Perspectives and Benefits of High-Throughput Long-Read Sequencing in Microbial Ecology

ABSTRACT

Short-read, high-throughput sequencing (HTS) methods have yielded numerous important insights into microbial ecology and function. Yet, in many instances short-read HTS techniques are suboptimal, for example, by providing insufficient phylogenetic resolution or low integrity of assembled genomes. Single-molecule and synthetic long-read (SLR) HTS methods have successfully ameliorated these limitations. In addition, nanopore sequencing has generated a number of unique analysis opportunities, such as rapid molecular diagnostics and direct RNA sequencing, and both Pacific Biosciences (PacBio) and nanopore sequencing support detection of epigenetic modifications. Although initially suffering from relatively low sequence quality, recent advances have greatly improved the accuracy of long-read sequencing technologies. In spite of great technological progress in recent years, the long-read HTS methods (PacBio and nanopore sequencing) are still relatively costly, require large amounts of high-quality starting material, and commonly need specific solutions in various analysis steps. Despite these challenges, long-read sequencing technologies offer high-quality, cutting-edge alternatives for testing hypotheses about microbiome structure and functioning as well as assembly of eukaryote genomes from complex environmental DNA samples.

INTRODUCTION

Modern microbial ecology research takes advantage of techniques that determine the molecular structure of nucleic acids. Because of their relative ease of use and lack of culturing requirements, high-throughput sequencing (HTS) of genomes, marker genes, and metagenomes has largely replaced microarrays and fingerprinting techniques, becoming the method of choice for understanding patterns in microbiome diversity, population structure, and evolution (1–3). HTS-based marker gene sequencing (i.e., metabarcoding) enables in-depth characterization of the community composition of both prokaryotic (bacteria and archaea) and eukaryotic (protists, fungi, and microfauna) microorganisms. Reconstruction of prokaryote genomes from complex environmental samples, such as soil, water, and sediments, using deep sequencing has greatly benefited our understanding about the potential function of so-far uncultivable microbial species and much expanded the tree of life (4, 5). Although the second-generation HTS methods have shed light on fundamental aspects of microbial ecology and function, they suffer from issues associated with short read lengths (up to 550 bp). This has forced a step back in biodiversity research that traditionally uses molecular marker genes exceeding 550 bp in length that provide greater taxonomic resolution (2, 6, 7). Furthermore, short read lengths limit the number of available target-specific primer binding sites that can be utilized when designing a marker gene amplicon. In genome and metagenome analyses, short reads cannot reliably reconstruct long repeats because of uncertainties in mapping reads; although protocols for mate-pair libraries are available, they have rarely been used.

To overcome these shortfalls, synthetic long-read sequencing and third-generation HTS methods have become increasingly used in microbiology (8, 9). The Pacific Biosciences (PacBio) and Oxford Nanopore Technologies (ONT) sequencing platforms both generate 30- to 50-kb reads on average but suffer from low raw read accuracy and higher per-read costs compared to those of short-read platforms (Table 1). Here, we provide an overview of the cutting-edge developments in laboratory and bioinformatics methods for long-read analysis from the microbial ecology perspective. We illustrate situations in which long reads have provided unique answers to research questions related to microbial identification, community and population ecology, and interspecific interactions. Finally, we outline the current limitations of long-read sequencing and evaluate their future prospects in microbiology research.

TABLE 1.

Comparison of recent long-read and short-read sequencing technologies^a

Sequencing technology (specifications)	Instrument cost	Library prep cost	Cost per lane/cell	Cost per million reads	Cost per billion bases	Runtime (h)	Throughput (k reads at maximum runtime)	Avg read length (maximum)	Read accuracy (%)	Error profile
PacBio Sequel I (maximum 8 cells)	300,000	150–400	1,000	2,000	100	10–20	500	20,000 (50,000)	85–88/99.9 (CCS)	Homopolymers
PacBio Sequel II (maximum 16 cells)	650,000	75–400	2,000	500	17	10–30	4,000	30,000 (100,000)	88–90/99.9 (CCS)	Homopolymers
ONT Flongle	1,300	40–90	80	400	13	0.1–12	200	30,000 (60,000)	96–99	Homopolymers
ONT MinION	900	90–130	600	600	12	0.1–48	1,000	50,000 (2.3M)	96–99	Homopolymers
ONT GridION (maximum 5 FCs)	45,000	90–130	600	600	12	48	1,000	50,000 (2.3M)	96–99	Homopolymers
ONT PromethION (maximum 48 FCs)	176,000	90–600	1,400	250	8	72	6,000	30,000 (330,000)	96–99	Homopolymers
SLR (example of NovaSeq 2 × 150 PE, 30× coverage, for 5 kb)	900,000	30–50	5,000	1,250	140	44	4,000	9,000 (12,000)	100	Unequal coverage
Illumina NovaSeq S4 (2 × 150 PE; maximum 4 lanes)	900,000	50–100	5,000	2.5	9	44	2,000,000	250 (290)	99.9	Low-quality ends
Illumina MiSeq (2 × 300 PE; maximum 2 lanes)	100,000	50–100	2,000	100	175	56	20,000	550 (590)	99.9	Low-quality ends
ION G5-S5 Prime, P550 chip (maximum 2 chips)	180,000	50	700	5	25	6.5	130,000	200 (250)	99.0–99.5	Homopolymers
ION G5-S5, P530 chip 600 SE	60,000	50	500	40	70	7	12,000	570 (650)	99.3–99.7	Homopolymers, low-quality ends
MGI Tech DNBSEQ-T7 (2 × 150 PE; maximum 4 cells)	990,000	50–100	6,000	1.2	4.5	24	5,000,000	250 (290)	99.9	Low-quality ends
MGI Tech DNBSEQ-G400RS (2 × 200 PE; 400 SE maximum 2 cells × 4 lanes)	480,000	50–100	1,800	4	11	108	450,000	350 (400)	99.9	Low-quality ends

