De-novo assembly of 82 bacterial genomes using Nanopore sequencing and prediction of biosynthetic capacity

Abstract

Historically, antibiotic development has been driven by research into secondary metabolism which is responsible for the biosynthesis of so-called natural products. This research has been revolutionized by our growing capacity to rapidly generate and analyze prokaryotic genome sequences, facilitating the identification of biosynthetic gene clusters (BGCs) involved in secondary metabolism. We present the genomic analysis of 82 bacterial isolates from our collection, covering a broad taxonomic spectrum. The bulk of analyzed isolates belong to the actinobacterial class of Actinomycetes, known for their remarkable ability to generate antimicrobial compounds with diverse chemistries. Whole genome sequencing was performed using the Oxford Nanopore technology and the majority of assembled sequences exhibit over 90% gene completeness, enabling reliable functional annotation. Genome mining analyses revealed variability in BGC content across different phyla, confirming isolates in the Streptomyces genus as preeminent producers of secondary metabolites. Utilization of the current dataset holds potential not only for the elucidation of the bacterial biosynthetic machinery, but also for phylogenetic comparison and biotechnological exploitation of the isolates under study.

Background & Summary

Microbes display tremendous biosynthetic capacities, producing a vast spectrum of secondary metabolites in response to environmental changes or interactions with other organisms¹. Beyond their ecological significance, these compounds, also known as natural products, serve as a rich source for the discovery of drugs, particularly antibiotics^2,3. Genes responsible for the biosynthesis of a specific secondary metabolite co-localize together in the microbial genome, forming biosynthetic gene clusters (BGCs)⁴. Actinomycetes, in particular, have been well-documented as prime producers of different antibiotic classes, mainly thanks to bacteria in the Streptomyces genus that typically harbor multiple (20 to 70) BGCs per genome^5,6. Considering the abundance of as yet unknown or uncultivated bacterial strains, the fact that the majority of BGCs remain “cryptic” in the absence of specific stimuli, as well as the appreciable diversity of BGC content even at the individual genome level, it is reasonable to assume that a great number of novel bioactive compounds remains untapped^7–9.

The current availability of numerous prokaryotic genomic sequences has fueled genome mining as a more targeted approach to elucidate bacterial biosynthetic capacity, enabling the identification and exploitation of microbial BGCs and significantly enhancing the potential for discovery of novel natural products^10–13. As expected, the efficiency of this approach is reliant on the quality of the sequenced genome, underscoring the need for contiguous and accurate genome assemblies^14–17. Short-read sequencing technologies have displayed some limitations in generating complete microbial genome assemblies, due to their difficulty in resolving repetitive and GC-rich regions^18–20. Conversely, long-read sequencing technologies are reported to improve bacterial genome assembly continuity by enabling the reconstruction of complex regions and by bridging the gaps between contigs^21–24. Although long-read technologies exhibit low base accuracy compared with short-read technologies, this disadvantage can be compensated for by production of high-coverage sequencing data^25,26. Error-prone reads can also be resolved through the implementation of various read correction tools^27,28. Such tools polish the bacterial genome assembly by performing raw read self-correction, thereby reducing sequencing errors and significantly improving the quality of the final genomic assembly. Another strategy to ameliorate quality of the sequencing output combines the high accuracy of short-read data with the sequence continuity of long-read data, thus producing hybrid genomic assemblies^29–35.

Given the undeniably higher cost and complexity of combining different sequencing technologies to generate hybrid assemblies, efforts have focused on maximizing the capacity of long-read methods to generate low-complexity prokaryotic genome sequences as accurately as possible^{15,22,26,36–39}. As reported, recent advances in Oxford Nanopore flowcell R10.4 have enabled the generation of reliable, near-complete, bacterial genome assemblies of 40x coverage without the need for short-read sequencing^40,41. Furthermore, investigators have reported accurate identification of antimicrobial resistance and virulence traits of pathogens by utilizing Nanopore-only data^42,43.

In an effort to further explore bacterial biosynthetic capacity, this study presents sequencing of 82 new genomes from our in-house collection of isolates spanning a wide taxonomic range, particularly species of the Streptomyces genus. Since information about BGCs can be more efficiently extracted from long-read sequencing data^14,15,44, we employed Oxford Nanopore sequencing for this study and the presented de-novo assemblies were based solely on data from this technology. To evaluate the completeness and accuracy of our genomic assemblies, various computational tools were employed, while all bacterial genomes were finally computationally screened for their BGC content. The introduced dataset is anticipated to serve as a useful resource to scientists who wish to investigate the broad biosynthetic diversity among these isolates, identifying potentially robust producers of secondary metabolites, including antibiotics and other bioactive molecules. It is further anticipated that the generated genomic sequences will more broadly facilitate research into the biology of these particular microorganisms, available to the community as pure strains from our in-house collection.

