Structural and functional insights of GSU0105, a unique multiheme cytochrome from G. sulfurreducens

Abstract

Geobacter sulfurreducens possesses over 100 cytochromes that assure an effective electron transfer to the cell exterior. The most abundant group of cytochromes in this microorganism is the PpcA family, composed of five periplasmic triheme cytochromes with high structural homology and identical heme coordination (His-His). GSU0105 is a periplasmic triheme cytochrome synthetized by G. sulfurreducens in Fe(III)-reducing conditions but is not present in cultures grown on fumarate. This cytochrome has a low sequence identity with the PpcA family cytochromes and a different heme coordination, based on the analysis of its amino acid sequence. In this work, amino acid sequence analysis, site-directed mutagenesis, and complementary biophysical techniques, including ultraviolet-visible, circular dichroism, electron paramagnetic resonance, and nuclear magnetic resonance spectroscopies, were used to characterize GSU0105. The cytochrome has a low percentage of secondary structural elements, with features of α-helices and β-sheets. Nuclear magnetic resonance shows that the protein contains three low-spin hemes (Fe(II), S = 0) in the reduced state. Electron paramagnetic resonance shows that, in the oxidized state, one of the hemes becomes high-spin (Fe(III), S = 5/2), whereas the two others remain low-spin (Fe(III), S = 1/2). The data obtained also indicate that the heme groups have distinct axial coordination. The apparent midpoint reduction potential of GSU0105 (−154 mV) is pH independent in the physiological range. However, the pH modulates the reduction potential of the heme that undergoes the low- to high-spin interconversion. The reduction potential values of cytochrome GSU0105 are more distinct compared to those of the PpcA family members, providing the protein with a larger functional working redox potential range. Overall, the results obtained, together with an amino acid sequence analysis of different multiheme cytochrome families, indicate that GSU0105 is a member of a new group of triheme cytochromes.

Significance

Geobacter species play an important biogeochemical role as they can sustain their growth by using a variety of extracellular electron acceptors. This ability has been explored in the development of biotechnological applications in the fields of bioremediation, microbial energy production, and sustainable electronic devices and materials. Geobacter's impressive respiratory versatility is associated with the unusual high number of cytochromes coded by its genome. Structural and functional characterization of these cytochromes is crucial to understand the extracellular electron transfer mechanisms of these bacteria. In this work, different biophysical techniques were used to characterize GSU0105, a periplasmic triheme cytochrome synthetized by G. sulfurreducens in Fe(III)-reducing conditions. The results obtained indicate that GSU0105 is a member of a new group of triheme cytochromes.

Introduction

Geobacter species are Gram-negative bacteria that play an important biogeochemical role in a diversity of natural environments, demonstrating an impressive respiratory versatility, as they are capable of sustaining their growth by using insoluble extracellular electron acceptors (such as metal oxides) and soluble intracellular electron acceptors (such as fumarate) (1). Some of these extracellular compounds are toxic or radioactive, making these organisms a potential target for bioremediation applications (2). In addition, these bacteria produce higher current densities in microbial fuel cells (MFCs) than any other group of microorganisms, and because of that, they are usually the go-to microorganism for energy production in MFCs (3, 4, 5, 6).

The genome of Geobacter sulfurreducens has an unprecedented number of putative c-type cytochromes, with the original publication of its genome revealing 111 coding sequences containing at least one match to the c-type cytochrome motif that identifies heme groups (CXXCH, where X corresponds to any amino acid) (7). A deeper analysis of the bacterium’s genome shows that its number of cytochromes may be higher (at least 132 genes contain one CXXCH motif), even if contracted and extended motifs observed in other c-type cytochromes are not considered (8, 9). If these types of motifs are considered, the number can go up to 180. The incredible number of cytochromes contained in G. sulfurreducens genome highlights the importance of electron transport for this microorganism and suggests that this bacterium possesses electron transfer networks with high flexibility and apparent redundancy, which allows the reduction of diverse metal ions in natural environments. In fact, no single gene deletion on the G. sulfurreducens genome was found to eliminate electron transfer to all electron acceptors, thus confirming the complexity of its electron transfer networks (10).

The cytochromes from G. sulfurreducens are strategically localized at the bacterial inner membrane, periplasm, and outer membrane, allowing the transfer of electrons from intracellular carriers, such as nicotinamide adenine dinucleotide (NADH), to extracellular electron acceptors. The structural and functional characterization of these cytochromes is crucial to understand the extracellular electron transfer (EET) mechanisms of this bacterium and to engineer improved forms of these electron transfer components. Because the EET pathways of G. sulfurreducens are highly redundant, a single engineered component of these chains may not have a great impact on the overall driving force of the process. However, these engineered components may be relevant in simplified, “stripped-down” strains of G. sulfurreducens, designed with a minimal number of electron transfer components (11). The simplification of the EET pathways in these strains allows a fine-tune of the specific electron transfer routes in which the engineered component is known to be upregulated in the wild-type strain, leading to an increase in the bacterial respiratory rates (for a review, see (12)).

Several cytochromes of G. sulfurreducens have been studied, and insights into the bacterium’s EET mechanisms have been presented (13). Out of this group of cytochromes, the five triheme periplasmic cytochromes of the PpcA family (PpcA-E) are in the frontline as potential targets to develop rational Geobacter-mutated strains (14). These cytochromes have an important and strategic position in the cell that allows them to function as capacitors and to control the electron flow that connects the inner membrane and outer membrane components of the bacterium (15). The PpcA family cytochromes have a low-molecular mass (∼10 kDa), ∼70 residues each, and their three heme groups are axially coordinated by two histidine residues (16).

A proteomic analysis on G. sulfurreducens cultures grown in different conditions showed that three cytochromes, GSU0105, GSU0701, and GSU2515, were not synthetized in cultures grown on fumarate but instead were highly synthetized in Fe(III)-reducing conditions (17). GSU0105 is a periplasmic cytochrome with a similar molecular mass and the same number of heme groups as the cytochromes from the PpcA family. However, a preliminary sequence analysis revealed a low amino acid sequence identity. Given the putative role of GSU0105 in Fe(III) reduction, this cytochrome is an interesting target. In this study, a biochemical and biophysical characterization of GSU0105 from G. sulfurreducens is presented.

