Infectious keratitis (IK), or commonly known as corneal infection, is the primary cause of corneal blindness globally (1). Depending on the geographical and temporal variations, the incidence of IK has been estimated to be 2.5–799 cases per 100,000 people annually, with a significantly higher incidence and prevalence observed in low- and middle-income countries (1, 2). Timely and accurate diagnosis followed by appropriate treatment is key to achieving a good clinical outcome (3). Studies have shown that visual outcome, corneal healing time and treatment success of IK are significantly influenced by the initial presenting severity of the infection (4–6).

IK is primarily diagnosed on clinical grounds, supplemented by microbiological investigations and/or imaging test (3). Corneal sampling for microscopy (with staining), microbiological culture and susceptibility testing are the current mainstay for IK diagnosis as it can determine the causative organisms and their antimicrobial susceptibility and/or resistance. Microbiological culture can be performed using either direct culture or indirect culture methods, with both showing similar diagnostic yields in many studies (7–10). The advantage of indirect culture is that it is clinically less arduous since it only requires one sampling by the front-line ophthalmologists/doctors (instead of multiple samplings as required in direct culture). The sample is then inoculated in a transport medium for further sub-inoculation on multiple agar plates at a microbiological laboratory, making it a more efficient and practical test in a hectic clinical environment.

However, microbiological culture (in the setting of IK) is hindered by its variably low diagnostic yield due to considerably lower infectious biomass (when compared to systemic infections) (11), potential inadequate sampling, prior use of antimicrobials (12), and poor growth of unusual pathogens (which can be further affected by a wrong choice of culture medium) (3). In addition, a positive culture result may take 2–3 days (and up to 14 days for some slow-growing pathogens) to become available, negatively impacting on the management of IK. As a result, newer molecular techniques such as polymerase chain reaction (PCR) have been increasingly explored and used for diagnosing IK (3, 13, 14).

PCR is a highly sensitive and time-efficient diagnostic test which can typically generate a positive (or negative) result within a few hours, making it an ideal diagnostic modality for IK. Among various assays and techniques, broad-range 16S/18S ribosomal RNA (rRNA)-based PCR has emerged as one of the most popular methods (3, 14–16). 16S rRNA is a major component of the 30S small subunit present in all prokaryotes, including bacteria. It contains multiple highly conserved regions (useful for universal PCR priming) and nine variable regions (V1-V9; pertinent for phylogenetic analysis and taxonomic resolution) (17). On the other hand, 18S rRNA (a core component of 40S small subunit) is highly conserved by eukaryotes, including fungi and Acanthamoeba, and it similarly contains nine variable regions (V1-V9) (18). Therefore, targeting the highly conserved, universal 16S/18S rRNA gene should theoretically identify almost any bacterial or fungal pathogens, including those that are difficult to grow in conventional culture media. In addition, PCR is able to detect microbial DNA in viable and non-viable organisms and is not affected by prior use of antimicrobials (19).

In our recent 12-year Nottingham Infectious Keratitis Study (20), we observed a low culture yield (37.7%) in IK. As a result, 16S/18S rRNA PCR was introduced to the local clinical care pathway in June 2021 as part of the standard IK workup to improve the diagnostic yield. To date, there were only two studies conducted in the UK specifically compared the diagnostic performance between culture and PCR for IK (14, 21). In view of the apparent gap in the literature, this study aimed to compare the performance among direct culture, indirect culture and 16S/18S rRNA PCR for IK and to examine the concordance among these tests.

Timely diagnosis and treatment serve as the key to achieving good outcomes in IK. To overcome the inherent limitations of microbiological culture, PCR test has increasingly gained traction for diagnosing IK in recent years in view of their high sensitivity and high specificity (3). So far, a wide range of PCR techniques have been described for IK, including species-specific PCR (21), semi-nested/nested PCR (31), touchdown PCR (32), multiplex PCR (33), real-time/quantitative PCR (rt-/q-PCR) (34, 35), and broad-range 16S/18S PCR (14, 18, 19, 21, 36, 37). To the best of our knowledge, this is the first study that had directly compared the performance of three different microbiological investigations, namely direct culture, indirect culture, and 16S/18S rRNA PCR, for diagnosing IK. We observed comparable diagnostic yield among all three tests (41.9%−53.0%), with moderate-to-substantial inter-test agreement between culture and PCR (especially in more severe infection cases).