Methods

Microbiological material

Bacterial isolates were collected and microbiologically characterized at the Department of Microbiology, ELTE, Budapest, Hungary. They represent the phyla of Pseudomonadota (proteobacteria), Actinomycetota (actinobacteria) and Bacillota (Firmicutes), as our dataset aimed to represent a wide taxonomic spectrum to enable comprehensive analyses across different bacterial lineages (Fig. 1). Strains with laboratory serial numbers preceded by the symbol “#” are available as frozen monocultures and have been described previously⁴⁵. Streptomyces strains with laboratory serial numbers preceded by the letter “M” were originally archived as lyophilized stocks in the early 1990s, prepared by late ELTE Professor István Mihály Szabó who isolated them from soil in Hungary. All strains are also maintained as frozen (−80 °C, 25% v/v glycerol) stocks in the corresponding author’s laboratory in Alexandroupolis. The strains are presented in Fig. 1 and excel files labelled “Strain Genomes File” and “Strain 16S rRNA File” available on Figshare⁴⁶.

Fig. 1.

Phylum-level distribution of sequenced bacterial strains. Each phylum is represented by a different bar color; concentric circles indicate 2 Mb intervals.

Starter liquid cultures were prepared from stocks by inoculating in 5 ml of Nutrient Broth No. 4 (NB, Sigma-Aldrich/Merck, Darmstadt, Germany) at 28 °C without shaking until bacterial growth was observed. Following two consecutive passages on the appropriate agar media (see “Strain Genomes File”)⁴⁶ to ensure purity of each strain, individual colonies were retrieved and used to inoculate 5 ml of fresh NB medium. Following incubation (28 °C, shaking at 180–200 rpm), bacterial cultures were ready for DNA extraction.

DNA extraction and quality assessment

High-molecular-weight genomic DNA was extracted using the MagAttract HMW DNA Kit (Qiagen, Hilden, Germany) following the manufacturer’s instructions for lysis of Gram-negative or Gram-positive bacteria. DNA size distribution and minimal obvious degradation were confirmed by gel electrophoresis (0.7% w/v agarose stained with ethidium bromide). DNA concentration and purity were initially assessed using Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). To enhance purity and deplete short DNA fragments, size-selection was carried out using 1:1 v/v solid-phase reversible immobilization (SPRI) magnetic beads (Mag-Bind TotalPure NGS by Omega Bio-tek, Norcross, GA, USA). Size-selected DNA preparations were then re-assessed on the Nanodrop and further quantified with a Qubit 3.0 fluorimeter, using the recommended Qubit dsDNA BR assay kit (both from Thermo Fisher Scientific). For quality of individual preparations, see “Strain Genomes File”⁴⁶.

Molecular phylogenetic classification

Molecular phylogenetic classification was already available for 37 bacterial isolates (those assigned serial numbers starting with #)⁴⁵, and those 16S rRNA gene sequences can be accessed through European Nucleotide Archive (ENA) Project ID: PRJEB47054⁴⁷.

Classification by 16S rRNA gene sequencing was also performed for an additional 49 bacterial isolates (see “Strain 16S rRNA File”)⁴⁶. PCR amplification with universal primers 27 F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 1492 R (5′-TACGGYTACCTTGTTACGACTT-3′)⁴⁸ was carried out as follows: each reaction in a final volume of 25 μl contained 1 U DreamTaq DNA Polymerase in 1x DreamTaq Buffer, 0.2 mM dNTP mixture, 10 μg of bovine serum albumin (all reagents from Thermo Fisher Scientific), 0.325 μM of each primer and 2 μl of genomic DNA. The reactions were carried out under the following conditions: 98 °C for 5 min, followed by 32 amplification cycles (94 °C for 30 s; 48 °C annealing temperature for 30 s; 72 °C for 60 s) and a final extension step at 72 °C for 10 min. Automated Sanger sequencing of purified amplification products was facilitated by LGC Genomics (Berlin, Germany). Sequences were assigned unique accession numbers and deposited to the ENA with Project ID: PRJEB71440⁴⁹ (see “Strain 16S rRNA File”)⁴⁶.

For taxonomic identification of bacterial isolates, the EzBioCloud database^50,51 was used to analyze all generated 16S rRNA gene sequences and determine the closest related species/strain that matched each one of our investigated isolates (see “Strain 16S rRNA File”)⁴⁶. The Contamination Estimator by 16S (ContEst16S) tool^52,53 was subsequently utilized to confirm purity of the strains and exclude those where 16S rRNA gene sequences indicated the presence of contamination. Assigned taxonomy identifiers (TAXIDs) are according to the Taxonomy database^54,55, while species names are according to the List of Prokaryotic Names with Standing in Nomenclature⁵⁶, as applicable at the time (Nov.–Dec. 2024).

