Bacterial meningitis and related invasive infections, including pneumonia, bacteraemia and septicaemia, are devastating diseases that represent a major public health concern worldwide (Rodgers et al., 2020), the main bacterial aetiological agents being: Neisseria meningitidis (the meningococcus); Streptococcus pneumoniae (the pneumococcus); Haemophilus influenzae; and Streptococcus agalactiae (group B streptococcus, GBS). Despite the availability of vaccines against many variants of these pathogens (Tsang, 2021), with the exception of S. agalactiae for which intrapartum antibiotic prophylaxis is used (Hughes et al., 2017), meningitis continues to affect populations globally, particularly in sub-Saharan Africa.

In sub-Saharan Africa, meningitis is endemic, with seasonal epidemics occurring unpredictably every 5–12 years (Hughes et al., 2017). A recent example being the meningitis outbreak among miners in the northeastern cities of the Democratic Republic of Congo, declared on 8 September 2021 (WHO Africa, 2022). These epidemics are associated with a high case fatality rate of 50% and many cases remain undiagnosed, increasing the delay in appropriate public health interventions. Control of meningitis requires improved access to and efficiency of diagnostic methods; indeed, diagnosis is one of the five pillars of the World Health Organisation (WHO) roadmap to defeating meningitis by 2030 (World Health Organization [WHO], 2021).

Efficient surveillance, outbreak investigation and clinical management of meningitis depends on laboratory confirmation of the causative pathogen from sterile sites, such as cerebrospinal fluid (CSF) and blood. The “gold standard” methods for confirmation of meningitis remain: (i) culture and (ii) polymerase chain reaction (PCR). Culture allows an isolate to be retained for further use; however, this can take at least 24 h and is often more challenging in sub-Saharan Africa due to long transportation time and/or previous antimicrobial treatment (Diallo et al., 2021). PCR is a rapid molecular diagnostic method that enables identification within a few hours. While PCR is sensitive, specific and does not depend on the presence of viable bacteria, it requires expensive equipment, reagents and expertise (Diallo et al., 2021; Feagins et al., 2020; Griffiths et al., 2018). The success of PCR depends on the presence of genomic regions specific to each pathogen. The genetic diversity of meningitis-associated pathogens (Spratt and Maiden, 1999) and their genomic variability indicates there is an on-going need to monitor the effectiveness of existing molecular diagnostic tests targeting pathogens associated with meningitis. It is also important to continue searching for novel genetic targets that are more sensitive and specific than those currently used, while also considering their genetic variability.

A narrative review identified 25 genetic targets used in the detection of H. influenzae, N. meningitidis, S. pneumoniae, and S. agalactiae (Diallo et al., 2021). Testing these targets in vitro is costly and time-consuming. An alternative is to identify suitable targets in silico through bioinformatic analyses using large genome datasets and then confirming their appropriateness in vitro using a reduced panel of bacterial isolates (van Weezep et al., 2019).

The development of high-throughput whole genome sequencing has led to the creation of genome databases such as PubMLST, which contain bacterial population sequence data and provenance metadata for over 100 species and genera (Jolley et al., 2018). This platform receives thousands of yearly submissions including new sequences, allele profiles and isolate records (Jolley et al., 2018). N. meningitidis, S. pneumoniae, H. influenzae, and S. agalactiae WGS deposited in PubMLST include data from healthy carriers, invasive disease cases and other clinical sources.

This study aimed to evaluate in silico published genetic targets used in PCR assays for the detection of N. meningitidis, S. pneumoniae, H. influenzae, and S. agalactiae and identify optimal genetic targets. These were then tested in vitro using a panel of bacteria strains representative of the genera present.

Bacterial meningitis remains a major public health threat, particularly in sub-Saharan Africa due to unpredictable epidemics and the urgent need to improve diagnostic methods for the rapid and accurate detection of the causative pathogens. This study sought to address these challenges using in silico approaches with PubMLST, a large nucleotide sequence database, providing a preliminary assessment of the specificity and sensitivity of diagnostic targets to guide in vitro validation tests. This approach enables a preliminary assessment of the specificity and sensitivity of diagnostic targets before extensive laboratory testing, significantly reducing the time, effort and costs associated with assay development (Santa Lucia et al., 2020; van Weezep et al., 2019). Promising assays identified in silico can then be validated by various laboratories, including those led by citizen scientists, using their available local strains. This collective effort increases the variety of isolates tested and reduces issues related to sample shipments. Additionally, using PubMLST to both select targets and evaluate their performance for H. influenzae may introduce a risk of overfitting, since the same dataset informs both steps. PubMLST is the largest and most diverse publicly available database for H. influenza genomes. However, it remains important to validate promising targets using independent genomic datasets or clinical isolates to ensure broader applicability and robustness.