^aCosts are given in EUR. Information is compiled from multiple recent literature sources and price requests from sequencing companies. FC, flow cell; PE, paired-end; SE, single-end.

LONG-READ SEQUENCING METHODS

PacBio sequencing.

PacBio was the first third-generation, long-read HTS technology to enter the market. This method is based on recording fluorescently labeled deoxyribonucleoside triphosphates (dNTPs) during complementary strand synthesis (Fig. 1). In a decade, PacBio raw error rates have declined from 25% to 10–12%, whereas the average sequence length of 2 kb has increased to 30 kb (early 2021). Over the past decade, sequencing throughput has increased from around 50,000 reads (RS I model) to 4 million reads per single-molecule real-time sequencing (SMRT) cell (Sequel II model). PacBio has also developed a much higher accuracy circular consensus sequence (CCS) approach, branded by the company as HiFi sequencing. In CCS sequencing, hairpin adaptors are ligated to the template DNA molecules, permitting individual molecules to be sequenced multiple times in a circular manner (10) (Fig. 1). The multiple “subreads” are then assembled into CCS reads with accuracies of >99.9% at >10-fold consensus.

FIG 1.

Comparison of workflows of the third-generation sequencing platforms for DNA and RNA sequencing. (Left) Pacific BioSciences PacBio SMRT sequencing; (right) Oxford Nanopore Technologies nanopore sequencing.

PacBio CCS sequencing is proven to be efficient for high-resolution marker gene profiling of bacterial and fungal communities. CCS sequencing with the earlier RS II model could produce full-length 16S rRNA gene sequences (1.5 kb) at >99% accuracy that improved species-level identification in bacteria (7, 11–14). In fungi, marker gene sequencing of the 5′ part of the 28S rRNA gene (650 bp) (15) and the full-length rRNA internal transcribed spacer (ITS) region (500 to 700 bp) (16) was the first step toward incorporating long-read HTS into fungal ecology research. Long reads spanning both the variable ITS region and more conserved rRNA genes further added to species-level taxonomic identification in both bacteria (17–20) and fungi (21–24). At present, long PacBio CCS sequences can distinguish bacterial genotypes and rRNA gene copies that differ by a single nucleotide (18, 19, 25, 26).

PacBio has been used for de novo genome assembly in viruses (27), bacteria (28, 29), fungi (30), plants (31), and animals (32). Initially, the best results were obtained when PacBio reads were coassembled with short Illumina reads (33). Since 2019, large-scale genome sequencing initiatives such as the 1000 Fungal Genomes Project use only PacBio sequencing (34). Furthermore, genomes as complex as a human’s no longer require complementary short-read sequencing (9, 35).

PacBio sequencing can be used for the analysis of transcriptomes and metagenomes. PacBio transcriptome sequences are sufficiently accurate and long to distinguish both single-nucleotide mutations and alternatively spliced isoforms (36, 37). Using a metagenomics approach, PacBio sequencing was effectively used for identification of bacterial species in mock communities (38) and complex environments (39) as well as assays of functional genes and the construction of high-quality metagenome-assembled genomes (MAGs) (40). Although early PacBio sequencing enabled direct RNA sequencing by using reverse transcription recording (41), later protocols use cDNA.

Nanopore sequencing.

ONT provides the best known and, so far, the only commercial nanopore-based sequencing method. This technique is based on recording signal disruptions of nucleotide oligomers when DNA molecules are translocated through nanopores (42) (Fig. 1). The mobile phone-sized MinION was the first instrument released in 2014 through an early access program that boosted improvement of laboratory and bioinformatics protocols by the research community. The early R7 series flow cells had an average read length of 4.7 to 6.7 kb (maximum, 40 to 300 kb) (43, 44) and 72 to 74% read accuracy. Currently, modal raw accuracy and sequence length reach 95 to 98% (2021 early access chemistry with 99% accuracy) and 50 kb (maximum, 2.3 Mb), respectively. Accuracy of individual reads can be further improved by various methods during sample preparation, sequencing, and bioinformatics. The outdated methods based on rolling circle amplification—INC-seq (45), NanoAmpli-seq (46), and R2C2 (47)—improve accuracy of ONT reads by up to 97 to 98%. Integration of unique molecular identifiers (UMIs) (see specific subsection below) has enabled researches to increase nanopore sequence accuracy to above 99.99% (26).

