Bioinformatics tools used for whole-genome sequencing analysis of Neisseria gonorrhoeae: a literature review

Abstract

Whole-genome sequencing (WGS) data are well established for the investigation of gonococcal transmission, antimicrobial resistance prediction, population structure determination and population dynamics. A variety of bioinformatics tools, repositories, services and platforms have been applied to manage and analyze Neisseria gonorrhoeae WGS datasets. This review provides an overview of the various bioinformatics approaches and resources used in 105 published studies (as of 30 April 2021). The challenges in the analysis of N. gonorrhoeae WGS datasets, as well as future bioinformatics requirements, are also discussed.

Introduction

Gonorrhoea, caused by the Gram-negative bacterium Neisseria gonorrhoeae (gonococcus), is the second most common sexually transmitted infection worldwide with a prevalence estimated at 86 million cases [1, 2]. Untreated infections in females, who bear the burden of this disease, and who are often asymptomatic, can lead to severe health outcomes such as pelvic inflammatory disease, ectopic pregnancy, infertility and occasionally maternal death [2]. Infections may be transmitted to children during childbirth; neonatal ophthalmia was a leading cause of blindness prior to antibiotic prophylaxis. Males, who are usually symptomatic, can also suffer complications such as epididymitis and infertility. Both sexes may experience complications such as arthritis, blood stream and joint infections. The progressive development of antimicrobial resistance (AMR) in N. gonorrhoeae isolates to all classes of antibiotics [i.e. sulphonamides, penicillins, tetracyclines, macrolides (e.g. azithromycin), fluoroquinolones, and even the last recommended monotherapeutic antibiotic class, third-generation cephalosporins], has compounded difficulties in treating gonorrhoea infections and may produce untreatable infections in the future [3, 4]. Because of AMR, N. gonorrhoeae has been placed on the antibiotic watch lists of the World Health Organization and many countries worldwide [2–5].

Whole-genome sequencing (WGS) of N. gonorrhoeae isolates has been widely used to predict AMR, to identify novel mutations and mechanisms associated with gonococcal AMR, to ascertain strain types (STs), for molecular epidemiological surveillance (micro and macro epidemiological analysis), to track gonococcal transmission and for phylogenetic analysis [3, 6–9]. Only one report reviewed the bioinformatics resources that have been used for the analysis of N. gonorrhoeae genome sequencing datasets [10]. The present article provides an overview of widely used, as well as specifically designed bioinformatics tools for the analysis of N. gonorrhoeae) whole genomes; 105 published studies (last literature search 30 April 2021) were reviewed. The majority of the studies used the Illumina sequencing platform, but the few which used Nanopore or PacBio are also referenced. This review provides insights into the varieties of analyses (e.g. AMR analysis and prediction, molecular epidemiology, strain transmission, strain typing, population structure analysis) used by investigators for analyzing gonococcal genomes. Also, this review is intended especially for those with moderate to no bioinformatics backgrounds who would like to analyze genomic sequences for N. gonorrhoeae. Current challenges, possible solutions and future directions for N. gonorrhoeae WGS data analysis are discussed.

Bioinformatics tools for the analysis of Neisseria gonorrhoea WGS data

Several steps can be included in the analysis of WGS datasets. Some are essential and others can be optional (Supplementary Table 1). These steps may include: (i) data cleaning, (ii) de novo assembly, (iii) scaffolding, (iv) quality assessment of assembled genomes and annotation, (v) read mapping, variant calling and core genome analysis, (vi) strain typing and AMR prediction, and (vii) phylogenetic analysis and visualization of trees.

Data cleaning (i.e. quality check, trimming and contamination check)

The quality (e.g. per-base sequence quality, adapter content, and sequence duplication level) of the DNA analyzed should be ascertained. Generally, WGS datasets are submitted in the FASTQ (or FASTA) [11] format, and quality is generally ascertained using FastQC [12], fastp [13], NGSQC toolkit [14] or PRINSEQ-lite [15] programs. Thirty-six (of 105) published gonococcal WGS studies [16–51] reported assessing the quality of N. gonorrhoeae raw reads before performing downstream analyses; 20 used FastQC [18–23, 28, 32–36, 40, 41, 43, 46, 48–51]; 2 each used fastp [42, 47] or PRINSEQ-lite [27, 30] and 1 used NGSQC toolkit [24]. Ten studies [16, 17, 26, 29, 31, 37–39, 44, 45] used pipelines that incorporate some of the above programs (i.e. Nullarbor [52], Gen2Epi [48], Explify [https://www.idbydna.com/explify-platform/], Shovill [https://github.com/tseemann/shovill] and INNUca [53]).

