Volume 29, Issue 64 e202302469
Research Article
Open Access

Mechanistic Characterisation of Collinodiene Synthase, a Diterpene Synthase from Streptomyces collinus

Kizerbo A. Taizoumbe

Kizerbo A. Taizoumbe

Kekulé-Institute for Organic Chemistry and Biochemistry, University of Bonn, Gerhard-Domagk-Straße 1, 53121 Bonn, Germany

Search for more papers by this author
Simon T. Steiner

Simon T. Steiner

Kekulé-Institute for Organic Chemistry and Biochemistry, University of Bonn, Gerhard-Domagk-Straße 1, 53121 Bonn, Germany

Search for more papers by this author
Prof. Dr. Jeroen S. Dickschat

Corresponding Author

Prof. Dr. Jeroen S. Dickschat

Kekulé-Institute for Organic Chemistry and Biochemistry, University of Bonn, Gerhard-Domagk-Straße 1, 53121 Bonn, Germany

Search for more papers by this author
First published: 14 August 2023

Graphical Abstract

Two homologs of the diterpene synthase CotB2 for cyclooctat-9-en-7-ol from Streptomyces collinus and S. iakyrus were investigated. The enzyme from S. collinus produces the new compound collinodiene besides cyclooctat-9-en-7-ol and other side products. Isotopic labelling experiments were performed to enlighten the cyclisation mechanism.


Two homologs of the diterpene synthase CotB2 from Streptomyces collinus (ScCotB2) and Streptomyces iakyrus (SiCotB2) were investigated for their products by in vitro incubations of the recombinant enzymes with geranylgeranyl pyrophosphate, followed by compound isolation and structure elucidation by NMR. ScCotB2 produced the new compound collinodiene, besides the canonical CotB2 product cyclooctat-9-en-7-ol, dolabella-3,7,18-triene and dolabella-3,7,12-triene, while SiCotB2 gave mainly cyclooctat-9-en-7-ol and only traces of dolabella-3,7,18-triene. The cyclisation mechanism towards the ScCotB2 products and their absolute configurations were investigated through isotopic labelling experiments.


The reactions catalysed by terpene synthases (TSs) belong to the most intricate single step transformations in natural product biosynthesis. These cyclisations of simple linear precursors such as geranyl (GPP), farnesyl (FPP), geranylgeranyl (GGPP) and geranylfarnesyl pyrophosphate (GFPP) proceed via cationic cascades initiated by diphosphate abstraction (class I and UbiA-related TSs)1-3 or by protonation (class II TSs).4, 5 In some cases two TS functionalities are combined in one enzyme with two domains.6-8 The rapid developments in genome sequencing technologies have resulted in the identification of TSs in various organisms including bacteria,1 fungi,9 plants,10 social amobae,11, 12 insects,13 octocorals,14, 15 and red algae.16, 17 For type I enzymes not only mono- and sesquiterpene synthases have been described, but more recently also several diterpene synthases,18, 19 sesterterpene synthases20-23 and even triterpene synthases24 have been reported. For the first characterised bacterial type I diterpene synthase (DTS), cyclooctat-9-en-7-ol synthase (CotB2) from Streptomyces melanosporofaciens,25 the gene is clustered with genes for the GGPP synthase (GGPPS) CotB1 and for two cytochromes P450, CotB3 and CotB4, that convert cyclooctat-9-en-7-ol (1) via cyclooctat-9-en-5,7-diol (2) into the lysophospholipase inhibitor26 cyclooctatin (3) (Scheme 1). A BLAST search revealed that the biosynthetic gene cluster for 3 occurs in at least 72 genome sequenced actinomycetes. A phylogenetic analysis based on the CotB2 amino acid sequences demonstrated that these enzymes fall into two clades, with clade A containing 48 CotB2 homologs including the known enzyme from S. melanosporofaciens, and clade B with 24 CotB2 homologs (Figure 1, an expansion for the branch of the CotB2 homologs is given in Figure S1). Here we report on the functional characterisation of two clade B enzymes from Streptomyces collinus and from Streptomyces iakyrus, and isotopic labelling experiments to investigate the cyclisation mechanism towards 1 and collinodiene, a newly obtained diterpene produced by the CotB2 homolog from S. collinus.

Details are in the caption following the image

Cyclooctatin (3) biosynthesis in Streptomyces melanosporofaciens.

