Metagenomic Next-Generation Sequencing Improves The Prognosis of Patients With Infectious Diseases on Mechanical Ventilation in The Intensive Care Unit (ICU)

Background: Metagenomic Next-Generation Sequencing (mNGS) has gradually shown its advantages in pathogen identication for clinical infectious disease. However, few studies were conducted on the evaluation between this technique and conventional methods like culture and PCR and the prognosis of patients with infectious diseases on mechanical ventilation in ICU Methods: We conducted this retrospective study from March 2018 to May 2020 in the rst Aliated Hospital of Guangzhou Medical University, a total of 228 patients with suspected infectious diseases on mechanical ventilation were included, including 104 cases of mNGS group and 124 cases of non-mNGS. Statistical analyses were performed between the two groups and subgroup of whether were immunocompromised. The concordance between mNGS, culture and PCR was also assessed. Results: The 28-day mortality rate of the patients in the mNGS group was lower after the baseline difference correction (19.23% vs. 29.03% (cid:0) p=0.039), indicating that mNGS may improve the prognosis of patients in ICU. And subgroup analysis showed that mNGS could improve the 28-day mortality of nonimmunosuppressive patients (cid:0) 14.06% vs. 29.82%, p=0.018 (cid:0) . According to the analysis of Logistic Regression, not performing mNGS, high APACHE II score and hypertension were independent risk factors for 28-day mortality, which strongly suggested that mNGS was one of the key factors affecting prognosis. A total of 157 samples performed mNGS, 116 of them received both mNGS and culture. mNGS presented advantages of positivity (69.8% double positive and 25.0% mNGS positive only) and concordance (79.0%, match and partly match). Conclusions: mNGS may improve the prognosis and reduce the 28-day mortality rate of patients with infectious diseases on mechanical ventilation in ICU. This technique has shown its advantages comparing with conventional methods, and will be wildly used as a promising technology for infectious disease. of transforming mNGS into a routine diagnostic test. studies practicality of mNGS in the work-up of undiagnosed infectious diseases. mNGS identifying rare, novel, dicult-to-detect and coinfected pathogens directly from clinical samples presents potential in resistance prediction by sequencing the antibiotic resistance genes, providing new diagnostic evidence that can be used to guide treatment options and improve antibiotic stewardship. mNGS last resort method to standardisation, and immunosuppression in patients with severe pneumonia remain scarce. In this study, we will discuss the performance of mNGS in clinical practice. It will include the composition of pathogens in the patients of severe pneumonia under the premise of immunosuppression or not, so as to provide clinical basis for early empirical antibiotic treatment, also include the comparison of clinical detection methods of Pneumocystis, and the correlation between mNGS and prognosis of patients. pleural effusion perform identication.

License:   This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License Introduction Metagenomic next-generation sequencing (mNGS) is increasingly being applied in clinical laboratories for unbiased culture-independent diagnosis. Whether it can be a next routine pathogen identi cation tool has become a topic of concern. We review the current implementation of this new technology for infectious disease diagnostics and discuss the feasibility of transforming mNGS into a routine diagnostic test. Since 2008, numerous studies from over 20 countries have revealed the practicality of mNGS in the work-up of undiagnosed infectious diseases. mNGS performs well in identifying rare, novel, di cult-to-detect and coinfected pathogens directly from clinical samples and presents great potential in resistance prediction by sequencing the antibiotic resistance genes, providing new diagnostic evidence that can be used to guide treatment options and improve antibiotic stewardship. Many physicians recognized mNGS as a last resort method to address clinical infection problems. Although several hurdles, such as work ow validation, quality control, method standardisation, and data interpretation, remain before mNGS can be implemented routinely in clinical laboratories, they are temporary and can be overcome by rapidly evolving technologies. With more validated work ows, lower cost and turnaround time, and simpli ed interpretation criteria, mNGS will be widely accepted in clinical practice. Overall, mNGS is transforming the landscape of clinical microbiology laboratories, and to ensure that it is properly utilised in clinical diagnosis, both physicians and microbiologists should have a thorough understanding of the power and limitations of this method.
In the intensive care unit (ICU), severe pneumonia diagnosis is particularly complex due to the similar clinical symptoms. It generally leads to the dysfunction of other organs and thus carries high mortality [1][2][3]. So it is critical to provide an effective symptomatic treatment early. However, the diagnosis is challenging because of the multitude of possible pathogens. Metagenomic next-generation sequencing (mNGS) is a novel approach that can theoretically detect all nucleic acids in a clinical sample. After the rst successful clinical application reported in 2014 [4], the clinical attention and acceptance of mNGS has been increasing, due to its high sensitivity and wide coverage, it can improve the positive rate compared with the traditional methods, which is very important for the detection of rare pathogens [5,6]. In the near future, mNGS might have the potential to become a routine diagnostic workup, especially in the diagnosis of severe pneumonia [7]. At present, literature relevant to clinical applications has mostly emerged as case reports or small-scale cohort studies, most of which have concerned about comparative analysis of infected patients [8][9][10][11]. However, reports on the relationship between infection and immunosuppression in patients with severe pneumonia remain scarce. In this study, we will discuss the performance of mNGS in clinical practice. It will include the composition of pathogens in the patients of severe pneumonia under the premise of immunosuppression or not, so as to provide clinical basis for early empirical antibiotic treatment, also include the comparison of clinical detection methods of Pneumocystis, and the correlation between mNGS and prognosis of patients.

