Research and analysis

Supplementary information: prediction of ciprofloxacin resistance from whole-genome sequencing data

Updated 13 November 2025

Applies to England and Wales

Analyses of whole-genome sequencing (WGS) and phenotypic data from 4,633 isolates from the Gonococcal Resistance to Antimicrobials Surveillance Programme (GRASP) indicated that ciprofloxacin resistance can reliably be predicted for Neisseria (N.) gonorrhoeae based on amino acid substitutions at GyrA codon 91.

Background

In N. gonorrhoeae, ciprofloxacin resistance is mainly attributed to mutations in the quinolone resistance-determining regions (QRDRs) of gyrA and parC. Although less common, resistance has also been attributed to mutations in parE and the norM promoter. The UK Health Security Agency (UKHSA) Sexually Transmitted Infections Reference Laboratory (STIRL) sought to determine whether WGS can be used to predict ciprofloxacin resistance for N. gonorrhoeae, providing an alternative to phenotypic testing for GRASP.

Methods

GRASP isolates collected in 2020, 2022, and 2023 that had both ciprofloxacin susceptibility testing results and WGS data, including full length sequences of gyrA, parC, parE, and the norM promoter, were included. Breakpoint plates were in use for ciprofloxacin susceptibility testing between 2020 and 2023, with isolates categorised as susceptible (minimum inhibitory concentration (MIC) ≤0.06 mg/L) or resistant (MIC >0.06 mg/L).

Isolates were partitioned into test (2022 to 2023) and validation data sets (2020). Exploratory data analyses were carried out on the test data set to characterise amino acid substitutions in GyrA, ParC, and ParE, and mutations in the norM promoter and their corresponding patterns of phenotypic ciprofloxacin susceptibility. Results were cross-referenced against known resistance determinants and a subset of substitutions considered for use as potential predictors of ciprofloxacin resistance. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy were calculated along with Wilson score 95% confidence intervals (CIs). Benchmarking was carried out on the validation data set to assess performance.

Results

Data sets

The test and validation data sets included 3,121 and 1,512 N. gonorrhoeae isolates, respectively (Table 1). Two isolates from 2023 were excluded because gyrA (n=1) or parC (n=1) sequences were not detected in the assembled genomes.

Table 1. Details of test and validation data sets

Data set (year)	Published GRASP isolates (n)	Published GRASP isolates with WGS (n)	Analysed isolates (n (% of all isolates))
Validation (2020)	1,534	1,512	1,512 (98.6%)
Test (2022)	1,460	1,409	1,409 (96.5%)
Test (2023)	1,762	1,714	1,712 (97.2%)

Source: Data from GRASP sentinel surveillance system.

Identification of predictors of ciprofloxacin resistance in the test data set

No resistance-associated mutations were identified in parE or the norM promoter, so these loci were excluded from further analyses. As shown in Figure 1, combinations of amino acid substitutions at key sites in the QRDRs of GyrA and ParC and associated patterns of phenotypic susceptibility were visualised to identify predictors of resistance. Consistent with previous studies, the amino acid present at GyrA codon 91 was predictive of resistance. A total of 1,816 out of 1,831 (99.1%) resistant isolates carried the GyrA S91F substitution compared to 1 out of 1,290 (0.1%) susceptible isolates.

Although present at a low frequency, GyrA S91I was also observed exclusively among resistant isolates (n=12, Figure 1). In contrast, wildtype GyrA S91 was identified in 1,287 out of 1,290 (99.8%%) susceptible isolates and just 2 out of 1,831 (0.1%) resistant isolates. The remaining GyrA and ParC substitutions were not detected independently of GyrA S91F, were observed in both susceptible and resistant isolates at low frequencies (for example, GyrA S91G), or were exclusively or primarily observed in susceptible isolates (Figure 1).