The first complete genomes of bacteria (44, 48–50) and viruses (51, 52) constructed from nanopore sequences were published in the mid-2010s. As in PacBio, these resolved many repeat-rich regions, but suffered from indels and hence frame shifts in genes (53). Therefore, ONT has been combined with short-read techniques to generate high-quality complete genomes (54). Other studies reported accurate bacterial species-level identification from mock communities and environmental samples (55, 56). In addition to cDNA-based (meta)transcriptomics (57), ONT enables direct RNA sequencing (58, 59). Furthermore, ONT instruments allow detection of base modifications during sequencing of DNA (60) and RNA (58).

MinION was followed by the benchtop GridION and PromethION models that increased throughput. In 2019, miniature Flongle flow cells were introduced to reduce costs of single-sample analyses. The main benefits of MinION and Flongle include small size, rapid analysis process, and low power requirements that enable analyses in field camps (61) and other nonlaboratory conditions (52, 62, 63).

Synthetic long-read sequencing.

Synthetic long reads (SLRs) represent long DNA molecules that are assembled from multiple short reads based on UMI barcodes located within the reads that link them to the DNA molecule from which they originated (Fig. 2). SLR protocols typically include indexing or capturing individual DNA molecules, a few rounds of amplification with UMI-tagged primers, enzymatic or sonication-based fragmentation, short-read HTS (Illumina, MGI Tech, or Ion Torrent), and bioinformatics-based reconstruction of long DNA molecules. SLR length and quality is limited by polymerase processivity (around 10 to 12 kb) and securing even coverage of UMIs and reads across long molecules. Compared with long-read HTS platforms, SLRs exhibit low capacity to resolve tandem repeats longer than 500 bases.

FIG 2.

Workflow of producing synthetic long reads (SLRs) using unique molecular identifiers (UMIs) (A) and the following three alternative short-read sequencing platforms: MGI Tech (B), Illumina (C), and Ion Torrent (D). The SLR preparation follows the LOOPseq protocol (74).

SLRs (1 to 3 kb) were initially developed to facilitate genome assembly of humans (64) and viruses (65). SLRs enabled reconstruction of full-length transcripts (66) and 16S rRNA genes from metagenomes (67) and amplicons (68) that exceeded PacBio and ONT sequencing in terms of accuracy, assembly length, and cost (69, 70). Moleculo, Inc. and 10× Genomics, Inc. commercialized methods to separate several up to 12-kb DNA molecules into 384-well plates and microfluidics (100,000 wells) chips, respectively, for compartmentalized sample preparation and Illumina sequencing (71, 72). These methods became limited by the number of wells and the difficulty of distinguishing homologous DNA molecules included in the same compartment. Recently, single-molecule SLR methods have been developed (73, 74). Loop Genomics, Inc. released the LoopSeq method to tag millions of individual molecules with UMI-indexed primers that are unique to each DNA molecule (74). LoopSeq amplicons of the full-length 16S rRNA gene (1.5 kb) had 10-fold higher accuracy compared to PacBio CCS reads and Illumina short reads in bacterial mock communities (75, 76) and maintained >99.99% accuracy (75). The UMI tagging used in the construction of SLRs also facilitates identification and removal of chimeric molecules produced during the PCR (68).

APPLICATIONS OF LONG-READ SEQUENCING METHODS

Identification of specimens and cultures.

Molecular identification of species relies on a suitable genetic marker (i.e., DNA barcode). These marker genes are chosen based on ease of amplification, taxonomic resolution, and the presence of a reference database. Longer marker genes contain more taxonomic and phylogenetic information but can be more difficult to amplify and sequence than shorter fragments (77). In spite of relatively low species-level resolution, the 16S rRNA gene is by far the most broadly used taxonomic marker (Fig. 3A). In fungi, the rRNA ITS region is the most commonly used amplicon for taxonomic classification (Fig. 3B), but 18S and 28S rRNA are alternatively used for several fungal phyla (2). In plants and photosynthetic protists, the ITS region, plastid genes, and their introns are all popular species-level identification markers, whereas various heterotrophic protist and animal groups rely on 18S or 28S rRNA gene, the ITS region, mitochondrial cytochrome c oxidase 1 (COI), or other mitochondrial genes (6). Because rRNA genes usually occur in operons, it is feasible to analyze genetic information from long rRNA gene fragments or the entire mitochondria or plastids for genotype-level identification and building reference information for marker gene sequencing (78, 79).

FIG 3.

Taxonomic resolution of rRNA marker gene fragments in short-read and long-read analyses. Distribution of taxonomic information in the rRNA operon of bacteria (A) and fungi (B). Relative sequence similarity (C, D) and number of mismatching nucleotides (E, F) in long-read and short-read rRNA gene sequence data of bacteria (left) and fungi (right) at the level of kingdom (open columns) and genus (shaded columns) based on pairwise alignments of fragments. The values are given as mean (box) ± standard deviation (whiskers) and median (asterisk). Different letters above bars depict statistically significant differences among groups (P_adj < 0.001). Mean fragment length ± standard deviation is indicated below the fragments. The methods of calculating distances are outlined in Item S1 in the supplemental material.

