner and co-workers [9] with 8 U Bst 2.0 WarmStart® DNA Polymerase (NEB) DNA polymerase, 1<sup>×</sup> SYBR I, 0.2/0.4/1.6 μM LAMP primers (outer/loop/inner), 10 nM 5 -DBCO LAMP target (positive control), 350 μM/dNTP, 65 ◦C in 1× isothermal amplification buffer (NEB) with 8 mM MgSO4. LAMP reactions in optimized conditions were carried out using 200 μM dNTPs, 0.2/0.4/1.6 μM primers (outer/loop/inner), 1× SYBR I, 4 U of Isotherm2G DNA polymerase (myPOLS Biotec, Konstanz, Germany) and 0.1 nM 5 -DBCO LAMP target (positive control) in a total of 10 μL of 1× Isotherm2G buffer at 55 ◦C for the indicated amount of time. Fluorescence of SYBR I was measured in 1 min intervals minute in a Bio-Rad CFX384 Touch™ Real-Time PCR Detection System. Following the amplification, melting point measurement was carried out with a gradient from 55 ◦C to 95 ◦C in 0.5 ◦C steps.

**Primer extension (PEx) and LAMP assay using immobilized primers.** Pierce Streptavidin coated 8-well strips (ThermoFisher, Waltham, MA, USA) were washed twice with 200 μL of 1× plate washing buffer. Subsequently, 1 μL of 500 μM 5 -biotin BRAF primer in 100 μL 1× PBS buffer were added to each well. After 15 min of incubation at room temperature, the primer solution was removed and 200 μL of 1 mM (D)-+-biotin in 1× PBS were added. After 5 min of incubation, the liquid was removed and the wells were washed once with 200 μL of PBS buffer and twice with 200 μL of 1× KTq reaction buffer. Following this, 50 μL of PEx reaction mixture (100 nM KlenTaq DNA polymerase, 200 nM BRAF template, 200 nM **dT15LAMPTP** in a total of 50 μL 1× reaction buffer (50 mM Tris-HCl, 16 mM ammonium sulfate, 2.5 mM magnesium chloride, 0.1% Tween 20, pH 9.2) were applied, the wells were sealed with PCR foil seal and incubated at 55 ◦C (measured with a digital thermometer) in a shallow water bath on a thermal block for 30 min. The supernatant was removed and the wells were washed first twice with 100 μL 1× reaction buffer, then three times with 1× PBS buffer and finally rinsed for 2 min under a water tap with MilliQ water. All liquid was removed and the wells were finally washed with 200 μL of 1× Isotherm2G DNA polymerase buffer. 50 μL of LAMP reaction mixture (200 μM dNTPs, 0.2/0.4/1.6 μM primers (outer/loop/inner), 1× SYBR I, 4 U Isotherm2G DNA polymerase) were employed in each well. For the positive control, 2 nM 5 -DBCO LAMP targets were added. The wells were incubated at 55 ◦C for the primer extension. Amplification was stopped by rapidly cooling the wells to 0 ◦C in an ice bath. Samples were instantly collected and run on a 2.5% agarose gel. The gel was read out using GelRed staining under UV light on a Chemidoc™ XRS system (Bio-Rad, Hercules, CA, USA).

#### **3. Results**

Here, we describe an approach towards the simplification of LAMP reactions based on a nucleoside-5 -*O*-triphosphate (dNTP) that is modified with a DNA strand serving as a LAMP target (Figure 1).

**Figure 1.** Depiction of the approach explored in this study. A primer is immobilized on a solid support via biotin-streptavidin interaction. After annealing of the template sequence, DNA polymerase and the LAMP template-modified dNTP (dTLAMPTP) are added. After incubation, the unbound conjugate is removed by repeated washing and the LAMP reaction mixture is added afterwards. Only in cases where matched primer template duplexes are present (top), the LAMP reaction is expected to be positive.