Details are in the caption following the image

Phylogenetic tree constructed from 4018 bacterial terpene synthase homologs. The largest branches with at least one functionally characterised enzyme are coloured in blue (sesquiterpene synthases) or green (diterpene synthases). The scale bar indicates substitutions per site.

Results and Discussion

The CotB2 homolog from S. collinus (ScCotB2) exhibited all highly conserved motifs required for functionality including the aspartate-rich motif 108DDMD and the NSE triad 218NDFYSYDRE, and showed 71 % amino acid sequence identity to CotB2 from S. melanosporofaciens (SmCotB2) (Figure S2). The same conserved motifs were observed in the CotB2 homolog from S. iakyrus (SiCotB2), with an identity of 75 % to SmCotB2 (Figure S3, an amino acid sequence alignment of all three enzymes is shown in Figure S4). The purified enzyme from S. collinus (Figure S5) converted GGPP into a mixture of diterpene hydrocarbons besides the alcohol 1 (Figure 2A). Compound 1 and three hydrocarbons were isolated from a preparative scale incubation and their structures were characterised by NMR spectroscopy (Tables S2–S5, Figures S6–S37), resulting in the identification of dolabella-3,7,18-triene (4, [α]D25=+47.4, c 0.02, CH2Cl2), a known natural product from the brown alga Dilophus spiralis ([α]D20=+41.0, c 0.10, CHCl3),27 dolabella-3,7,12-triene (5, [α]D25=+25.0, c 0.03, CH2Cl2) for which only the enantiomer has been obtained by synthesis before ([α]D25=−25.0, c 0.12, CHCl3, 90 % ee),28 and the new natural product 6 for which we propose the name collinodiene. In contrast, SiCotB2 produced mainly 1 besides traces of 4, but no 5 or 6 (Figure 2B). None of the two enzymes accepted GPP, FPP or GFPP as substrate.

Details are in the caption following the image

Total ion chromatograms of the products obtained from GGPP with A) ScCotB2 and B) SiCotB2. Asterisks indicate acyclic products arising through spontaneous (hydro)lysis of GGPP in the presence of Mg2+.

The proposed cyclisation mechanism for the products of ScCotB2 starts from GGPP with 1,11- and 10,14-cyclisations to A followed by a 1,5-hydride shift to B (Scheme 2). A third cyclisation to C and two sequential 1,2-hydride migrations yield E that can undergo another 1,5-hydride transfer to F. A subsequent 1,2-Me migration to I, a 1,4-hydride shift to J and final deprotonation result in 6. Notably, intermediate F is a branching point towards 1, from which a surprising mechanism by cyclisation to G, rearrangement to H and ring opening with attack of water was determined in previous isotopic labelling experiments.29 The side products 4 and 5 are explainable from intermediate A by deprotonation or a 1,2-hydride shift to K and deprotonation.

Details are in the caption following the image

A) Structure elucidation of 6 (bold: COSY correlations, single headed arrows: HMBC correlations, double headed arrows: NOESY correlations). B) Cyclisation mechanism from GGPP to the ScCotB2 products.

The ScCotB2 cyclisation mechanism was deeply investigated in a series of in vitro experiments using isotopically labelled terpene precursors (Table S6). While compound 6 was the main product of ScCotB2 in a preparative scale incubation with 80 mg of GGPP (Figure 2A), small scale incubation experiments with labelled substrates (1 mg scale) usually gave 1 as the main product and smaller amounts of 46. The enzymatic conversion of all twenty isotopomers of (13C)GGPP29, 30 with ScCotB2 followed by extraction with C6D6 and product analysis through 13C NMR spectroscopy demonstrated the incorporation of labelling into positions of 1 and 46 in agreement with the mechanism of Scheme 2 in all cases (Figures S38–S57). Specifically, these experiments confirmed the previously reported rearrangement from the cyclopropylcarbinyl cation G to the cyclopropylcarbinyl cation H, resulting in a positional exchange of C-8 and C-9 of 1 (Figures S45 and S46).31 The results also supported the 1,2-Me group migration from F to I (Figure S55) and revealed a strict stereochemical course for the deprotonation of A from C-17 – and not from C-16 – to form 4. Also, the face selectivity at C-15 in the 1,5-hydride migration from A to B is high, as the labelling from C-16 and C-17 is not distributed over two positions in the product, but each selectively incorporated into one of the two geminal Me groups of 1, 5 and 6 (Figures S53 and S54).

