Patient characteristics

During the study period, 246 patients with sepsis were admitted to the ICU, 42 of whom had pus samples collected and sent for culture and mNGS. Seven patients were excluded because the interval between sample culture and mNGS was more than 24 h, and 35 patients were finally included in the study. Table 1 presents the clinical characteristics of these patients. Sum of all comorbidities is higher than 100%, which is caused by more comorbidities in some of the patients. For example, hypertension, digestive system disease and diabetes are the three most common complications in this study, which were often combined in the same patient. Sepsis caused by intra-abdominal infection (37.1%) was the most common among patients included in this study, followed by thoracic (28.6%) and skin and soft tissue (22.9%) infections. Upon ICU admission, the acute physiology and chronic health evaluation II (APACHE II) and sequential organ failure assessment (SOFA) scores were 19.1 ± 4.6 and 8 (4, 10), respectively. Thirty-four (97.1%) patients had already received anti-infective therapy at the time of mNGS. Nine (25.7%) patients died in the ICU.

Table 1 Clinical characteristics of the 35 patients included in this studyIdentification of Pathogens using the mNGS and culture methods

For culture methods, a positive result was reported in 19 patients (54.2%), whereas the mNGS results were positive in all 35 patients. A total of 125 potential causative pathogens were reported in 35 pus samples by mNGS, of which three were considered false positives, including two cases of Ralstonia and one case of Mycobacterium avium, which were considered contamination. Although mNGS reported the presence of one case of Epstein-Barr virus and one case of Cytomegalovirus in pus specimens from two patients, neither was considered a causative pathogen. Culture detection methods revealed 29 pathogens among 20 patient samples, of which six were considered false positives, including one Klebsiella pneumoniae, one Streptococcus anginis, and one Aeromonas hydrophila, which were assumed to be species identification errors, and one strain of Escherichia coli, one strain of Acinetobacter baumannii, and one strain of Sphingomonas paucimobilis, which were considered to be contamination. After comprehensive analysis, a total of 120 pathogenic bacteria were identified by the final gold standard, including 112 bacterial strains (53 aerobic or facultative anaerobes, 59 obligate anaerobes) and seven fungi. In addition, one strain of oral Mycoplasma was detected.

Figure 1 lists all bacteria and fungi (at the genus level) identified by the gold standard. Bacteria were detected by mNGS in all specimens except for one Streptococcus, whereas culture detected only 23 out of 112 bacterial strains. There was a statistically significant difference in the positive rate of bacterial detection between the two methods (P < 0.001). Among aerobes and facultative anaerobes, Streptococcus was the most frequently detected bacterial genus (n = 16, 30.2%), followed by Enterococcus (n = 8, 15.1%) and Klebsiella (n = 6, 11.3%). Specifically, the positive rate of mNGS in detecting Streptococcus (93.8% vs 37.5%; P = 0.012) and Enterococcus (100.0% vs 0.0%; P = 0.021) was higher compared with culture methods. Although the difference was not statistically significant, mNGS detected all Klebsiella, whereas the culture method only detected 1 of 6 patients. Only 11 obligate anaerobes were detected by the culture method, whereas all of the obligate anaerobes were reported by mNGS. Among them, Bacteroides was the most frequently detected bacterial genus (n = 15, 25.4%), followed by Prevotella (n = 13, 22.0%). Specifically, the positive rate of mNGS in detecting Bacteroides (100.0% vs. 26.7%; P = 0.026), Prevotella (100.0% vs. 23.1%; P = 0.044) and Fusobacterium (100.0% vs. 33.3%; P = 0.025) was higher compared with culture methods. The seven fungi detected were all Candida species (six Candida albicans and one Candida glabrata). mNGS detected three strains of Candida, culture also detected three strains of Candida, but only one case of Candida was reported by the two methods at the same time, and the rates of detection did not show a statistically significant difference (P = 1.000).

Fig. 1

The overlap of positivity between mNGS technique and culture method for different pathogens. Due to the excessive classification at the species level, the genus level is used for classification. Blue indicates detection by mNGS only, gray indicates detection by culture method only, and orange indicates detection by both methods simultaneously. *The pathogens were observed to have a higher positive rate by mNGS than that by culture method, and the difference was significant (P < 0.05)

Diagnostic performance comparison of mNGS and culture

Using the gold standard results as a reference, the diagnostic efficacies of mNGS and culture methods for aerobic or facultative anaerobic, obligate anaerobic, and fungal infections were compared (Fig. 2). It was found that mNGS increased the sensitivity of diagnosing aerobic or facultative anaerobic infections from 44.4% to 94.4%, and the accuracy rate increased from 71.5% to 97.2%; mNGS also increased the sensitivity of diagnosing obligate anaerobic infections from 52.9% to 100.0%, and the accuracy rate increased from 77.1% to 100.0%. Moreover, the specificity of mNGS remains at 100%, irrespective of whether they are bacteria or fungi. However, mNGS did not show any advantage in terms of fungal infections.