Based on our findings, we recommend using sodC for N. meningitidis, cfb for S. agalactiae, SP2020 for S. pneumoniae, and dmsA for H. influenzae due to their high sensitivity, specificity, and consistent prevalence in WGS data. The gene sodC, also recommended by WHO/CDC, is highly specific and sensitive for detecting meningococci (Thomas et al., 2011). This gene, which encodes Cu-Zn superoxide dismutase, is ubiquitous in N. meningitidis and less likely to undergo antigenic variation due to selective pressure (Thomas et al., 2011). This study also revealed the equivalent efficacy of porA (an outer membrane porin). However, rectal and pharyngeal N. gonorrhoeae isolates from Australian and Swedish patients have been found to harbor an N. meningitidis porA sequence, presumably acquired through horizontal genetic exchange and recombination (Golparian et al., 2012; Whiley et al., 2011). Thus, caution is advised when using porA as a target, although detecting N. gonorrhoeae in invasive meningococcal cases would be unlikely.

The gene cfb, which encodes the extracellular pore-forming toxin (CAMP factor), has been demonstrated as an effective target for GBS detection (Carrillo-Ávila et al., 2018). It is noteworthy that other rtPCR tests such as (i) the Becton Dickinson MAX GBS assay; (ii) the ARIES GBS assay from Luminex Corporation; and (iii) the Xpert GBS LB assay produced by Cepheid Inc. also prioritize cfb as the primary target gene (Diallo et al., 2021). However, there is no WHO/CDC recommendation for this gene as GBS is not routinely tested in surveillance.

Despite lytA (the major autolysin of pneumococcus) being widely recommended by WHO/CDC and routinely used in surveillance of S. pneumoniae (Satzke et al., 2013), concerns arise due to its homologs in closely related Streptococcus species (Tavares et al., 2019), potentially increasing the false positivity rate (Ganaie et al., 2021; Llull et al., 2006; Simões et al., 2016). Furthermore, the work of Martín-Galiano and García (2021) revealed that pneumococcal prophages harbor lytA-like genes homologous to S. pneumoniae lytA, and that there were recombination events between the pneumococcal and phage lytA homologs, further questioning its reliability. In contrast, SP2020 (a putative transcriptional regulator gene of the GntR family) has been shown to be a better target than lytA for S. pneumoniae diagnosis, consistent with the findings of Tavares et al. (2019). In this study, lytA was included in the in silico analysis to enable direct comparison with SP2020 due to its common use. Given its limitations, lytA was not evaluated further in vitro. The results support the use of SP2020 as a preferred diagnostic target.

The gene fucK (gene encoding fuculokinase) demonstrated the best overall performance in silico for the identification of H. influenzae. However, this target gene performed poorly in comparison to the ones used for the three other pathogens, suggesting a need for improvement of H. influenzae diagnostic determinants. Furthermore, more than 1% of non-Hi sequences had the fucK gene and this gene did not detect all H. influenzae, as reported by de Gier et al. (2015). We were unable to amplify fucK in vitro using the published conditions (Meyler et al., 2012). Therefore, although in silico analyses suggest that it is a promising target, we cannot confidently assert that it is the best target without experimental validation. It would thus be beneficial to optimize the existing primers or design new ones. In contrast, dmsA performed better than hpd, the gene recommended by WHO/CDC (Wang et al., 2011), suggesting its efficacy for H. influenzae identification. The dmsA gene is required for fitness of H. influenzae (Dhouib et al., 2021) and appears to be the most efficient test for the identification of H. influenzae in our study. Nasreen et al. (2024) showed that the DmsABC complex protects H. influenzae against oxidative stress, particularly from host-derived hypochlorite. Expression of dmsA increases under such stress, and its deletion impairs bacterial survival and intracellular persistence. These findings suggest that dmsA contributes to both stress adaptation and host interaction.