The 1,5-hydride shift from A to B in the biosynthesis of 1 and 6 was investigated using dimethylallyl pyrophosphate (DMAPP) and (E)- or (Z)-(4-13C,4-2H)isopentenyl pyrophosphate32 (IPP) in conjunction with GGPP synthase (GGPPS) from Streptomyces cyaneofuscatus33 and ScCotB2 (Figure S58). With these substrates stereoselective deuterations are introduced at C-8 of GGPP with a known stereochemical course,34 and with (E)-(4-13C,4-2H)IPP an upfield shifted triplet was observed for C-8 of 1 (Δδ=−0.37 ppm, 1JC,D=22.8 Hz), demonstrating that the deuterium incorporated from this substrate remains bound to C-8 (Figure S58C; no signal was detected for C-8 of 6). With DMAPP and (Z)-(4-13C,4-2H)IPP an unchanged singlet was obtained for C-8 of 6, while for C-8 of 1 a doublet as a result of a 3JC,C coupling was observed at the original chemical shift of this carbon (Figure S58D), indicating that the deuterium from (Z)-(4-13C,4-2H)IPP shifts away from its original position. The target position C-15 for this shifting hydride was evident for compound 1 from a GC/MS analysis (Figure S59). For the unlabelled compound a fragment ion is observed at m/z 229 arising by the neutral loss of water and iPr group cleavage. With (E)- and (Z)-(4-13C,4-2H)IPP the corresponding fragment ions are observed at m/z 235 and m/z 234, respectively, confirming the loss of one deuterium in the iPr group cleavage and thus the hydride shift from C-8 to C-15 with (Z)-(4-13C,4-2H)IPP. Analogous results were obtained for compound 6 (Figure S60).

The 1,2-hydride shift from C to D in the biosynthesis of 1 and 6 was studied through the enzymatic transformation of (3-13C,2-2H)GGPP33 with ScCotB2 (Figure S61). As a result of the hydride shift from C-2 to C-3 a direct 13C-2H bond is formed, evident from the upfield shifted triplets for C-3 of 1δ=−0.49 ppm, 1JC,D=19.5 Hz) and of 6δ=−0.47 ppm, 1JC,D=19.5 Hz). Along similar lines, the 1,2-hydride shift from D to E in the biosynthesis of 1 was investigated using (2-2H)FPP33 and (2-13C)IPP35 in combination with GGPPS and ScCotB2, resulting in the detection of an upfield shifted triplet for C-3 of 1δ=−0.42 ppm, 1JC,D=19.2 Hz, Figure S62; the corresponding signal for 6 was too weak in this experiment). The 1,2-hydride shift from D to E was also evident from an experiment with (2-2H)FPP33 and (3-13C)IPP.33 With this substrate combination the migrating deuterium ends up in a neighbouring position of the labelled carbon C-3, which influences its chemical shift in 1 and 6δ=−0.11 ppm in both cases, Figure S62).

This method was also used to investigate the 1,5-hydride shift from E to F in the biosynthesis of 1 (Figure S63). The enzymatic conversion of (2-2H)GPP36 and (3-13C)IPP leads to a product in which the deuterium ends up in between the labelled carbons C-3 and C-7. The latter one is one position away from the target position C-6 of the migrating deuterium and thus influenced by an upfield shift of Δδ=−0.06 ppm, while the small upfield shift for C-3 (Δδ=−0.01 ppm) is typical for a deuterium atom in a distance of two positions. Therefore, the labelling of both C-3 and C-7 allows to track down the final destination of the deuterium atom with confidence, which confirms the 1,5-hydride shift. For 6 the same hydrogen atom is supposed to undergo a second relocation from C-6 to C-11 in the 1,4-hydride transfer from I to J. As a result of the combined hydride migrations from E to F and from I to J the hydrogen at C-10 of GGPP should end up at C-11 of 6. The results obtained from an incubation of (2-2H)GPP and (1-13C)IPP with GGPPS and ScCotB2 supported this mechanistic hypothesis through the observation of an upfield shifted doublet for C-1 (Δδ=−0.10 ppm, 3JC,C=2.1 Hz), indicating deuterium in a neighbouring position (C-11, Figure S64). As for the terminal deprotonation step from J to 6, the incubation of (R)- and (S)-(1-13C,1-2H)IPP37 with IDI,38 GGPPS and ScCotB2 demonstrated the specific loss of the proton from (R)-(1-13C,1-2H)IPP (Figure S65).

