Complete genome sequence of Streptomyces cyanogenus S136, producer of anticancer angucycline landomycin A

Abstract

Streptomyces cyanogenus S136 is the only known producer of landomycin A (LaA), one of the founding members of angucycline family of aromatic polyketides. LaA displays potent anticancer activities which has made this natural product a target of numerous chemical and cell biological studies. Little is known about the potential of S136 strain to produce other secondary metabolites. Here we report complete genome sequence of LaA producer and how we used this sequence to evaluate for this species its phylogenetic position and diversity of gene clusters for natural product biosynthesis.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13205-021-02834-4.

Introduction

Streptomyces cyanogenus S136 (= DSM5087) has first become known in 1990 as a producer of orange-colored angucycline compounds of the landomycin family (Henkel et al. 1990). Since that time S136 remains the only natural source of landomycin A (LaA), the largest member of the family, as well as one of the most potent angucyclines in terms of activity against a range of cancer cell lines (Ostash et al. 2009). An in-depth understanding of genetics of S. cyanogenus is believed to provide more facile access to landomycins for various cell biology studies (Yushchuk et al. 2019). Besides the published sequences of biosynthetic gene clusters for LaA (Westrich et al. 1999; Rebets et al. 2003), and, recently, polyene lucensomycin (Yushchuk et al. 2021), nothing is known about the genetic capacity of S. cyanogenus S136 for the production of other secondary metabolites. Furthermore, the relatedness of S136 to the other streptomycetes and information on global regulators of LaA biosynthesis remain vestigial. As a necessary step towards addressing these issues, here we report generation and bioinformatics analysis of complete S136 genome. Our data reveal that S136 strain possesses an unusual set of 16S rRNA genes and a significant number of natural product biosynthetic gene clusters (BGCs). The latter is in line with what is observed for other Streptomyces spp. (Nett et al. 2009). The generated sequence was used as a test case to compare different computational tools for BGC mining. Results of our findings are presented and discussed here.

Materials and methods

Genomic DNA manipulations, sequencing and initial annotation

Streptomyces cyanogenus S136 (DSM5087) was kindly provided by Prof. A. Bechthold (Freiburg University, Germany). Its genomic DNA was purified from a 24 h culture grown at 30 °C in tryptic soy broth according to salting out procedure N4 as described in (Kieser et al. 2000). DNA concentrations and quality were determined using Trinean Xpose (Gentbrugge, Belgium) and Agilent RNA Nano 6000 kit on Agilent 2100 Bioanalyzer (Agilent Technologies, Böblingen, Germany). The DNA sample that has passed all quality control checks was used to make two WGS libraries, one for Illumina (TruSeq DNA PCR-Free Kit) and one for Oxford Nanopore Technologies (ONT) sequencing (SQK-LSK109 ligation sequencing kit). The Illumina library was sequenced on HiSeq platform, and ONT library was subjected to two GridION runs. Further data handling and hybrid assembly was carried out exactly as described in (Tippelt et al. 2020). The assembled S136 genome was annotated with Prokka v1.11 (Seeman et al. 2014) and deposited into GenBank under accession number CP071839.

Agar plug assay of antibacterial activity

S. cyanogenus S136 and its LaA-nonproducing mutant ∆lanI7 (Rebets et al. 2008) were grown for 5 days at 30 °C on following agar plates: SMMS, R5, SFM (Kieser et al. 2000) and ISP3 (g/L: oatmeal flour - 34, agar - 18, tap water to 1 L, pH prior to sterilization 8.0). The agar plugs were cut off the lawns and assayed against Bacillus cereus ATCC19637 as described in (Yushchuk et al. 2021).

Phylogenetic reconstruction

Phylogenetic analysis (maximum likelihood algorithm) was performed on concatenated amino acid sequences of conserved proteins using IQtree software on CIPRES server [www.phylo.org]. List of conserved S. coelicolor genes used in this work (Fig. S1, Electronic Supplementary Materials, ESM), was obtained from work of Gao and Gupta (2012). Using reciprocal best BLASTP hit (RBH) strategy, we identified orthologues in 82 complete Streptomyces genomes (species and corresponding GenBank accession numbers are given in Table S1), as well as in S. cyanogenus genome. BLAST v 2.9.0 minimum parameters to consider RBH orthologues were as follows: E-value ≤ 1e − 10, identity percentage ≥ 30%, coverage per high-scoring pair, no less than 50%. Evolutionary model LG + I + F + G4 was selected as the most optimal one for our dataset on the basis of minimal BIC score. Reliability of topology of the resulting trees was estimated with SH-aLRT test. For 16S phylogenetic tree GTR model was used for single gene nucleotide sequences.