WGS reads of poor quality (adapter sequences not removed and low-quality bases) could lead to longer computational times and possibly inaccurate results (e.g. Single Nucleotide Polymorphisms (SNPs) analysis) [54]. Trimming (eliminating poor quality regions) may be performed using programs such as Trimmomatic [55], Cutadapt [56], PRINSEQ-lite [15], FASTX (http://hannonlab.cshl.edu/fastx_toolkit/index.html), Sickle (https://github.com/najoshi/sickle), ConDeTri [57], SolexaQA [58], ERNE-FILTER [54], Trim Galore [59], Nesoni (https://pypi.org/project/nesoni/) and fastp [13]. Thirty-nine of 105 published N. gonorrhoeae WGS studies [16, 17, 21–24, 26–39, 41, 42, 44–46, 48, 49, 51, 60–70] performed trimming; 13 used Trimmomatic [23, 32, 34, 36, 48, 51, 60–62, 66, 67, 69, 70]; 5 used Cutadapt [21, 35, 41, 44, 46]; two each used fastp [42, 63], Trim Galore [33, 64] or PRINSEQ-lite [27, 30]; and 1 each used Nesoni [22], Sickle [28] or NGSQC toolkit [24]. Eleven reports [16, 17, 26, 29, 31, 37–39, 45, 65, 68] used pipelines with in-built functions for trimming (i.e. Gen2Epi, CLC Genomics Workbench [https://digitalinsights.qiagen.com/products-overview/discovery-insights-portfolio/analysis-and-visualization/qiagen-clc-genomics-workbench], INNUca, Shovill, Exemplify, JEKESA and Nullarbor). Sixty-six (of 105) studies did not report trimming, possibly because the raw reads were of good quality. Nevertheless, ideally quality check reports should be included in Supplementary data.

Contamination removal (i.e. identifying reads present from species other than N. gonorrhoeae) is generally accomplished using programs such as Kraken [71], DeconSeq [72], MetaPhlAn2 [73], KmerFinder [74, 75] and ConFindr [76] (Supplementary Table 2). These programs are k-mer (a subsequence of length k) based taxonomic identification tools. Eighteen of 105 published gonococcal WGS studies in which DNA was extracted from pure cultures [16, 17, 21, 26, 31, 32, 36, 37, 41, 42, 46, 48, 51, 60, 62, 77–79] performed a contamination check; 10 applied Kraken [21, 32, 41, 42, 46, 48, 51, 62, 78–79] and 3 used either DeconSeq [60] or MetaPhlAn2/ MetaPhlAn [40]. Five studies used pipelines (i.e. Nullarbor, Gen2Epi), which include Kraken [16, 17, 26, 31, 37] for contamination checks. The contamination removal step may not be required when gonococcal DNA is extracted from pure cultures or when the majority (~90% if possible) of the WGS reads are N. gonorrhoeae. Ideally, authors should include data cleaning reports as supplementary data.

Data cleaning is especially relevant when analyzing gonococcal genomes directly purified from clinical specimens (e.g. urinary, vaginal, cervical and body fluids). Different species may occur in unknown abundance, mainly in low quantity, which could make the identification of gonococcal reads in such specimens problematic. Further, the presence of multiple gonococcal strains or closely related species could lead to genome assemblies containing chimeric contigs. Hardick et al. [80] performed N. gonorrhoeae WGS from clinical swabs (oral, rectal and urinary) and found that aside from a low concentration (0.001–0.178%) of gonococcal DNA in some clinical samples, mixed bacterial species in rectal and urinary swabs complicated the bioinformatics analyses steps; they used KmerFinder to remove contaminations. In another study, Graham et al. [81] removed human sequences from urine specimens using Kraken before generating the de novo assembly of N. gonorrhoeae short reads.

De novo assembly

De novo assembly is the process of constructing longer continuous stretches, called contigs, from clean reads without prior knowledge of the genome sequence. Commonly used software for de novo assembly are SPAdes [82], Velvet [83], Abyss [84], Unicycler [85], CLC Genomics workbench, Nullarbor, Gen2Epi, INNuca, Shovill, JEKESA (https://github.com/stanikae/jekesa), GS de novo assembler, IDBA-UD [86], Newbler (https://cals.arizona.edu/swes/maier_lab/kartchner/documentation/index.php/home/docs/newbler), Sanger Institute Assembly pipeline [87] or A5-Miseq (https://github.com/koadman/docker-A5-miseq) (Supplementary Table 2).

