From genome structure to function: insights into structural variation in microbiology

Abstract

Structural variation in bacterial genomes is an important evolutionary driver. Genomic rearrangements, such as inversions, duplications, and insertions, can regulate gene expression and promote niche adaptation. Importantly, many of these variations are reversible and preprogrammed to generate heterogeneity. While many tools have been developed to detect structural variation in eukaryotic genomes, variation in bacterial genomes and metagenomes remains understudied. However, recent advances in genome sequencing technology and the development of new bioinformatic pipelines hold promise in further understanding microbial genomics.

Next generation sequencing and high throughput technologies have advanced the field of microbiology dramatically over the last two decades. Whereas bacterial genomes were thought to be fairly static in the days-to-weeks context, the application of high throughput sequencing approaches has revealed that bacterial genomes are, in fact, highly plastic. Bacteria can adapt rapidly to environmental fluctuations by modulating genomic content through genomic rearrangements, sometimes reversibly. The functional consequences of these types of genomic adaptation have been demonstrated elegantly in many bacterial pathogens, particularly in regions of the genome that have roles in host-microbe and microbe-microbe interactions. However, many questions remain as to exactly where, when, and how frequently these variations occur.

In the first part of this review, we will highlight a few key examples of structural variation (SV) in bacteria and describe how these genomic alterations affect phenotype. Then we will discuss the challenges of discovering new variations using next generation sequencing technology. Finally, we will highlight how long read sequencing identifies novel genomic variations in both isolates and in complex communities.

Mechanisms of SV in bacterial genomes

In bacteria, SV is typically defined as a rearrangement that occurs in a region of the genome greater than 50 base pairs in size. While the different types of SV can be defined in many different ways, we will break down the most common into three separate, broad categories: inversion, duplication, and insertion.

Inversion

An inversion occurs when a segment of DNA is excised and recombined in the opposite direction. Generally, the inverted region is flanked by high homology sequences or shared recognition sites and are primarily driven by specific enzymes or through prophage activation. In many cases, these inversions are reversible and appear to be part of a preprogrammed scheme to generate heterogeneity, alter gene expression, or adapt to specific niches. In bacteria there are two types of widely observed inversion: site specific recombination and large chromosomal inversions.

Site specific recombination is a targeted, preprogrammed, and reversible inversion of genomic segments. These segments can range in size from 100s to 1000s of base pairs. This process is enzymatically driven by recombinases that recognize two palindromic flanking sequences, often referred to as inverted repeats, which can vary in length between 10 and 30 base pairs and may contain mismatches [1–5]. In bacteria, site specific recombination is mediated by tyrosine or small serine-type recombinases, named for the catalytically active amino acid that carries out the nucleophilic attack on the DNA backbone [4,6,7]. Enzyme complex activity can be uni- or bi-directional and may be influenced by additional small proteins or accessory factors [2]. Depending on the prevalence of the recognition sites, recombinases can act on a single locus or at multiple loci.

Site specific recombination is widespread and found across many different bacterial phyla [8,9]. Most well characterized inversions flip regulatory regions or promoter sequences, forming an ON/OFF switch. When the region is in the same orientation as the downstream locus, transcription occurs. However when it is flipped to face the opposite direction, transcription no longer happens. Site specific recombination can also occur within and across coding regions resulting in domain switching [10–13]. When this occurs, enzymes can change specificity by swapping modular specificity-determining domains (Fig. 1A). Many well studied examples of site specific recombination have been identified in regions important for host-microbe interactions and microbe-microbe interactions [1]. For example, the phase variable Bacteroides capsular proteins both modulate the immune response and protect against phage predation (Round et al. 2011; Porter et al. 2020). Large chromosomal inversions occur in regions greater than 10,000 base pairs. Similar to site specific recombination regions, they often have homologous regions at their ends; however, unlike site specific recombination, they are hypothesized to be prophage driven or are a result of faulty recombination [14–17]. Intriguingly, these large chromosomal rearrangements can facilitate niche adaptation and in some instances, may be reversible [18]. They are also fairly common among clinical isolates [19].