Additional tools for S. cyanogenus genome annotation

To find oriC and replication-related genes, Ori-Finder2 server (Luo et al. 2019) was used. Pseudofinder (Syberg-Olsen et al. 2020) was used to predict pseudogenes. BlastKOALA and GhostKOALA (Kanehisa et al. 2016) were applied for K number assignment. Average nucleotide identity (ANI) calculator was accessed from http://enve-omics.ce.gatech.edu/ani/.

Tools used for analysis of secondary metabolite BGCs

Analysis of secondary metabolite BGC was performed with antiSMASH (Blin et al. 2019) with subsequent manual inspection. If region contained several putative BGCs (further referred to as subclusters in this paragraph), each of them was extracted from the region and re-annotated with antiSMASH. If similarity score of a subcluster to known BGC was higher than the score of the region, the subcluster was considered an independent BGC. If the similarity score was high (> 60%), BGC borders assignment was done using comparison to the known BGCs from MiBIG (Kautsar et al. 2020). For those high-scoring BGCs additional BLAST run was performed with reciprocal hits retrieval. There were cases when similarity score was below 60% or similarity cannot be inferred on the basis of resemblance to the known BGC. In these cases, Cluster Blast results were applied to retrieve the shared BGC block with an assumption, that this block is genuine BGC, taking into the synteny between the latter and the other automatically annotated regions. For the BGCs that can be compared to the known ones, manual inspection was performed to highlight missing genes, using publicly available articles where the reference BGCs were described.

Default setting were applied when we run S136 genome through DeepBGC, SEMPI, PRISM for BGC annotation, RRE-Finder, RiPPMiner, DeepRiPP, BAGEL4, NeuRiPP for ribosomally produced peptides (RiPP) annotation, ARTS and BiG-FAM for additional annotation. For RRE-Finder and NeuRiPP runs the S136 genome was re-annotated with Prodigal short. The DeepBGC data was filtered on such variables: deepbgc score > 50, cluster_type score > 50, activity score > 50, # domains > 5, # biodomains > 1, #  proteins > 1. For NeuRiPP analysis we used the CNN-parallel architecture, as it yielded the best results, according to the published data (Los 2019).

Results

General structure of S. cyanogenus S136 genome

A single high-quality (Phred score > 30) 8,773,899-bp long contig of S. cyanogenus S136 genome was generated as a result of sequencing and assembly strategy outlined in the Methods section. The contig was subsequently analyzed with BUSCO software (Seppey et al. 2019) to assess the completeness of the generated sequence. This analysis showed that the sequence truly belongs to the Streptomycetales lineage with two missing and one fragmented nearly-universal single copy orthologues (out of 1579; Fig. S2). Given the result, the sequence can be confidently considered a complete genome. Computed features of S136 genome are presented in Table 1 in comparison with several selected Streptomyces species. We included S. reticuli into the comparison because it is the closest neighbor of S136 strain in the phylogenetic tree (see next paragraphs).

Table 1.

Features of S. cyanogenus genome in comparison with model Streptomyces

Organism	Genome length (Mbp)	GC (%)	CDS	Average CDS length (bp)	Average coding density (%)	rRNA operons	tRNA	tRNA (tRNAscan-SE)a
S. cyanogenus	8.8	71.84	7590	992.09	87.07	6	90a	79
S. coelicolor	8.7	72.12	7767	992.03	88.9	6	65	72
S. bingchenggensis	11.9	70.76	10,096	1046.8	88.54	5	66	76
S. albidoflavus	6.8	73.32	5895	1014.93	87.45	7	66	69
S. xiamenensis	5.92	72.02	5348	987.43	88.58	5	56	61
S. reticuli	8.3	72.72	7411	990.29	87.85	6	72	74