Here, the combined usage of several spectroscopic techniques, including ultraviolet (UV)-visible, circular dichroism (CD), electron paramagnetic resonance (EPR), and nuclear magnetic resonance (NMR) allowed the elucidation of the structural and functional features of GSU0105’s redox centers. The heme’s spin-states were determined, and the nature of their axial ligands is discussed based on amino acid sequence analysis and site-directed mutagenesis studies. The working functional redox range of the cytochrome at the bacterium’s physiological pH range was determined, showing that GSU0105 has the necessary features to carry electron transfer in a wider range compared to the other triheme periplasmic cytochromes. Altogether, it is proposed that GSU0105 may belong to a new group of triheme cytochromes.

Materials and methods

Amino acid sequence analysis and DNA manipulation

The gsu0105 gene sequence was retrieved from G. sulfurreducens PCA genome (GenBank: AE017180.2). The amino acid sequence was analyzed using the Uniprot bioinformatic tool (accessed January 2017) (18) to predict the signal peptide cleavage site. The software predicted it to be located between residues Leu21 and Ala22. The gene encoding for GSU0105 was amplified from G. sulfurreducens genomic DNA using Phusion High-Fidelity DNA polymerase (Thermo Fisher Scientific, Waltham, MA) and the primers listed in Table S1, which contain restriction sites for the enzymes NotI and HindIII. The resulting DNA fragment and the pVA203 vector (19) were digested with the referred restriction enzymes, purified using the E-gel electrophoresis system (Invitrogen, Waltham, MA) and ligated using T4 DNA ligase (Fermentas, Waltham, MA). The DNA was transformed into Escherichia coli DH5α cells, and a colony polymerase chain reaction screen was performed using Taq DNA polymerase (VWR, Radnor, PA). Positive clones were grown in liquid media, and the plasmid was purified using the NZYMiniprep kit (NZYTech, Lisbon, Portugal). The final plasmid was sequenced by STAB VIDA (Caparica, Portugal) and designated pGSU0105.

The primers used for the substitution of the putative axial ligand residues Met37, Met40, and Met60 for histidine residues were designed using the QuikChange Primer Design program (Agilent Technologies, Santa Clara, CA) and are listed in Table S1. The mutations were introduced in the pGSU0105 plasmid following the instructions of the NZYMutagenesis kit (NZYTech). The presence of the desired mutations was confirmed by DNA sequencing.

Protein sequence analysis

A Protein Basic Local Alignment Search Tool (BLAST) at the National Center for Biotechnology Information (NCBI) (20) was performed with the GSU0105 sequence (WP_010940781.1, including its signal peptide), using the non-redundant protein sequences database and the default blastp algorithm. From the 38 sequences obtained (December 2020), the ones from unclassified organisms were excluded, and the resulting 12 sequences were aligned using the Clustal Omega tool (21).

For sequence analysis, the triheme G. sulfurreducens c₇ cytochrome PpcA (WP_010941274.1), Desulfuromonas acetoxidans c₇ cytochrome (WP_176290204.1), and Shewanella oneidensis cytochrome (WP_011074175.1) were selected to perform individual BLAST searches. The top hits were selected and, together with GSU0105 sequences, were used for multiple sequence alignment.

Synthesis and purification of GSU0105 from G. sulfurreducens

The GSU0105 cytochrome was synthetized using E. coli JM109 (DE3) cells as host. The host cells were transformed using the heat shock method and contained two plasmids: 1) pEC86, encoding for the cytochrome c maturation gene cluster ccmABCDEFGH, required for the hemes incorporation (22), and for a chloramphenicol resistance marker, and 2) pGSU0105, encoding the cytochrome GSU0105 after the OmpA leader sequence, a lac promoter, and an ampicillin resistance marker (19). These cells were grown at 30°C in 2× yeast extract media, supplemented with 34 μg/mL chloramphenicol and 100 μg/mL ampicillin, to an optical density at 600 nm of ∼1.5. At this stage, protein synthesis was induced with 10 μM of isopropyl β-D-thiogalactoside, and the cell culture was grown overnight at 30°C. Cells were harvested by centrifugation at 6400 × g for 20 min at 4°C. The cell pellet was gently resuspended in lysis buffer (100 mM tris(hydroxymethyl)aminomethane-HCl (pH 8), 20% sucrose, and 0.5 mM EDTA, containing 0.5 mg/mL of lysozyme), and after 15 min of incubation at room temperature, precooled water was added to the cell suspension, which was then left incubating on ice for 15 min to stop the lysozyme’s activity. After that step, the suspension was centrifugated at 14,700 × g for 20 min at 4°C. The supernatant constituting the periplasmic fraction was further ultracentrifugated at 225,000 × g for 1 h at 4°C to remove any remaining membrane debris. The final supernatant was dialyzed twice against 4.5 L of 20 mM tris(hydroxymethyl)aminomethane-HCl (pH 7.5) using a Spectra/Por dialysis membrane (molecular weight cutoff: 3.5 kDa) and then loaded onto 2 × 5 mL Bio-Scale Mini UNOsphere S cartridges (Bio-Rad Laboratories, Hercules, CA), equilibrated with the same buffer. The protein was eluted with a sodium chloride gradient (0–300 mM). The obtained fractions were evaluated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis, 15% acrylamide/bis-acrylamide, stained either for hemes (tetramethylbenzidine staining) and/or with BlueSafe (NZYTech). The fractions containing the protein were concentrated in Amicon Ultracentrifugal filter units (Ultra-4, molecular weight cutoff: 3 kDa) and equilibrated with 100 mM sodium phosphate buffer (pH 8) before being injected in either a Superdex 75 XK 16/70 or a Superdex 75 Increase 10/300 GL molecular exclusion columns, equilibrated with the same buffer. Both chromatography steps were performed on an ÄKTA Pure system (GE Healthcare, Chicago, IL), and the final protein purity was evaluated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis, 15% acrylamide/bis-acrylamide. The concentration of the cytochrome was determined by measuring the absorbance of the reduced form at 552 nm, using the molar extinction coefficient of 97.5 mM⁻¹ cm⁻¹, determined for PpcA from G. sulfurreducens (23). GSU0105’s mutants (M37H, M40H, and M60H) were synthetized and purified using the same methodology.

UV-visible spectroscopy

The UV-visible absorption spectra of GSU0105 were acquired in a Thermo Fisher Scientific Evolution 201 spectrophotometer. The measurements were made with a quartz cuvette (Hellma Analytics, Plainview, NY), with 1 cm of path length, at room temperature, using protein samples with a concentration of ∼1.25 μM, prepared in 32 mM sodium phosphate buffer (pH 7) with NaCl (100 mM final ionic strength). The spectra were obtained between 300 and 750 nm. The sample reduction was achieved with the addition of sodium dithionite.