SPAdes [82], uses a De Bruijn graph algorithm to assemble short reads into contigs. Forty-four (of 105) published N. gonorrhoeae WGS studies (Supplementary Table 1) [18–23, 32–36, 40–43, 46, 48, 49, 51, 60–63, 66, 69, 78, 88–105] generated de novo contigs using SPAdes. Other studies (n = 45) used programs such as Velvet (n = 9) [27, 106–113], CLC Genomics workbench (n = 11) [65, 77, 114–122]. Seven used Unicycler [47, 64, 67, 70, 123–125]; four used the Nullarbor pipeline [16, 26, 31, 37]; three used the Sanger Institute Assembly pipeline [25, 79, 126]; two each used INNuca [29, 45] and Abyss [28, 127]; and one each used Shovill [38], the Gen2Epi pipeline [17], JEKESA [68], GS de novo assembler [128], IDBA-UD [30], Newbler [129] and A5-Miseq [130]. Sixteen (of 105) studies [24, 39, 44, 79, 131–142] did not use this step, as the WGS were was only used for either variant calling or multivariant linear regression analysis. De novo assembly should be performed for questions such as identifying variants or complex rearrangements in the genome, preforming genome comparisons, and AMR prediction.

Scaffolding

De novo contigs of N. gonorrhoeae can be assembled into longer sequences called scaffolds. Scaffolding is optional; most studies have no requirement for this step. Scaffolding is usually performed to refine assembled genomes because draft assemblies may contain genes split over several contigs [125]. For example, Brynildsrud et al. [125] found that in some N. gonorrhoeae WGS assemblies, the modB gene was split over several contigs [125]. In another study, Suwayyid et al. [143] reported that the Nf1 gene was split across two contigs. The main reason for such fragmentation is the high number of repetitive sequences in N. gonorrhoeae [144].

Different algorithms and technologies used for scaffolding were reviewed by Ghurye and Pop [145]. Only a few pipelines for N. gonorrhoeae include scaffolding features (i.e. Sanger Institute Assembly pipeline [87] and Gen2Epi [48]). Parmar et al. [17, 146] applied Gen2Epi [48] to assemble N. gonorrhoeae de novo contigs into scaffolds. Harris et al. [25], Town et al. [79] and Sánchez-Busó et al. [126] used the Sanger Institute Assembly pipeline [87] to generate and improve gonococcal genome assemblies (Supplementary Table 2).

Scaffolding may introduce assembly problems such as missing information caused by the presence of long sequences of ambiguous nucleotides, as observed in analyses performed using Gen2Epi [17, 48, 51]. For gonococcal WGS analysis, methods to assemble contigs into longer, relatively error-free scaffolds should be developed to provide longer genomes. Alternatives for obtaining longer reads include using sequencing methods such PacBio or Oxford Nanopore [47, 50, 67, 103, 147].

Quality assessment of assembled genomes and annotation

After the assembly/scaffolding steps, genomes may be assessed for quality using the program Quality Assessment Tool for Genome Assemblies (QUAST) [148]. QUAST [148], (https://github.com/ablab/quast), compares and evaluates genome assemblies on the basis of ‘N’ statistics (N50, NA50, NGA50) values, the total number of contigs and misassemblies. This important step is performed to check the completeness and accuracy of the assembled genomes. Twenty N. gonorrhoeae WGS studies used QUAST to evaluate the quality of assembled genomes [21, 30, 32–34, 36, 38, 42, 47, 48, 51, 60, 69, 88, 95–98, 100, 104]. Pipelines used to assess the quality of assembled genomes included Nullarbor (n = 4) [16, 26, 31, 37], INNUca (n = 2) [29, 45], Gen2Epi (n = 1) [17] and JEKESA (n = 1) [68]. Preferably, data on quality assessment could be included in Supplementary materials when submitting peer-reviewed articles.

Annotation of assembled genomes is used to identify key features (e.g. various genes with known and unknown functions and their products, mobile genetic elements, genome duplications). Frequently used programs for annotation are Prokka [149], Prodigal [150], and the pipelines NCBI prokaryotic genome annotation [151] and DFAST [152] (Supplementary Table 2). Prokka was used to annotate N. gonorrhoeae assemblies in 25 studies [16, 18, 19, 22, 23, 26, 30, 31, 34, 37, 38, 40, 44, 49, 50, 60, 62, 69, 78, 89, 90, 101, 103, 111, 130]; 4 used the NCBI prokaryotic genome annotation pipeline [27, 33, 47, 68]; 3 used Prodigal [17, 48, 51] and 1 used DFAST [64]. Although this step is optional, annotation results, such as total number of predicted genes and their positions, could be helpful to assess the completeness of the assembled genome and could be included in Supplementary data of papers.