^aAs annotated with barrnap in Prokka

^aAs annotated with barrnap in Prokka

Several peculiarities of S136 genome are worth further comments. Regardless of the search algorithm, S136 genome appears to possess tRNA gene set larger than in the other Streptomyces spp. genomes (Table 1). In the same time, differences between Prokka and tRNAscan-SE results show that this issue is worth studying in more detail. For example, after cursory analysis we revealed that some of the tRNA genes annotated by Prokka actually are parts of the protein-coding sequences. Out of 7590 annotated S136 CDS, 2534 fell into one of the KEGG categories. Results of BlastKOALA and GhostKOALA are similar with 33.4% and 32.2% of annotation success rate (Fig. S3, ESM). There was no observed GC skew in 960-bp oriC region (GC = 66.35%); nine genes involved in replication are scattered over the genome. We found 295 ORFs too short to be classified as genes, and 93 ORFs interrupted with stop codons (Syberg-Olsen et al. 2020). In total 388 elements could be classified as pseudogenes, which are evenly spread over the genome (Fig. S4, ESM).

Phylogenetic position of S136

Six genes for 16S rRNA were found within the genome of S. cyanogenus S136. Five of these genes were identical, the sixth one (namely S1361_18950) diverged significantly, sharing only 97.4% of nucleotide sequence identity with the other homologues (Fig. 1). Notably, such identity is lower than general taxonomic threshold, often used for species discrimination (98.7%, Yarza et al. 2014). We then first reconstructed the phylogeny of S136 strain and 82 other Streptomyces spp. (Table S1) using nucleotide sequence of S1361_07650 (one of the five identical 16S rRNA genes). The other tree was reconstructed using nucleotide sequence of the divergent S136 16S rRNA gene S1361_18950. The phylogenetic position of S136 appeared to be different in both trees, clearly depending on the 16S rRNA being used for the reconstruction (Fig. 2). S. puniciscabiei TW1S1 appeared to be a sister clade for S136 strain when S1361_18950 was used for reconstruction, whereas S. pluripotens MUSC135 was found to be the closest to S136 in the other reconstruction. Indeed, the 16S rRNA gene set of S. puniciscabiei aligned well to S1361_18950 (Fig. S5A, ESM), sharing 98.5% nucleotide sequence identity. Hence, S136 genome is a rare case of presence of disparate 16S rRNA gene whose origin and functional consequences deserve further investigations. Neither of the two reconstructions placed S136 strain in one clade with the other 4 S. cyanogenus strains, whose 16S rRNA genes are available in GenBank. (Fig. S5b, ESM).

Fig. 1.

Alignment of 480-bp segments of six 16S rRNA genes of S. cyanogenus encompassing variable V4-V6 region of 16S rRNA gene. Gene S1361_18950 is the most divergent one

Fig. 2.

Comparison of topologies of the tree (tanglegram) built on the dataset containing typical (left) and atypical (right) S. cyanogenus 16S rRNA genes. S136 position is marked with red asterisk. Multiple sequence alignment for these trees was prepared with MAFFT (maxiter = 1000, ginsi method)

Given that 16S rRNA genes might be suboptimal proxy to phylogeny of S. cyanogenus, we decided to compare the phylogenetic position of S136 on the trees based on protein and 16S markers. We followed the approach of Gao and Gupta who proposed a set of highly conserved proteins for phylogenetic reconstruction within class Actinobacteria (Gao and Gupta 2012). Unfortunately, genomic records for the other S. cyanogenus species were unavailable and thus were not included into reconstruction. There was a strong discordance in topology of 16S- and protein-based trees (Fig. S6). Given that a set of protein markers provides a more robust phylogenetic signal to infer intrageneric relationships (Kämpfer et al. 2012), we assume that the protein tree (Fig. 3) represents the most likely phylogenetic position of S. cyanogenus within Streptomyces genus. In this case S. reticuli TUE45 came out as the closest relative of S. cyanogenus. An average nucleotide identity between S136 and TUE45 genomes (using two-way ANI approach; Goris et al. 2007) was 91.44% which is below 95% threshold level for the members of one species. Therefore, S. cyanogenus S136 likely represents new genomic species of Streptomyces.

Fig. 3.