Redox titrations of GSU0105 monitored by visible spectroscopy

The heme’s reduction potential values were determined by titrations monitored by visible spectroscopy, using a methodology (24,25) that has been applied in several previous studies (15). The redox titrations were carried out at 288 K inside an anaerobic glove box (MBRAUN, Stratham, NH) with O₂ levels kept under 1 ppm with argon circulation. The visible spectra were recorded in an Evolution 300 spectrophotometer (Thermo Fisher Scientific) placed inside the glovebox. GSU0105 solutions with 10 μM protein concentration were prepared in 80 mM sodium phosphate buffer with NaCl (250 mM final ionic strength) at pH 7 and 8.

The solution potentials were measured using a combined Pt/Ag/AgCl electrode (Crison Instruments, Barcelona, Spain), calibrated with quinhydrone-saturated solutions at pH 4 and 7. The following mixture of redox mediators was added to the solution with a final concentration of 1.3 μM each, as described in the literature (24), to ensure good equilibrium between the redox centers of the protein and the working electrode: potassium ferricyanide (E^0’ = +430 mV), p-benzoquinone (E^0’ = +280 mV), tetramethyl-1,4-phenylenediamine (E^0’ = +260 mV), 1,2-naphtoquinone-4-sulphonic acid (E^0’ = +215 mV), 1,2-napthoquinone (E^0’ = +143 mV), trimethylhydroquinone (E^0’ = +115 mV), phenazine methosulfate (E^0’ = +80 mV), phenazine ethosulfate (E^0’ = +55 mV), gallocyanine (E^0’ = +21 mV), methylene blue (E^0’ = +11 mV), indigo tetrasulfonate (E^0’ = −30 mV), indigo trisulfonate (E^0’ = −70 mV), indigo disulfonate (E^0’ = −120 mV), 2-hidroxy-1,4-naphthoquinone (E^0’ = −145 mV), anthraquinone-2,6-disulfonate (E^0’ = −185 mV), anthraquinone-2-sulfonate (E^0’ = −225 mV), safranine O (E^0’ = −280 mV), neutral red (E^0’ = −325 mV), benzyl viologen (E^0’ = −345 mV), diquat (E^0’ = −350 mV), and methyl viologen (E^0’ = −440 mV). To verify hysteresis and reversibility, each redox titration was performed in both oxidative and reductive directions, using sodium dithionite as the reducing agent and potassium ferricyanide as the oxidizing agent. The experiments were performed at least two times, and the reduction potentials (relative to the normal hydrogen electrode) were found to be reproducible within ±5 mV. The reduced fraction of GSU0105 was determined by integrating the area of the α-peak (552 nm) above the line connecting the flanking isosbestic points (542 and 560 nm) to subtract the optical contribution of the redox mediators (25). The fitting of the experimental data was made using a model that takes into account the possible macroscopic oxidation stages for a triheme cytochrome (see Supporting materials and methods, section S1) and considering that the hemes of GSU0105 have different molar extinction coefficients at 552 nm (α-band). The fitting was obtained using the Solver tool from the Microsoft Excel software.

CD spectroscopy

CD spectra of GSU0105 (25 μM protein prepared in 20 mM NaCl at pH 7.4) were recorded at 25°C in the far-UV region using an Applied Photophysics Chirascan qCD spectropolarimeter (Leatherhead, UK). The assays were performed using 0.02 cm-path length quartz cells. Spectra were recorded over a 190–260 nm wavelength range in 1 nm steps at a scan speed of 20 nm per minute. Three scans were acquired and averaged for each spectrum. The buffer contribution was corrected for all spectra. The temperature was controlled to ±1°C. The conformational stability of GSU0105 was assessed by performing temperature denaturation, monitored in the 190–260 nm wavelength range. For thermal-induced denaturation, a heating rate of 1°C per minute was used, and temperature was increased from 10 to 95°C. After the temperature ramp, the sample was quickly cooled to 25°C, at which a final spectrum was registered.

The Beta Structure Selection (BeStSel) deconvolution method (26) was used to estimate the secondary structure elements of GSU0105.

EPR spectroscopy

X-band EPR spectra were obtained using a Bruker EMX spectrometer (Billerica, MA) equipped with an Oxford Instruments ESR-900 continuous flow helium cryostat (Abingdon, UK) and a high sensitivity perpendicular mode rectangular cavity. An as-purified GSU0105 sample was prepared in 32 mM sodium phosphate buffer (pH 7) with NaCl (100 mM of final ionic strength), to a final protein concentration of 200 μM. The fully reduced sample was prepared anaerobically inside a glovebox by incubation with a slight excess of sodium dithionite. All spectra were recorded using a microwave frequency of 9.39 GHz, a modulation amplitude of 1.0 mT, and a microwave power of 2 mW. Spectra were analyzed and simulated using SpinCount (27).

NMR spectroscopy

Samples preparation

Protein samples were prepared with ∼100 μM concentration in 32 mM sodium phosphate buffer (pH 7) with NaCl (100 mM of final ionic strength). The buffer was prepared either in pure D₂O or in 90% H₂O/10% D₂O. The pH values of the samples were measured with a glass microelectrode and were not corrected for isotope effects. For sample reduction, the NMR tubes were sealed with a gas-tight serum cap, and the air was flushed out from the sample to avoid possible oxidation of the samples. Then, the samples were reduced directly in the NMR tube with gaseous hydrogen (Air Liquide, Paris, France) in the presence of catalytic amounts of hydrogenase from Desulfovibrio vulgaris Hildenborough. To ensure that the protein samples were kept in the completely reduced state, the NMR tubes were left in a hydrogen atmosphere.

NMR experiments

The NMR experiments were acquired in a Bruker Avance III 600 MHz spectrometer equipped with a triple-resonance cryoprobe. The ¹H chemical shifts are referenced to sodium trimethylsilylpropanesulfonate at 0 ppm, as previously described (28). The different spectra obtained were processed using TopSpin3.5.7 (Bruker BioSpin, Karlsruhe, Germany). One-dimensional (1D) ¹H-NMR spectra were acquired for the protein’s oxidized state at 298 K, with 32,000 data points, a spectral width of 96 kHz, with a total of 2048 transients, and water presaturation. For the cytochromes in the reduced state, 1D ¹H-NMR spectra were acquired at 298 K, with 16,000 data points, a spectral width of 19 kHz, with a total of 1024 transients, and water presaturation.