Oxford Nanopore sequencing

Library preparation was performed using the Native Barcoding Kit 24 V14 (SQK-NBD114.24, Oxford Nanopore Technologies, Oxford, UK) according to the manufacturer’s recommended protocol. Briefly, for library preparation, DNA samples of similar concentration and purity were grouped together, so as to favor more balanced occupancy of the nanopores by DNA fragments during sequencing. Each run involved multiplexing of 24 barcoded DNA samples, as instructed by the manufacturer. For the clean-up step after adapter ligation, the Long Fragment Buffer was used to enrich for DNA fragments of 3 kb or longer. Libraries were loaded onto R.10.4 flowcells for sequencing using the MinION Mk1B device (Oxford Nanopore Technologies). Four separate sequencing experiments were performed, during which raw data acquisition and basecalling were carried out simultaneously with sequencing, using the MinKNOW software incorporating Guppy, enabling real-time data processing and demultiplexing. The first two sequencing experiments were controlled by MinKNOW version v23.04.6, while the software was updated to v23.07.12 for the third and fourth run.

Read processing

Each sequencing run generated 24 directories corresponding to each demultiplexed barcode, containing raw reads stored as multiple.fast5 files. To achieve the highest possible raw read accuracy, the Super accurate (SUP) model of Guppy software v6.5.7⁵⁷ (Oxford Nanopore Technologies) was utilized, employing the “dna_r10.4.1_e8.2_400bps_sup.cfg” configuration file. Quality control assessments were conducted using MinIONQC v1.4.1^58,59, FastQC v0.12.1^60,61, and Nanoplot v1.41.6^62,63, with default parameters, to evaluate read quality and detect any potential issues. Adapters were trimmed using Porechop v0.2.4⁶⁴ with default settings. The reads were filtered with Filtlong v0.2.1⁶⁵ to keep 90% of the longest and most accurate reads. Then, an additional round of quality control was performed using the same tools.

De novo assembly pipeline and evaluation

Assembly of all quality-checked Nanopore sequencing reads was performed using Flye v2.9.2^66,67 with the “–nano-raw” option and default parameters. The Flye assemblies were subsequently polished using Medaka v2.0.1⁶⁸ (Oxford Nanopore Technologies) with model “r1041_e82_400bps_sup_v5.0.0” and all QC-passed sequencing reads per barcode. Assembly quality assessment was conducted using Quast v5.2.0^69,70, with default parameters, to evaluate assembly quality metrics such as N50, L50, and GC content. Genome coverage statistics were evaluated using bamdst v1.0.9⁷¹. The completeness of genome assemblies was assessed using Benchmarking Universal Single-copy Orthologs (BUSCO) v5.8.1^72–74 with default parameters and the “–lineage_dataset” option enabled for order-specific analysis.

To streamline the analysis process and ensure reproducibility, all tools were integrated into a single Python script, except for bamdst and BUSCO. To avoid conflicts among different libraries and software dependencies, all software was containerized using Anaconda⁷⁵ virtual environments, creating isolated environments for each tool. This approach ensured robust analysis pipelines, allowing seamless integration and execution of the various bioinformatics tools used in this study.

Genome annotation and investigation of bacterial secondary metabolism

Genome assemblies were annotated using Prokka v1.14.6^76,77, employing a protein dataset for each sample derived from the most closely related reference proteome (obtained from UniProt and NCBI databases). Prokka annotation process was performed using default parameters along with the “–proteins” and “–addgenes” options.

The bacterial genome assemblies were computationally analyzed using antiSMASH v7.1.0^78,79 to assess the biosynthetic capacity of investigated bacterial strains via prediction of putative BGCs. The analyses were conducted enabling features ClusterBlast and Minimum Information about a Biosynthetic Gene cluster (MIBiG) cluster comparison.

Data Records

The whole genome sequences of 82 bacterial isolates were deposited in the ENA database with Project ID: PRJEB71438⁸⁰ (see “Strain Genomes File” for individual accession numbers per strain)⁴⁶. The 16S rRNA gene sequences for streptomyces strains reported in this study were also deposited in the ENA database with Project ID: PRJEB71440⁴⁹ (see “Strain 16S rRNA File” for individual accession numbers per strain)⁴⁶. Details of bacterial isolates and their molecular classification are provided in the “Strain Genomes File” and “Strain 16S rRNA File”⁴⁶. The results of antiSMASH analyses are presented in “Strain BGCs File”⁴⁶. All accompanying material associated with this manuscript is available on Figshare⁴⁶. All other data and results are included in the main manuscript.