So far, there were limited studies in the UK (14, 21) and few outside the UK (19, 27, 36, 38) that had evaluated the performance of culture and 16S rRNA PCR for bacterial keratitis. When compared to Somerville et al. study (14), we observed a considerably higher diagnostic yield for both culture (53.0% vs. 23.0%) and PCR (41.9% vs. 26.0%) methods in our study. The discrepancy might be attributed to different proportion of patients with prior antimicrobial use, heterogeneous patient cohorts (with different presenting severity of infection), and corneal sampling techniques. There was a significantly higher use of prior antimicrobial reported in Somerville et al. study (34.0%) as opposed to our study (8.1%), which might have negatively impacted on the diagnostic yield in their study. Similar to their study, our study did not demonstrate any superiority in the diagnostic performance of PCR over culture in cases with prior antimicrobial use. Although the performance of PCR is reported to be less affected by prior use of antimicrobials than microbiological culture, the impact may still be clinically significant, particularly when there is a long interval between antimicrobial exposure and sample collection (39, 40). In addition, there is a significantly lower infectious biomass in IK (i.e., the entire corneal surface area is only 1.3 cm²) as opposed to systemic infections (e.g., pneumonia or sepsis), and the constant tear drainage and increased tearing in IK setting might further dilute the microbial load, negatively impacting on the diagnostic yield (3).

Since the publication of our previous studies (20, 23), we had also implemented further training on corneal sampling and a change in the sampling technique/instrument (from needle/blade to flocked swabs, known to improve the diagnostic yield) (25, 41), which might have accounted for the improvement in the culture yield by 15% within our practice (from 37.7% to 53.0%). In addition, previous studies, including our Nottingham studies, have shown that culture positivity is significantly influenced by the initial severity of the infection (4, 5, 42). As the clinical severity of IK was not reported in Somerville et al. study, direct comparison with their study was not possible from this aspect.

Based on the reference standard used (either direct culture standard or composite reference standard), 16S rRNA PCR was shown to exhibit a good sensitivity (73.5%−74.3%) and specificity (90.3%−100%) for diagnosing bacterial keratitis in our study. This was comparable to the previous findings reported in the literature, with a sensitivity of 63.6%−100% and specificity of 67.5%−100% (3). Interestingly, a recent UK study by Hoffman et al. reported a significantly lower sensitivity in PCR (25.0%) than microbiological culture (95.6%) for diagnosing bacterial keratitis (21). As both studies utilized the same commercial company (MicroPathology Ltd, Coventry, UK) for PCR-Sanger sequencing, the higher sensitivity of PCR observed in our study is likely attributed to the differences in corneal sampling techniques, sample transportation method, and inclusion criteria/clinical threshold for performing corneal sampling for PCR.

One of the inherent issues with high sensitivity of PCR is that the test can produce false positive results as the test picks up environmental or internal contaminants (e.g., from latent host DNA); (19, 38, 43) for instance, some universal primers of 16S rRNA genes may amplify eukaryotic rRNA genes non-specifically, producing false positive results (44). Somerville et al. (14) demonstrated 16S rRNA amplicons in negative controls, highlighting the possibility of background contamination resulting in false positives. To address this issue, the PCR total cycle number was set at 30 (but not higher) to avoid false positive results, which could potentially explain the significantly lower rate of Propionibacterium spp. (a common ocular surface commensal) diagnosed by the PCR (2.8%) when compared to the direct and indirect culture methods (12.2%−22.2%). Despite having the lowest but comparable sensitivity, 16S rRNA PCR was able to improve the diagnostic yield in direct culture-negative cases by ~10%, highlighting its adjuvant role for the diagnosis of IK.

On the other hand, some studies have shown that PCR has a higher yield than culture/smear, which might have been due to the differences in types of infection (e.g., ocular vs. non-ocular infections), sampling techniques and sample transport/storage methods, laboratory techniques (e.g., choice of universal primers, extraction and amplification techniques), and patient cohort (e.g., varied clinical severity and types of infectious keratitis) (16, 19, 27, 38, 40). Although our study was set out to include fungal keratitis cases, none of the culture or PCR yielded any fungal pathogen, limiting the interpretation of 18S rRNA PCR for diagnosing fungal keratitis. That said, the lack of false positive results and the perfect concordance between culture and 18S rRNA PCR suggests a high specificity of this test for ruling out fungal keratitis (though the sensitivity remains to be elucidated in our patient cohort).