Fig. 2

Diagnostic performance comparison of mNGS and culture. Two by 2 contingency tables showing the diagnostic performance of mNGS and culture method with gold standard results as the reference standard for aerobic and facultative anaerobic infection (A), obligate anaerobic infection (B) and fungal infection (C). Accuracy refers to how many ratios are correct in all judgments, calculated using the formula (True Postive + Ture Negative True Positive + Ture Negative + False Positive + False Negative). GS, gold standard; PPV, positive predictive value; NPV, negative predictive value

Mixed infection identified by mNGS and culture

Infection with more than one pathogen is defined as a mixed infection. Out of the 35 patients, culture diagnosed a mixed infection in only one patient, whereas mNGS identified two (5.7%) patients with two infecting pathogens, seven (20.0%) patients with three infecting pathogens, and 15 (42.9%) patients with at least four infecting pathogens (Supplementary Table 1). Patients can be infected with up to nine pathogens. The most common type of mixed infection is bacteria-bacteria, followed by bacteria-fungus. In addition, 15 patients (42.9%) had a mixed infection comprising non-obligate anaerobes and obligate anaerobes.

Concordance between the mNGS and microbiological culture methodologies

For bacterial and fungal detection, mNGS and culture methods both gave positive results in 19/35 (54.3%) cases. For three patients, consistent results were obtained with both detection methods, whereas completely inconsistent results were obtained for 3 patients. For the remaining 13 patients, “partially matched” results were obtained, indicating that at least one pathogen was detected by both methods (Fig. 3).

Fig. 3

Concordance analysis between mNGS and culture method for bacterial and fungal detection. For the double-positive subset, the results of the two methods were divided into completely matched, partial matched (at least one pathogen detected by the two methods overlapped), and completely mismatched

Impact on antibiotic treatment

Twelve patients underwent antibiotic adjustment after the mNGS results were obtained, all of which were beneficial adjustments, with nine patients initiating targeted therapy and the remaining three patients successfully de-escalating therapy (Supplementary Table 2).

Source of microorganisms

Some bacteria (especially obligate anaerobes) have specific natural habitats in the human body. We observed that site-specific bacteria tended to be detected in pus in groups and inferred from this that the source of the infectious bacteria could be identified in 20 patients, including 12 cases of odontogenic infection, seven cases of gut-derived infection, and one case was infected by bacteria from vagina. Interestingly, the odontogenic bacteria all caused empyema (n = 7) or skin and soft tissue infections (n = 5), whereas the gut-derived microbes all caused intra-abdominal infections.

Comparison of sepsis caused by obligate and non-obligate anaerobic bacteria

Among the 35 patients, 17 patients were mainly infected with obligate anaerobic bacteria. Using patients infected with non-obligate anaerobes (n = 18) as controls, we compared the disease severity at ICU admission, treatment course, and prognosis between the two groups (Table 2). The results showed that there was no significant difference in white blood cell count, neutrophil percentage, and C-reactive protein between the two groups at admission, but the SOFA score [9.0 (7.5, 14.3) vs. 5.0 (3.0, 8.0), P = 0.005] and procalcitonin value [4.7 (1.8, 39.9) vs. 2.5 (0.7, 8.0), P = 0.035] in the non-obligate anaerobic infection group were significantly higher than those in the obligate anaerobic infection group. The proportion of septic shock (66.7% vs. 35.3%, P = 0.044) and acute liver injury (66.7% vs. 23.5%, P = 0.018) in the non-obligate anaerobic infection group was higher than that in the obligate anaerobic infection group. In addition, the ICU mortality (38.9% vs. 11.8%, P = 0.121) and 28-day mortality (44.4% vs. 11.8%, P = 0.06) of patients in the non-anaerobic group were also higher than those in the anaerobic group, but the difference was not statistically significant.

Table 2 Comparison of clinical characteristics of non-obligate anaerobic infection and obligate anaerobic infection groups