Results and discussion

GSU0105 amino acid sequence analysis

The amino acid sequence of the c-type cytochrome GSU0105 contains three typical CXXCH heme binding motifs (Fig. 1). The analysis of the sequence shows that the heme axial coordination of the cytochrome is necessarily different from those belonging to the c₇ family, which all have bis-histidine (bis-His) axial coordinated c-type hemes (Fig. S2) (16). In fact, the distal ligands in these proteins are fully conserved between all the sequences. For GSU0105, the sequence shows that besides the three histidine residues included in the CXXCH binding motifs, there is only one histidine residue left in the mature sequence (His41, Fig. 1), which indicates that only one of the heme groups can have bis-His coordination. The distal coordination position of c-type hemes can be occupied by methionine, histidine, asparagine, tyrosine, cysteine, and lysine residues, the protein’s N-terminal amino group, a water molecule, and/or be transiently vacant (29,30). A BLAST search retrieves 12 sequences (out of a total of 38, from which the ones from unclassified organisms were excluded), and their alignment shows that, besides His41, there are only six conserved amino acids that can act as axial ligands to the heme groups: three methionine (Met37, Met40, and Met60) and three lysine (Lys33, Lys62, and Lys66) residues (Fig. 1). It is important to notice that although residues Lys33 and Met40 are in close proximity to the other axial ligands in the amino acid sequence (histidine residues in the binding motifs and His41), they are still possible axial ligands, considering the relative position of the axial ligands in tetraheme cytochromes from Desulfovibrio species (31).

Figure 1.

GSU0105 amino acid sequence analysis. Alignment of GSU0105 sequence with homolog sequences from different bacteria obtained from a BLAST analysis (NCBI access number): GSU0105 G. sulfurreducens (WP_010940781.1), G. soli (WP_039647343.1), Geomonas oryzae (WP_129125168.1), Geom. ferrireducens (WP_136524092.1), G. pickeringii (WP_039744778.1), G. lovleyi (WP_012469124.1), G. bremensis 1 (WP_026841093.1), G. uraniireducens (WP_011941028.1), G. bemidjiensis (WP_012532384.1), G. sp. (strain M21) (WP_015839232.1), G. pelophilus (WP_085814588.1), and G. bremensis 2 (WP_185243631.1). Sequences were aligned with Clustal Omega and their % pairwise identity for mature sequence is indicated. Residues are colored by sequence similarity (dark green 100%, green 80–100%, light green 60–80%, and white <60% similarity). Heme binding motifs CXXCH are identified, and the fully conserved putative axial ligands (Lys33, Met37, Met40, His41, Met60, Lys62, and Lys66) are marked with ^∗. The scissors identifies the signal peptide cleavage site. To see this figure in color, go online.

Multiheme cytochromes containing heme groups with mixed axial coordination have been described previously. For example, the G. sulfurreducens dodecaheme cytochrome GSU1996 has four triheme domains, in which, in each domain, two hemes are bis-His, whereas the other heme is His-Met coordinated (32). Also, the triheme cytochrome DsrJ, which belongs to a transmembrane redox complex found in D. desulfuricans ATTC 27774 (33) and in the purple sulfur bacterium Allochromatium vinosum (34) possesses three hemes that all have different types of coordination: a bis-His, a His-Met, and an unusual His-Cys coordination. More recently, a tetraheme cytochrome from the annamox bacterium Kuenenia stuttgartiensis with an unusual contracted heme binding motif was characterized, and complementary spectroscopic techniques showed that the protein has four low-spin hemes: two with bis-His coordination, one with His-Lys, and one with His-Cys (8). However, such heme mixed coordination is unprecedent in small soluble periplasmic cytochromes in which the ratio of amino acid residues to heme groups is low, suggesting that GSU0105 may belong to a different group of cytochromes within the class III of c-type cytochromes (35). This class is formed by multiheme cytochromes that present low reduction potentials (see Redox properties of GSU0105) and only around 30 residues per heme group (36). Nevertheless, no biochemical or biophysical studies have been reported for any of GSU0105’s homologs, and further studies are needed to identify additional distinctive features of this group of cytochromes.

Spectroscopic characterization of GSU0105 from G. sulfurreducens

The purified GSU0105 was studied using complementary spectroscopic techniques, including CD, UV-visible, EPR, and NMR to probe the protein’s folding and secondary structure elements, the spin-state of the hemes, and the nature of their axial ligands.

GSU0105 is composed mainly of random coil secondary structure elements

The far-UV CD spectrum of the cytochrome at 25°C in the oxidized state is presented in Fig. 2 A. The overall spectrum indicates the net presence of random coil structures, but it also contains features that point to the existence of a mixture of α-helix and β-sheet secondary structural elements, namely 1) an intense positive band at 190 nm and 2) an intense negative band at 206 nm and one less negative band at 223 nm. The positions of the mentioned bands at 190, 206, and 223 nm, with slightly different wavelengths compared to the values commonly seen for α-helices (190, 208, and 222 nm), probably result from substantial contributions related with the protein’s heme groups, as previously observed for other triheme cytochromes (37,38), but also because of the presence of significant percentages of other types of secondary structural elements.

Figure 2.

Far-UV CD spectral features of oxidized GSU0105. (A) Fittings (performed with the model inserted on BeStSel - see Materials and methods) of the far-UV spectra of GSU0105 at 25°C before (blue line) and after (black line) the temperature ramp and at 95°C (red line). The CD units are automatically converted to Δε (M⁻¹ cm⁻¹) by the BeStSel software. (B) Monitorization of the thermal stability of GSU0105. The spectra were acquired between 10 and 95°C (color gradient from blue to red). To see this figure in color, go online.

The percentage of secondary structural elements was estimated with the BeStSel deconvolution method (26), which has some limitations, namely the unaccountability of rare secondary structural elements and aromatic contributions. Furthermore, in highly disordered proteins, the software may incorrectly attribute disordered regions to specific structured features. GSU0105 only contains two aromatic residues (Phe25 and Phe50), and the results obtained show that it is not highly disordered, thus validating the usage of this method.