Figure 1: SV reversibly alters restriction modification and methylation enzymes.

Restriction modification systems can both methylate and cleave DNA. DNA cleavage is one of the main mechanisms of bacteriophage defense. Whereas DNA methylation patterns are important both for self DNA recognition and transcriptional regulation. The restriction modification loci contains hsdR (restriction), hsdM (modification), and hsdS (specificity) genes. The hsdS gene contains two target recognition domains (TRD) which determine the specificity of the enzymes. The hsdS gene can be reversibly modified affecting its specificity. (A) In regions that vary by site specific recombination, loci contain multiple copies of hsdS. Generally, only one copy of hsdS (hsdS_A) is transcribed [13]. The additional copies of hsdS (hsdS_B, hsdS_C) function as alternative alleles. Site specific recombinases flip DNA sequences in between inverted repeats (dashed lines) and shuffle TRDs between the transcribed and non-transcribed hsdS genes [66]. Invertible systems have been observed in many bacteria including Streptococcus pneumoniae [12] and Bacteroides fragilis [67]. (B) In other organisms, hsdS gene specificity varies by slipped-strand mispairing. In many hsdS genes, simple nucleotide repeats separate TRDs [32]. Increases or decreases in the number or repeats can cause premature truncation. Truncated HsdS form dimers altering specificity. A well characterized example of slipped-strand mispairing is found in Neisseria gonorrhoeae [68].

Duplication

Genomic duplications occur when regions of the genome are copied. Duplication is found across all domains of life and can contribute to a large percentage of an organism’s genome [20]. Duplication events that involve the replication of genes are often referred to as copy number variation.

Evolutionarily, there can be multiple outcomes resulting from gene duplication. If a higher level of protein production is advantageous, identical genes are maintained. For example, many bacteria harbor multiple copies of ribosomal RNA which likely enables these bacteria to ramp up protein production relatively easily when needed for rapid adaptation to changing environments [21]. Duplication and divergence of gene sequences can also result in specialization. Alternatively, if multiple gene copies are detrimental, pseudogene or gene loss can occur.

In contrast to genomic duplication that is preserved for the long term, short term adaptive amplification has also been described. Often referred to as multicopy duplication or gene accordions, rapid copy generation has been observed transiently in response to stress [22–25]. The rapid accumulation of identical genes increases the amount of protein produced. Adaptive amplification can cause antibiotic resistance, niche adaptation, and altered host-microbe interactions [26,27]. The exact mechanisms that drive this phenomenon in vivo are not known, but it is postulated to be driven by homologous recombination after the initial duplication event [25].

Finally, slipped-strand mispairing, which can occur in regions of repetitive DNA sequences, can result in expansion or deletion of repeated sequences of DNA. These repetitive elements can influence downstream gene expression by altering regulatory elements, causing frameshifts, or generating alternative start sites [28–32] (Fig. 1B).

Insertion

Insertion is the addition of a DNA sequence into a region of the genome where it was not present prior. There are many different types of insertions that can occur in bacteria. Some involve DNA taken up from outside of the cell via conjugation or natural competency [33], while others involve the duplication or movement of genetic elements from within the genome to another position. Insertion can be carried out by homologous or non homologous recombination. Consequences of insertion vary depending on the location of insertion and what elements are carried within it.

A main driving force of insertional events are mobile genetic elements. Common types of these elements are insertion sequences (also known as transposases), integrons, prophages, and transposons. Mobile genetic elements can excise themselves and move in a site-specific or non-specific manner throughout the host genome [34–36]. Additionally, many mobile genetic elements contain cargo genes that cause antibiotic resistance, toxin production, or enhanced metabolic capabilities [35]. Downstream consequences of insertion largely depend on where in the genome the mobile elements insert. If insertion is within a coding region, that gene may become inactivated. Alternatively, many mobile elements contain regulatory elements or promoters, which may upregulate adjacent loci [37,38].

