Comprehensive analysis of microorganisms in severe septic patients in the intensive care unit by mNGS and microbial culture

Abstract

Background

Severe sepsis is a life-threatening condition involving dysregulated systemic inflammatory responses and acute organ dysfunctions. Timely and accurate pathogen identification is critical for the effective treatment of severe septic patients in the intensive care unit (ICU). Although metagenomic next-generation sequencing (mNGS) enables the sensitive and unbiased detection of pathogens, its clinical implications in identification of causative pathogens, the association with the outcomes and the development of treatment regimens in such patients remain underexplored.

Methods

184 clinical samples were collected from 81 severe septic patients and subjected to mNGS analysis. Blood and bronchoalveolar lavage fluid (BAL) samples were collected from the majority of patients, while sputum, cerebrospinal fluid (CSF), pleural effusion (PE), ascites, urine, hydropericardium (HPC) and blister effusion (BE) samples were also collected from select patients. Microorganisms were detected by DNA and RNA mNGS and the top 5 microorganisms detected by mNGS in each sample were used for analysis. Microbes were isolated from most patients and the isolates were tested for drug susceptibility.

Results

mNGS identified 183 top 5 microorganisms, with bacteria (92.3%), viruses (3.3%) and fungi (2.7%) as the major detected microbes. Among them, 40, 98 and 45 were pathogenic (21.9%), opportunistic (53.6%) and non-pathogenic (24.6%), respectively. Significantly more pathogenic microbes were detected in the sputum (83.3%) and the bronchoalveolar lavage fluid (BAL; 75.0%) than the blood (36.1%) by RNA mNGS (87.4%) than DNA mNGS (58.2%). Patients having the top-1 pathogenic microorganism detected by DNA mNGS or 4–5 pathogenic microorganisms detected by RNA mNGS had a poor association with clinical outcomes. Moreover, detection of R. pickettii, C. difficile and S. enterica were significantly associated with high mortality. The majority of patients (89.5%) were positive for microbial cultures. Each of these patients had at least one drug-resistant organism and nearly half (45.1%) were infected with two or more drug-resistant strains.

Conclusions

Detection of predominant pathogenic microorganisms or particular bacteria in the sputum or BAL samples by mNGS are associated with poor clinical outcomes among severe septic patients in ICU in this cohort. The high prevalence of multidrug-resistant bacteria among these patients underscores the importance of integration of mNGS with antimicrobial susceptibility assessment in the clinical practice to develop the most effective treatment regimens.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12866-026-04730-4.

Background

Severe sepsis is a life-threatening complication characterized by systemic inflammatory response and acute organ dysfunction. It accounts for approximately 20% of global mortality annually [1, 2]. The mortality rate of severe septic patients in the intensive care unit (ICU) is nearly up to 30%, more than twice that of non-septic patients [3, 4]. Thus, it is critical to diagnose and treat these patients. The predominant cause of sepsis is due to infections by a variety bacteria but it can also by viral, fungal, and sporadic parasitic infections [5]. Because of the large number of different kinds of causative pathogens, it has been challenging to determine which microbes cause severe sepsis in a patient using conventional methods.

Conventional microbial culture (CMC) has been widely used for the diagnosis of sepsis, but nearly half of samples are often negative for CMC [6, 7]. Three serious limitations of CMC are poor sensitivity in detecting certain types of bacteria (such as fastidious or anaerobic pathogens), long turnaround time and frequent contamination from normal flora [8–10]. To address these limitations, other approaches such as the polymerase chain reaction (PCR) and mass spectrometry have been used to detect the causative microbes. PCR allows for more sensitive and rapid detection of pathogens. However, the requirement of a large number of specific primers for detection of a plethora of microbes impedes its clinical use. The mass spectrometry method is hindered by microbial culture which is imperative for enriching detection materials [11, 12]. Because of the lack of methods to accurately and quickly detect pathogens that cause sepsis, the majority of severe septic patients are often empirically treated with antibiotics, often resulting in septic shock and sepsis-associated organ dysfunction [13]. Consequently, there is an urgent need for approaches with superior efficiency and precision in identifying the causative pathogens to significantly improve the outcomes of severe septic patients [14].