Phenotypic ciprofloxacin susceptibility results were confirmed for 3 discrepant isolates with the following combinations of substitutions (Figure 1):

• GyrA S91F, GyrA D95A, and ParC S87R (susceptible)
• wildtype at all key sites (resistant)
• wildtype at GyrA S91 and ParC S87R (resistant)

There is conflicting evidence in the literature regarding the impact of a sole ParC S87R substitution; therefore, phenotypic ciprofloxacin susceptibility testing was also repeated for susceptible isolates in this category (n=77). Results were confirmed for all but 2 isolates that were no longer viable, suggesting that ParC S87R is not sufficient to confer resistance in this setting.

Figure 1. Combinations of GyrA and ParC amino acid substitutions and associated ciprofloxacin phenotypes for the test data set (2022 to 2023; n=3,121) [note 1]

Source: Data from GRASP sentinel surveillance system.

Note 1: the central heatmap depicts the presence (black) or absence (white) of substitutions at GyrA codons 91 and 95 and ParC codons 86 to 91 with rows representing individual substitutions and columns combinations thereof. The lower heatmap indicates ciprofloxacin antimicrobial susceptibility testing results for each combination of substitutions.

Benchmarking use of GyrA codon 91 as a predictor of ciprofloxacin resistance

A simple classification scheme was implemented to predict ciprofloxacin resistance in N. gonorrhoeae based on the amino acid detected at GyrA codon 91:

• gyrA not detected: quality control failure, no susceptibility prediction provided

• serine (S): wildtype, predicted susceptible
• phenylalanine (F) or isoleucine (I): non-wildtype, predicted resistant based on test and validation data sets
• other amino acids: non-wildtype, conservatively predicted resistant with a warning that phenotypic testing is required for confirmation

Benchmarking indicated that the GyrA codon 91 classification scheme performed well, with sensitivity of 99.9% and 97.6% in the test and validation data sets, respectively, and specificity of 99.8% and 100% (Table 2).

Table 2. Results of benchmarking use of GyrA codon 91 as a predictor of ciprofloxacin resistance based on GRASP test and validation data sets

Data set (n)	Sensitivity (95% CI)	Specificity (95% CI)	PPV (95% CI)	NPV (95% CI)	Accuracy (95% CI)
Test (3,121)	99.9% (99.6% to 100%)	99.8% (99.3% to 99.9%)	99.8% (99.5% to 99.9%)	99.8% (99.4% to 100%)	99.8% (99.6% to 99.9%)
Validation (1,512)	97.6% (96.1% to 98.5%)	100% (99.5% to 100%)	100% (99.4% to 100%)	98.1% (97.0% to 98.9%)	98.9% (98.3% to 99.3%)

Source: Data from GRASP sentinel surveillance system.

The classification scheme was also applied to each year individually to compare the proportion of resistant isolates identified by phenotypic ciprofloxacin testing and predicted based on WGS data (Table 3). The largest difference was 1.0%, observed for the 2020 collection.

Table 3. Number and percentage of N. gonorrhoeae isolates in GRASP with phenotypic and inferred ciprofloxacin resistance, England and Wales, 2020 and 2022 to 2023 [note 2]

Ciprofloxacin susceptibility method (breakpoint (mg/L))	2020 n=1512	2022 n=1409	2023 n=1712
Agar dilution, breakpoint plates (>0.06)	666 (44.0%)	828 (58.8%)	1003 (58.6%)
Inferred based on GyrA codon 91	650 (43.0%)	828 (58.8%)	1004 (58.6%)

Source: Data from GRASP sentinel surveillance system.

Note 2: data is shown for isolates with both phenotypic ciprofloxacin susceptibility testing results and WGS data. The overall reported ciprofloxacin resistance was, however, very similar for each year: 44.3% in 2020, 58.6% in 2022, and 58.6% in 2023.

Conclusions

Predicting ciprofloxacin resistance from WGS based on a simple GyrA codon 91 classification scheme appears to be reliable and could replace GRASP phenotypic testing. Phenotypic testing will, however, be required to confirm results for unusual amino acids detected at GyrA codon 91.