Streptomyces species tree built on the basis of conservative proteins. S. cyanogenus is at the bottom of the tree (left, red asterisk). ML algorithm was used for the tree reconstruction. On the left is the cladogram, on the right-phylogram of conserved proteins tree. Branch support values (%) are computed with the help of SH-alRT test

Secondary metabolome of S136

Within S136 genome antiSMASH annotated 33 secondary metabolite BGCs with default settings. Highest score was observed for several BGCs directing the production of ectoine, spore pigment, desferrioxamine B, albaflavenone, lagmysin, landomycin, geosmin, citrulassin E, hopene (more than 80% confidence). Only landomycin BGC is known to be actively expressed in S136 (Yushchuk et al. 2019). The number of BGCs predicted by antiSMASH should be treated with caution due to known problems with correct identification of BGC borders and inefficient recognition of carbohydrate-based BGCs (Olano et al. 2014). An example of artificial fusion of disparate BGCs into one region was clearly present in S136 genome, where well-studied LaA BGC was lumped together with the other two putative BGCs (Fig. S7, ESM) Furthermore, the alternative tools for BGC annotation, such as PRISM and DeepBGC, predict somewhat different sets of BGCs in S136 genome. We therefore manually compared the outputs of different software tools in an attempt to assess, as precisely as possible, the true number of different secondary metabolite BGCs in S136 genome (see Methods section and Fig. S8, ESM). As a result, a total of 41 BGCs were revealed (Table 2). Almost half of these BGCs (19 out of 41) significantly resemble known BGCs. 30% of all S136 BGCs are distinct from the BGCs families present in BiG-FAM database, a vast (over 1.2 million entries) collection of BGCs from different taxa (Kautsar et al. 2021). Particularly, their distance from BiG-FAM families was more than 900, a proposed threshold above which the BGC of interest is considered novel. The novelty of natural products encoded within S136 genome could also be viewed from a perspective of BGC percentage that belong to small BiG-FAM families (e.g. those that consist of less than 100 BGCs). In S136 genome 12.5% of the BGCs fell into this category. This could be interpreted as an additional layer of confidence in BGC prioritization, especially if those families have no experimentally characterized BGCs. A more detailed description of how different software tools portray genetic blueprint of the secondary metabolic landscape of S136 strain can be found in fig. S8–17, ESM, and the text that accompanies the figures.

Table 2.