Recently, the metagenomic next-generation sequencing (mNGS) has shown the great promise for the unbiased, sensitive and accurate detection of pathogenic microorganisms in septic patients [15]. Importantly, mNGS can simultaneously detect a wide variety of microorganisms in a single sample. It does not rely on microorganism culture or prior knowledge of the pathogens and thus can serve as a promising tool to significantly improve the detection of microorganisms in sepsis and other infectious diseases [16]. Moreover, mNGS is highly sensitive in detecting even limited copies of microbial genomes, with a rapid turnaround time of 1 to 2 days [17]. Different kinds of clinical samples like blood, bronchoalveolar lavage fluid (BAL), sputum and cerebrospinal fluid (CSF) can all be used for the mNGS analysis [18]. However, since a large number of predominant and subdominant microorganisms can be simultaneously detected by mNGS, it has been challenging to determine which of them are responsible for the clinical severeness and outcomes of severe septic patients.

In this study, we conducted mNGS analysis of severe septic patients in ICU to investigate the most frequently detectable microorganisms, the influence of sample sources and genome types on detection of microorganisms and the association of sepsis outcomes with particular microorganisms. Our results will be informative for future development of prognostic tools of sepsis and treatment regimens.

Methods

Clinical samples

A total of 184 specimens were collected from 81 severe septic patients who were admitted to the ICU in the First Affiliated Hospital of Jinan University from October 2020 to August 2021. These specimens were collected from blood, BAL, sputum, CSF, pleural effusion (PE), ascites, urine, hydropericardium (HPC) and blisters effusion (BE). The criteria for enrollment were: (i) the sepsis 3.0 criteria jointly issued by the Society of Intensive Care Medicine and the European Society of Intensive Care Medicine [1], (ii) sepsis-related organ failure assessment (SOFA) score ≥ 2 [19], (iii) ≥ 20 years old, (iv) both CMC and mNGS test results and (v) agreement for mNGS analysis. Clinical outcomes were categorized in two groups: improvement for those survived to hospital discharge and death for those died in hospital. The written consent was obtained from all patients or their authorized family members, and the study was approved by the Institutional Review Board and Ethical Committee of the First Affiliated Hospital of Jinan University (KY2023-047).

Metagenomic next-generation sequencing

The mNGS analysis was carried out for the clinical samples as described in literature [20]. Briefly, genomic DNA was extracted from the specimens using the QIAamp^® UCP Pathogen DNA Kit (Qiagen, Hilden, Germany) according to the manufacturer’s protocol. Total RNA extraction was performed using the QIAamp^® Viral RNA Kit (Qiagen) and ribosomal RNA was removed using the Ribo-Zero rRNA Removal Kit (Illumina, San Diego, USA). The obtained RNA was used for cDNA synthesis using reverse transcriptase and dNTPs (Thermo Fisher Scientific, Waltham, USA). Libraries were constructed with the prepared DNA and cDNA materials using the Nextera XT DNA Library Prep Kit (Illumina). The library quality was assessed using the Qubit dsDNA HS Assay kit and analyzed using the High Sensitivity DNA kit (Agilent Technologies, Santa Clara, USA) on an Agilent 2100 Bioanalyzer. Library pools were then loaded onto an Illumina Nextseq 550Dx sequencer for 75 cycles of single-end sequencing to generate raw sequence reads for each library.

No-template controls (NTCs; sterile deionized water) were processed concurrently with clinical samples, following the identical protocol. Low abundance of only two microbes (Enterococcus faecium and Staphylococcus aureus) was detected in 6 of 10 NTCs. Enterococcus faecium was detected in all 6 NTCs while Staphylococcus aureus was detected in only two NTCs as a minor form. Importantly, the raw reads were less than 100 for each microbe in each NTC. More importantly, none of the other 181 taxa detected in our study were detected in NTCs.

The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2021) in National Genomics Data Center (Nucleic Acids Res 2022), China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (GSA: CRA021369) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa [21, 22].