The PubMLST database includes isolates from various sources, some known to be from asymptomatic carriage and others from unspecified sources. Our assays are able to detect isolates from these different sources. Additional analysis was performed to evaluate the efficacy of the targets for bacteria isolated from cases of invasive disease and/or meningitis. All four targets exhibited high sensitivity for the target bacteria from invasive disease (97.2%–99.7%) and for bacteria from meningitis specimens (98.4%–100%). Further evaluation with more specimens from invasive diseases from diverse geographical regions, which was beyond the scope of the present work, would be useful to confirm the results presented here. Although some bacterial variants are currently more prevalent than others, the inclusion of non-invasive isolates in the different test panels remains important. Indeed, these strains have no intrinsic factors that prevent them from causing disease, and also need to be monitored because they can acquire virulent genes especially as they are exposed to new vaccine pressures such as for N. meningitidis and potentially for S. agalactiae. This has been shown with the non-virulent N. meningitidis carriage strain that acquired both a serogroup C capsule and the filamentous bacteriophage MDAΦ, which has been shown to enhance colonization of nasopharyngeal epithelial cells, increasing virulence, and leading to epidemics first reported in 2013 in the Tambuwal area of Nigeria, with the strain spreading to different regions of Niger (Brynildsrud et al., 2018). Similarly, in S. pneumoniae, non-encapsulated strains (NESp) typically cause non-invasive pneumococcal diseases. However, NESp strains have recently been identified as causative agents of invasive disease. Bradshaw et al. (2020) demonstrated that NESp are highly transformable, capable of acquiring large DNA segments that increase their persistence and virulence during invasive disease. Group B streptococci (GBS), commensals of the vagina and gastrointestinal tract, can become invasive, particularly in newborns, through GBS adaptation to environmental changes under the control of the CovRS two-component regulatory system (Patras et al., 2013). Genomic mutations, including those affecting capsule synthesis regulator (CovR), also appear to influence the transition of GBS from a commensal state to a pathogen and its ability to persist in mothers before and after delivery (Shabayek and Spellerberg, 2018). The assays failed to detect the sequences of the target genes analyzed in isolates belonging to certain serotypes or capsules in silico due to complete or partial deletion of these genes (Khatami et al., 2018; Whyte et al., 2020) in some strains and due to the stringent conditions applied (no mismatches allowed in the primers). This issue needs to be monitored in real life, as some missed genotypes, such as serogroups B and C in N. meningitidis, serotypes Ia, Ib, III, and V in S. agalactiae, serotypes 4, 14, 7F, 9V, and 18C in S. pneumoniae, and non-typeable H. influenzae (NTHi), can cause disease (Lambertsen et al., 2010; Levy et al., 2010; Moreno et al., 2020; Resman et al., 2011; Sleeman et al., 2006; Tazi et al., 2011).

Our in silico analysis was conducted using WGS databases, which only include culturable bacteria. All available genome sequences are derived from cultured bacteria, which can limit the diversity captured, especially for strains that are difficult to grow. This approach may introduce bias as the need to culture pathogens for whole genome sequencing (WGS) limits the representativeness of the data. Primer design depends on the available sequence data, which is currently mostly from cultured bacteria. As a result, the diversity of uncultivable strains may be overlooked and PCR may lack the sensitivity to detect them. One solution to improve representativeness would be to use culture-independent sequencing methods, such as metagenomics, which can explore a wider bacterial diversity without relying on specific primers. These allow detection of both culturable and unculturable strains and may reveal additional genetic targets for more sensitive molecular diagnostics. Furthermore, according to the WHO report, laboratory data from weeks 1 to 30 of 2024 (January 1 to July 28) indicated that 4,926 cerebrospinal fluid (CSF) samples tested by PCR out of 7,468 suspected cases were negative, despite strong clinical suspicions of meningitis (World Health Organization [WHO], 2024). These results suggest either the presence of other pathogens that current tests do not detect, or a lack of sensitivity in the current diagnostic methods. In this context, our study is particularly relevant. By improving diagnostic tools, we aim to enhance the detection capacity of the four most virulent meningitis pathogens. However, this approach must be expanded to identify other genes for diagnosing additional pathogens responsible for meningitis, such as viruses or other infectious agents known to be difficult to detect with current methods. WHO data highlights the critical need to develop more sensitive diagnostic tests adapted to the contexts of low- and middle-income countries (LMICs). Given the healthcare challenges posed by the burden of infectious diseases in these regions, implementing tools suited to local conditions is essential.

In conclusion, the genes sodC, cfb, SP2020, and dmsA have allowed for the in silico identification of N. meningitidis, S. agalactiae, S. pneumoniae, and H. influenzae, respectively, from various clinical sources, including invasive cases, and specifically in cases reported clinically as meningitis, and have shown promising results in vitro despite the limited number of samples tested. The diagnostic measures should nevertheless be interpreted with caution given the absence of confidence intervals and formal statistical testing. These genes thus have the potential to significantly enhance the precision of molecular diagnostics for meningitis. However, laboratory confirmation with a larger number of samples, including patient samples such as CSF or blood, remain necessary. Additionally, the performance of these targets in cases of co-infection or samples with low pathogen loads was not assessed in this study due to limited data. Future work should evaluate diagnostic accuracy under these conditions to ensure reliability in diverse clinical scenarios. The in silico approach, utilizing extensive WGS databases such as PubMLST, combined with the in vitro approach, enables efficient and cost-effective preliminary evaluation of diagnostic targets. This can be particularly beneficial in situations characterized by variability in etiological agents and potential changes in their relative prevalence due to collective immunity induced by vaccines, especially in resource-constrained environments.