An additional type of insertional event is translocation. Translocations involve the transfer of genetic material from one part of the chromosome to another. Similarly to inversions, these types of SV are balanced and do not result in the net loss or gain of genomic content. However, these rearrangements may be biased to occur in specific locations of the genome (Darling et al. 2008). Phages are a common source of both insertion and translocation events in many bacterial genomes [39]. Depending upon their location of insertion and region translocated, transcription and expression patterns may be altered both inside and out of the translocation (Nagy-Staron et al. 2021).

Challenges of detecting SV in bacterial genomes

Despite the demonstrated importance of SV, these phenomena remain understudied in bacteria and especially in the context of microbiomes. In particular, the prevalence of SV across the domain Bacteria and its resulting effects on phenotype and evolution are not well understood.

A recent attempt to systematically identify invertible sequences in bacteria revealed their presence in species spanning 10 bacterial phyla, whereas there were only a handful of examples prior [8]. A general search for deletions and duplications in bacteria from the human microbiome revealed thousands of events across 56 different bacterial species, and at least one event in every species analyzed [40]. An analysis of bacterial SVs in bacteria from the microbiome samples of 1,437 individuals revealed strong associations between bacterial SVs and host bile acid composition [41]. Despite these recent studies demonstrating the prevalence and potential importance of these events, there is relatively little attention paid to SVs when compared with work performed on single nucleotide polymorphisms [42].

The neglect of SV is, in part, due to the technical difficulty of identifying SV with short read sequencing data. Short read sequences and sequencing library insert sizes are significantly shorter than many structural variants, limiting the accuracy and methodologies available for identification of these elements from short read sequencing data. Specifically, de novo assembly from short read sequencing data often fails to assemble through repetitive and low complexity regions, which frequently flank SVs, sometimes causing ambiguity in structural variant identification (for example: making it difficult to distinguish an insertion from a tandem duplication [43].

Detection SV via short read sequencing

Despite these challenges, many effective short read-based approaches for SV detection exist (Table 1), although most are targeted towards eukaryotic genomes. We highlight a subset of these approaches in this review (Table 1), however many more have been developed [47,57]. SV detection methods generally fall into two categories: (i) mapping based approaches and (ii) assembly based approaches [43]. Assembly based approaches leverage either an assembly graph or a de novo assembly for a given sample that is then used to compare to a reference genome, where SVs can be identified as differences between the assemblies. Cortex [44] and SGVar [45] are successful examples of assembly graph-based approaches, although both were developed and tested for use in eukaryotic isolate sequencing samples. Mapping based approaches map raw sequencing reads to a reference assembly, identifying potential SVs as regions of unexpected mapping patterns. For example, LUMPY [46] and Manta [47] analyze read depth, paired end read discordance, and split reads to identify potential SVs. MGEfinder looks for clipped read alignments to identify possible insertions and is optimized for use with bacterial genomes [48]. Most individual approaches specialize or perform better in detecting specific types of SVs [49], and picking a suite of approaches to broaden the range of detectable SVs is common, but the particular combination of approaches selected should be considered carefully [49].

Table 1:

Selected computational SV detection methods described in this review. For more methods, see [49] and [62].

Name	Publication	Approach	Types of SV	Long-read compatible	Benchmarked in bacteria	Benchmarked in shotgun metagenomes
Cortex	[44]	Assembly	Insertions, deletions
SGVar	[45]	Assembly	Insertions, deletions
LUMPY	[46]	Mapping	Insertions, deletions, duplications, inversions, translocations
Manta	[47]	Mapping	Insertions, deletions, duplications, inversions, translocations
MGEfinder	[48]	Mapping	Insertions		yes
PhaseFinder	[8]	Mapping	Inversions		yes	yes
SGVFinder	[40]	Mapping	Duplications		yes	yes
breseq	[69]	Mapping	Insertions, deletions		yes
cuteSV	[61]	Mapping	Insertions, deletions, duplications, inversions, translocations	yes
Sniffles	[50]	Mapping	Insertions, deletions, duplications, inversions, translocations	yes
combiSV	[62]	Multi-method	Insertions, deletions, duplications, inversions, translocations	yes

Detecting SV in metagenomic sequencing data

Shotgun metagenomic sequencing is an indispensable tool in bacterial genomics, and metagenomes offer the unique possibility to observe the full range of SVs present in natural bacterial populations. However, certain characteristics of metagenomes, relative to isolate genomes, require methods to either be developed with metagenomes in mind or benchmarked appropriately. In a metagenomic sample, a mixed population of organisms from the same strain or species are often present, introducing possible SV heterogeneity. In addition, mis-mapping reads from related species may influence false positive discovery rates. A subset of methods for SV detection have been designed for metagenomic data, including PhaseFinder for identifying inversions [8] and SGVfinder for deletions and duplications [40]. Methods that consider low frequency alleles may also be applicable, such as Sniffles [50].