Subregions of bacterial and fungal rRNA operons differ greatly in the number and proportion of pairwise differences that are informative for discrimination among species. In bacteria, all widely used rRNA short-read markers have much lower taxonomic resolution than the full-length 16S rRNA gene, ITS region, or 23S rRNA gene (Fig. 3C and E). Combining the ITS region with any (or both) of the flanking genes provides the best resolution at the level of strains and species. However, the ITS marker is of highly variable length, which may introduce PCR and sequencing biases. Long fragments covering the ITS region may miss multiple taxa, in which the rRNA genes are not arranged in a conventional manner. In fungi, the 18S rRNA gene subregions and the entire gene has very limited species-level resolution (Fig. 3D and F). On average, ITS1 and ITS2 subregions are nearly equally variable and of comparable length albeit with great among-group variation. In many fungal taxa (e.g., Sordariales, Glomerales, Hypocreales, Helotiales), variability and capacity to identify species are limited to one of the ITS subregions; therefore, full-length ITS sequences provide greater overall taxonomic resolution. Supplementing 18S and/or 28S sequence data with ITS sequences may add some species-level resolution (21), but it mainly improves above order-level identification and allows construction of phylogenies, addressing questions about phylogenetic diversity and community phylogeny.

Sanger sequencing of marker genes is still broadly used for species identification. However, long-read HTS approaches provide more accurate information at the level of alleles and genotypes (80) and enable retrieval of marker gene sequences from samples that are contaminated with cooccurring organisms (81). Using a single PacBio Sequel SMRT cell for sequencing, Hebert et al. (82) recovered high-quality, 658-base COI barcodes from 86% of 9,800 arthropod individuals. Similar results were obtained for analyzing 3,500 fly individuals in a single ONT MinION run (83).

For fungi, protocols covering the entire rRNA operon in one or two amplicons were successfully implemented for 77 samples using both PacBio and ONT sequencing (78). A part of this approach was later used to generate ca. 4,700-base 18S-ITS-28S barcodes for 430 cultures of early diverging fungal lineages, which improved the understanding of molecular diversity, inter- and intraspecific differences, and phylogenetic relations among these taxa (80). ONT sequencing of rRNA intergenic spacer (IGS) sequences (2.5 kb) was suggested as a preferential method for reliably distinguishing among closely related human pathogenic and nonpathogenic species of Cryptococcus (84).

Multiplexing hundreds to thousands of specimen samples in a single HTS run allows cost-savings of up to 100-fold compared with Sanger sequencing. For microorganisms, the high barcoding capacity of third-generation HTS has remained underutilized despite its potential value for routine screening of culture collections and fungaria. Furthermore, long-read marker gene sequencing may simultaneously recover the identity of the target organism along with the associated microbiota and reveal novel biotic interactions (85).

Biodiversity assessment.

Marker gene sequencing (metabarcoding) has become a standard procedure for determining richness and composition of microbial communities (86). At the depth of tens to hundreds of thousands of reads per sample, short reads offer a comprehensive semiquantitative view about the taxa present but usually provide low resolution at species level (Fig. 3). Long reads improve taxonomic resolution to species and genotype levels, especially when covering both variable and conserved loci in fungi (22–24) and bacteria (17, 18, 20). For example, long PacBio reads of the Cyanobacteria rbcL gene improved taxonomic and phylogenetic resolution and facilitated the discovery of two novel cyanobacterial clades associated with hornworts; cyanobacterial communities were driven by dispersal limitation, plant species, and temporal variation (87). The gyrB gene sequencing of rooibos (Aspalathus linearis) root nodulating bacteria showed that the distribution of symbionts is dispersal limited, whereas cooccurring nonnodulating bacteria are influenced by drift (88).

In eukaryotes, including fungi, accurate PacBio CCS reads provide a relatively more reliable phylogenetic placement of taxa that remain unidentified to order or phylum level based on sequence similarity comparisons (89, 90). Phylogenies of reads belonging to unculturable taxa revealed multiple novel genera of anaerobic Neocallimastigomycota from feces of herbivores, with a substantial host species effect on richness and composition (91). The ITS-28S 1,500-base amplicons enabled the resolution of much of the phytopathogenic genus Fusarium to species level and revealed that certain cover crops may act as reservoirs of harmful crop pathogens (22). Furthermore, use of such amplicons improved identification of arbuscular mycorrhizal fungi and revealed that genotypes present in commercially inoculated soils may replace native strains (21).

Molecular diagnostics.