Read mapping, variant calling and core genome analysis

In the read mapping step, short WGS reads are aligned against (an) N. gonorrhoeae reference genome(s) using programs such as Burrows–Wheeler Alignment (BWA) [153], Bowtie [154], Stampy [155], ParSNP [156], Galaxy SNVPhyl Workflow [157], Gen2Epi [48], CLC Genomics workbench, Snippy [158] and SMALT [159] (Supplementary Table 2). This step is important if the objective is to perform SNP analysis (for core genome and AMR prediction) and to find the total number of N. gonorrhoeae reads in WGS samples. Thirty-one studies (of 105) used BWA [16, 24, 30, 31, 37, 38, 40, 42, 44, 49, 50, 62, 78, 79, 91, 93, 94, 100, 104, 105, 107, 110, 113, 117–119, 123, 124, 132, 133, 138]; 10 used SMALT [18–20, 106, 114, 77, 121, 129, 126, 131]; 3 each used ParSNP [101, 102, 160] or Stampy [108, 109, 136]; 2 each used Galaxy SNVPhyl Workflow [89, 90], CLC Genomics workbench [112, 122], Snippy [26, 92] and Bowtie [48, 51]; and one used Gen2Epi [17].

The variant calling step is required if the objective of a study is to perform evolutionary analyses (e.g. to study gonococcal AMR and/or gonorrhea transmission). Frequently used programs for variant calling are: SAMtools [161], GATK [162], Pilon [163], Snippy [158], Pathogenwatch [36], SNVPhyl [157], CLC Genomics workbench, SSAHA pipeline [164] or Freebayes [165] (Supplementary Table 2). SAMtools [161], used for post-processing (such as indexing, sorting, data extraction and format conversion) of alignment files (output files generated after aligning short reads to the reference genomes), has been used for variant calling [20, 30, 42, 44, 77–79, 109, 113, 114, 116, 117, 121, 123, 124, 126, 131, 132, 136], SNP verification by comparing reads or assemblies to the reference genome [94], per-base position information determination [91], sorting and indexing the alignment file [20], format conversion of the alignment file [79] and determining base count and nucleotide ratio [112, 113]. Twenty-one reports used Samtools [20, 24, 30, 42, 44, 77–79, 108, 109, 113, 114, 116, 117, 121, 123, 124, 126, 131, 132, 136]; six used ParSNP [41, 46, 61, 101, 102, 112]; four each used Freebayes [18, 19, 37, 62] or GATK [91, 94, 134, 142]; two each used Pilon [40, 100] or SNVPhyl [89, 90]; and one each used Snippy [92], Pathogenwatch [36], CLC Genomics workbench [122] and the SSAHA pipeline [129]. Sixty-two studies did not require this step for analysis (Supplementary Table 1).

Core, pan- and accessory genomic analysis of assembled genomes is optional and is performed using Roary [166] and ParSNP [156]. Eight of 33 N. gonorrhoeae WGS studies in which core genomes were analyzed performed phylogenetic analysis using Roary [25, 49, 50, 60, 78, 103, 111, 126], and 8 used ParSNP [21, 22, 32, 35, 63, 95, 99, 125]. Three studies [32, 35, 99] did not describe whether WGS reads or assembled contigs were used to generate core genome alignments; this information is critical to ascertain whether alignments were based on whole or core genomes.

Strain Typing and AMR prediction

Various schemes have been used to type N. gonorrhoeae isolates using molecular methods [i.e. N. gonorrhoeae multi-antigen sequence typing (NG-MAST), multi-locus sequence typing (MLST) and N. gonorrhoeae sequence typing for AMR (NG-STAR)] [167–169] for purposes such as ascertaining gonococcal population structure, strain transmission and AMR identification [16–37, 60–65, 77–79, 88–100, 106, 108–112, 114, 116–121, 123, 124, 126–132, 134–138]. NG-MAST [168], an online database available at http://www.ng-mast.net, includes sequence data for two highly polymorphic antigen-encoding loci, porB (porin B) and tbpB (transferrin-binding protein B). The sequences of both loci are assigned individual allele numbers, and a distinctive ST is assigned to each allelic profile. This typing scheme has also been implemented in various programs such as NGMASTER [170], short read sequence typing (SRST) [171], Gen2Epi [48], Gen2EpiGUI [51] and pubMLST [167] to assign NG-MAST STs to the assembled queries (Supplementary Table 2).

For MLST typing, the Bacterial Isolate Genome Sequence Database (BIGSdb), a widely used database hosted by pubMLST.org [167, 172, 173], contains 9206 N. gonorrhoeae of 618 426 genomes (as of 20 May 2021). The BIGSdb platform exploits stored sequence data (submitted via the pubMLST submission system) of gonococcal specimens ranging from short Sanger sequencing reads to whole genomes (assembled contigs or complete genomes). The pubMLST platform also hosts several web-based and stand-alone tools to identify MLST sequence types (i.e. alleles of abcZ, adk, aroE, fumC, gdh, pdhC and pgm) directly from sequence reads. Information present in the pubMLST typing databases is also used in other programs, such as antimicrobial resistance identification by assembly (ARIBA) [174], SRST [171], Gen2Epi [48] Gen2EpiGUI [51] and MLST [175], to assign MLST STs.