The results obtained with the BeStSel deconvolution method demonstrated that GSU0105 accounts for 41.3% of folded conformations (29.9% α-helices and 11.4% β-sheets) and 58.7% of disordered conformations. By comparing these data with other triheme cytochromes from G. sulfurreducens (Table 1), one can verify that these proteins usually possess very low percentages (<50%) of folded conformations. The PpcA family cytochromes have a conserved general fold, mainly composed of random coil motifs, a two-strand β-sheet in the N-terminal, and several α-helical segments (Fig. S3). In these proteins, the low amino acid number per heme ratio (between 23 and 25) hinders the possibility of the existence of complex secondary structure elements, most likely to promote solvent exposure and large surface areas that allow the formation of low affinity complexes with other redox partners, necessary for efficient electron transfer (39, 40, 41). GSU0105 presents an even lower percentage of folded conformations, which is another indication that the protein may belong to a different group of triheme cytochromes, as discussed in the previous section.

Table 1.

Secondary structural elements of GSU0105 and other triheme cytochromes from G. sulfurreducens

Protein	Secondary Structural Elements (%)
	α-Helix	β-Sheet	Turn	Others
GSU0105 (25°C)a	29.9	11.4	15.0	43.6
GSU0105 (95°C)a	12.4	16.5	18.4	52.6
GSU0105 (25°C)a,b	18.0	15.1	16.8	50.1
PpcA Gsc	42.3	9.9	15.5	32.3
PpcB Gsc	38.0	15.5	14.1	32.4
PpcC Gsc	30.7	13.3	18.7	37.3
PpcD Gsc	38.0	8.5	11.3	42.2
PpcE Gsc	36.6	9.9	9.9	43.6

The percentages of secondary structural elements were either estimated from CD data using the BeStSel deconvolution method (for GSU0105) or obtained from the Dictionary of Secondary Structures of Proteins (DSSP) data deposited in the respective Protein Data Bank (PDB) identification codes (for the remaining triheme cytochromes from G. sulfurreducens (Gs)).

^aAccording to the results obtained from the BeStSel deconvolution method.

^bAfter temperature ramp and quick cool down.

^cAccording to the DSSP data deposited in PDB: 1OS6 (PpcA Gs) (42), PDB: 3BXU (PpcB Gs) (43), PDB: 3H33 (PpcC Gs) (16), PDB: 3H4N (PpcD Gs) (16), and PDB: 3H34 (PpcE Gs) (16).

To monitor the thermal stability of the protein, a characterization of the temperature-induced unfolding was carried out. The spectra obtained show that the secondary structural elements of GSU0105 undergo significant changes during the temperature variation (Fig. 2 B). In fact, by plotting the variation of the ellipticity at 206 and 223 nm as a function of temperature, a complex behavior is observed, which cannot be explained by considering a two-state transition of a monomer from a folded to unfolded state. This means that GSU0105 goes through several rearrangements of its secondary structural elements during the denaturation process, which can include multiple steps and intermediary states linking the native state to the final thermally unfolded state. Similar results have been observed in a monoheme ferricytochrome c, in which the protein undergoes a series of reversible conformational switches, resulting in alterations of the distal ligand position of the heme (44). The main variations of the spectra throughout the temperature variation are observed in the signals corresponding to the α-helices, which show a blueshift in their wavelengths. This is a result of an increase in the percentage of random coil motifs in the protein. Moreover, other slight changes may be due to small alterations of the conformation of the heme groups. At 95°C, GSU0105 still possesses marked ellipticities and is not fully unfolded (Fig. 2; Table 1).

Upon cooling to 25°C, the final spectrum reveals that GSU0105 partially refolds (Fig. 2; Table 1). Altogether, the data indicates that GSU0105 is stable at high temperatures as seen for other triheme cytochromes from G. sulfurreducens (37,38). In these cytochromes, the three covalently bound c-type hemes keep the overall proteins’ structural integrity, whereas the few secondary structure elements undergo small variations.

UV-visible spectra of GSU0105 present features of low-spin hemes

The optical absorption spectrum of the oxidized cytochrome (Fig. 3) has maxima at 354, 408, and 527 nm. Upon reduction, the protein shows the Soret, β- and α-bands maxima at 417, 523, and 552 nm, respectively. These spectral patterns are similar to those shown by low-spin hexacoordinated hemes (45).

Figure 3.

UV-visible spectral features of cytochrome GSU0105 in the oxidized and reduced states. The maxima of the UV-visible spectra of the cytochrome in the oxidized (solid black line) and reduced states (dashed black line) are labeled.

Methionine-coordinated hemes typically display a band at 695 nm in the UV-visible spectrum of the oxidized form, which commonly features a half width of 30–40 nm and an extinction coefficient around 20 times smaller than the α-band of low-spin hemes in the reduced state (46, 47, 48). This band is not present in the oxidized spectrum of GSU0105 (Fig. 3), even when difference spectra are measured. This is a common practice because the extinction coefficient associated with this band may be even lower than usual when the hemes have a high solvent exposure (49). In these cases, the measurement of difference spectra in cytochromes allows the deconvolution of low-intensity Q and charge-transfer bands as most of these have a significant variation of their signal intensity with the oxidation state of the heme iron (50).

The absence of a band at 695 nm in the UV-visible spectrum does not exclude the possibility that one of the hemes of GSU0105 is coordinated by a methionine residue, as proven in different works (51,52). To gather more information on this and on the hemes spin-states, EPR and NMR experiments were performed for GSU0105.

EPR and NMR features of GSU0105

X-band EPR spectroscopy of the as-purified oxidized GSU0105 exhibits a complex spectrum, dominated by two main sets of features that arise from the contribution of high- and low-spin ferric heme species (Fig. 4 A). A very small contribution of adventitious Fe(III) is also observed at g = 4.23. The simulation of the EPR spectrum indicates an approximate 1:2 ratio of the high- and low-spin species, respectively (Fig. 4 A). To simulate the high-spin species, a low value of the rhombicity was used (E/D = 0.005), and because the zero-field splitting D is unknown, a value of 20 cm⁻¹ was assumed so that at the temperature used, only the ground Kramers doublet, |±1/2⟩, is populated. In the fully reduced state, the cytochrome is EPR silent (Fig. 4 A).

Figure 4.