For most sequencing platforms, sample preparation and sequencing take days or weeks if an external service is used. Here, ONT offers the following several unique properties suited for rapid diagnostics: low cost of apparatus, portability and low power requirements, technically nondemanding library preparation steps, affordability, and real-time access to sequencing data from DNA or RNA molecules. High-speed, accurate identification greatly benefits detection of pathogenic or hazardous organisms for rapid countermeasures. For a striking example, the Ebola virus was diagnosed from symptomatic patients using ONT MinION instruments in field camps and hospitals in West Africa within a day since sample collection based on cDNA (51, 52). A similar workflow was previously used to trace a salmonellosis outbreak at the pathotype level (43). Furthermore, ONT sequence data were used for accurate real-time identification of antibiotic resistance indicators in Neisseria gonorrhoeae isolates to predict susceptibility to therapeutic antimicrobials (92). Using 3.5- to 6-kb amplicons covering the rRNA operon, putative fungal pathogens were identified from external otitis in dogs (93).

In plant pathology, ONT has been used to detect cDNA of “Candidatus Liberibacter asiaticus” and plum pox virus from infected plants and insect vectors (94). Various fungal pathogens were detected from vegetables and pine needles using marker gene sequencing and metagenomic approaches with variable success (95). Llontop et al. (96) developed a protocol for species-level and occasionally genotype-level identification of bacterial tomato pathogens based on metagenomics data. With respect to other environmental applications, toxic dinoflagellates were identified to species based on a 3,000-bp 18S-ITS-28S rRNA gene amplicon (97). Similarly, invasive bivalves were diagnosed from water samples based on a 600-base mt-16S rRNA gene amplicon (98).

For rapid diagnosis potential, ONT MinION is unrivalled among sequencing platforms. Shortened DNA extraction and library preparation protocols revealed that metagenome-based pathogen diagnosis from plant samples can be achieved within 150 min of sample collection (95). Hundreds of clinical samples can be tested for SARS-CoV-2 using a single instrument in 1 day (99). Unfortunately, ONT sequencing has become routinely established in only a few human and plant pathology laboratories, probably due to lack of standard workflows and unwillingness to implement novel techniques. Also, the benefit of low cost compared with that of other HTS methods has remained poorly utilized, as very few publications using ONT emerge from developing countries.

Population biology.

Long reads are advantageous over short reads in intraspecific population genetics analyses because of their capacity to distinguish among genotypes, phase haplotypes, and resolve repeat-rich regions. PacBio sequencing has been used for developing primers and sequencing long microsatellite-rich markers (100) as well as detecting multiple antiretroviral drug-resistant genotypes in an HIV⁺ patient (101). SLRs revealed 10 genotypes of the hepatitis B virus in a single patient (65) and population structure of gut bacteria of fruit flies (102). Combined SLR and PacBio shotgun sequencing of Acropora coral holobionts revealed no biogeographic structuring of the coral or its Symbiodiniaceae algal mutualists, but symbiont composition explained much of the vulnerability to bleaching (103). PacBio 16S-ITS sequencing of lake bacteria revealed that different genotypes of the same species dominate in Japan and Europe, but there is very low regional-scale difference in Japan and Central Europe (19). In humans, PacBio-based detection of SNPs in alleles revealed that introgressed genomic fragments of the Denisovans and Neanderthals were positively selected in the evolution of Melanesian population (104). These methods offer great potential for tracking hybridization in microorganisms, such as pathogenic fungi.

In pathogens, all long-read sequencing methods have been successfully used to detect and monitor genotypes and virulence genes. SLRs revealed that long-term use of antibiotics may lead to evolution of Escherichia coli strains with up to 50 antibiotic resistance genes (105). PacBio genomic sequencing of Treponema pallidum isolates revealed >100 different, nonoverlapping copies of the tprk gene that encodes a membrane-bound protein, which prevents efficient resistance and vaccine development (106). PacBio single-cell sequencing revealed >100 karyotypes harboring mosaic aneuploidy (partial polyploidy) among a clonal population of Leishmania, suggesting that local variability in chromosome number contributes to the high adaptability of the pathogen (107). ONT-based phylogenomics of Mycoplasma bovis revealed no genetic differentiation among host ungulates but common cross-continental transportation of strains belonging to geographical subgroups with calves (108).

Long-read HTS has revealed the organization and copy number of rRNA genes and evolutionary mechanisms of their persistence. In strains of the arbuscular mycorrhizal fungal genus Rhizophagus, multiple isoforms with distinct secondary structure of the 28S rRNA genes are present, and their transcription depends on host associations rather than other biotic and abiotic factors, suggesting that these variants may be related to host selectivity in arbuscular mycorrhizal symbiosis (109). Overall, the presence of different ITS copies points to introgression and ineffective concerted evolution; divergent gene copies can be fixed by inbreeding (110). In prokaryotes, long reads have been used to show that unlinked rRNA genes are widespread (111).

Long reads are increasingly used in genome analysis and population genomics (112), with the greatest benefits in diploid and polyploid macroorganisms (9, 113) because of their capacity to solve repeat-rich regions and high variability suitable for distinguishing genotypes. Thus, long-read population genomics tools are useful for assessing functional differences among genotypes in addition to in-depth quantitative recording of genetic differences.

Metagenomics.