In addition, we observed a very high percent agreement (>80%) between culture and PCR. Although good PCR-culture concordance (77.0%−90.6%) has been previously reported (14, 36), low concordance has also been noted in some retrospective studies (21, 45). It is noteworthy to highlight that almost all studies reported test concordance in the form of percent agreement. While percent agreement serves as an easy way to interpret inter-test agreement rate, such analysis does not consider chance agreement and may erroneously overestimate inter-test agreement (30, 46). In view of this issue, we performed additional Cohen's k analysis, which showed substantial pairwise agreement on microbiological results among all three tests (mean k = 0.67–0.72). To dissect the potential reasons influencing test concordance, we further performed clinico-microbiological correlation, which showed that positive culture-PCR concordance was most likely to be achieved in more severe IK cases, corroborating the findings observed in the Shimizu et al. study (40). This suggests that 16S rRNA PCR may have more additional clinical values in culture-negative IK cases or those that have lesser severity of infection where it can provide additional information that may not be detected by the culture method.

Another strength of this comparative study lies in the inclusion of an indirect culture group. The reasons for including an indirect culture group were twofold. First, we aimed to explore the diagnostic potential of indirect culture in replacing the direct culture method (which was the gold standard in our practice and many others) as indirect culture is more time-efficient since it only requires one sampling at the front-line ophthalmic setting, with further sub-inoculation being done at the microbiology laboratory. Our study demonstrated similar diagnostic performance and inter-test concordance between direct and indirect culture, supporting its potential use in a busy clinical setting. This also supports the findings of previous similar studies (8, 9). Secondly, both indirect culture and PCR test were performed based on the same sample from the inoculated transport medium, allowing a direct comparison without any influence of the sampling sequence. Although a few studies have demonstrated that the sequence of sampling did not affect the diagnostic yield (14), the theoretical impact and concern of having less infectious biomass in later samplings still exist. Therefore, having an indirect culture group (which utilized the same inoculated sample for PCR) addressed this issue.

One of the limitations of this study was the moderate sample size, an issue that was commonly noted in many previous studies (14, 19, 27, 38). However, the number of cases included in our study (n = 65 samples per test group) satisfied the minimum required sample size (n = 61 per test group) based on our calculation. Our study observed a moderate diagnostic yield in both culture and PCR methods (ranged 40%−50%), which might be related to the inclusion of less severe infection cases (~60% of the cases have an infiltrate size of < 3 mm). Nonetheless, the diagnostic yield was higher than our previous study (37.7%), (20) suggesting an improvement of our sampling technique. A cost-benefit analysis of PCR use in IK diagnosis would also be beneficial as earlier accurate diagnosis could improve clinical outcomes and reduce the risk of complication and need for surgical interventions. For example, Knight et al. (47) investigated three different PCR and culture combinations to diagnose Enterobacteriaceae infection and found the combination of PCR and culture to be the most cost-effective. In the future, multiplex PCR or metagenomic next generation sequencing may also further improve the diagnosis of IK, though its routine use in clinic is currently limited by the cost and the availability of technical expertise and resources (11, 48–51). Artificial intelligence may also serve as a valuable tool for assisting the diagnosis of IK in the future, though robust validation of such technology is still required before deployment to the clinical settings (52).

In conclusion, this study demonstrated good diagnostic performance of all three different microbiological diagnostic modalities, namely direct culture, indirect culture and PCR. We also highlight the potential adjuvant role of 16S rRNA PCR for bacterial keratitis, particularly in culture-negative and less severe IK cases. However, larger studies and cost-benefit analysis are required to determine its role for routine use in future clinical practice.

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

The studies involving humans were approved by the Clinical Governance team at the Nottingham University Hospitals NHS Trust. The studies were conducted in accordance with the local legislation and institutional requirements. The ethics committee/institutional review board waived the requirement of written informed consent for participation from the participants or the participants' legal guardians/next of kin because corneal sampling formed part of the standard clinical care for infectious keratitis, therefore no additional informed patient consent (over and above the usual verbal consent for corneal sampling) was required.

YH: Data curation, Formal analysis, Investigation, Methodology, Project administration, Visualization, Writing – original draft. LS: Data curation, Methodology, Writing – review & editing. ZO: Data curation, Investigation, Writing – review & editing. ML: Data curation, Formal analysis, Methodology, Project administration, Writing – review & editing. IM: Methodology, Writing – review & editing. DJIT: Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – review & editing. JC: Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – review & editing. JH: Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – review & editing. DS: Conceptualization, Investigation, Methodology, Project administration, Resources, Writing – review & editing. HD: Conceptualization, Investigation, Methodology, Project administration, Resources, Writing – review & editing. DSJT: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Visualization, Writing – original draft.