The LAMP template-modified dNTP (dTLAMPTP) may be a substrate for DNA polymerases in sequence selective primer extension reactions on solid support. Single nucleotide incorporation is highly sequence selective [17,18] and allows discrimination of single nucleotide variations by the mere difference of incorporation efficiencies of a matched versus mismatched nucleotide. In turn, the immobilized LAMP template (green, Figure 1) can be targeted by strand displacement-proficient DNA polymerases in LAMP reactions. This approach holds promise offering the advantage that—in principle—with one single LAMP sequence an infinite number of targets can be analyzed without tedious redesign of the LAMP sequence, since the immobilized primer strand (orange, Figure 1) is responsible for the sequence selective capture of the target (blue, Figure 1).

The approach depicted above is based on the covalent connection of the LAMP template to a nucleotide. A LAMP template, however, has to harbour six distinct binding sites in fixed spacing. Typical sequences hence consist of around 200 or more nucleotides (nt), which vastly exceed the oligonucleotide-modified nucleotides that have been reported for successful incorporation and which were modified with up to 40 nucleotides [19]. We chose a LAMP template (sequence see Table 1) with a 245 nt sequence taken from the genome of the Lambda phage [9,20]. The sequence was shortened by 61 nucleotides in the middle area and split up into two halves with lengths of 91 nt and 93 nt. For conjugation to the nucleotide, the oligonucleotide representing the 5 -end of the target sequence was equipped with a 5 -dibenzocyclooctyne (DBCO) modification (Figure 2). The 3 -half of the sequence was phosphorylated on its 5 -end to allow splint ligation with T4 DNA ligase, which was carried out at 16 ◦C in presence of two equivalents of a 30nt splint.


**Table 1.** Employed DNA sequences.

**Figure 2.** Synthesis of LAMP template-conjugated dT15LAMPTP. Left: reaction conditions: 1, DMF, HATU, Et3N, rt, 12 h; right: the two modified halves of the LAMP template are ligated by splint ligation with T4 DNA ligase yielding a 5 -DBCO-modified 184mer. The click reaction between the azide-functionalized nucleotide and the LAMP template is carried out in PBS buffer at room temperature 12 h and the product is purified by ion-exchange HPLC.

To react with the 5 -DBCO modified oligonucleotide that harbors the LAMP sequence, an azide-functionalized nucleotide was prepared starting from the known dTTP analog 1 (Figure 2) [16]. We chose a linker length that has been demonstrated before to be suitable for appending large "cargo" to dNTPs without greatly compromising DNA polymerase activity [19]. Employment of 16-azidohexadecanoic acid [21], HATU, and Et3N in DMF yields compound 2. Conjugation of 2 and the LAMP template by strain-promoted 1,3-dipol cycloaddition (SPAAC) [22] was achieved in PBS buffer and the product dT15LAMPTP was purified by ion exchange HPLC and centrifugal filtration.

Next, primer extension experiments were conducted with dT15LAMPTP in comparison with natural dTTP and the dTTP derivative 2 with KlenTaq DNA polymerase using a template containing the B type raf kinase (BRAF) T1796A point mutation, which is strongly associated with carcinogenesis [23]. After incubation, samples of the primer extension (PEx) reaction were quenched and analyzed by denaturing polyacrylamide gel electrophoresis (PAGE). For dTTP, the expected shift for single nucleotide incorporation (Figure 3) was observed.

**Figure 3.** Primer extension experiment employing the dT15LAMPTP. (**A**) The LAMP target sequence, KlenTaq DNA polymerase, natural dTTP and compound 2 drawn to scale. (**B**) Partial sequence of the incorporation site in the BRAF sequence context and the three different nucleotides used in this experiment. (**C**) PEx experiment with KlenTaq DNA polymerase at 55 ◦C with 1 μM dT15LAMPTP. P: Primer, 1: dTTP, 1 min, 2: dTTP, 30 min; 3: Compound **2**, 1 min, 4: Compound 2, 30 min; 5: dT15LAMPTP, 1 min, 6: dT15LAMPTP, 30 min; M: Marker.