Detecting SV via long-read sequencing

Long read sequencing technologies offer the potential to expand our ability to identify and characterize SV. For example, long read sequencing revealed SV in the human genome missed for years by short read sequencing [51]. Long reads can sequence through entire SV events, including large, multi-kb events, and assemble through flanking repetitive regions, allowing for SV detection with precision, simplifying detection approaches, and even resolving SV events undetectable with short read sequencing data [52–54].

Two technologies currently dominate long read sequencing [55], Oxford Nanopore and PacBio SMRT sequencing. SMRT sequencing utilizes a polymerase tethered to the bottom of a well that fluoresces upon addition of a nucleotide [56]. Nanopore sequencing passes a molecule through a protein nanopore, where the nucleotide can be inferred from the resulting ionic current fluctuations [57]. Both technologies have historically suffered from a sequencing error rate much higher than that of short read sequencing. A high error rate poses problems for detecting some SVs by convoluting identification of rare and low coverage events. In nanopore data, indels and substitutions introduced by sequencing error are frequent but not uniform; homopolymers are particularly difficult to resolve [58]. SMRT sequencing similarly struggles with resolving homopolymers [59]. Despite issues posed by higher error rates, long reads are invaluable in detecting SV events currently missed by short read sequencing, and long-read sequencing technologies continue to improve, with overall error rates at the read level asserted to have been reduced <1% for both SMRT sequencers [59] and nanopore sequencers [60].

Methods for SV detection utilizing long read sequencing exist but are fewer in number. Notably, none have been tested for use with shotgun metagenomic data. cuteSV utilizes long read mapping to extract a variety of alignment signatures for identification of SVs [61]. Sniffles also uses a long read mapping approach, combined with subsampling of read alignments to adjust mismapping parameters for each individual dataset, resulting in higher precision [50]. combiSV [62] combines six different long read SV callers to improve overall precision and recall. In addition, both mapping and assembly based methods for SV detection rely on high quality genome assemblies to make inferences. Here long read sequencing has already made large inroads by enabling reliable reconstruction of high quality genomes from isolate [63,64] and metagenomic [54,65] sequencing. This is particularly exciting for metagenomic sequencing experiments, because short read metagenomes typically produce more fragmented genomes than isolate sequencing, resulting in missed SV events.

Conclusion

Over the last two decades, next generation sequencing coupled with computational analyses have revolutionized bacteriology. Much of the increase in bacterial genomic data has relied on short read sequencing technology. Despite the explosion in available data and genomic information, the limitations of short read sequencing have been unable to address fundamental questions about SV in bacterial genomics. Advancements in long read sequencing technologies are helping address this gap in knowledge. Additionally, microbial public sequencing datasets and assembled reference genomes represent a largely untapped resource for SV mining, and can be combined with newly generated datasets to make new inferences. Further development of methodologies for the identification of SVs in both bacterial isolate and metagenomic sequencing samples will be needed to fully make use of these resources.. Given the rich diversity of SVs that exist in bacteria, ranging from inversions to duplications to insertions, we anticipate that the application of improved tools to study these events will enable us to better understand how SVs contribute to bacterial phenotypic variability and adaptability.

Highlights.

• Bacteria can adapt to environmental fluctuations through genome rearrangement

• Inversions, duplications, translocations, and insertions are common methods of adaptation

• Short read sequencing can identify various structural variants

• Long read sequencing enables detection of a greater breadth of structural variants and variants undetectable via short read sequencing

• Structural variation plays a significant and largely underappreciated role in biology