Patients of This Study
This retrospective study was conducted between March 2018 and May 2020. We reviewed the medical records of patients admitted to the intensive care unit (ICU) of the First A liated Hospital of Guangzhou Medical University. The study was approved by the Institutional Ethics Committee of the First A liated Hospital of Guangzhou Medical University (No.2020K-42). Patients with infectious diseases on mechanical ventilation were admitted to ICU during this period were included or excluded for our study according to the following criteria.
Exclusion: 1)pregnant woman; 2)patient unable to ful ll the required medical follow-up; 3)patients participating in a clinical trial implicating a new drug; A total of 246 patients were enrolled, including 112 in the mNGS group and 134 in the non-mNGS group. While 8 cases in mNGS group and 10 cases in non-mNGS group were excluded due to incomplete data. 50 of the remaining 228 cases were considered as immunocompromised when clinically diagnosed as: 1)individuals on immunosuppressive therapy (e.g., cytotoxic agents, glucocorticoids, etc.); 2)transplantation individuals (e.g., solid organ transplantation or bone marrow transplantation); 3)hematologic cancers; 4)certain trauma or surgery (e.g., splenectomy); 5) secondary to metabolic diseases (e.g., malnutrition, non-controlled diabetes, uremia).
All 228 cases have complete medical treatment record including Clinical diagnosis and symptoms, APACHE II score, SOFA score, comorbidities and Blood laboratory tests when transferred to ICU, 28-day mortality, In-hospital mortality, Length of ICU stay, Days ventilator-free at day 28, and Total cost.
Statistically analysis was performed based on whether performed mNGS or whether immunocompromised. Since the data were anonymous, there is no need for informed consent. The owchart of this comparative study described in Figure 1.

Metagenomic Next-generation Sequencing and Analysis
Nucleic Acid Extraction In our study, 104 cases used blood (41), bronchoalveolar lavage uid (BALF,80), sputum (31), cerebrospinal uid (1); and pleural effusion (4) to perform mNGS for pathogen identi cation. Blood samples were drawn from patients into ethylenediaminetetraacetic acid tubes with a volume of 5 ml, then stored at 4°C. The plasma separation experiment should be completed within 8 hours, with the method of centrifugation at 1600g for 10min, the supernatant was transferred to new sterile tubes for DNA extraction. Samples of 1-4 ml of sputum, bronchoalveolar lavage uid, pleural effusion were collected from patients according to standard protocols and stored in -80°C before DNA extraction. For sputum samples, 0.1% dithiothreitol (DTT) was used for liquefying at room temperature for 30min. BALF and pleural effusion could go directly to the DNA extraction procedures.
To begin with DNA extraction, 0.5 ml of the samples was transferred to a new 1.5ml microcentrifuge tube, DNA was extracted using the MAPMI Sample preparation kit (360120, CapitalBio Corporation, Beijing, China) according to the manufacturer's recommendation.

Library Preparation
The extracted DNA was quanti ed before the preparation of sequencing libraries with a total DNA less than 500ng. DNA libraries were constructed through end repair, adapters adding, and PCR ampli cation. The quality of the DNA libraries was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, California) combined with qPCR based on Applied Biosystems 7500 Real-Time PCR System (Thermo Fisher, USA), the primers based on the sequences of the adapters were used for the qPCR. Then quanti ed DNA libraries were pooled and analyzed on BioelectronSeq 4000 (CapitalBio Corporation, Beijing, China) sequencing platform. The whole process requires 48~72 hours from DNA extraction to issuing the reports.