Bioinformatics analyses

Raw reads from mNGS were preprocessed by FastQC to remove short and low-quality reads. The human genomic DNA sequence was identified and removed using KneadData 0.35 (https://github.com/biobakery/kneaddata) with the GRCh37 human reference genome. The reads were then filtered and trimmed using Trimmomatic 0.33 (https://github.com/timflutre/trimmomatic). Clean data were analyzed using Kraken2 to classify sequences at the species level, and Bracken was used to estimate species abundance, thereby identifying changes in community composition at the species level (https://github.com/DerrickWood/kraken2). The taxonomic composition and relative abundance of microorganisms were depicted using Krona charts using the ktImportText command of the KronaTool v2.7 (https://github.com/marbl/Krona). Positive detection was reported when the Reads Per Million (RPM) from a given species or genus was ≥ 10 [23].

Microorganisms detected in each sample were ranked based on the relative abundance (RPM). Only the top five detected microorganisms (named top 1 through 5) which accounted for an average of 70.3% of the total detected microorganisms. The rest microorganisms in each sample were not used for further analysis since they account for minor populations. Pathogenic microorganisms were defined as those causing disease. Opportunistic microorganisms were defined as those generally not causing diseases but causing diseases in immunocompromised individuals or after their translocations. Non-pathogenic microorganisms were defined as those not reported for causing any diseases.

The sample-level analyses were conducted for comparisons among the microorganism in each sample, whereas the patient-level analyses were done for association between the microorganism detection and the clinical outcomes, for which each patient only had one outcome and therefore only one representative mNGS result per patient was used for analysis. The multiple data from each patient were not aggregated for the outcome association analysis. Instead, only the data from the first time point were used for analysis.

Conventional microbial culture and drug susceptibility testing

Clinical specimens were processed according to standard microbiological procedures. Blood cultures were performed using the BacT/ALERT 3D system (bioMérieux, France) by incubating for up to 6 days. All other specimens were cultured using specific procedures, such as Columbia blood agar, MacConkey agar, chocolate agar, and Sabouraud dextrose agar (Kailin, Jiangmen, China) by incubating at 35 °C, 5% CO₂ (when applicable) for 2–6 days according to the standard protocols. Antimicrobial susceptibility testing was conducted using VITEK 2 Compact (bioMérieux) with FUNGUS3, N335, and P639 cards for fungi, Gram-positive bacteria, and Gram-negative bacteria, respectively.

Statistical analysis

Statistical analyses were performed using STATA (version 16.0) and R software (the R Foundation for Statistical Computing). Continuous numerical variables were described as medians and interquartile ranges. Categorical data were analyzed by the chi-square testing or continuous corrected chi-square testing. Firth’s bias-reduced logistic regression models were fitted to identify impact factors associated with sepsis prognosis. All statistical tests conducted were two-tailed, with a value p < 0.05 considered as statistically significant.

Results

Identification of the most common microorganisms in severe septic patients

Detection of causative pathogens in septic patients is critical to formulate treatment regimens and improve clinical outcomes. To investigate what microorganisms were most frequently present, we studied 81 severe septic patients in ICU (Table 1). All these patients experienced severe function failure of multi-organs, with their SOFA score ≥ 2. Most of them (71.3%) had pneumonia and more than half of them (59.3%) did not survive. Other demographic characteristics of these patients were summarized in Table 1. To understand what causative microorganisms were more frequently detected among different sample sources, we collected 184 samples from nine different tissues (Table S1). Since the majority of the patients had severe pneumonia, 95 BAL samples were obtained from nearly half of the patients (38) and 46 sputum samples were collected from 16 patients. Twenty-five CSF samples were obtained from 11 patients with meningitis. A few samples were also obtained from five other tissues: PE (8), ascites (8), urine (7), HPC (2), BE (2). DNA was extracted from all samples (184) and RNA was extracted from the majority of the samples (143) to investigate whether DNA or RNA materials would yield different results for the detection of microorganisms among severe septic patients (Table S1). In total, 327 samples were subjected to mNGS analysis.

Table 1.