Technical Validation

The reconstruction and functional characterization of bacterial genomic sequences is pivotal for understanding the bacterial biosynthetic machinery. Such information can be exploited to advance scientific progress, including pharmaceutical development. Since our primary motivation for this study was to explore bacterial BGCs involved in secondary metabolism, a subset of our in-house collection of bacterial isolates was subjected to whole-genome sequencing analysis using Nanopore technology. The collection displays diversity in the phylogeny of selected isolates, but we prioritized the sequencing of Streptomyces species considering their enhanced biosynthetic capabilities. Initially, the sequenced subset included a total of 96 bacterial isolates from our collection, however, 14 of those were abandoned (and are not reported here), either because the collected data was insufficient, or because the analysis of 16S rRNA gene sequences with the ContEst16S tool on EzBioCloud demonstrated that the sequencing output contained genomic segments of different bacterial origin, indicating contamination of the original culture. This highlights the importance of incorporating a contamination check in the sequencing workflow, ensuring the purity of starting cultures and DNA samples. The 82 genomic assemblies that passed those initial tests are reported here and encompass 50 Actinomycetota, 24 Pseudomonadota and 8 Bacillota (Fig. 1 and “Strain Genomes File”⁴⁶).

We performed de-novo assembly of bacterial genomes, facilitated by the use of Nanopore long-read sequencing technology, as this circumvents some limitations of reference-based approaches. Specifically, reference-based methods may exclude sequence reads from the strain under study, if these do not align with the available reference genome, potentially overlooking genomic regions important for capturing microbial diversity²². The assemblies were generated using Flye, a high-performance assembler for Nanopore long-read data. While we did not perform an exhaustive benchmark of all available assembly tools, our pipeline followed the established best practices for the technology platform at the time of the study. Moreover, depositing of the raw reads to the ENA database ensures that, as bioinformatics tools and assembly algorithms continue to evolve, the community will be able to perform more sophisticated re-analyses or hybrid assemblies to further improve these results in the future.

The output of the sequencing process (Tables 1, 2; Figs. 2, 3) was subjected to quality assessment per DNA sample, examining parameters such as the number of contigs, N50, Q-score and genome coverage for all sequencing runs. Regarding the number of contigs, 42.6% of the bacterial genome assemblies contain less than 20 contigs, while approximately 24% were assembled into 1 to 3 contigs. A greater number of contigs was observed for assemblies generated from the first and second sequencing runs (Table 1). This was caused by multiple interruptions of the sequencing device during prolonged (up to 36 h) continuous usage, an inherent flaw that MinKNOW v23.04.6 software reports attributed to device errors, such as failed script or data transfer from the device, even though hardware check was successfully passed during experiments. Those issues were resolved with the updated version v23.07.12 of the MinKNOW software that was used during the third and fourth sequencing runs (Table 2).

Table 1.

Quality assessment metrics of 42 barcoded de novo genome assemblies generated from sequencing runs 1 and 2. As discussed in the Technical Validation section, assemblies generated from sequencing runs 1 and 2 were more fragmented, due to technical issues of MinKNOW v23.04.6 which were resolved after updating to version v23.07.12 of the software. Lower sequencing coverage was also attributed to inconsistent performance of the MinION and pore decline in reused flowcells.