Biosynthetic Gene Clusters in Streptomyces cyanogenus S136 genome

BGCs #	Coordinates (left–right edges)	Size, bp	Chemical core	Most similar (%, if available) BGC directs the production of	Notes
1	246,653–289,035	42,382	Polyketide, type I PKS	Macbecin (39)	Some genes for transport, regulation, precursor supply and polyketide modification are missing
2	335,166–341,095	5929	Saccharide	–	–
3	738,785–742,279	3494	Polyphenols, melanin	Melanin (57)	No transport gene in the BGC as compared to the reference
4	1,016,423–1,017,610	1187	Pyranone, type 3 PKS	Germicidin (100)	–
5	1,027,563–1,051,082	23,519	Peptide + nucleoside	–	–
6	1,121,707–1,212,402	90,695	Polyketide, type I PKS	Chlorothricin (55)	Similar to chlorothricin BGC; several structural genes absent
7	1,237,690–1,273,267	35,577	Peptide, NRPS	Acyldepsipeptide 1 (15)	Only core is similar to acyldepsipeptide
8	1,473,681–1,476,488	2807	Naphthquinone, type III PKS	Flaviolin (75)	Absent regulatory gene
9	1,620,953–1,667,528	46,575	Polyketide, type I PKS	–	–
10	2,301,274–2,304,405	3131	Amino acid (ectoine)	Ectoine (100)	Known ectoine BGC
11	2,700,072–2,709,366	9294	Polyketide, type II PKS	spore pigment (83)	100% similarity to curamycin BGC (BGC0000215)
12	3,187,463–3,191,286	3823	Terpene	–	No known BGC; clusterblast reveals BGCs with similar terpene subBGCs
13	3,323,050–3,324,394	1344	Polyphenols, melanin	Melanin (100)	Known melanin BGC
14	3,411,660–3,417,647	5987	Hydroxamate, NRPS	Desferrioxamin B (83)	Known siderophore BGC with missing transport gene
15	4,115,479–4,164,034	48,555	RiPP	Linaridin	Identified only by RiPPMiner
16	4,509,957–4,535,739	25,782	Peptide, NRPS	–	No similar BGCs in MiBIG
17	4,554,282–4,563,340	9058	Other, lincomycin	–	Core gene similar to LmbU from lincomycin BGC. No other hits
18	4,671,291–4,678,868	8480	Peptide, bacteriocin		Novel bacteriocin BGC
19	4,997,513–5,003,746	6233	RiPP	–	Core similar to BGCs for thiopeptide biosynthesis
20	5,735,976–5,738,290	2314	Terpene	Albaflavenone (100)	–
21	6,374,219–6,384,127	9908	Other, siderophore	–	–
22	6,453,759–6,520,752	66,993	Peptide, NRPS	–	Likely two BGCs; first one carries core genes and second one accessory (+ MeAsp subBGC). Considered one BGC until further verification
23	6,522,403–6,527,344	4941	RiPP	Lagmysin (80)	Highly similar to lagmysin BGC
24	6,552,992–6,560,907	7915	Terpene	Isorenieratene (75)	No regulatory gene and geranylgeranyl diphosphate synthesis genes
25	6,588,268–6,597,540	9272	Peptide, bacteriocin	–	–
26	6,606,883–6,632,271	25,388	Peptide, NRPS	–	Often located near bacteriocin BGC
27	6,643,964–6,676,435	32,471	Polyketide, type II PKS	landomycin A (100)	Active BGC
28	6,731,997–6,741,890	9893	Peptide, bacteriocin	–	–
29	6,770,569–6,772,731	2162	Terpene	Geosmin (100)
30	6,952,316–6,959,533	7217	Other, siderophore	–	–
31	6,985,105–6,991,275	6170	Butyrolactone	–	Butyrolactone BGC present in different species; no known hits
32	7,019,743–7,027,247	7504	Betalactone	–	–
33	7,293,687–7,298,086	4399	lanthipeptide, RiPP	citrulassin E (100)	Similar to citrulassin BGCs with different degree of gene losses
34	7,298,435–7,307,214	8779	lassopeptide, RiPP	–	No known BGC hit, but shared synteny across several species
35	7,427,159–7,440,932	13,773	Terpene	Hopene (92)
36	7,480,261–7,485,226	4965	Lassopeptide, RiPP	–	–
37	7,571,920–7,665,592	93,672	Polyketide, type I PKS	Naphthomycin A (68)	Naphtomycin BGC
38	7,705,775–7,792,970	87,195	Polyketide, type I PKS	Lucensomycin	(Yushchuk et al. 2021)
39	8,000,207–8,077,410	77,203	bacteriocin + NRP	–	–
40	8,414,168–8,439,406	25,238	Polyketide, type I PKS	–	–
41	8,461,058–8,490,615	29,557	Saccharide	–	Carbohydrate synthesis subBGC (42% similarity to d-olivose pathway from clorothricin BGC -)

Given the large complement of BGCs revealed in S. cyanogenus genome, we wondered as to whether all of them are unconditionally silent. Variation of cultivation conditions is one of the simplest ways to induce the expression of specialized pathways in Streptomyces; indeed, even landomycin A production by S136 is medium-dependent (Yushchuk et al. 2018, 2019). We therefore explored how several common agar media influence the production of antibacterial compounds by two S. cyanogenus strains, wild type (S136) and its landomycin-nonproducing mutant ∆lanI7. While the wild type showed no activity under our test conditions, the ∆lanI7 accumulated some antibacterial compounds when grown on SMMS, ISP3 and especially R5 agars (Fig. 4). The level of antibacterial activity observed under these conditions appeared to be higher than the one observed by us previously under different cultivation conditions (Yushchuk et al. 2021).

Fig. 4.

S. cyanogenus produces antibacterially active compounds under certain conditions. Agar plugs of S136 and ∆lanI7 were grown and assayed against B. cereus as described in Methods. Agar media used to grow the strains are denoted on the bottom of the figure. Halos of B. cereus growth inhibition are visible around agar plugs cut off the lawns of ∆lanI7 grown on SMMS, R5 and ISP3 media. Photos represent typical results of three independent expreriments