EPR and NMR features of GSU0105. (A) Experimental continuous-wave X-band EPR spectra of GSU0105 at the oxidized (orange) and reduced (light blue) states, recorded at 10 K, with a microwave frequency of 9.39 GHz, a modulation amplitude of 1.0 mT, and a microwave power of 2 mW. The black line below the oxidized experimental spectrum is the simulated spectrum, accounting for a 1:1:1 contribution of the high-spin species (a) and the two low-spin species (b and c), for which each spectrum is represented in gray. The high-spin species spectrum (a) was simulated with E/D = 0.005 and a D = 20 cm⁻¹. The two low-spin species spectra were simulated with g-values of 3.11, 2.27 and ≤ 1.40 (b) and 2.98, 2.22, and 1.49 (c). The signal with a g-value of 4.23 corresponds to a small amount of high-spin ferric iron adventitiously present in the sample. (B) 1D ¹H-NMR spectrum of oxidized GSU0105. The low-field region is zoomed in to show the resonances of the heme methyl substituents. (C) 1D ¹H-NMR spectrum of reduced GSU0105. A close-in of the spectrum (×10) is represented. The typical high-field region of axial methionine resonances is shown for the wild-type (light blue) and mutant (dark blue) proteins. To see this figure in color, go online.

The intense and quasi-axial signal with g_max of 6.10 can be attributed to a high-spin S = 5/2 ferric heme. The g_z contribution of this signal, expected at g = 2.00, is underneath the resonance attributed to adventitious copper (II). The existence of this high-spin species is consistent with what is observed in the NMR spectrum (see below) of the oxidized cytochrome.

The species at a higher magnetic field are characteristic of low-spin ferric hemes. The resonance at g ~ 2.98 reveals a shoulder, indicating that two very similar but not identical species contribute to this set of resonances. It was not possible to deconvolute the spectra by varying either the microwave power or the temperature, suggesting that both hemes have quasi-identical relaxation properties. An analysis of the data revealed two species, with g-values at 1) 3.11, 2.27, and ≤1.40 and 2) 2.98, 2.22, and 1.49.

Similarly to EPR, NMR is a very powerful technique to identify the spin-state of heme groups in cytochromes because their signals appear in quite distinct spectral regions, depending if the hemes are high- or low-spin. In the paramagnetic oxidized state, 1D ¹H-NMR spectra of high-spin cytochromes display extremely broad signals and resonances above 40 ppm (usually from heme methyl substituents). Low-spin cytochromes, on the other end, present narrower spectral windows, with the main heme substituents frequencies ranging from 8 to 35 ppm. In the reduced state, 1D ¹H-NMR spectra are also quite distinct for high- and low-spin cytochromes. In fact, high-spin hemes present wider spectral regions (from −15 up to 30 ppm) than low-spin ones (from −5 up to 10 ppm).

The 1D ¹H-NMR signals of GSU0105 in the oxidized state are very broad and cover a wide spectral region, namely from −10 ppm to above 45 ppm (Fig. 4 B). Therefore, the cytochrome is paramagnetic in the oxidized state, with at least one high-spin heme (Fe(III), S = 5/2), as observed in the EPR experiments. The NMR signals of the cytochrome in the oxidized state are very broad because of the strong paramagnetic contribution of the high- and low-spin heme(s), which makes it very difficult to analyze any other spectral features of the protein in this redox state.

In the reduced state, the 1D ¹H-NMR spectrum of GSU0105 displays a narrow spectral window between −5 and 11 ppm (Fig. 4 C), indicating that all the hemes from GSU0105 are low-spin (Fe(II), S = 0). Furthermore, the low-frequency region of the reduced state 1D ¹H-NMR spectrum can be used as a fingerprint to detect axial ligands because of the heme ring-current effects (45). Histidine ligands do not show resonances in this region (their signals are usually located in the protein’s signal envelope), but methionine ligands display characteristic patterns (53). The GSU0105 spectrum in the reduced state displays a typical signal pattern for the side-chain signals of a distal coordinated methionine, which includes a three-proton intensity peak at approximately −3 ppm and up to four resolved one-proton intensity peaks in this region of the spectrum (Fig. 4 C). The possibility of the existence of two hemes with His-Met axial coordination cannot be discarded based on the analysis of the diamagnetic 1D ¹H-NMR spectrum of GSU0105 because their signals may be overlapped.

To identify the axial ligands of GSU0105, three mutants of the fully conserved methionine residues were produced (M37H, M40H, and M60H) and analyzed by 1D ¹H-NMR spectroscopy in both oxidation states. As observed for the wild-type GSU0105, the oxidized spectra of all mutants also display a large spectral width and broad low-field resonances typical for high-spin heme substituents (data not shown). In the reduced state, the high-field axial methionine resonance pattern observed for the wild-type cytochrome is observed for the M37H and M60H mutants but not for the M40H mutant (Fig. 4 C). These results unequivocally identify Met40 as one of the distal ligands of GSU0105. Moreover, the absence of any axial methionine resonance patterns in the 1D ¹H-NMR spectrum of the M40H mutant indicates that the axial ligand of the third heme is not a methionine residue.

The candidates for the unidentified distal ligand are the three fully conserved lysine residues (Lys33, Lys62, and Lys66). However, and in contrast with axial methionine residues, the NMR signal pattern for an axial lysine residue in the 1D ¹H-NMR spectrum is not straightforwardly identifiable (54,55). This feature hinders the assignment of the third heme distal ligand of GSU0105 using the same strategy, based on site-directed mutagenesis and NMR spectroscopy. Therefore, a different approach would have to be used to identify this ligand. The best methodology would be structure determination by x-ray crystallography or NMR. However, the low protein yield obtained for this cytochrome (0.2 mg per liter of cell culture) currently hampers this approach. Crystallization trials demand high concentrations of protein, and NMR requires isotopically labeled proteins synthetized in minimal media, which would result in even lower yields.

Furthermore, there are other features of GSU0105 that hinder its structural characterization in more detail by NMR. Contrary to the triheme cytochromes from the PpcA family of G. sulfurreducens (56), the spectrum of GSU0105 in the reduced state has signals with unexpectedly large linewidths for a ∼10 kDa protein (Fig. 4 C). These spectral features may arise from inherent properties of GSU0105. Taking a closer look into the backbone amide region (7–11 ppm) of the spectrum (Fig. 5 C), the signals are well dispersed, which is typical of a folded protein, meaning that the protein is not aggregated in the experimental conditions used. A possibility is the existence of exchange processes. If most of the protein signals are in a slow to intermediate exchange rate between two conformations, their signal widths would be significantly affected. This phenomenon would not affect the results of the remaining experiments, and its effects may only be detectable by NMR. There are several experimental conditions that can be varied to optimize the exchange rates of a certain chemical event, namely the magnetic field strength, temperature, and ionic strength (57). 1D ¹H-NMR spectra of GSU0105 in the reduced state were acquired at different temperatures and ionic strength conditions, but all spectra kept the signal broadening initially observed (data not shown).