Metagenomics is broadly used for analyses of taxonomic and functional potential of microbial communities and recovery of individual viral and prokaryote genomes (i.e., metagenome-assembled genomes [MAGs]). MAG assemblies benefit from long reads by greater continuity in repeat-rich regions and reduced incorporation of contaminant sequences from other organisms (40, 105, 114). Using ONT long-read assemblies polished with short-read data, Singleton et al. (115) generated 1,083 high-quality prokaryote genomes from 23 sewage sludge samples. Many of these MAGs represented the first full genomes of various bacterial orders and candidate phyla, adding unique information about the potential functioning of these enigmatic taxonomic groups. Furthermore, long reads improve assignment of functional genes to specific taxa (116) and reduce the overall proportion of taxonomically and functionally unassigned contigs (117). Finally, long reads with sufficiently low error rates, such as PacBio CCS, allow direct determination of full-length functional genes without the need for assembly, thus offering better sensitivity to rare genes (13, 40).

Quantification accuracy of target groups using long-read approaches may outperform Illumina-based quantification for bacteria depending on GC content and library preparation methods (38, 118). ONT sequencing has been used to uncover the predominately vegan diet of rats (122) and quantify plant species’ contribution to pollen samples (120). However, quantification accuracy is constrained by among-taxon and among-tissue differences in DNA extractability, amount of DNA per biomass, as well as mitochondrial, plastid, and rRNA gene copy numbers (121, 122).

Gene expression.

Sequence-based (meta)transcriptomics methods can be used as proxies to estimate community activity (123). Most RNA molecules, especially in eukaryotes, are alternatively spliced or have highly similar copies; therefore, long reads are more efficient than short reads in recovering these RNA isoforms (36, 57, 124). For example, the PacBio Iso-Seq technique revealed that mutations in viruses cause a variable immune response among cells (125), and alternative splicing of a maize gene secures resistance to the root parasitic worm Diabrotica virgifera (126).

RNA sequencing is traditionally performed by reverse transcription and sequencing of cDNA molecules. Currently, only ONT supports direct RNA sequencing (58). This method revealed that the dominant zooplankton species, their parasites, and transcript profiles may all differ seasonally (127). In double stranded DNA (dsDNA) herpes simplex virus 1 (HSV-1), direct RNA sequencing enabled novel fusion transcripts that encode chimeric proteins (128). Direct RNA sequencing has several potentially important advantages over cDNA approaches because it allows for epigenetic investigation of native RNA and avoids biases related to reverse transcriptase activity potentially leading to codon switch artifacts and exclusion of certain RNA molecules.

Epigenetics.

DNA and RNA methylation influence nucleic acid-protein interactions and affect processes such as gene expression, chromosome stability, and self-recognition. The signal of PacBio and ONT sequencing is inherently affected by modified bases, which can be used for explicit recording of these epigenetic modifications (Fig. 1) (60, 129–131). Through regulating transcription, DNA methylation is related to the development, pathogenicity, and secondary metabolism of antagonists and viruses (131–133). Thus, epigenetics provides a yet little-studied dimension to functionality of populations and species within microbial communities.

Third-generation HTS techniques have become methods of choice in epigenetics, with important discoveries in phage-bacteria and host-pathogen interactions. PacBio analyses revealed that DNA methylation in certain genes is related to expression of phage-encoded Shiga toxin genes in an E. coli pathogenic isolate (134). The m4C methylations in Helicobacter pylori are essential for the bacterial association to host cells (135). In Campylobacter jejuni, methylations affect cell adherence, invasion, and biofilm formation, and their variation among cells promotes overall adaptability of the population (136). Matching methylation signatures may also improve binning of metagenomic contigs among closely related strains and linking plasmids or phages to their host (137, 138).

COMBINING OTHER MOLECULAR TOOLS WITH LONG-READ SEQUENCING

Target enrichment and single-cell methods.

In environmental samples, metagenomics and metatranscriptomics approaches may require enrichment of low-abundance target organisms or genes. Several methods exist to selectively enrich certain community members. DNA or proteins in microbial cells can be selectively labeled with fluorescent or heavy isotopes for stable isotope probing (SIP) (139), fluorescence in situ hybridization (FISH) (140), Raman spectroscopy (141), etc. Sorted cells or small groups of cells can be subjected to separate DNA extraction and molecular analysis in nano-wells or droplets (142, 143). Combined ONT and Illumina single-cell sequencing improved detection of full-length transcript isoforms in various mouse cell lines (144).

Target capture techniques are used to enrich chromosomes, genes, or RNA molecules by linking these targets to biotin-attached probes, followed by separation onto streptavidin-coated beads or wells. For example, the major histocompatibility complex of human DNA (20-kb fragments) was enriched 83-fold, yielding 10% on-target Illumina and PacBio sequences (145). Using large enriched fragment targeted sequencing (LEFTseq), reads from DNA fragments of the symbiotic bacterial genus Wolbachia were enriched 11-fold to 200-fold in invertebrate samples (146). Experiments using biotin-16-aminoallyl-2′-dUTP-labeled probes were used to enrich RNA isoforms (1 to 2.5 kb) around 7,600-fold in hundreds of human genes (147). Combined with SLR HTS, single-cell and target enrichment methods can be used for selective genome or transcriptome sequencing of certain microbial taxa, which may be particularly important for understanding the structure and function of the rare biosphere.