Processing of compound **2** and incorporation of the respective modified nucleotide into the nascent DNA strand led to a pronounced shift of the product by PAGE analysis due to the long alkyl chain-modification impeding migration through the gel matrix. Finally, the usage of dT15LAMPTP led to a very pronounced shift of the product in PAGE analysis, similar to that observed when protein-conjugated nucleotides and shorter oligonucleotide-conjugated nucleotides were used [19,21,24–26]. The band corresponding to the LAMP sequence-conjugated nucleotide runs at approx. 225 nt, which is consistent with the combined size of the 21-mer primer, the 184 nt LAMP sequence, and the connecting alkyl linker. Therefore, not only was the conjugation between the LAMP sequence and the nucleotide confirmed, but it was also shown that this DNA polymerase is able to incorporate nucleotides equipped with ssDNA, being considerably longer than the sequence context used for incorporation.

With the LAMP target-modified nucleotide dT15LAMPTP in hand, the LAMP reaction itself was optimized. The assay was conducted as reported in the original publication with Bst 2.0 DNA polymerase, 350 μM for each dNTP, and 0.2/0.4/1.6 μM primers (outer/loop/inner, respectively) at 65 ◦C and monitored by real-time SYBR green I fluorescence detection [9]. Using these conditions, the positive control containing 10 nM of the ligated LAMP target was amplified at a cycle quantification value (Cq) of 15.4, but all non-template controls (NTC) showed a similar amplification ranging from Cq 26.4 to 46.6 (Figure 4A). To ensure that this behavior was not caused by a contamination with LAMP target, the experiment was repeated several times with freshly prepared reagents. However, amplification within non-template controls was persistent in all runs. Similar issues were reported in other studies with false positive amplification in LAMP assays [14,15].

**Figure 4.** Real-time monitoring of the LAMP with SYBR green I using dT15LAMPTP. (**A**) LAMP Assay using the conditions reported by Tanner et al. [9] with Bst 2.0 DNA polymerase at 65 ◦C, 0.2/0.4/ 1.6 μM outer/loop/inner primers, 8 mM MgSO4 and 350 μM dNTP each. (**B**) Optimized conditions with Isotherm2G DNA Polymerase at 55 ◦C, 0.2/0.4/1.6 μM outer/loop/inner primers, 2 mM MgSO4 and 200 μM dNTP each.

To overcome these issues, a screening for appropriate LAMP conditions was carried out including different DNA polymerases, incubation temperatures, primer ratios and concentrations of dNTPs, Mg2+, SYBR green I and betaine. In the end, the assay conditions were changed to Isotherm2G DNA polymerase, 55 ◦C reaction temperature, 200 μM dNTP each with SYBR green I and remaining primer concentrations at their original levels (0.2/0.4/1.6 μM outer/loop/inner primer). Primers were denatured prior to the addition to the master mix in order to minimize the effect of primer dimers or the presence of any self-primed secondary structure. With the optimized conditions, 0.1 nM of the ligated LAMP target were detected at Cq 28 with no to minimal false positive reactions, which was amplified at sufficiently delayed time points (Cq 51, Figure 4B).

Having optimized the LAMP conditions, we next investigated immobilized primers in a primer extension of the LAMP target-modified nucleotide and subsequent LAMP reaction. Therefore, we used streptavidin coated 8-well plates as the solid phase to immobilize biotin-modified primer strands. The wells were first incubated with 5 biotinylated primer in PBS and subsequently blocked with biotin. After several washing steps, 50 μL of primer extension reaction mixture containing 100 nM KlenTaq DNA polymerase, 1 μM dT15LAMPTP, and 200 nM BRAF template was added (Figure 5, lane 1).

**Figure 5.** LAMP detection of DNA targets. A PEx reaction employing dT15LAMPTP and KlenTaq DNA polymerase was performed on an immobilized primer in the presence or absence of the matched PEx template. Subsequently, the wells were washed and a LAMP reaction mixture was added. Samples of each well were taken and analysed on a 2.5% agarose gel. Picture colours were inverted to improve contrast. 1: 200 nM (10 pmol) matched BRAF template, 2: no BRAF template, 3: 200 nM BRAF template but natural dTTP instead of dT15LAMPTP, 4: no PEx reactions but LAMP reaction spiked with LAMP target sequence as a positive control, M: marker.