Discussion

In this work we report complete sequence of Streptomyces cyanogenus S136 genome, the only known producer of anticancer polyketide landomycin A. Two features stand out in this genome as compared to the genomes of the other Streptomyces species: increased number of tRNA genes and presence of highly divergent 16S rRNA gene. While the fact and relevance of inflated number of tRNA genes for S136 physiology await experimental scrutiny, the existence of one dissimilar 16S rRNA gene in S136 genome readily raises the issues of taxonomical kind. Indeed, we showed in this work that the use of different S136 rRNA genes in phylogenetic reconstruction led to incongruent species trees. One therefore has to be careful in the choice of the molecular markers to infer phylogeny for such strains. We recommend to use a set of protein-encoding genes in this case, as they would be less susceptible to reconstruction artifact. In all, our phylogenetic efforts led us to believe that S. cyanogenus S136 is a new genomic species, and its closest known relative is S. reticuli TUE45. 16S rRNA phylogeny reconstructions also showed that S. cyanogenus S136 is distantly related to four other S. cyanogenus strains.

The genome sequence was subjected to a number of bioinformatics tools that identify natural product BGCs, and a round of manual curation of the output data. This approach yielded a list of 41 BGCs, a significant portion of which overlap with the previously experimentally characterized BGCs present in MiBIG database. Some S136 BGCs lack MiBIG hit; in this case their identity was inferred from combined data annotation approach, which is graphically summarized in ESM with the help BGCViz software (Fig. S9, ESM). Therefore, even in absence of comparable MiBIG data, novel BGCs can be pinpointed by taking into account the combined annotation results from a set of search tools other than antiSMASH.

According to our findings (see ESM, Fig. S9–17), BGCs # 9, 16, 22, 26, 39, 40 and 41 deserve special attention due to large distance between them and BGCs deposited in BiG-FAM database, implying their novelty. Nevertheless, we cannot rule out at the moment that our assumption of novelty can be undermined by the incompleteness of the BGCs within S136 genome. Attempts to induce the expression of these BGCs in S136 genome are underway in our laboratories. Genome of S136 also carries two BGCs similar, yet not identical, to well-studied gene clusters whose activation could be achieved in a relatively straightforward way given the presence of all necessary cluster-situated regulatory genes. These are BGCs #6 (for chlorothricin-like compound) and #37 (naphthomycin A). Indeed, our recent efforts in this regard allowed us to activate the BGC #38 leading to production of expected polyene compound lucensomycin (Yushchuk et al. 2021). Several BGCs seem to be silent because of the absence of regulatory or transport genes. These are BGCs #1 (macbecin), #3 (melanin), #8 (flaviolin), #14 (siderophore) and #24 (isorenieratene). Respective or similar BGCs have been studied in other species, and they carry regulatory or transport genes that can be expressed heterologously in S136 to induce the aforementioned BGCs.

Specialized RiPP-finding tools reveal a swath of RiPP “dark matter” potentially encoded within S136 genome. RiPPMiner and BAGEL4 were the most promising tools displaying good balance between ability to identify putative BGCs within novel genomic regions and identification reliability. Full list of potential RiPP BGCs, as predicted by the aforementioned tools, are given in ESM, fig. S15–17; below we discuss a few of them, which could be confidently described. BGCs #19, 34 and 36 encode novel RiPPs and are identified by multiple bioinformatic tools (see Table 2). The cluster #15, initially detected by RippMiner, wasn’t pinpointed by any other search tools, except NeuRiPP, and so it can be considered a novel RiPP. BGC #1 from DeepRiPP (Fig. S16) intersects with NRPS clusters of PRISM and DeepBGC. Given the high reliability of PRISM identification of NRPS clusters, this RiPP hit should be treated with caution. The antiSMASH-identified BGC (#7) does not overlap with this gene cluster.

To conclude, S. cyanogenus S136 likely represents novel species that displays a large potential to produce various natural products besides landomycins and lucensomycin. Its genome harbors numerous BGCs directing the biosynthesis of reduced polyketides, peptides of ribosomal and non-ribosomal origin, terpenes and saccharides. Some of them are worth further experimental efforts due to high likelihood of chemical novelty of the corresponding natural products. Our initial bioassays readily reveal that some of these BGCs are expressed and lead to copious production of some as-yet-unknown antibacterially active compounds. This emphasizes biotechnological potential of S. cyanogenus S136.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

The work was supported by grants BG-80F and BG-09F from the Ministry of Education and Science of Ukraine (to B.O. and V.F.).

Declarations

Conflict of interest

The authors declare that they have no conflict of interests.