The properties of CAS9, an RNA-guided DNA endonuclease, enable the enrichment of target molecules to yield nearly 200-kb DNA fragments for PacBio and ONT sequencing (148, 149). Given the site-specificity of restrictases, this method is applicable for population-level studies but is more difficult to use for gene or community profiling of environmental samples because of high chances of nontarget restriction and systematic bias in loss of material.

ONT nanopore sequencing offers a unique opportunity for target enrichment by aborting the recording of reads during the sequencing process. The nontarget DNA molecules are recognized by predefined motifs and ejected from the active nanopores. The Read Until (150) and UNCALLED (151) software enable severalfold enrichment of target sequences.

Unique molecular identifiers.

UMIs are short unique index sequences that are integrated into DNA molecules, allowing reads to be reassociated with the same DNA molecule using bioinformatics approaches (68, 152). UMIs are essential to SLR technologies by linking together the short reads derived from a common DNA molecule and thereby allowing the long sequences of that DNA molecule (SLR) to be reconstructed (Fig. 2). Recently UMIs have also been implemented in third-generation HTS to enable >99.99% accuracy and a <0.02% chimera rate (26, 75). Furthermore, as UMIs label individual molecules, they are also useful for improving relative quantification (152).

BIOINFORMATICS TOOLS AND DATABASES

The emergence of long-read sequencing technologies required optimization of bioinformatics tools initially developed for short reads and generation of new software to realize the full potential of long-read sequencing in microbial ecology (153). An overview of bioinformatics tools used for long-read analyses is regularly updated in the Long Read Tools database (https://long-read-tools.org/).

In analysis of metagenomes, the recently developed metaFlye outperforms software generated for short-read genome assembly (154). For gene-level analyses of metagenomics data, the use of highly-accurate long reads allows assembly to be avoided entirely because most sequenced genes are covered by a single read, simplifying bioinformatics workflows and reducing assembly and binning biases (155, 156). New taxonomic classification methods, such as MetaMaps, outperform simple applications of short-read tools (157, 158). Purpose-built classification methods may be most valuable for ONT metagenomic reads due to the potential application of that technology to real-time surveillance for pathogens or antimicrobial resistance (96, 159).

In long-read marker gene sequencing, bioinformatics tools developed for short reads have been updated to handle long sequences (7). High-resolution “denoising” tools such as DADA2 and unoise3 have been extended for highly accurate long reads (25, 160, 161). New taxonomic assignment approaches have shown marginally higher classification performance from the common full-length 16S or ITS gene targets (162, 163) than standard short-read methods, but a void still exists in methods that can exploit the below-species level resolution that might be achievable from such data.

Most popular marker gene databases contain sequences that are of variable length but often shorter than long-read amplicons, which limits the taxonomic resolution that can be obtained. SILVA contains near full-length sequences of small and large subunits of rRNA operon of prokaryotes and eukaryotes, but these are insufficient for amplicons spanning the ITS region. Based on genomic data in RefSeq (https://ftp.ncbi.nlm.nih.gov/genomes/refseq/), Graf et al. (18) constructed the Athena database, which comprises around 56,000 sequences spanning >2,000 bases, i.e., much of the 16S rRNA gene, ITS, and around 600 bases of 23S rRNA gene. Since high-quality long reads often distinguish between different strains within a species and between different alleles of multicopy marker genes within a genome, reference databases need to account for such infraspecific variation. Such genotype-level information about bacterial 16S rRNA genes has been gathered in more local and habitat-specific databases (161, 164). For fungi and other eukaryotes, the UNITE database includes user-annotated information about multiple marker genes (2) with a specific reference subset developed for long 18S-ITS-28S sequences in version 9. The EukRef holds nuclear 18S rRNA and plastid 16S rRNA sequences for most eukaryotes, with a specific focus on protists (165).

LIMITATIONS OF LONG-READ SEQUENCING

One of the principal challenges to incorporating long-read HTS technologies is the relatively high requirements on the quality and quantity of the input DNA or RNA, which can be difficult to obtain from some samples. The higher rate of chimera formation in long amplicons and high raw error-rate of nanopore and PacBio sequencing represent another challenge, but this can be circumvented through the use of UMIs and proper bioinformatics tools (26).

Long-read sequencing methods are still relatively new compared to Illumina short-read sequencing. As standard practices continue to evolve, we strongly recommend that researchers routinely include quality control measures, such as mock communities comprising organisms with high-quality sequenced genomes (e.g., ZymoBIOMICS Microbial Community DNA Standard), to evaluate the technical biases and quantitative performance of their long-read sequencing protocols. In this rapidly progressing field, continued evaluation and improvement of long-read workflows for microbial ecology are of vital importance.