Demographic characteristics of the patients

Characteristic	Number of patients
Age, median years (range)	63 (52,72)
Gender
Male	15 (18.5%)
Female	66 (81.5%)
SOFA score (≥ 2)	81 (100%)
Comorbidities
Gastrointestinal disorders	12 (13.8%)
Pneumonia	62 (71.3%)
Cardiovascular disease	36 (41.4%)
Autoimmune diseases	23 (26.4%)
Diabetes	22 (25.3%)
Trauma	7 (8.0%)
Organ failure	45 (51.7%)
Hypertension	33 (37.9%)
Tumor	20 (30.0%)
Hypoproteinemia	12 (13.8%)
Chronic obstructive pulmonary disease	3 (3.4%)
Surgical drainage	25 (28.7%)
Outcome
Improvement	33 (40.7%)
Death	48 (59.3%)

SOFA Sequential organ failure assessment, Improvement Survival to hospital discharge, Death Died in hospital

One advantage of mNGS is its ability to unbiasedly detect all microorganisms in the sample, even those present at low abundances. Thus, more than dozens of microorganism strains from different kingdoms can be identified from each sample. It is daunting to analyze all those detected microorganisms from each sample. In fact, results for microorganisms detected in very low abundance in a sample are most likely unreliable and not associated with sepsis. When the top five microorganisms were analyzed in this study, the total raw reads from them accounted for 70.3% of all the raw reads detected by mNGS in each sample (Fig. 1A). Importantly, the abundance of the raw reads for each of the top 5 microorganisms relatively high: all more than 1000 RPM for DNA samples and all but five more than 1000 RPM for RNA samples (Fig. S1). The high abundance of each detected microorganism showed that the mNGS data for all top 5 microorganisms were reliably detected. Among severe septic patients, it is likely that only the top few microorganisms were the causative pathogens. Therefore, to simplify the analysis and reliably determine the impacts of the microorganisms on the clinical outcomes, we analyzed only the top 5 detected microorganisms for each sample in the subsequent analyses.

Fig. 1.

Microorganisms detected by mNGS. A The proportion of the combined top five microorganisms among all detected microorganisms by mNGS in each sample. B The proportion of each microorganism species among the total detected microorganisms by mNGS. C The most frequently identified microorganisms by DNA and RNA mNGS. Only the microorganisms present at ≥ 10% in each sample are shown. The bacteria identified by both DNA and RNA mNGS are indicated by circles

A total of 183 microorganisms were detected from 327 DNA and RNA samples (Fig. 1B and table S2). Bacteria (169) accounted for the largest proportion (92.3%), while six viruses (3.3%) and five fungi (2.7%) were the next highest proportion microorganisms (Fig. 1B). One protozoan, one chlamydia and one mycoplasma were also identified. Nine and twelve predominant microorganisms (> 10%) were detected by DNA and RNA mNGS analyses, respectively. Four bacteria (E. coli, A. baumannii, P. aeruginosa and K. pneumoniae) were detected by both DNA and RNA mNGS, while others were preferentially detected by DNA or RNA mNGS alone (Fig. 1C). This indicates that results from the mNGS analyses of DNA and RNA samples can be different. Very low background reads (< 100) were detected in only 6 of 10 no-template controls, suggesting no serious contamination concerns in the data analyses in this study (Table S3).

Detection of more pathogenic microorganisms by RNA mNGS than DNA mNGS

Among all detected 183 top 5 microorganisms, 40 were pathogenic (21.9%), 98 were opportunistic (53.6%) and 45 were non-pathogenic (24.6%) (Fig. 2 and Table S2), based on previously established classifications [24, 25]. Significantly higher proportion of samples containing at least one pathogenic microorganism for RNA mNGS (87.4%) than DNA mNGS (58.2%) (p < 0.001) (Fig. 2; Table 2). All samples [5] having four or five pathogenic microorganisms were detected by RNA mNGS (Fig. 2). Significantly higher levels of samples containing 3 pathogenic microorganisms were detected by RNA mNGS than DNA mNGS (p = 0.004). No significant differences were observed for samples containing nonpathogenic microorganisms between DNA and RNA mNGS analyses.

Fig. 2.

The top five microorganisms identified by mNGS in each sample. Each row represents the top 5 microorganisms detected by mNGS in a sample. The top five microorganisms were ranked based on their relative abundances in each sample. The left panel shows the samples with at least one pathogenic microorganism in the top five panel; samples are grouped first by the number of pathogenic microorganisms (5/5, 4/5, 3/5, 2/5, 1/5) and then by the positions at which pathogenic microorganism(s) appear in the panel. The right panel shows the samples without detected pathogenic microorganisms in the top five panel, grouped by the number of opportunistic microorganisms. ND, no data

Table 2.