Barcoded samples	Species name	Assembly length (bp)	Number of contigs	GC (%)	N50 (bp)	Average depth (x)
1_nb01	Bosea thiooxidans	6,349,271	256	67.37	120,849	24.65
1_nb03	Lysinibacillus sphaericus	4,935,994	140	37.55	227,665	19.14
1_nb05	Bacillus licheniformis	4,364,058	33	45.84	1,326,042	48.26
1_nb06	Ureibacillus acetophenoni	6,229,690	195	34.99	62,021	43.67
1_nb07	Alishewanella longhuensis	8,329,063	343	47.22	42,005	21.32
1_nb09	Castellaniella denitrificans	4,565,760	148	67.84	96,483	68.25
1_nb10	Castellaniella defragrans	6,436,768	249	63.73	61,759	34.17
1_nb11	Oligella ureolytica	3,710,395	137	43.70	63,017	9.94
1_nb12	Stutzerimonas stutzeri	6,609,220	256	64.11	67,114	14.68
1_nb13	Rhodanobacter lindaniclasticus	7,309,924	304	67.16	46,451	18.15
1_nb14	Streptomyces griseus	8,377,519	118	72.30	1,533,214	28.94
1_nb16	Streptomyces tendae	9,986,999	387	72.10	120,464	18.63
1_nb17	Streptomyces avidinii	8,824,473	260	71.80	282,487	18.47
1_nb18	Streptomyces canarius	11,234,897	396	71.92	142,725	16.67
1_nb19	Streptomyces griseus	8,381,104	133	71.71	527,544	37.80
1_nb20	Streptomyces alboniger	10,458,132	443	71.76	74,823	13.48
1_nb21	Streptomyces chartreusis	10,572,294	45	70.79	2,955,546	45.64
1_nb22	Streptomyces xanthochromogenes	8,355,972	65	71.28	2,428,440	46.65
1_nb23	Streptomyces narbonensis	8,940,517	338	72.06	143,075	23.00
1_nb24	Streptomyces spiroverticillatus	9,516,703	98	71.20	544,295	40.23
2_nb01	Lysinibacillus sphaericus	7,961,604	396	37.25	36,231	11.57
2_nb02	Bacillus safensis	6,058,678	321	41.83	36,550	17.97
2_nb04	Streptomyces microflavus	13,075,107	672	71.23	50,866	13.67
2_nb05	Streptomyces aurantiogriseus	10,721,168	232	70.93	2,089,405	42.30
2_nb06	Streptomyces badius	11,908,868	585	71.65	43,706	13.03
2_nb07	Streptomyces californicus	9,863,829	432	72.65	43,000	6.99
2_nb08	Streptomyces cyaneofuscatus	8,933,790	480	71.53	60,290	15.37
2_nb09	Streptomyces violarus	10,285,877	633	71.18	42,177	9.19
2_nb10	Streptomyces glaucescens	9,965,271	557	72.27	44,522	8.11
2_nb11	Streptomyces griseomycini	10,083,991	509	72.65	99,455	17.31
2_nb12	Streptomyces hirsutus	12,515,227	545	71.13	49,492	8.52
2_nb13	Streptomyces antimycoticus	10,408,690	408	70.95	56,074	6.83
2_nb14	Streptomyces shaanxiensis	10,208,071	363	71.12	213,402	26.39
2_nb15	Streptomyces lateritius	8,855,071	420	71.33	39,910	7.23
2_nb16	Streptomyces rochei	9,912,515	573	72.56	64,306	13.57
2_nb17	Streptomyces griseoloalbus	10,807,963	569	72.59	41,114	12.55
2_nb18	Streptomyces tendae	11,908,553	681	72.47	42,905	13.14
2_nb19	Streptomyces fumanus	11,427,027	562	73.01	38,469	9.16
2_nb20	Streptomyces microflavus	11,451,493	448	71.28	43,483	7.62
2_nb22	Streptomyces alboniger	12,178,729	638	70.12	30,785	7.84
2_nb23	Streptomyces tendae	8,847,876	335	72.04	45,050	7.22
2_nb24	Streptomyces violaceorubidus	10,097,966	441	72.34	39,630	7.98

Table 2.

Quality assessment metrics of 40 barcoded de novo genome assemblies generated from sequencing runs 3 and 4. Samples generating genome assemblies with <20 contigs and/or >30x coverage are shown in bold.