Bioinformatic Analysis
The original sequencing data were subjected for quality control, and the reads with length less than 50bp, low-quality, and low complexity were removed. The remaining high-quality sequencing data mapped to the human reference genome grch38 for human host sequencing depletion use Bowtie2.
Subsequently, the non-human sequences were classi ed by simultaneous alignment to the databases of viruses, bacteria, fungi and parasites for annotation. The nal pathogen detection results include a list of suspected pathogens, the number of hit reads and genome level coverage statistics. All the pathogenic genomic sequences were downloaded from NCBI and PATRIC databases. At present, the bacterial database contains 13992 species, the fungal database contains 1659 species, the virus database contains 13000 species, and the parasite database contains 287 pathogen genomic data.
In order to judge the suspected pathogens, we reviewed data of different sample type of healthy people, and calculated the relevant reference values, including hit reads number and coverage of all bacteria, fungi, viruses and parasites detected [6,12,13]. The quantiles, median and maximum values at 5%, 75%, 95% of these samples were calculated, and the 5% and 95% quantiles were used as reference ranges for normal people. The criteria for judging suspected pathogens are as follows: 1) the number of hit reads is greater than 95% quantile in the reference range; 2) the length of target genome covered is greater than 95% quantile in the reference range; 3) the suspected pathogen has potential infection ability.

Statistical Analyses
Statistical analyses were performed using R software v.4.0.2. Baseline characteristics were presented as the median (interquartile range) and the count (proportion), then analyzed using the Mann-Whitney U test and the Chi-square test. P values < 0.05 were considered statistically signi cant. Since the mNGS group and non-mNGS group had a signi cant difference in APACHE II score, ARDs, Lymphocyte and LDH (P<0.1; Table 1), comparisons of the outcome data (28-day mortality and in-hospital mortality) was corrected using characteristics mentioned above as covariate. A multivariate logistic regression analysis was used for multiple characteristics evaluation on survival.

1.Comparison of whether to perform mNGS
In the total 228 patients, 104 of them performed mNGS and 124 patients not. Characteristics and baselines of patients who whether to performed mNGS, statistical analysis result list in Table 1. Statistical analysis information includes age, sex, APACHE II scored and SOFA score, Major diagnosis of ICU admission, Comorbidities, Blood laboratory tests when transferred to ICU. Compared with non-mNGS group, mNGS patients had higher proportion of ARDS (56.7% vs 38.7%, P=0.007), lower lymphocytes (0.4 vs 0.6, P<0.001), higher LDH level (446.3 vs 307.8, P<0.001) and higher APACHE II score (although there was no statistical difference between the two groups). All data were presented as the median (interquartile range) unless otherwise stated.
To evaluate the outcomes of patients who whether to perform mNGS, the result of 28-day mortality, In-hospital mortality, Length of ICU stay, Days ventilator-free at day 28, Total cost, list in Table 2. No signi cant statistical difference was found in 28-day mortality and in-hospital mortality.  The total 228 patients who were survival or non-survival at day 28, statistical analysis result list in Table 3. The APACHE II score, SOFA score, AKI proportion, concentration of procalcitonin (ng/ml) showed signi cant differences between the patients who were survival or non-survival at day 28, but whether perform mNGS show no signi cant difference, this is probably because the mNGS test does not be widely used in clinical hospital or this test has not accepted wide recognition because its high cost or other limitations (Table 4).  From Table 4, we selected the indices present P<0.1 for logistic regression. The result show that without performing mNGS, higher APACHA II score and Hypertension may be high risk factors, with the Pr(>|z|) 0.007, 0.027 and 0.036 separately (Table 5).

2.Comparison of whether were Immunocompromised
Characteristics and baselines of patients who whether were immunocompromised and whether performed mNGS, statistical analysis result shows in Table 6.   Characteristics and baselines of patients who were survival or non-survival at day 28, immunocompromised or not, perform mNGS or not. statistical analysis base on indices of clinical factor list in Table 8.
In the subgroup immunocompromised of total 50 patients, 37 survived and 13 were non-survival, no statistical difference found based on the indices of clinical factors. Maybe the cases of Immunocompromised patients are not enough for this study.
In the subgroup of nonimmunocompromised, 135 of total 178 survived, and 43 not survived. APACHE II score, SOFA score, proportion of Acute kidney injury, Hypertension, Procalcitonin, LDH, and whether performed mNGS (40.74% vs 20.93%, P=0.018 ) showed signi cant difference. Same as the analysis method of Table 5, we select the indices present P<0.1 for logistic regression. The analysis results show that the conclusion is almost consistent. Without performing mNGS, higher APACHA II score are high risk factors for the nonimmunocompromised subgroup, with the Pr(>|z|) 0.026, and 0.005 (Table 9). Table 9 Logistic regression of indices P<0.1 in Table 8 APACHE

Concordance Between mNGS and Culture
The main sample types collected for pathogen identi cation are blood, BALF, sputum, only one sample was collected for testing in 63 cases, and the rest  Table 10.    (Table 11) [14][15][16][17][18], and the result revealed 100% positive. Compared with culture, the rapid feedback of mNGS and high positive rate hasten clinical decision making and guide clinical laboratories to improve culture conditions for fastidious organisms. Moreover, for coinfections, unculturable microorganism infections, and new pathogenic infections, mNGS shows great proven advantages [19,20].