Number of pathogenic or opportunistic microorganisms detected by mNGS in each patient

	No. of microorganisms	DNA	RNA	Total
				Number	Frequency
Pathogenic	5	0	2	2	0.9%
	4	0	3	3	1.3%
	3	8	21	29	12.5%
	2	29	35	64	27.6%
	1	70	64	134	57.8%
Total		107	125	232	100.0%
Opportunisti	5	31	5	36	37.9%
	4	14	10	24	25.3%
	3	16	3	19	20.0%
	2	12	0	12	12.6%
	1	3	0	3	3.2%
Total		77	18	95	100.0%

184 DNA samples and 143 RNA samples were analyzed

Fewer pathogenic microorganisms detected in the blood than other specimens

When the detection of microorganisms from the blood, sputum, BAL and CSF were analyzed separately, RNA mNGS identified more pathogenic microorganisms than DNA mNGS (Figs. 3A and S2). However, only the difference in the blood samples were statistically significantly (p < 0.001). The samples from PE, ascites, urine, HPC and BE were very few and they were excluded for further analysis (Fig. S4).

Fig. 3.

Comparison of the detection rates of pathogenic microorganisms among different sample types. A Comparison of detection rates between DNA and RNA mNGS among the blood, sputum, BAL and CSF samples. B Comparison of detection rates among the blood, sputum, BAL and CSF samples. All these samples were analyzed by DNA and RNA mNGS

When the RNA genomes were analyzed, the pathogenic microorganisms detected in the sputum, BAL and CSF were only slightly higher than in the blood and none of the differences were statistically significant (Fig. 3B). However, when the DNA genomes were analyzed, more pathogenic microorganisms were detected in the sputum (83.3%), BAL (75.0%) and CSF (57.1%) than the blood (36.1%). The likelihood to detecting pathogenic microorganisms in the sputum and BAL were significantly higher than in the blood (p < 0.001) (Fig. 3B). Although more pathogenic microorganisms were also detected in the CSF samples, the difference was not statistically significant. This demonstrated that the specimens from lower respiratory tract (sputum and BAL) had higher detection rates, indicating more abundant pathogenic microorganisms in the primary infection sites, consistent with previous studies [26–28].

Comparison of pathogen detection via mNGS and CMC

CMC was carried for 624 samples from 81 severe septic patients in ICU. The sputum samples accounted for half of them (51.4%), while the blood and urine samples accounted for 22.9% and 13.0%, respectively (Table S4). Samples from other sources were very few (less than 3%). Sixteen microorganisms were detected by CMC. Among them, 13 were bacteria (81.2%) and three were fungi (18.8%) (Fig. 4A). They were dramatically fewer than those (183) detected by mNGS (p < 0.001). No viruses, protozoan, chlamydia and mycoplasma were detected by CMC. Compared to the top 16 microorganisms detected by mNGS, only five (E. coli, A. baumannii, P. aeruginosa, K. pneumoniae and B. cenocepacia) were identified by both mNGS and CMC (Fig. 4B). Culture success rates were low and required multiple attempts to obtain a positive culture result. In addition, each CMC test usually identified only 1–2 microorganisms in the 91% of patients, although as many as six microorganisms could be detected in one patient.

Fig. 4.

Comparison of the mNGS and CMC methods for detection of microorganisms. A The total number of microorganisms detected by CMC. B Detection of microorganisms by CMC and mNGS. All 16 microorganisms detected by CMC are ordered based on the detection frequencies in all samples. The 16 most frequently detected microorganisms detected by mNGS are ordered based on their detection frequencies. C The disparity of the detected pathogenic microorganisms by CMC and mNGS among 81 patients. Both positive are the patients for whom pathogenic microorganisms were detected by CMC and mNGS. Matched are the patients for whom all microorganisms identified by CMC were also detected by mNGS. Partially matched are the patients for whom some microorganisms identified by CMC were also detected by mNGS but others were not. Mismatched are the patients for whom all microorganisms detected by CMC were not identified by mNGS. The bacterial identified by CMC and mNGS are indicated by circles