Barcoded samples	Species name	Assembly length (bp)	Number of contigs	GC (%)	N50 (bp)	Average depth (x)
3_nb01	Labrys miyagiensis	6,451,965	1	63.68	6,451,965	61.31
3_nb02	Mesorhizobium amorphae	6,475,598	1	62.07	6,475,598	82.44
3_nb03	Bosea minatitlanensis	4,732,811	1	68.57	4,732,811	102.87
3_nb05	Sphingobium scionense	6,142,515	9	64.00	4,819,090	53.97
3_nb06	Rhizobacter fulvus	5,725,747	1	68.59	5,725,747	62.13
3_nb07	Pseudomonas fragi	5,054,805	12	59.32	2,861,027	48.50
3_nb08	Bacillus nitroreducens	4,617,405	4	37.45	3,812,000	109.94
3_nb09	Rhizobium arsenicireducens	4,439,158	3	62.15	4,285,843	167.69
3_nb10	Frigidibacter albus	4,556,185	6	66.91	4,350,424	36.23
3_nb11	Thauera mechernichensis	4,805,311	5	65.21	4,283,565	88.78
3_nb12	Afipia felis	4,130,270	4	62.84	3,880,815	118.52
3_nb13	Aquamicrobium terrae	5,911,167	10	65.23	2,273,974	78.54
3_nb14	Eoetvoesia caeni	4,953,590	2	58.42	4,715,377	30.79
3_nb15	Castellaniella caeni	3,389,029	1	64.70	3,389,029	181.01
3_nb17	Streptomyces atroolivaceus	8,170,221	2	70.70	8,046,083	37.09
3_nb18	Streptomyces rochei	7,700,368	17	72.57	5,336,130	21.76
3_nb19	Streptomyces pilosus	7,567,575	2	72.44	7,309,760	47.28
3_nb20	Streptomyces diastaticus	7,070,005	4	73.38	6,616,679	64.87
3_nb21	Streptomyces canus	9,876,360	2	70.30	8,721,601	31.71
3_nb22	Streptomyces parvulus	7,765,051	2	72.74	7,149,483	18.74
3_nb23	Streptomyces tanashiensis	8,510,880	2	72.34	7,250,060	35.36
3_nb24	Streptomyces avidinii	8,455,439	2	72.31	8,451,836	57.30
4_nb01	Streptomyces diastaticus	6,729,465	8	73.51	4,488,046	14.16
4_nb02	Streptomyces abikoensis	6,029,806	5	72.72	5,371,170	97.29
4_nb04	Streptomyces griseorubiginosus	9,693,545	74	70.89	311,467	12.14
4_nb05	Streptomyces purpurascens	8,834,081	162	71.06	88,743	6.86
4_nb06	Kitasatospora aureofaciens	9,623,598	66	72.52	548,223	15.72
4_nb07	Streptomyces viridochromogenes	7,910,144	6	72.02	7,030,946	30.52
4_nb08	Streptomyces lavendulae	8,858,801	20	72.66	1,269,958	23.57
4_nb09	Streptomyces rimosus	7,436,011	140	71.84	85,432	5.20
4_nb11	Leifsonia shinshuensis	4,389,968	3	71.05	4,235,312	14.01
4_nb13	Corynebacterium variabile	3,207,583	19	67.33	240,792	9.87
4_nb14	Tsukamurella paurometabola	4,510,532	27	68.46	448,140	9.83
4_nb15	Cellulomonas persica	3,677,918	1	73.54	3,677,918	31.99
4_nb16	Bacillus tequilensis	4,149,298	2	43.76	4,145,740	328.68
4_nb17	Brevibacillus borstelensis	5,366,844	1	51.69	5,366,844	79.49
4_nb18	Thauera mechernichensis	4,599,847	2	65.55	4,426,618	66.74
4_nb19	Alishewanella longhuensis	4,147,163	2	47.17	4,143,592	42.72
4_nb20	Alcaligenes phenolicus	4,568,329	5	56.87	4,414,899	90.76
4_nb21	Ottowia pentelensis	4,007,106	1	68.43	4,007,106	142.74

Fig. 2.

Distribution of mean read quality across 82 bacterial genome assemblies. Each bar is divided into three segments representing the proportion of reads that satisfy the Q15 (blue), Q12 (green), or Q10 (coral) quality threshold; colored dots next to the barcodes indicate phyla (as in Fig. 1).

Fig. 3.

Evaluation of genome assembly completeness by BUSCO analysis. Assemblies were analyzed employing the closest available lineage for each species. Panels a-d represent barcoded samples across four sequencing runs; colored dots next to the barcodes indicate phyla (as in Fig. 1).

Similar observations, i.e. regarding the continuity of assemblies, were made when evaluating the N50 metric. Specifically, the third sequencing run yields an overall N50 value of 2 to 8 Mbp, which (considering the small size of bacterial genomes) is indicative of highly contiguous assemblies. A greater diversity in N50 values is observed for the fourth run, with the majority of barcodes hovering at around 4 Mbp (Table 2). In contrast, substantially lower N50 values are reported for the first two runs. Specifically, in the first sequencing run, N50 values fall within the range of hundreds of thousands base pairs, suggesting a minor yet not concerning decline in assembly quality. However, in the case of the second run, most N50 values range around 50 Kbp, indicating a significant drop in assembly continuity (Table 1). Such limitations of the technology are important to keep in mind, for example, when interpreting experiments conducted with older MinKNOW versions.

To assess basecalling accuracy, Q-score metrics were examined (Fig. 2). Approximately 90% of the total reads of each barcode had a quality score greater than 15, signifying an accuracy of over 96.84%. A rather lower base accuracy relative to short-read sequencing was anticipated, considering that Nanopore is an error-prone technology and no short-read polishing was applied.

Another important quality metric is the average sequencing depth, which could mitigate the low-per-base accuracy that is an inherent limitation of Nanopore sequencing. In this study, around 46% of the genome assemblies have a coverage exceeding 30x, with 7 assemblies surpassing 100x coverage. No correlation between the sequencing coverage and the length or GC content of genome assemblies was observed (Tables 1, 2).

The completeness of the bacterial sequences was evaluated with BUSCO analysis, using the corresponding order-level dataset for every species (Fig. 3). For non-model genomes, the accepted BUSCO score ranges from 50% to 95% completeness, contingent upon the biological complexity of each species. In the current study, 63 assembled genomes exceeded the threshold of 90% completeness, with the highest levels recorded during the third sequencing run. No sample ranked below 50% completeness. On the other hand, the relatively higher rate of missing, fragmented, or complete and duplicated BUSCO genes in the first two runs, can serve as an indicator of issues during the sequencing or assembly processes⁷⁴, discussed above.