Figure 5.

Redox titrations of GSU0105 and PpcA family triheme cytochromes monitored by visible spectroscopy. The solid (pH 7) and dashed (pH 8) lines indicate the results of the fits to the Nernst curves for the three macroscopic reduction potentials of GSU0105 (in black), PpcA (in red), PpcB (in blue), PpcD (in orange), and PpcE (in green), which are indicated in Table 2. The open and filled symbols in the GSU0105 curves represent each experimental point of the reductive and oxidative titrations at pH 7 (●) and 8 (▲), respectively. The bottom left inset shows a close-in on the apparent midpoint reduction potentials of the cytochromes. For simplification, in this inset, the experimental points are not represented. The top right inset shows the variation of the α-band region of the visible spectra of GSU0105 throughout the visible redox titration at pH 7. To see this figure in color, go online.

Redox properties of GSU0105

Redox titrations of GSU0105 followed by visible spectroscopy were performed at pH 7 and 8 (Fig. 5) to cover the physiological pH range for G. sulfurreducens growth (58). The reductive and oxidative curves are superimposable (no hysteresis), indicating that the redox process is fully reversible. The redox window of GSU0105 spans from −315 to 85 mV, which is compatible with the presence of bis-His, His-Lys, and His-Met coordinated hemes. Typically, c-type heme groups axially coordinated by a methionine present more positive reduction potentials compared to bis-His coordinated hemes (47). This relates to the fact that the methionine’s side-chain sulfur is a good electron acceptor, which favors the electron-rich reduced state of the heme, resulting in more positive reduction potentials compared to bis-His coordinated heme groups. However, experimental data show that in many cases, the network of residues in the vicinity of the heme groups has a higher effect on their reduction potential values than the nature of the axial ligand. For example, there are reports of His-Met coordinated hemes with reduction potential values ranging from −251 to +358 mV (59, 60, 61). Similarly, although there are few reports for His-Lys coordinated hemes, their reduction potential values can be as different as −80 or 365 mV (54,55).

Macroscopic redox behavior of GSU0105

For other triheme cytochromes, the fitting of experimental data from redox titrations followed by visible spectroscopy can be obtained with a sequential model (see Supporting materials and methods, section S3), considering that each heme has the same contribution (0.3(3)) for the α-band absorbance. In the case of GSU0105, the contribution for the α-band absorbance of the last heme to oxidize (the one with the most positive reduction potential) is smaller compared to the other hemes. For this reason, the experimental data were fitted by considering different contributions for the three hemes (Supporting materials and methods, section S3).

An equal contribution of 0.39 ± 0.02 for the α-band absorbance was determined for the first two hemes to oxidize, whereas a contribution of 0.22 ± 0.01 was determined for the last heme to oxidize. The differences in these values are most likely related with the different types of axial coordination of the hemes or with the fact that one of them is high-spin in the oxidized state.

Looking at the macroscopic reduction potentials for the three ($E_1^0$–$E_3^0$) oxidation steps (Fig. S4; Table 2), GSU0105 has negative but distinct reduction potential values for two of the oxidation steps and a positive value for the other one. Although these values are macroscopic in nature and thus do not correspond to the individual heme reduction potentials, those of the first two oxidation steps are significantly separated from the reduction potential of the last oxidation step, which can be considered to have a nearly independent oxidation (Fig. 5; Fig. S4). This means that the reduction potential of the last oxidation step can be attributed to the last heme to be oxidized.

Table 2.

Comparison of the apparent midpoint and macroscopic reduction potentials (relative to the normal hydrogen electrode) of GSU0105 and other triheme cytochromes from G. sulfurreducens at pH 7 and 8

	Eapp (mV)		$E_1^0$ (mV)		$E_2^0$ (mV)		$E_3^0$ (mV)
	pH 7	pH 8	pH 7	pH 8	pH 7	pH 8	pH 7	pH 8
GSU0105	−154	−154	−223	−219	−137	−138	8	35
PpcA Gs (15)	−117	−138	−171	−182	−119	−139	−60	−93
PpcB Gs (15)	−137	−143	−185	−192	−140	−145	−84	−103
PpcD Gs (15)	−132	−148	−181	−191	−133	−152	−78	−94
PpcE Gs (15)	−134	−139	−191	−194	−133	−138	−82	−85

The apparent midpoint reduction potentials (E_app) correspond to the point at which the oxidized and reduced fractions are equal. $E_1^0$, $E_2^0$, and $E_3^0$ are the macroscopic reduction potentials for the first, second, and third oxidation steps, respectively (for a review, see (62)). The values for the PpcC cytochrome from the PpcA family of G. sulfurreducens (Gs) are not presented because the cytochrome possesses two conformations in solution that hinder its thermodynamic characterization (63). The values presented have an experimental estimated error of ±5 mV.

The last heme to be oxidized has a smaller contribution to the α-band because of the low- to high-spin interconversion and concomitant change in its molar extinction coefficient. Therefore, the last oxidation step $\left(E_3^0\right)$ is dominated by the oxidation of a heme that turns high-spin in the oxidized state. Considering the observations made in Spectroscopic characterization of GSU0105 from G. sulfurreducens, this heme has a putative His-Lys axial coordination. Thus, the first and second oxidation steps are dominated by the oxidation of the His-Met40 and His-His41 coordinated hemes. Nevertheless and despite the different coordination, it is not possible to infer which of the hemes has the most negative reduction potential. Interestingly, in a triheme domain from the dodecaheme protein GSU1996, which possesses two bis-His and one His-Met coordinated hemes, the reduction potential of the His-Met coordinated heme is not the most positive (64).

The individual heme reduction potentials have been determined for several low-spin triheme and tetraheme cytochromes by combining two-dimensional ¹H-EXchange SpectroscopY (2D ¹H-EXSY) NMR data and potentiometric redox titrations followed by visible spectroscopy (15,65). However, in the case of the triheme cytochrome GSU0105, such studies cannot be implemented because of the severe broadening of the NMR signals caused by the high-spin heme, which impairs the assignment of the individual heme oxidation patterns along the different oxidation stages (for a review, see (12)).