To assess the concordance between the mNGS and CMC results, a total of 81 patients with positive results from either method were analyzed. Among these 81 patients, no microorganisms were successfully cultured by CMC in 23 (28.4%) patients (Fig. 4C). In all other 58 cases (71.6%), the microorganisms were identified by CMC and mNGS (both positive). In the “matched” group, all microorganisms identified by CMC in 28 patients (48.3%) were also detected by mNGS. In the “partially matched” group, only some microorganisms identified by CMC in 10 patients (17.2%) were detected by mNGS. In the “mismatched” group, all the microorganisms detected by CMC in 20 patients (34.5%) were different from those identified by mNGS, indicating that some microorganisms detected by CMC were not detected by mNGS in some cases, consistent with the previous findings [29, 30]. These results suggested that mNGS exhibited substantially greater sensitivity in detecting microorganisms than CMC. In some cases, however, microorganisms detected by CMC were not detected by mNGS, indicating that some minority microorganisms could be expanded during CMC.

Poor clinical outcomes of microbial infections among severe septic patients

When the DNA samples were analyzed, no samples were found to contain 4 or 5 pathogenic microorganisms. Among the DNA samples from 81 patients, the pathogenic microorganisms that were the most predominant (top-1) were identified in 15 samples (Fig. 5). Among them, three and two pathogenic microorganisms were detected in two and seven patients, respectively, while only one pathogenic microorganism was found in six patients. Importantly, 14 of these 15 patients (93.3%) patient did not survive. This rate was significantly higher than the rate (52%) for the 66 other patients in whom the top microorganisms were not pathogenic, including the patients without any pathogenic microorganisms (p = 0.0029). When RNA samples were analyzed, four patient samples having 4 or 5 pathogenic microorganisms all died (Fig. 5). Among seven patients in whom the most predominant microorganism (top-1) was pathogenic and contained at least three pathogenic microorganisms, five of them also died. Overall, the majority (81.8%) of these patients (9 out of 11) died. However, no statistically significant differences were detected between any groups, likely due to the small numbers in the groups for comparison. We also did not find any statistically significant differences among different groups classified based on the CMC results. These results showed that detection of the pathogenic microorganisms as the top-1 microorganism in a patient for DNA mNGS analysis or detection of 4–5 pathogenic microorganisms or at least three pathogenic microorganisms, with one of them as the top-1 microorganism, in a patient for RNA mNGS analysis would be associated with a high mortality. Importantly, the majority of these combinations of pathogenic microorganisms (73%) were found in the sputum and BAL samples (Fig. 5), demonstrating the importance of detection of microbes by mNGS in both types of the samples.

Fig. 5.

Clinical outcomes of pathogenic microorganisms detected by mNGS. Each row represents the top 5 microorganisms detected in a sample from a distinct patient. The samples are grouped and ordered as described in Fig. 2. The sample types are indicated at the far right for some samples. The numbers of patients are indicted at the top of each group. The asterisk indicates the group with significant difference (p = 0.029)

We next investigated which of the microorganisms were more likely to lead to poor clinical outcomes. Nearly three times (2.64) more C. difficile was detected by DNA mNGS analysis in the death group than the improvement group (Fig. 6A). Among RNA samples, the proportions of S. enterica, A. faecalis, E. faecium and R. pickettii were 2.25, 2.25, 4 and 9 times more, respectively, in the death group than the improvement group (Fig. 6A). Overall, the death-to-recovery ratio (DtR) of the infection of these five bacteria were exceeded two (Fig. 6A). The Firth’s bias-reduced logistic regression analysis showed that the risk factors of R. pickettii, C. difficile and S. enterica were significantly associated with mortality, with odds ratios of 6.785, 2.994, and 2.455, respectively (Fig. 6B).

Fig. 6.

Association between detected microorganisms and clinical outcomes of severe septic patients. A The detection frequencies of the microorganisms in the death and improvement groups. Only the death-to-recovery (DtR) value above two are shown; one from DNA mNGS and four from RNA mNGS. B The odds ratio analysis of pathogen types and select microorganisms. Significant association between a particular microorganism and death is highlighted in red