CHOICE OF SEQUENCING PLATFORMS

Both short-read and long-read technologies advance rapidly, as several sequencing instruments and important supporting methodological advances become available every year. Upon choosing the optimal method, microbial ecologists need to consider the required accuracy, amount of data needed, and expenses afforded (Table 2). Although the long-read sequencing is currently cheaper than the obsolete 454 pyrosequencing and early versions of Illumina sequencing, short-read sequencing costs have dropped >10-fold in the last decade. Because generation of sequencing libraries and data are costlier for long reads, short-read technologies may be preferable, for example, in marker gene sequencing, when short reads are expected to possess sufficient resolution, or when high species-level taxonomic resolution is of limited concern.

TABLE 2.

Feasibility of DNA and RNA sequencing using various sequencing platforms, their combination with UMIs (in parentheses), and hybrid approaches using short reads and long reads based on cost-efficiency, read length, and accuracy^a

Sequencing capability	Illumina/MGI Tech	SLRs (30×)	PacBio CCS (+UMIs)	ONT (+UMIs)	Hybrid approach
Rapid diagnostics	+	−	− (−)	+++ (++)	−
(Meta)genomes/ transcriptomes <0.5 kb	+++	−	+ (−)	− (+)	−
DNA (meta)barcoding <0.5 kb	+++	−	+ (−)	− (+)	−
(Meta)genomes/ transcriptomes 0.5–3.5 kb	+	++	+++ (+)	++ (++)	+
DNA (meta)barcoding 0.5–3.5 kb	+	++	+++ (+)	− (+++)	−
(Meta)genomes 3.5–10 kb	−	+++	++ (+)	++ (++)	++
DNA (meta)barcoding 3.5–10 kb	−	+++	++ (++)	− (+++)	+
(Meta)genomes >10 kb	−	++	++ (++)	+++ (+)	+++
Allele/isoform detection, phasing	+	++	++ (+)	++ (++)	+++
Base methylation	−	−	+++ (−)	+++ (−)	−
Direct RNA sequencing	−	−	− (−)	+++ (−)	−

^a+++, preferred; ++, suboptimal; +, useable but not recommended; −, not applicable or not competitive.

Besides the unique direct RNA sequencing and epigenome sequencing options, long-read protocols offer several advantages over short reads in many routine applications. With careful DNA extraction and library preparation, ONT and PacBio platforms can produce ultralong reads (>100 kb) that exceed those produced by short-read instruments by 3 to 4 orders of magnitude. These ultralong reads may be critical for circularizing bacterial chromosomes and solving long repeats such as entire rRNA loci in eukaryotes (40, 166). Currently, the ONT PromethION in-house application produces the largest amount of data per euro, which enables reconstruction of hundreds of high-quality bacterial genomes from metagenomes (115). However, the MGI Tech DNBSEQ-T7 short-read platform is the cheapest commercial solution in terms of throughput per base and per read (Table 1). Given the amounts of data and sequence accuracy, MGI Tech sequencing of SLRs is currently the most cost-efficient solution for high-quality sequencing of fragments >3.5 kb (73). UMI-based methods outperform other methods when there is scant starting material or a need for ultrahigh accuracy (>99.99%) of a single long read, e.g., for detecting rare genotypes and frameshifts (74, 75). Because of its relatively lower raw read length compared with ONT, PacBio Sequel II may be best suited for sequencing DNA fragments of 0.5 kb to 3.5 kb to take full advantage of the CCS option for the quality of individual reads (Table 2). We anticipate that any major shift in quality, throughput, or read length of HTS platforms may strongly alter the method preferred for testing specific scientific hypotheses.

CONCLUSIONS

Long-read HTS offers multiple cutting-edge options for understanding microbiome structure and functioning. Direct RNA sequencing has proved its efficiency in detecting transcriptomes of various DNA viruses (167) and discovery of single-stranded RNA (ssRNA) viruses (168), including coronaviruses (169), without the reverse transcriptase bias. Direct RNA sequencing has a potential to recover the diversity and function of double-stranded RNA (dsRNA) viruses and establish the roles of small RNA molecules in the functioning of organisms and interacting with viruses or microbial pathogens and symbionts. cDNA sequencing and potentially direct RNA sequencing offer a possibility to detect the rRNA gene products of highly divergent organisms that have issues with primer matching and introns (70). DNA and RNA methylome sequencing has been successfully used to link plasmids and associated phages to bacterial organisms (137, 167), with a further potential for detecting various epigenomic modifications in eukaryotes that are little understood thus far (170). We speculate that using these epigenetic patterns along with GC content, codon use, intron types, and relative read coverage may improve assembly of chromosome-sized scaffolds and linking chromosomes to reconstruct nearly full eukaryote genomes from complex environmental samples in the near future. Beside natural base modifications, nucleotide analogues such as bromodeoxyuridine and biotin-tagged DNA molecules can be directly recorded using the ONT technology to track DNA replication (171), suggesting that these methods may offer great promise for real-time detection of functionally enriched taxa based on DNA and RNA. Although long-read HTS methods have a potential to outperform short-read approaches in nearly all subfields of microbiology in terms of analytical performance, the requirement for relatively large amounts of input material and higher costs (compared to short-read sequencing) may remain major bottlenecks for certain research applications.