Diagnostic value of next-generation sequencing to detect periprosthetic joint infection

Periprosthetic joint infection (PJI) is the most common cause of total knee arthroplasty failure (20.4%) [1] and one of the main reasons for revision after total knee arthroplasty (12.8%) [2]. Such a revision usually demands multiple surgical interventions and has a higher incidence of complications. The mortality rate of patients with PJI undergoing arthroplasty is five times that of patients undergoing aseptic arthroplasty [3, 4].

The main difficulty associated with PJI is prompt and accurate diagnosis, along with the detection of pathogens. Due to complicated reasons such as the existence of biofilms on the surface of artificial joints, inconsistent and nonoptimal processes for sampling pathogenic bacteria, small number of bacteria, and complex symbiotic relationship involving bacteria, pathogens cannot be easily detected in approximately 50% patients with PJI. Compared with culture-positive cases, the rate of infection recurrence is 4–5 times higher in culture-negative cases [5].

Several biomarkers have been studied in recent years to improve the accuracy of diagnosing PJI, such as C-reactive protein (CRP) [6], α-defensin [7], leukocyte esterase [8], interleukin-6 (IL-6) [9], vascular endothelial growth factor [10], and granulocyte colony-stimulating factor [11], but the overall sensitivity (77–97%) and specificity (86–96%) remain inadequate [12]. PCR was once considered a reliable method for diagnosing PJI [13]. Studies on the use of PCR to diagnose PJI continue to be extensively conducted, but the sensitivity of the method has been observed to widely vary. A meta-analysis reported that the overall sensitivity of PCR was only 76%, which is comparable to that of traditional culture [14].

Next-generation sequencing (NGS) is an emerging microbial diagnostic method that can detect all nucleic acids present in a specimen at once, including those from the host. For example, NGS has been reported to successfully detect pathogenic bacteria in samples from patients with nervous system infections [15] and sepsis [16]. Further, it has been used for identifying pathogenic organisms in medical systems and is associated with a good clinical value. We conducted this study because at present, very few studies have explored the use of NGS for diagnosing PJI [5, 17], and thus, its diagnostic value in case of PJI remains unclear.

Based on the MSIS criteria, the infection and non-infection groups included 15 and 20 patients, respectively (Fig. 1). There were no significant differences in age, gender, body mass index, and affected joints between the groups. There were seven cases (46.7%) of hip joints and eight (53.3%) of knee joints in the infection group, and there were 12 cases (60.0%) of hip joints and eight (40.0%) of knee joints in the non-infection group. No statistically significant difference was noted in the composition ratio between the groups (P > 0.05; Table 1).

In the infection group, 14/15 (93.3%) patients showed positive results via NGS and 7/15 (46.7%) had positive bacterial culture results, whereas in the non-infection group, 18/20 (90.0%) patients showed negative results via NGS and 1/20 (5.0%) had bacterial culture results. The mNGS positive results of infection group included 6 cases of Staphylococcus aureus, 2 cases of Pseudomonas aeruginosa, 2 cases of Staphylococcus epidermidis, 2 cases of Streptococcus pyogenes, 1 case of Staphylococcus lugdunensis, and 1 case of Clostridium perfringens. The culture positive results of the infection group included 3 cases of Staphylococcus aureus, 2 cases of Streptococcus pyogenes, and 1 case of Clostridium perfringens,1 case of Staphylococcus. The mNGS positive results of non-infection group included 1 case of Candida glabrata and 1 case of Clostridium perfringens. The culture positive results of the infection group included 1 case of Candida glabrata (Table 2). NGS and bacterial culture results showed significant differences between the infection (P < 0.01) and non-infection groups (P = 0.011).

CRP levels were significantly higher in the infection group [18.3 (8.9–32.2) mg/dL] than in the non-infection group [5.7 (3.7–10.8) mg/dL)] (P = 0.001). The difference in ESR levels was not statistically significant between the groups [32.3 (11.7–50.5) mm/h vs. 17.2 (12.6–22.8) mm/h, P = 0.099]. The levels of IL-6 and PCT were 16.8 (9.5–37.6) pg/mL and 3.5 (0.3–4.8) ng/mL in the infection group, respectively, which were significantly higher than those in the non-infection group [8.4 (3.3–15.5) pg/mL and 0.2 (0.1–0.3) ng/mL, P = 0.007 and P = 0.001, respectively]. There was no statistically significant differences in D-dimer levels between the groups [0.6 (0.2–1.9) mg/L vs. 0.5 (0.2–0.9) mg/L, P = 0.479] (Table 3, Fig. 2).