Extensive antimicrobial resistance among severe septic patients in ICU

Two hundred and two successful cultures were obtained from 57 patients by CMC. On average, 3.5 drug resistance tests were done for each patient. Drug-resistant microorganisms were identified from the majority of patients (89.5%), indicating extensive antimicrobial resistance (AMR) in this cohort (Fig. 7A). About half of the patients (55%) were infected with one drug resistance microorganism while 31%, 12% and 2% of the patients were infected with 2, 3 and 6 drug-resistant microorganisms, respectively (Fig. 7B). These results demonstrated that nearly half of the patients were infected with two or more drug-resistance microorganisms. Twelve of 13 bacteria were found resistant to at least one drug (Fig. 7C). Importantly, all 10 bacterial isolates were resistant. Wild-type isolates were only found for three bacteria. For two bacteria (K. pneumoniae and S. maltophilia), drug resistance isolates were found predominant (83.3% and 62.5%, respectively) among the patients (Fig. 7C). Only one isolate was obtained S. agalactiae from one patient, and it was sensitive to tested drugs. Among three fungi, the minority drug resistance isolates (28.6%) were identified in C. albicans, while no drug resistance isolates were detected for C. glabrata and C. tropicalis. These results showed that drug resistance was highly prevalent among bacteria but much less frequent among fungi.

Fig. 7.

Antimicrobial resistance analysis of the samples from the severe septic patients in ICU. A Number of patients carrying microorganisms resistant or sensitive to drugs. B Proportion of patients harboring different numbers of drug-resistant microorganisms. C Number of patients infected with drug-resistant microorganisms. Drug-resistant and drug-sensitive bacteria and fungi are indicted by red and green, respectively (D) The frequency of the microorganisms detected by CMC and mNGS. The frequency of the microorganisms detected by CMC are for the drug resistance microbes, while the frequency of the microorganisms detected by mNGS are for all microorganisms

E. coli was the most frequently detected bacterium by mNGS (70.03%) and often accounted for the majority of microbial populations in both DNA and RNA samples (Figs. 1C and 7D). However, E. coli was not frequently detected by CMC (Fig. 7D) but all detected were drug resistant (Fig. 7C). This indicates that all E. coli strains might have developed drug resistance. Overall, it is likely that the vast majority of the bacteria detected in this cohort developed drug resistance which might have contributed significantly to treatment failure of severe septic patients.

Discussion

Accurate and timely identification of causative pathogens is of great importance in improving the survival of septic patients in the ICU [3]. However, the complexity of causative pathogens in the samples from septic patients poses a huge challenge [31]. Recently, mNGS has been used for detection of pathogens and provided favorable clinical impact in the treatment of septic patients in the ICU because of its rapid, unbiased and simultaneous detection of diverse types of pathogens [32, 33]. Here we developed a new analysis strategy that identified 183 common microorganisms from analysis of 327 samples by analyzing only the top 5 microorganisms in the individual samples. The top 5 microorganisms accounted for at least 70% of the microorganisms detected by mNGS. With this new strategy, we have identified nine common microorganisms as determined by DNA mNGS and 12 common microorganisms as determined by RNA mNGS. In each case, these common predominant microorganisms accounted for more than 10% of the total detected microorganisms. This new tool is rapid, reliable, accurate and easy to implement for providing better care for severe septic patients in ICU.

By analyzing the top 5 microorganisms in each sample, 183 microorganisms, including bacteria, fungi, mycoplasma and viruses, were identified. In contrast, CMC only isolated 13 bacterial strains and three yeast strains. No pathogenic microorganisms were detected by CMC in 28% of patients after repeated attempts. However, 40 pathogenic microorganisms were successfully detected in 80 patients by mNGS. Discordance between the results from mNGS and CMC analyses is often observed [34, 35]. In 20 patients, 10 microbes were detected by CMC but not by mNGS among top 5 microorganisms. Among those 10 microbes, 7 were still not detected by mNGS when all microorganisms are analyzed. This observation was similar to those found in previous reports [36, 37]. This suggests that some very low abundancy microbe that were missed by mNGS could be selectively enriched by CMC or that these microbes were missed by mNGS, although very unlikely. If it is later, it would suggest that CMC will be a necessary complementary method for mNGS analysis.

The most of severe septic patients (71.3%) in this cohort had pneumonia and had life-threatening conditions. Thus, lower respiratory tract specimens, including sputum and BAL, had much higher positive rates for identification of various microorganisms than the other specimens by the mNGS analysis. Overall, more microorganisms were detected among different specimen types by RNA mNGS than DNA mNGS, suggesting that RNA mNGS can serve as a more favorable approach for accurate pathogen detection in sepsis. Significantly fewer pathogenic microorganisms were detected by DNA mNGS in the blood and CSF samples than the other samples.