Several general databases, such as Comprehensive Antibiotic Resistance Database (CARD), Pathosystems Resource Integration Center (PATRIC) and Lactamase Engineering Database (LacED) [176, 177, 178] catalogue antibiotic resistance determinants [179, 180]. A variety of tools have also been used to predict AMR profiles in bacteria, including N. gonorrhoeae [129, 176] e.g. Resistance Gene Identifier (RGI), Basic Local Alignment Search Tools (BLAST), Rapid Annotation Using Subsystems Technology (RAST) [180], Antibiotic Resistance Gene-ANNOTation (ARG-ANNOT) [179], ResFinder [181], ARIBA [174], ABRicate [182], ResFinder, PointFinder and NG-STAR [183–187].

CARD [176] was developed to store sequences of genes carrying resistance to various antibiotics along with relevant metadata. Other tools available in CARD for quick identification and visualization of antibiotic resistance genes in new, unannotated genomes [23], include RGI and BLAST. A recent release (3.0.3) of CARD included N. gonorrhoeae AMR information collected from systematic reviews [176]. Kubanov et al. [23] used the RGI web portal from CARD to search for and analyze AMR genetic determinants in genomic sequences of N. gonorrhoeae strains.

PATRIC [177] is an online resource created to store and integrate different types of data such as genomic, transcriptomic, protein–protein interaction, three-dimensional structure, taxonomic, and typing data. Currently, PATRIC contains 419 985 bacterial genomes including 7741 genomes (as of 20 May 2021) of N. gonorrhoeae. PATRIC uses RAST [180] for the annotation of genomes. Kivata et al. [183] identified the determinants of fluoroquinolone resistance in N. gonorrhoeae whole genomes with RAST implemented in PATRIC. Other studies used the independently hosted RAST server (https://rast.nmpdr.org/) to annotate N. gonorrhoeae genomes for AMR [129, 183–185]; Ang et al. [184] reported determinants associated with resistance to ciprofloxacin, penicillin and tetracycline in a strain of N. gonorrhoeae. Marri et al. [185] and Didelot et al. [129] utilized RAST for annotation, without AMR prediction. The program ARG-ANNOT [179] was used by Ryan et al. [92] to detect AMR determinants in 609 genomes of N. gonorrhoeae isolates.

ARIBA (https://github.com/sanger-pathogens/ariba) identifies AMR associated variants and genes directly from short sequencing reads [174]. Ten studies used ARIBA to analyze N. gonorrhoeae AMR [22, 29, 42, 45, 62, 63, 126, 130, 140, 188]. For example, Town et al. [188] identified mutations in gyrA, mtrR, porB1b, ponA, parC, bla_TEM and penA alleles implicated in AMR and also used the ARIBA micplot module to associate the Minimum Inhibitory Concentrations (MICs) of 1277 N. gonorrhoeae isolates [188]. Sánchez-Busó et al. [126] detected the presence of bla_TEM and tetM in 419 N. gonorrhoeae isolates and Yahara et al. [130] and Williamson et al. [62] identified known AMR genetic markers and variants in 271 and 2186 N. gonorrhoeae genomes, respectively, using ARIBA.

ABRicate [182], an AMR gene detection program used in four published gonococcal WGS studies [16, 26, 31, 37], links assembled genomes to AMR profiles. Other tools, such as ResFinder (https://cge.cbs.dtu.dk/services/ResFinder/) and PointFinder (https://bitbucket.org/genomicepidemiology/pointfinder/src/master/), were used to identify chromosomal mutations and horizontally acquired AMR genes in N. gonorrhoeae [181, 189]. Four studies used ResFinder to investigate AMR genes [60, 81, 95, 109] and Bailey et al. [60] determined AMR-associated genes from N. gonorrhoeae WGS datasets using both ResFinder and PointFinder.

A web-based application specific to N. gonorrhoeae, NG-STAR, (https://ngstar.canada.ca/pages/welcome?lang=en) analyzes the nucleotide sequences of seven genes (penA, mtrR, porB, ponA, gyrA, parC and 23S rRNA) associated with AMR [169]. NG-STAR performs AMR prediction of specific gene sequences and provides allele types and an NG-STAR ST. Forty-three of 105 published gonococcal WGS studies used the NG-STAR scheme for AMR prediction.