GSU0105 possesses a local redox-Bohr effect

The redox-Bohr effect reflects on the pH dependence of the protein’s reduction potential value and is caused by the protonation/deprotonation of an acid-base group (redox-Bohr center) in the vicinity of the heme(s). Considering pure electrostatics, it is expected that the deprotonation of the redox-Bohr center (removal of a positive charge) would stabilize the oxidized form of the heme(s), leading to the concomitant decrease of the reduction potential value. In the case of GSU0105, the apparent reduction potential value (E_app) obtained for GSU0105 was −154 mV at pH 7 and 8 (Table 2). In fact, the redox titration curves are superimposable in this region, indicating that there is not a net redox-Bohr effect in the physiological pH range. Nonetheless and regardless of the similarity of the E_app values at pH 7 and 8, from the analysis of the redox titration curves there is a notorious separation at lower reduction potential values in the region dominated by the oxidation of the heme with the highest reduction potential (Fig. 5). This result indicates that there is a local redox-Bohr effect near the last heme to oxidize. However, contrary to the expected electrostatic behavior of the redox-Bohr effect, the reduction potential of the last heme to oxidize is higher at pH 8 (Table 2). This effect has been observed in other multiheme cytochromes (25,66,67) and attributed to mechano-chemical couplings (such as movement of charges, rearrangement of hydrogen bond networks, and other phenomenon connected to acid-base transitions and structure modifications) that mask the electrostatic cooperativity between the opposite charges of the hemes and the deprotonated redox-Bohr center(s).

Considering that the heme with the redox-Bohr effect is the one hypothesized as being high-spin in the oxidized state, there may be a physiological relevance for this effect. The pH-dependent stabilization of the reduced form of this heme (low-spin), compared to the oxidized one (high-spin), suggests that the pH might modulate the heme spin-state interconversion. At a pH below the pK_a of the redox-Bohr center, the high-spin configuration (oxidized state) will be stabilized. The opposite effect is attained at pH values above the pK_a of the redox-Bohr center. Similar pH-dependent spin-state interconversion mechanisms have been observed in other cytochromes (68, 69, 70, 71); however, further experiments are needed to pinpoint the physiological relevance of this pH-linked feature of GSU0105.

The redox-Bohr effect is particularly relevant in other triheme cytochromes from G. sulfurreducens (see inset of Fig. 5; Table 2), such as PpcA (21 mV) and PpcD (16 mV), which were proposed to couple electron and proton transfer at the bacterium’s physiological pH range (72). In fact, in these proteins, it was observed that the deprotonation of the redox-Bohr center (removal of a H⁺) facilitates the concomitant oxidation (e⁻ transfer) of the nearest heme (15). On the other hand, PpcB and PpcE have negligible pH modulation of their heme reduction potential values (6 and 5 mV, respectively) and were suggested to exclusively perform electron transfer at slightly different redox windows. In the case of GSU0105, the redox-Bohr effect does not affect the E_app, but instead, it is a particular feature of the highest redox potential heme. Consequently, the pH dependence of the redox potential value of this heme allows the protein to reduce or expand its working functional range (Table 2) with the concomitant regulation of the low- to high-spin interconversion of the heme group.

Finally, it is worth noting that the redox curves of the PpcA family from G. sulfurreducens are much steeper compared to those of GSU0105, a consequence of the more similar heme reduction potential values (Fig. 5; Table 2). Considering the differentiated redox window of GSU0105 compared to the cytochromes from the PpcA family, it seems this cytochrome has the necessary properties to bridge electron transfer between a wider range of electron donors and acceptors in the periplasm of G. sulfurreducens.

Conclusions

The outgoing interest in electrochemically active microorganisms and their potential for several fields of biotechnology is prompting the investigation and exploration of their unique metabolism and molecular components. G. sulfurreducens is in the forefront of such investigations because of its high current production in MFCs, metabolic versatility, and ability to produce protein nanowires. The intricacies of this bacterium’s EET routes are slowly being dissected; however, it is still unclear which are the main electron transfer pathways and how they relate with different final electron acceptors. In this work, we presented a biophysical characterization of GSU0105, a periplasmic triheme cytochrome involved in Fe(III) respiration. The amino acid sequence comparison with other triheme cytochromes leads to the proposal that GSU0105 does not belong to the c₇ family of triheme cytochromes but, instead, is a member of a group of homologous cytochromes with mixed heme axial coordination.

GSU0105 possesses a low percentage of folded structural elements and three low-spin hemes in the reduced state (Fe(II), S = 0) and a mixture of low- (Fe(III), S = 1/2) and high-spin (Fe(III), S = 5/2) hemes in the oxidized state as confirmed by CD, UV-visible, EPR, and NMR spectroscopic techniques. The amino acid sequence analysis indicates that one of the hemes in GSU0105 contains a bis-His coordination, using the conserved His41 as the sixth ligand (Fig. 1), whereas the NMR experiments performed on GSU0105’s mutants indicate that the other heme contains a His-Met coordination (Fig. 1). The third heme is likely coordinated by one of the three fully conserved lysine residues (Lys33, Lys62, and Lys66). The data also indicate that this heme turns into a high-spin electronic configuration in the oxidized state, in which the distal position of the heme may become vacant or, alternatively, coordinated by a weak-field ligand (such as a water molecule, for example).

Finally, the effect of the pH on the redox properties of the protein (redox-Bohr effect) revealed an unprecedented mechanism for a triheme cytochrome. In fact, in the physiological region, the E_app value is unaffected by the pH, in opposition to that of the heme of the highest reduction potential. Interestingly, this is the heme that undergoes a low- to high-spin interconversion, which indicates that this interconversion is pH driven. Moreover, this effect also has an impact on the modulation of the working functional redox range of the protein. Compared to the other triheme periplasmic cytochromes from G. sulfurreducens, the working redox functional window of GSU0105 is wider and covers more positive regions, adding even further versatility to the periplasmic components of the bacterium.

Author contributions

C.A.S. and L.M. conceived, designed, and supervised the project. F.F. and M.T. performed the EPR experiments. T.M.F. performed the remaining experiments. All authors analyzed the data and co-wrote the manuscript.

Editor: Wendy Shaw.

Supporting citations

Reference (73) appears in the Supporting material.