The area under the curve (AUC) of ESR and D-dimer was 0.667 and 0.572, respectively (P > 0.05). The AUC of CRP, IL-6, and PCT was 0.827, 0.767, and 0.808, respectively (P < 0.05). The corresponding clinical diagnostic cut-off values were 11.2 mg/dL for CRP, 23.05 pg/mL for IL-6, and 2.29 ng/mL for PCT (Table 4, Fig. 3). According to this critical point, the sensitivity of CRP, IL-6, and PCT was 0.667, 0.467, and 0.467, respectively, and the specificity was 0.850, 0.95, and 0.95, respectively. The sensitivity of NGS and bacterial culture was 0.933 and 0.467 (P < 0.05), respectively, and the specificity was 0.90 and 0.95 (P > 0.05), respectively. The accuracy of NGS, bacterial culture, CRP, IL-6, and PCT were 0.91, 0.74, 0.77, 0.74, and 0.83, respectively.

On comparing NGS with CRP, IL-6, PCT, and bacterial culture, the overall test results were statistically significant (P < 0.01). Moreover, when comparing IL-6 and PCT with CRP, no significant differences were found in the overall test results (P > 0.05), and on comparing NGS with CRP, IL-6, PCT, and bacterial culture, the difference in sensitivity was statistically significant (P < 0.05). Compared with CRP, the specificity difference of NGS was statistically significant (P < 0.05), and compared with IL-6, PCT, and bacterial culture, the specificity of NGS was statistically insignificant (P > 0.05; Table 5).

We herein explored the value of NGS for diagnosing PJI. Our results showed that the sensitivity of NGS to diagnose PJI (93.3%) was significantly higher than that of bacterial culture (47%); moreover, NGS showed a higher positive diagnostic rate for culture-negative cases (87.5%). We also compared the diagnostic value of NGS and commonly used serological biomarkers to detect PJI. We found that the accuracy and sensitivity of NGS for diagnosing PJI was significantly higher than those of serological biomarkers.

NGS is widely used to perform metagenetic analyses, and it can simultaneously detect sequences from millions of bases in a single biochemical reaction. In 2005, Margulies et al. performed genome-wide sequencing of Mycoplasma genitalium and Streptococcus pneumoniae, marking the beginning of the “NGS revolution.” [20] Further, in 2014, Wilson et al. used NGS for the first time to diagnose infectious diseases [16]. NGS could reportedly diagnose clinically uncommon pathogenic bacterial infections and markedly improve the detection of clinically important pathogens, revolutionizing microbial diagnosis. In 2018, Tarabichi et al. reported the use of NGS for diagnosing PJI for the first time. Based on the MSIS criteria, they found that the positive diagnostic rate of NGS was 89.3% for infected cases and 25% for non-infected and culture-negative cases [5]. In the present study, the positive diagnostic rate of NGS was 93.3% for infected cases and 5.3% for non-infected and culture-negative cases. This disparity in results can be attributed to differences in threshold standards. Different NGS platforms, reagents, and databases tend to lead to variances, and NGS detection results also greatly vary. NGS can be not only qualitative but also quantitative, and it can detect the sequence number of bacterial fragments in a specimen. Further studies are warranted to explore the diagnostic significance of the sequence number.