AMR prediction tools for N. gonorrhoeae WGS analyses are based on prior knowledge of resistance mutations associated with resistance phenotypes; they do not identify novel mutations conferring AMR, which could be located in intergenic regions or novel genes. Shi et al. [190] proposed a deep learning (DL)-based workflow named Deep Neural Pursuit-Average Activation Potential to identify existing as well as novel point mutations possibly conferring AMR in N. gonorrhoeae using WGS data paired with MICs for five antibiotics. Eyre et al. [113] used multivariate linear regression to predict genetic determinants of MICs for penicillin, tetracycline, azithromycin, cefixime and ciprofloxacin in 681 WGS datasets of N. gonorrhoeae [113]. Alternatively, using seven N. gonorrhoeae WGS datasets (n = 4358) with associated ciprofloxacin and azithromycin MICs, Hicks et al. [104] evaluated the performance of various machine learning (ML) algorithms (e.g. set covering machine, random forest classification and regression) to predict AMR phenotypes from WGS data. To improve ML-based AMR predictions in future, Hicks et al. [104] suggested that more diverse WGS datasets paired with antibiotic susceptibility data be analyzed, a consistent standard protocol to measure the MICs be followed, and more ML algorithm types (e.g. Bayesian framework and neural networks) be implemented and evaluated on larger N. gonorrhoeae WGS datasets. Recently, a Prediction of Antimicrobial Resistance by MAPping (PARMAP; https://github.com/452990729/PARMAP) program was developed to identify AMR in the pan genomes of bacteria [191]. The authors validated this method using 1597 N. gonorrhoeae genomes [160] and identified existing as well as novel mutations associated with ciprofloxacin resistance.

Genome wide association studies (GWAS) is a technique used for the analysis of multiple variants in datasets to test associations between genotype variants and phenotype. Three research articles used GWAS to identify novel AMR determinants in N. gonorrhoeae [49, 105, 160]. Ma et al. performed multiple regression-based GWAS in N. gonorrhoeae isolates (n = 4852) and found variations due to unknown genetic loci in the predicted values of MIC for azithromycin, ciprofloxacin and ceftriaxone, as compared to the original MIC values [49]. Ma et al. [105] also implemented the same approach to identify novel mutations associated with increased azithromycin resistance in N. gonorrhoeae . Using single-site GWAS analysis, Schubert et al. [160] discovered 35 genomic loci and 240 epistatic pairs (i.e. where the effect of a gene mutation depends on the absence or presence of the mutation in other genes) associated with increased resistance to five antibiotics in a dataset of 1102 N. gonorrhoeae isolates.

LacED [178] is an extension of the Jacoby and Bush ß-lactamase list [192] (a curated database of ß-lactamases, which is no longer updated) which allows users to detect TEM (named after the patient, Temoniera, from whom it was isolated) ß-lactamases with novel mutation profiles [186, 187]. This database provides structural, annotation and alignment information related to different TEM ß-lactamases (numbered according to the scheme proposed by Ambler [193]). LacED was used by Muhammad et al. [186] and Stefanelli et al. [187] to compare gonococcal ß-lactamase sequences and assign bla_TEM numbering according to the Ambler scheme [186, 187]. Singh et al. [194] characterized a novel ß-lactamase-producing plasmid in N. gonorrhoeae, pJRD20 (Canada type), compared its TEM sequence using the Ambler scheme and identified a unique 6 bp deletion responsible for a unique truncation of TEM.

Phylogenetic analysis and visualization of trees

Because N. gonorrhoeae acquires genetic diversity in its core genome by homologous recombination [195], recombinant fragments must be removed before generating phylogenies; recombined regions within a genome may lead to incorrect phylogenetic trees and inferences. Artifacts that can be introduced include loss of molecular clock, overestimation of the number of mutations and substitution rate heterogenicity, wrong clonal relationships, and star-shaped tree topology with long terminal branches [196]. Genealogies Unbiased By recomBination In Nucleotide Sequences (Gubbins), a tool to identify and remove recombinations from bacterial genomes [197], uses a spatial scanning statistical framework to remove recombined regions by identifying the loci containing elevated base substitution densities. Gubbins is very effective in identifying mosaic recombination and recombined regions presenting mobile genetic elements, and analyzing a large collection of bacterial WGS datasets. Gubbins was used in 37 gonococcal WGS studies [16, 21, 22, 31, 32, 35, 37, 40–42, 46, 49, 61–63, 69, 70, 78, 79, 91–93, 100–102, 105, 107, 114, 116–119, 121, 125, 126, 129, 130] to detect and/or delete recombination events.

ClonalFrame [198, 199], suitable for smaller WGS data sets, establishes the clonal relationship between bacterial isolates by identifying regions with homologous recombination and point mutations. A new implementation of this program, ClonalFrameML, was developed to handle a large WGS datasets. Seven studies used ClonalFrameML [26, 103, 108, 123, 133, 134, 136], two used Galaxy SNVPhyl [89, 90] and one used Harvest Suite (https://github.com/marbl/harvest) [95] to remove recombination.