The DNA mNGS analysis showed that when the most predominant microorganism was pathogenic irrespective of the number of pathogenic microorganisms in a patient, the chance for patients to die was significantly higher than those without the pathogenic microorganisms as the most predominant one. Importantly, these pathogenic microorganisms were most often detected in the sputum and BAL samples (73%), demonstrating the importance of detection of microbes by mNGS in these kinds of samples. When individual microorganisms were analyzed, the infections with R. pickettii, C. difficile and S. enterica had significantly higher Odds ratios (6.785, 2.994, and 2.455, respectively). Previous studies have showed that immunocompromised populations are vulnerable to opportunistic infections, including C. difficile and R. pickettii infections, leading to an increased risk of sepsis [38, 39]. These results show that identification of a pathogenic microorganism as the most predominant by DNA mNGS and any one of R. pickettii, C. difficile and S. enterica can be associated with poor clinical outcomes.

The development of the next-generation sequencing technique has significantly lowered the cost of mNGS testing, enabling its widespread use in identification of causative pathogens, particularly for septic patients in the ICU who require prompt treatment and intervention [40]. In this study, we have found that the mNGS analysis can identify the most frequent microorganisms and show that the detection of selected pathogen patterns can be associated with clinical outcomes. However, the drug susceptibility tests showed that the vast majority of isolated microbial strains (89.5%) have developed drug resistance. The all isolates from 10 of 13 bacteria were resistant to antimicrobials. Although these drug resistance microbes were detected at high abundances by mNGS, drug resistance genes could not be deciphered because only the fragmental genome sequences were obtained from each microbe. Thus, the mNGS analysis should be combined with drug resistance tests to develop the most effective treatment regimens for severe septic patients in ICU.

This study has some limitations. It is a single-center ICU cohort with a modest sample size (81 patients; 184 specimens) and may limit analysis power. Given the observational nature of this study, associations between the detection of microbes and the clinical outcomes should be treated cautiously. More works are warranted to distinguish infection from colonization, especially for respiratory samples.

Conclusion

Analysis of the top 5 microorganisms detected by mNGS represents a rapid, reliable and easy-to-implement approach for providing better care of the severe septic patients in ICU. The detection of predominant or multiple pathogenic microorganisms in lower respiratory tract samples such as sputum and BAL in these patients is significantly associated with poor clinical outcomes. The high prevalence of multidrug-resistant microbes in these patients underlines the critical role of drug resistance in treatment failure. These results demonstrate that the integration of mNGS with drug resistance testing will be key for saving lives of the severe septic patients in ICU.

Abbreviations

mNGS

Metagenomics next-generation sequencing

SOFA

Sequential organ failure assessment

RPM

Reads per million

BAL

Bronchoalveolar lavage fluid

CSF

Cerebrospinal fluid

PE

Pleural effusion

HPC

Hydropericardium

BE

Blister effusion

DtR

Death-to-recovery ratio

Authors' contributions

F. G., H. Y. and Q. Z. designed and supervised the study. F. G., Q. Z. and Z. C. wrote the manuscript. Z. C., Y. Y., X. C., W. C., L. Z. and L. G. performed mNGS analysis and analyzed CMC and drug resistance tests. H. Y. and Y. Y. collected clinical samples, analyzed clinical data and treated the patients.

Funding

This research was supported by grants from the Fundamental Research Funds for the Central Universities (21624318), the Guangdong Medical Research Foundation (A2024458), the China Postdoctoral Science Foundation (2021TQ0126), and the Program for Guangdong Introducing Innovative and Entrepreneurial Teams (No. 2021ZT09Y552).

Data availability

The newly obtained sequences have been deposited in the Genome Sequence Archive under accession number CRA021369. All other data are included in the manuscript.

Declarations

Ethics approval and consent to participate

The study was approved by the Institutional Review Board and Ethical Committee of the First Affiliated Hospital of Jinan University (KY2023-047), following the principles of the Declaration of Helsinki. Written informed consent was obtained from all participants or their authorized family members.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.