At present, bacterial culture remains the standard method for the definitive diagnosis of PJI; however, the sensitivity of this technique is highly inconsistent (58–95%), which can be attributed to the use of antibiotics before sampling, inconsistency in standardization of sampling and specimen transmission, variances in cultivation time, and presence of unusual bacteria [21, 22]. Berbari et al. found that 53% patients with culture-negative PJI had been treated using antibiotics before sampling. On using large quantities of broad-spectrum antibiotics before sampling, bacterial count reduces or secretions contain antibiotics, consequently affecting bacterial growth and propagation [23]. Moreover, the current bacterial culture time may not be sufficiently long, typically ranging from 3 to 5 days, rather than 2 weeks [24]. In addition, other medical institutions use conventional culture media, which makes the cultivation of microorganisms such as fungi and mycobacteria challenging. Studies have reported that 46% culture-negative PJIs are caused by fungal infections and 43% by mycobacteria [25]. Diagnosing culture-negative PJI is very difficult and may involve a comprehensive evaluation with clinical methods, radiology, serology (inflammatory markers), histopathology, and microbiology. We herein found that the sensitivity of the bacterial culture method to diagnose PJI was 47%, which is lower than the data reported in the literature. This could be because among the eight culture-negative patients with PJI, NGS results indicated that one patient was infected with Candida albicans. NGS results also showed that Staphylococcus aureus was the most common pathogenic bacteria in prosthetic joint infections. Further, two patients had a history of treatment using broad-spectrum antibiotics. Our findings indicate that bacterial culture and NGS results were generally consistent.

To the best of our knowledge, such an investigation has not been conducted as yet. The ideal method to screen PJI should be highly sensitive so as to minimize false negatives. Serological markers are still the most commonly used method for the clinical diagnosis of PJI. Our results showed that the AUC of ESR and D-dimer was 0.667 and 0.572, respectively, but the results showed no statistical significance. D-dimer, a coagulation-related indicator, has recently been used as a tool for the diagnosis of periprosthetic joint infection (PJI), but its reliability is uncertain. A meta-analysis reported that D-dimer has limited performance for the diagnosis of PJI [26]. The results of our research are the same as reported in the literature. The sensitivity of CRP to diagnose PJI is 67%, which was lower than that reported by the most recent study (85.1%) [27]. The main reason for this is that in this study, the diagnostic cut-off value calculated by the Youden index for CRP was > 11.2 mg/L, which is higher than the cut-off value recommended by the MSIS (> 10 mg/L). A higher cut-off value can lead to a decrease in sensitivity. IL-6 was initially considered to be a highly sensitive and specific marker of PJI [28]. However, because pertinent studies did not consider the confounding effects of previous antibiotic use and related inflammatory conditions on IL-6 and other inflammatory markers, there may have been a selection bias [29]. PCT is often used as a marker of systemic infection, but its role in the diagnosis of local infections (such as PJI) is limited because the threshold of PCT in patients with local infections substantially overlaps with its normal range [30]. We herein found that with regard to PJI diagnosis, in comparison with CRP, there was no significant difference in the overall test results of IL-6 and PCT. This result is the same as that previously reported in the literature, i.e., studies have not confirmed that IL-6 and PCT are superior to traditional blood biomarkers for diagnosing PJI [31].

As with other methods to diagnose PJI, NGS also has both advantages and disadvantages. The main advantage of NGS is unbiased sampling, leading to the identification of known as well as novel organisms [32]. Further, NGS can provide auxiliary genomic information needed to predict drug resistance [33]. It can also generate quantitative or semiquantitative data pertaining to bioconcentration by counting sequencing reads, which is very useful for samples containing multiple microorganisms or in case of disease processes involving more than one pathogen [34]. The main inherent disadvantage of NGS is that microbial nucleic acids in a sample can be affected by the presence of human nucleic acids. The vast majority of sequences (usually > 99%) come from human hosts, consequently limiting the overall analytical sensitivity of pathogen detection methods [35]. Another potential challenge is that testing samples, reagents, or contaminating microorganisms in the laboratory environment can affect the accuracy of results [36, 37]. In addition, while metagenes can be used to detect pathogenic bacteria, they cannot be used for drug sensitivity tests at the same time, which remains a persistent issue. The third disadvantage is that NGS is not a feasible approach for routine testing, because the price is still high at $500 or more per test.

This study had some limitations: (1) The sample size was relatively small. (2) Choosing the MSIS criteria as the gold standard for diagnosing PJI may cause grouping errors. This can explain the results that bacteria were detected in the samples of two separate patients in non-infection group. (3) Benefit analysis was not performed for NGS.

Our results nevertheless clearly indicated that NGS had higher accuracy and sensitivity than the bacterial culture method and commonly used serological biomarkers, and thus, we report that its diagnostic value to detect PJI is relatively much higher.