For compound 1 the hydride shifts from A to B and from E to F had been investigated previously using the deuterated substrates (8,8-2H2)GGPP and (10-2H)GGPP.31 The advantage to place additional 13C-labels in the substrate is that these can be used as probes to detect the presence of deuterium at or near the target position in the product without the need of compound isolation, albeit on the expense of a more complex synthesis of the labelled materials, while the usage of only deuterium often requires compound isolation and full structure elucidation by NMR spectroscopy to identify the target positions of migrating deuterium atoms. Furthermore, the stereoselective labelling experiments performed here have clarified which of the two hydrogen atoms migrates in the step from A to B (Hα), which confirms a prediction based on three computational studies.39-41 The same computational studies propose for the unique rearrangement from G to H in the biosynthesis of 1 a bicyclobutonium cation transition state.42 This rearrangement can be expected to proceed with inversion of configuration at C-9 (Scheme 3). The deuterium incorporation observed from (R)- and (S)-(1-13C,1-2H)IPP after conversion with IPP isomerase (IDI), GGPPS and ScCotB2 confirmed a stereochemical course with inversion of configuration (Figure S66).

Details are in the caption following the image

The cyclopropylcarbinyl-cyclopropylcarbinyl rearrangement from G to H in the biosynthesis of 1 proceeds with inversion of configuration at C-9.

The interpretation of the results from this experiment require the knowledge about the absolute configuration of 1 that has originally been assigned through the modified Mosher method.31 During the course of this study stereoselective labelling experiments through enzymatic conversions of DMAPP and (E)- or (Z)-(4-13C,4-2H)IPP with GGPPS and ScCotB2 have been performed to assign the absolute configuration of 6. These stereoselectively labelled precursors introduce stereochemical anchors of known configuration into GGPP and subsequently into 6. The relative configuration with respect to the naturally present stereogenic centres can be resolved from NOESY assignments which ultimately allows to conclude on the absolute configuration of 6 as shown in Scheme 2 (Figure S67). The additional 13C-labels in the stereoselectively deuterated probes enable a sensitive detection of the incorporation pattern through HSQC spectroscopy without the need of compound isolation. The same experiment confirmed the absolute configurations of 1, 4 and 5 (Figures S68–S70), and also the results from two more experiments using (R)- and (S)-(1-13C,1-2H)IPP in conjunction with IDI, GGPPS and ScCotB2 were in line with the same absolute configurations of 6, 1 and 5 (Figures S71–S73).


Taken together, we have demonstrated that the CotB2 homologs from clade B not only produce cyclooctat-9-en-7-ol (1), which represents the single product of the clade A enzyme SmCotB2 from S. melanosporofaciens, but also form additional side products as observed for SiCotB2 from S. iakyrus that produces minor amounts of dolabella-3,7,18-triene (4). For ScCotB2 from S. collinus the product profile is shifted towards collinodiene (6), with additional formation of 1, 4 and dolabella-3,7,12-triene (5). Whether this functional switch is specific for clade B enzymes remains to be investigated by studying the products of more CotB2 homologs. It has been shown by site-directed mutagenesis of SmCotB2 that exchanges of single amino acid residues can completely change the enzyme product.43-45 Notably, targeting the active site residue Trp288 in the W288G variant results in 4 (the same enantiomer as obtained with ScCotB2),44 while the N103A variant makes a main product with the same planar structure as 5.45

We have also investigated the cyclisation mechanism towards 1 in all detail and could experimentally clarify a prediction based on DFT calculations in the gas phase39, 40 and QM/MM simulations in the crystallised enzyme,41 which of the diastereotopic hydrogens at C-8 migrates towards C-15. Also all other results of the labelling experiments were in line with all previous computational work, e. g. regarding a sequence of two 1,2-hydride shifts (C-D-E), instead of an originally proposed 1,3-hydride shift.31 This result is also in line with our recent study on guaianes showing that 1,3-hydride shifts can only occur in trans-substituted cyclopentyl cations, but not in cis-systems.46 Furthermore, the unique cyclopentylcarbinyl-cyclopentylcarbinyl rearrangement in the biosynthesis of 1 was shown to proceed with inversion of configuration. Also for 6 the cyclisation cascade was investigated experimentally in all elementary steps. For the early steps the cascade is the same as for 1, and unique steps were shown to include a Me group migration and a 1,4-hydride shift. None of the enzyme variants reported so far resulted in compounds with the rearranged skeleton of 6, but it will be interesting to investigate in a future site-directed mutagenesis study which differences in the amino acid sequences of CotB2 homologs are important to produce this skeleton.