Discussion
In our single center retrospective study, patients in mNGS group had a higher proportion of ARDS, lower absolute number of peripheral blood lymphoid bacteria, higher LDH level, and higher APACHE II score (although there was no statistical difference between the two groups). This indicating that in clinical work, we prefer to mNGS detection for patients with more severe disease severity and lower immune level [21]. In contrast, the 28-day mortality rate of the patients in the mNGS group was lower after the baseline difference correction, indicating that mNGS could improve the prognosis of patients in ICU [22,23]. Moreover, subgroup analysis showed that mNGS could improve the 28-day mortality of nonimmunosuppressive patients. While for immunosuppressive patients, the expected conclusion has not been reached. That's probably because the cases of nonimmunosuppressive patients are not enough in this retrospective study, and more cases of immunosuppressive patients are needed for further clarifying [24].
Logistic regression showed that, not performing mNGS, high APACHE II score and hypertension were independent risk factors for 28-day mortality, which strongly suggested that mNGS was one of the key factors affecting prognosis [23]. The improvement of prognosis may be related to the sensitivity of mNGS to pathogen detection. Despite prolonged hospital stay and increased costs, this is more likely to be associated with the severity of the illness.
mNGS can improve the sensitivity of pathogens detection, including bacteria, fungi, viruses and parasite are more sensitive than the conventional culture and PCR methods at the same time. Although it is still unable to distinguish whether the pathogens detected by mNGS are pathogenic microorganisms, combined with prognostic indicators, it can provide more enlightenment for clinical work [4][5][6][7]. Theoretically, it can detect all pathogens in a clinical sample unbiased at one time, and shows proven advantages for rare, novel, di cult cases of infectious diseases. However, there is still a lot of improvements for this technique, for example, to formulate negative and positive criteria, improve the turnaround time and optimize the method for drug resistance prediction for clinical.
For different sample type like blood, sputum, cerebrospinal uid, tissue, swab and other miscellaneous specimen, how to establish a uni ed nucleic acid extraction standard makes the detection e ciency of pathogens maximization di cult. Host nucleic acid contamination is also a major factor that interferes with pathogen detection. In the process of nucleic acid extraction, how to remove host effectively, thereby enrich pathogen nucleic acid and improve the sensitivity of mNGS, this still needs relevant research to be solved. In terms of bioinformatics, recent studies have evaluated the merits and demerits of different tools. The existing methods not only have different algorithms, but also use different databases, which leads to differences in pathogen identi cation ability and relative abundance calculation [25]. In the absence of a uni ed standard in the process of mNGS, the user's choice of the software based on personal experience, accessibility and convenience will affect the repeatability and reliability of the results, which will become an obstacle to the clinical standardization of mNGS.
This technique takes a long time comparatively, usually 24 to 48 hours of turnaround time at present. This needs to be further improved to meet the needs of ICU. In addition, the application of mNGS for patient in ICU is suitable for carrying out in the hospital, while expensive instruments and di culty of operation present to be a problem.
Presently, mNGS is mainly based on short read length sequencing, for the characteristics of sequencing platforms (Illumina, Thermo Fisher, MGISEQ). In 2014, Hasman et al clari ed that mNGS can be used for the detection of pathogens and drug-resistant genes in urine samples [26]. However, samples with low abundance of microbial sequences, it is di cult to determine the attribution of the drug-resistant genes accurately. Improvement of this technique is expected in the future so as to provide proven precise medical service for infectious pathogens identi cation for clinical practice.

Conclusion
We found that more pathogens could be detected using mNGS compared to traditional microbial detection methods. mNGS may improve the prognosis and reduce the 28-day mortality rate of patients with infectious diseases on mechanical ventilation in ICU. This technique has shown its advantages comparing with conventional methods, and will be wildly used as a promising technology for infectious disease. Additional studies are required to con rm the usefulness of this technology. The owchart of case inclusion and exclusion, with a total of 246 cases admitted to ICU, 224 cases were enrolled in this study. These cases divided into mNGS group, non-mNGS group, immunocompromised subgroup, and nonimmunocompromised subgroup.157 samples used for further analysis while 116 samples both received mNGS and culture result.

Figure 2
Pathogen distribution of mNGS result. With 38 of 157 samples found Acinetobacter baumannii by mNGS, and the results detected less than 2 were not displayed in illustration.