Phylogenomic trees are generated using Randomized Axelerated Maximum Likelihood (RAxML) [200], PhyML [201], MEGA [202], FastTree [203], IQ-Tree [204], BEAST [205], Phylip [206], Phyloviz [207], Interactive Tree of Life (iTOL) [208], DNAml [201], GrapeTree [209], SplitsTree [210], PathoBacTyper (http://halst.nhri.org.tw/PathoBacTyper/) and RapidNJ [211] (Supplementary Table 2). RAxML[200] uses the maximum likelihood method to build phylogenetic trees from large datasets. Seventy-six of 105 papers performed phylogenetic analyses (Supplementary Table 1). Thirty-two reports utilized RAxML [21–24, 26, 31, 32, 35, 40–42, 46, 60, 61, 69, 77–79, 88, 91, 93, 94, 99, 100, 103, 112, 114, 116–118, 121, 126]; 10 used PhyML [18, 19, 92, 102, 106, 108, 111, 129, 130, 133]; 9 used IQ-Tree [16, 30, 37, 44, 62, 63, 70, 101, 119]; 5 used MEGA [20, 28, 50, 120, 138]; 3 each used ITOL [49, 95, 105], Phylip [27, 134, 142] or Phyloviz [29, 45, 139]; 2 each used BEAST [123, 124] or Galaxy SNVPhyl [89, 90]; and 1 each used DNAml [127], GrapeTree [112], SplitsTree [110], FastTree [43], CLC Genomics workbench [34], PathoBacTyper [17] or RapidNJ [125].

Phylogenetic trees can be visualized using tools such as iTOL [208], Figtree [212], Phyloviz [207], Pathogenwatch [36] and ggtree [213] (Supplementary Table 2). Fifteen studies used iTOL [16, 17, 24, 32, 35, 40, 42, 44, 49, 60, 69, 93, 95, 105, 126]; 8 used Figtree [18, 19, 22, 23, 78, 103, 112, 130], 2 used MEGA [134, 142] and 1 used GrapeTree [125].

BLAST and computational pipelines for analyzing N. gonorrhoeae WGS data

BLAST [214] has been utilized in many analytical tools, pipelines, and databases as outlined previously. However, stand-alone BLAST searches followed by manual examination were used in 34 N. gonorrhoeae WGS studies [18–21, 25, 30, 34, 36, 44–46, 48–50, 61–63, 69, 70, 91, 93, 94, 98, 100, 103, 105, 111, 113, 125–127, 129, 130, 188]. For example, Yahara et al. [130] extracted nucleotide sequences in order to perform in silico MLST typing, penA genotyping, identification of bla_TEM and tetM genes, and to search for recombinant fragments within penA. BLAST was used to compare gonococcal genomes [50], for AMR determinant identification [20, 25, 48, 51, 61, 98], for the detection of mobile genetic elements [79, 129] and for in silico NG-MAST genotyping [18, 48, 51, 91, 94].

Several pipelines and platforms combine multiple WGS analysis steps (Gen2Epi, Nullarbor, INNUca, Sanger Sequencing assembly Pipeline and Pathogenwatch). Gen2Epi automatically links genomic scaffolds to AMR and molecular epidemiology data (NG-MLST, NG-MAST and NG-STAR) for N. gonorrhoeae [48]. Gen2Epi is a command-line tool in which WGS analysis of N. gonorrhoeae is completed in 5 simple integrated options (i.e. data cleaning, de novo assembly, scaffolding, assembly quality assessment and annotation, plasmid-type identification, and AMR and epidemiological analysis). A user-friendly interface for Gen2Epi, Gen2EpiGUI, [51] is available at ftp://www.cs.usask.ca/pub/combi [51]. Novel functions to automatically download and update the underlying databases (i.e. NG-MAST [168], NG-MLST [167] and NG-STAR [169]), which are not available in Gen2Epi, have also been implemented in Gen2EpiGUI [51]. The code for Gen2Epi and Gen2EpiGUI is publicly available at https://github.com/DillonLab and https://github.com/ReemaSingh; GitHub, a code-hosting platform for collaboration and version control, allows researchers from around the world to improve these programs.