Experimental Section

Construction of phylogenetic tree: Based on the amino acid sequences of 4018 bacterial terpene synthase homologs that were collected by a BLAST search in bacteria with sequenced genomes a phylogenetic tree was constructed (Figure 1). Every sequence included in this analysis was individually inspected for the presence of the highly conserved motifs of type I terpene synthases (including the aspartate-rich motif, the NSE triad, the pyrophosphate sensor and the RY pair). The tree was constructed through the tree builder function of Geneious (alignment type: global alignment with free end gaps, cost matrix: Blosum45, genetic distance model: Jukes-Cantor, tree build method: neighbor-joining, gap open penalty: 8, gap extension penalty: 2).

Strains and culture conditions: Streptomyces collinus Tü 365 and Streptomyces iakyrus DSM 40482 were obtained from DSMZ (Braunschweig, Germany) and were cultivated on medium 1496 (5.0 g tryptone, 3.0 g yeast extract, 1.0 g glucose, 1.0 L H2O, pH 7.0, for agar plates 15 g agar-agar was added) at 28 °C. Saccharomyces cerevisiae FY834 was grown in liquid YPAD medium (10.0 g yeast extract, 20.0 g peptone, 20.0 g glucose, 400 mg adenine sulphate, 1.0 L H2O) or on SM-URA agar plates (1.7 g yeast nitrogen base, 5.0 g ammonium sulphate, 20.0 g glucose, 770 mg nutritional supplement minus uracil, 20.0 g agar-agar, 1.0 L H2O) at 28 °C. Escherichia coli BL21 (DE3) was grown in LB broth (10.0 g tryptone, 5.0 g yeast extract, 5.0 g NaCl, 1.0 L H2O, for agar plates 15 g agar-agar was added. Kanamycin was used at a concentration of 50 μg mL−1 at 37 °C. All media were autoclaved at 121 °C for 20 min prior to use.

Gene cloning: The desired genes encoding terpene synthases were obtained via polymerase chain reaction (PCR) and were amplified using Q5® -High-Fidelity DNA polymerase (New England Biolabs, Ipswich, MA, USA) together with the isolated genomic DNA and the primer pairs WP020942948_F and WP020942948_R (for ScCotB2, accession number WP_020942948), and WP033306056_F and WP033306056_R (for SiCotB2, accession number WP_033306056), respectively (Table S1). The temperature program was: initial denaturation at 98 °C for 5 min, followed by 33 cycles with melting at 98 °C for 10 s, annealing at 60 °C for 30 s and elongation at 72 °C for 40 s, and then a final extension at 72 °C for 5 min. The primary amplificates were used as a template in a second PCR with the primers containing homology arms (homologous sequence to the end sequences of linearised pYE-Express vector, digested with EcoRI and HindIII) for homologous recombination in yeast using the standard PEG/LiOAc/salmon sperm protocol.47, 48 S. cerevisiae cultures were grown on SM-URA plates for 2 days and cells were collected to isolate the plasmid DNA with the Zymoprep Yeast Plasmid Miniprep II kit (Zymo Research, Irvine, CA, USA). The isolated plasmid was used to transform E. coli BL21 (DE3) electrocompetent cells. Cells were plated and cultivated overnight on LB agar plates containing kanamycin. Single colonies were picked and used to inoculate 6 mL LB with kanamycin. After 24 h incubation time plasmids were extracted using the PureYield Plasmid Miniprep System (Promega) and checked by analytical digest and by sequencing. Thus, transformants carrying the plasmids pYE_ScCotB2 and pYE_SiCotB2.