Through the Prokka whole genome annotation process, it was possible to determine the number of coding (CDS), rRNA and tRNA genomic sequences per assembly (Table 3). Out of 82 genome assemblies, annotation was successfully applied to 79, as a reference genome, proteome or both were not available for the remaining 3 samples. The 79 annotated genomes contain an average of 7,417 CDS, 16 rRNA, and 87 tRNA sequences. A surge in the number of predicted genes observed in the second sequencing run could be due to possible overestimation, considering the high fragmentation of the genome assembly leading to the scattering of genes across multiple contigs⁸¹.

Table 3.

Number of coding (CDS), rRNA, and tRNA sequences per barcoded bacterial assembly, as predicted by Prokka genome annotation.

Barcoded sample	CDS	rRNA	tRNA	Barcoded sample	CDS	rRNA	tRNA
1_nb01	6,264	6	62	2_nb24	10,505	16	109
1_nb03	5,410	27	101	3_nb01	5,951	6	50
1_nb05	4,536	15	80	3_nb02	6,227	6	48
1_nb06	6,729	64	125	3_nb03	4,509	6	58
1_nb07	8,671	29	151	3_nb05	5,932	12	63
1_nb09	4,168	6	64	3_nb06	5,290	3	53
1_nb10	6,316	10	71	3_nb07	4,563	4	69
1_nb11	4,591	7	54	3_nb10	4,430	6	53
1_nb12	7,008	15	92	3_nb11	4,381	12	64
1_nb13	7,440	10	75	3_nb12	3,910	3	51
1_nb14	7,262	18	78	3_nb13	5,756	3	50
1_nb16	9,512	21	95	3_nb14	4,669	6	49
1_nb17	8,205	19	88	3_nb15	3,125	9	56
1_nb18	10,220	10	99	3_nb17	7,051	18	75
1_nb19	7,412	23	101	3_nb18	6,771	18	89
1_nb20	10,079	17	100	3_nb19	6,660	18	79
1_nb21	9,238	16	86	3_nb20	5,926	21	95
1_nb22	7,356	25	82	3_nb21	8,853	18	83
1_nb23	8,268	26	93	3_nb22	6,867	18	78
1_nb24	8,443	24	91	3_nb23	7,497	21	85
2_nb01	10,138	35	165	3_nb24	7,369	21	93
2_nb02	7,075	38	115	4_nb01	5,770	21	93
2_nb04	13,149	22	138	4_nb02	5,277	21	85
2_nb05	9,833	17	91	4_nb04	8,658	15	82
2_nb06	12,193	20	158	4_nb05	8,447	6	82
2_nb07	10,240	20	102	4_nb06	8,415	21	97
2_nb08	8,586	13	88	4_nb07	7,029	18	79
2_nb09	10,447	15	108	4_nb08	8,012	21	95
2_nb10	10,580	15	96	4_nb09	7,319	14	80
2_nb11	9,672	14	105	4_nb11	4,190	3	54
2_nb12	13,273	20	106	4_nb13	3,029	3	60
2_nb13	10,960	16	91	4_nb14	4,416	3	54
2_nb15	9,651	20	91	4_nb15	3,267	6	52
2_nb16	9,437	23	89	4_nb16	4,133	29	86
2_nb17	10,780	15	107	4_nb17	5,168	32	126
2_nb18	11,844	19	142	4_nb18	4,115	12	62
2_nb19	11,352	35	146	4_nb19	3,600	15	75
2_nb20	12,100	30	121	4_nb20	4,228	9	57
2_nb22	12,986	12	102	4_nb21	3,817	6	48
2_nb23	9,418	10	89

Overviewing the quality assessment metrics mentioned above, several factors should be taken into account when troubleshooting Nanopore sequencing which remains a less reliable technology compared with more established short-read platforms. Firstly, sequencing performance is largely dependent on the quantity and quality of the starting genetic material used⁸². Given that the four sequencing runs differed in the concentration and purity of the bacterial DNA preparations pooled together in the same library, variations in the sequencing throughput were anticipated across runs. Specifically, the first and third runs were performed using DNA of highest quality in terms of quantity and purity, while the second and fourth runs were carried out with DNA preparations of less optimal quality (see “Strain Genomes File”)⁴⁶. It must be stressed, however, that variability in DNA preparation quality was largely due to inherent biological properties of different bacterial strains, affecting the concentration and purity of recovered DNA, rather than its integrity which remained high (see “Strain Genomes File”)⁴⁶. For example, gamma-proteobacteria typically reach high cell densities quickly, allowing for higher prep concentrations compared to alpha-proteobacteria that grow more slowly and to lower densities; Gram-negative cells are typically easier to lyse compared with Gram-positive cells; and the streptomycetes are Gram-positive bacteria, typically growing in tight clumps that are difficult to disperse during lysis.