Nullarbor [52], using a command-line interface, comprises multiple WGS analysis steps including NG-MLST analysis, AMR prediction, and variant calling, although NG-MAST and NG-STAR analysis is not available. Four studies performed various WGS analysis of N. gonorrhoeae using Nullarbor to investigate gonorrhoea transmission, AMR profiles and to perform various assembly and annotation functions [16, 26, 31, 37]. INNUca (https://github.com/B-UMMI/INNUca) provides analysis functions for quality check, contamination detection, de novo assembly, and quality assessment of assembled contigs and has in-built capacity for MLST analysis. Two studies used the INNUca pipeline [29, 45], and three studies used the Sanger Institute Assembly pipeline [25, 79, 126] for WGS data analysis. Pathogenwatch was used for molecular typing, AMR prediction and whole-genome phylogenetic clustering (and visualization) using epidemiological and phenotypic data [25, 36, 61]; its main limitation is that it does not provide resources for functions such as data cleaning, de novo assembly, NG-STAR analysis and annotation.

Conclusions and recommendations

This review is the first report to catalogue the computational tools used for the analysis of N. gonorrhoeae WGS data. A variety of approaches have been taken for various analyses.

For accurate WGS analysis the following approaches should be taken to (i) clean N. gonorrhoeae WGS reads by removing adapters and contaminated sequences from other microorganisms, (ii) identify and remove recombinant regions from the genome before performing phylogenetic analysis, and (iii) analyze all samples using identical commands and parameters during the processing steps. Authors should consider including complete analysis strategies [including software versions, scripts, or workflows, such as provided by Galaxy (https://galaxyproject.org/) and Jupyter notebooks (https://jupyter.org/)] in the methods section of each manuscript.

Many pipelines require advanced bioinformatics skills (e.g. Gen2Epi, the Sanger Institute Assembly pipeline, INNUca and Nullarbor). If pipelines are not accessed for analysis, multiple individual programs must be implemented to achieve steps such as data cleaning, assembly, gene prediction and annotation, strain typing, and AMR prediction. These programs and pipelines often employ a Linux-based command-line interface using different languages such as R, Perl and C++, which can be difficult for those with minimal training in bioinformatics. Analyses of gonococcal genomes using user-friendly tools such as Gen2EpiGUI, Pathogenwatch, pubMLST simplifies bioinformatics analyses for those with more limited bioinformatics skills. Laboratories that conduct analysis of large data sets, requiring repetitive steps (i.e. de novo assembly and strain typing) may benefit from these pipelines and platforms rather than accessing individual programs. On the other hand, the requirement for transmission and evolutionary analyses may require a unique combination of tools for variant calling, recombination removal and phylogenetic analysis tools, such as SAMtools [161], Freebayes [165], Gubbins [197], ClonalFrameML [198, 199], RAxML [200] and PhyML [201]. For example, Koboldt et al. [215] listed best practices for accurate variant calling in clinical samples, such as removing false positives and manually visualizing the alignment in clinical samples.

Currently, there is no single platform that contains all functions required for complete N. gonorrhoeae WGS data analysis. For instance, Gen2Epi can perform data cleaning, de novo assembly, scaffolding and annotation, plasmid assembly, strain typing and AMR prediction, but analysis steps like variant calling, recombination removal and core genome phylogeny are missing. Likewise, pubMLST integrates third-party tools to perform multiple steps for WGS analysis (i.e. strain typing, core-genome MLST, genome comparison, AMR prediction and network analysis) but lacks data cleaning, assembly and scaffolding processes. The development of novel user-friendly frameworks, particularly for those with limited bioinformatics skills, would facilitate the more widespread N. gonorrhoeae WGS applications in both clinical and research settings.

Key Points

• All (n = 105) published Neisseria gonorrhoeae WGS studies (as of 30 April 2021) were reviewed.

• Computational databases and analytical tools used for N. gonorrhoeae WGS data analysis were summarized.

• Analytical resources for strain typing (i.e. NG-MAST, MLST and NG-STAR) and antimicrobial resistance determination for N. gonorrhoeae WGS data were reviewed.

• Analytical limitations (e.g. missing information, split genes or contaminating reads from closely related species/strains) of some tools were identified.

• Only a few computations pipelines for N. gonorrhoeae WGS analysis are available. The development of more user-friendly bioinformatics tools and algorithms would facilitate the analysis of WGS data from N. gonorrhoeae, especially for clinical diagnostic settings.

Reema Singh is a Postdoctoral Fellow in the Department of Biochemistry, Microbiology and Immunology (College of Medicine) at the University of Saskatchewan.

Anthony Kusalik is a Professor in the Department of Computer Science at the University of Saskatchewan. He is also the Director of the Bioinformatics Program at the University of Saskatchewan.

Jo-Anne R. Dillon is a Distinguished Professor in the Department of Biochemistry, Microbiology and Immunology (College of Medicine) at the University of Saskatchewan. She is also a Research Scientist at the Vaccine and Infectious Disease Organization.

Acknowledgments

This manuscript is published with the permission of the Director of VIDO-InterVac as journal series no. 928.

Funding

This work was partially supported by funding from the Canadian Institutes of Health Research (grant # 148996 to JRD).