Gene expression and protein purification: A preculture of LB medium (10 mL) supplied with kanamycin (50 μg/mL) was inoculated with the E. coli expression strain and grown with shaking overnight at 37 °C. The preculture was used to inoculate expression cultures (1 mL L−1) that were grown with shaking at 37 °C until an OD600 of 0.4–0.6 was reached. The cultures were cooled to 18 °C and IPTG solution (400 mm, 1 mL L−1) was added to induce protein expression. The cultures were shaken overnight and harvested by centrifugation (3.600×g, 30 min). The pelleted cells were resuspended in binding buffer (10 mL L−1 culture; 20 mm Na2HPO4, 500 mm NaCl, 20 mm imidazole, 1 mm MgCl2, pH=7.4, 4 °C) and lysed under ice cooling by ultra-sonication (5×1 min). The cell debris was removed by centrifugation (14.610×g, 2×7 min), the supernatant was filtered through a syringe filter and loaded onto a Ni2+-NTA affinity chromatography column (Protino Ni-NTA, Macherey-Nagel, Düren, Germany). The column was washed with binding buffer (2 column volumes) and the desired His6-tagged protein was eluted using elution buffer (20 mm Na2HPO4, 500 mm NaCl, 500 mm imidazole, 1 mm MgCl2, pH=7.4, 4 °C). Protein concentrations were determined by Bradford assay.49

Incubation experiments and compound isolation: Protein fractions obtained from Ni2+-NTA purification were used directly for test enzymatic incubation experiments with geranylgeranyl diphosphate (1 mg) in incubation buffer (1 mL; 50 mm Tris/HCl, 10 mm MgCl2, 10 % glycerol, pH=8.2) containing β-cyclodextrin (10 mm) and enzyme (0.5 mg mL−1). Incubations were performed at 30 °C with shaking overnight. The reaction mixtures were extracted with 200 μL n-hexane, followed by drying with MgSO4 and GC/MS analysis. Preparative scale incubations were performed under the same conditions using 80 mg GGPP, followed by extraction with n-hexane (3×150 mL). The pooled organic fractions were dried with MgSO4, evaporated under reduced pressure and subjected to column chromatography on silica gel using pentane 100 % and pentane/diethyl ether (2 : 1) to yield the respective terpenes as colourless oils.

Isotopic labelling experiments: Isotopic labelling experiments were performed with substrates (ca. 1.0 mg each, in 1 mL 25 mm aq. NH4HCO3), incubation buffer (5 mL) and enzyme elution fractions (1 mL each). The substrates and enzyme preparations are listed in Table S6. After incubation at 30 °C with shaking overnight, the products were extracted twice with C6D6 (500 μL and 200 μL). The extracts were dried with MgSO4 and analysed by NMR and GC/MS.

GC/MS analyses: GC/MS analyses were carried out on a 7890B/5977 A series gas chromatography/mass selective detector (Agilent, Santa Clara, CA, USA). The GC was equipped with an HP5-MS fused silica capillary column (30 m, 0.25 mm i. d., 0.50 μm film; Agilent) and operated using the settings: 1) inlet pressure: 77.1 kPa, He at 23.3 mL min−1, 2) injection volume: 1 μL, 3) temperature program: 5 min at 50 °C, then increasing by 5 °C min−1 or 10 °C min−1 to 320 °C, 4) splitless or split ratio 50 : 1, 60 s valve time, and 5) carrier gas: He at 1 mL min−1. The MS was operated with the same settings: 1) source: 230 °C, 2) transfer line: 250 °C, 3) quadrupole: 150 °C and 4) electron energy: 70 eV. Retention indices (I) were determined from a homologous series of n-alkanes (C7−C40).

NMR spectroscopy: NMR spectra were recorded on a Bruker (Billerica, MA, USA) Avance III HD Cryo (700 MHz) and AV III HD Prodigy (500 MHz) NMR spectrometer. Spectra were referenced against solvent signals (1H NMR, residual proton signals: CDCl3 δ=7.26 ppm, C6D6 δ=7.16, D2O δ=4.79; 13C NMR: CDCl3 δ=77.16 ppm, C6D6 δ=128.06 ppm).50


This work was supported by the Deutsche Forschungsgemeinschaft DFG (project number 513548540). Open Access funding enabled and organized by Projekt DEAL.

    Conflict of interest

    The authors declare no conflict of interest.

    Data Availability Statement

    The data that support the findings of this study are available in the supplementary material of this article.