Controls were set up without a BRAF template (lane 2) or with dTTP instead of the LAMP target modified nucleotide dT15LAMPTP (lane 3). Following 30 min of incubation at 55 ◦C, the reaction mixture was removed. After the plates were washed intensively with 1× KlenTaq reaction buffer, PBS buffer, MilliQ water and 1× Isotherm2G reaction buffer, the LAMP reaction mixture was applied. Furthermore, an additional sample (Figure 5, lane 4) was treated equally as the other samples, but incubated in KlenTaq reaction buffer instead of a primer extension reaction mixture. Here, the LAMP reaction was spiked with 2 nM of LAMP target to serve as a positive control. The LAMP reaction was incubated at 55 ◦C for 20 min and stopped by rapid cooling of the plates in ice water. Samples were taken and directly subjected to analysis by agarose gel electrophoresis. Upon agarose gel electrography analysis, a ladder-like pattern of bands was observed for the positive reaction (Figure 5, lane 1). The pattern is consistent with the positive control in well 4, which proves the specific amplification starting from the LAMP target. No amplification was observed for both negative controls analysed in lanes 2 and 3. Hence, a LAMP reaction can be utilized to detect the presence of a PEx template using the LAMP target-conjugated dTTP derivative dT15LAMPTP.

#### **4. Conclusions**

In summary, a shortened LAMP target sequence derived from the genome of the Lambda phage was generated and conjugated to an azide-functionalized dTTP derivative via click chemistry to yield dT15LAMPTP. Primer extension experiments with the LAMP conjugate revealed that KlenTaq DNA polymerase is able to incorporate the modified nucleotide into a primer strand in spite of the length of the attached oligonucleotide. Next, dT15LAMPTP was employed in primer extension reactions in solid phase in which the modification, if covalently connected to the primer, served as a reporter for the presence of the template in the primer extension reaction. Indeed, we found that amplification starting from the immobilized LAMP template was only observed if the template for the preceding primer extension was present in the reaction due to processing of dT15LAMPTP. Thus, the results demonstrate proof-of-concept that the robustness and simple setup of primer extension-based assays with the rapid and sensitive amplification of LAMP is feasible. The herein depicted results might advance and simplify LAMP-based applications.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2624-8549/2/2/490\T1\ textendash498/s1, Figure S1: Nucleic acid amplification by LAMP. The two inner primers (green in Step I and orange in Step II) comprise of a site complementary to a sequence in the target oligonucleotide and a 5 overhang that is complementary to a site within the elongated primer (F1c, B1c). After the inner primer is elongated, the outer primer (black) binds upstream of the inner primer at the target sequence and its elongation by the strand displacement DNA polymerase releases the prolonged inner primer (Step I). The procedure is then repeated at the other side of the released, elongated inner primer (Step II), generating a new sequence that is similar to the target sequence, but instead of the outer primer binding site, it is now on both sides equipped with a sequence complementary to an area inside the oligonucleotide (orange, Step III). Annealing of these complementary sequences will lead to a dumb bell-like structure in which first, the self-primed 3 end is elongated by the DNA polymerase to open the dumb bell-end on the other side and second, the annealing and elongation of new inner primer releases the stem-loop generated in the first step (Step IV). Thus, a new self- primed 3 -end is formed with which the cycle of self-primed elongation and release by an inner primer is continued. In the end, a mixture of stem-loop like DNA concatemers and cauliflower-like structures with various repeat counts are obtained (Step V).

**Author Contributions:** Conceptualization, M.W. and A.M.; methodology, M.W. and A.M.; validation, M.W. and A.M.; formal analysis, M.W.; investigation, M.W.; resources, A.M.; writing—original draft preparation, M.W. and A.M.; writing—review and editing, M.W. and A.M.; visualization, M.W. and A.M.; supervision, A.M.; project administration, A.M.; funding acquisition, A.M. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by DFG Deutsche Forschungsgemeinschaft, grant number MA 2288/16-2.

**Acknowledgments:** We thank Samra Ludmann for assisting in the preparation of the manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