Secondly, it should be noted that, in line with the manufacturer’s specifications, two flowcells were used, each one for two 36-hour runs. Consequently, it is reasonable to expect a decline in sequencing throughput, as available flowcell pores progressively deteriorated during the second and fourth runs. Indeed, lower sequencing coverage was more likely linked to inconsistent performance of the MinION and pore decline in reused flowcells, although the low concentration of DNA preparations in runs 2 and 4 also contributed. Lastly, as already discussed, it is important to keep in mind that the output of Nanopore sequencing is still heavily reliant on the software performance and its stable connectivity to the sequencing device, factors that may be beyond user’s ability to fully control.

To gain insight into the biosynthetic potential of the 82 sequenced bacterial isolates, their genomic assemblies were subjected to antiSMASH analysis, so as to predict the presence of BGCs organized in so-called “Regions” defined by the antiSMASH expert community (Fig. 4 and “Strain BGCs File”⁴⁶). Moderate correlation was observed between the number of contigs per assembly and the corresponding number of antiSMASH predicted Regions (Fig. 5), supporting that, even though assembly fragmentation may influence BGC prediction to some extent, antiSMASH annotations can be considered as reasonably reliable with regard to possible BGC fragmentation during genome assembly. Indeed, upon visual inspection of antiSMASH results for possible BGC disruption, only a limited number of clusters were found to localize at the edges of contigs. An increase in BGC fragmentation was nevertheless detected in the second sequencing run, and this can be attributed to the high degree of contig fragmentation in the corresponding genome assemblies, as discussed above.

Fig. 4.

BGC type distribution per barcoded bacterial sample relative to the total number of predicted BGC Regions. Each BGC type is color-coded and the following abbreviations were used: NRPS for non-ribosomal peptide synthetase, PKS for polyketide synthase and RiPP for ribosomally synthesised and post-translationally modified peptides. Colored dots next to the barcodes indicate phyla (as in Fig. 1).

Fig. 5.

Correlation between the number of predicted BGC Regions and the number of contigs per bacterial genome assembly (y = 0.0522x + 14.67, R² = 0.4186).

As expected, the BGC content varies according to the taxonomic classification of each bacterial isolate. In line with existing literature, isolates in the Streptomyces genus exhibited the greatest number of BGCs compared to all other bacteria, harboring 21–68 BGCs per genome which are predicted to produce a whole range of secondary metabolites belonging to diverse chemical classes. Search of the MIBiG database further demonstrates the potential medicinal value of the predicted secondary metabolites, many of which are likely to have antimicrobial or antitumor activity. Identifying specific isolates as proficient producers of secondary metabolites through genome mining of putative BGCs may advance their targeted investigation and biotechnological exploitation towards the discovery of new compounds with unprecedented functional and medicinal properties.

Furthermore, the generated dataset is anticipated to more broadly support investigations into the biology of microbes in our collection, advancing academic collaboration and scientific progress. To this end, even suboptimal assemblies remain valuable as genomic maps for the design of “wet-lab” research involving unique non-model strains in our collection. For example, our group is currently using this genomic dataset to guide whole BGC cloning and heterologous expression of antibiotic secondary metabolites in genetically engineered hosts. We believe that providing genomic assemblies alongside transparent quality measures – and making raw reads available for re-analyses as computational tools evolve – is balancing bioinformatics rigor with “wet-lab” utility, encouraging future reuse of these data associated with our collection of strains.

Author contributions

D.B.: Microbiological work, Nanopore sequencing (wet-lab), BGC analyses; S.Z.: 16S rRNA gene sequencing, bioinformatics (MinION operation and advanced genomic analyses); A.K. & D.T.: bioinformatics (MinION operation, software pipeline and initial genomic analyses); G.F.: Resources, co-supervision (D.B.) and expertise; T.F.: Resources, co-supervision (S.Z.) and expertise; K.M.: Resources and expertise; S.B.: Conceptualization, team supervision, resources, investigation, project administration, funding acquisition. D.B., S.Z., A.K., D.T. and S.B. wrote the manuscript which was reviewed and approved by all authors.

Data availability

ENA Project ID: PRJEB71438⁸⁰ includes the whole-genome sequences of investigated strains. ENA Project ID: PRJEB71440⁴⁹ includes the 16S rRNA gene sequences of Streptomyces strains reported in this study. Details of bacterial isolates and their molecular classification, as well as the results of antiSMASH analyses are available on Figshare⁴⁶. All other data is presented in the main manuscript.

Code availability

The complete bioinformatic pipeline, including steps for data acquisition, software requirements, and the final Python code, is available through Figshare and GitHub^83,84.

Competing interests

The authors declare no competing interests.