We retrospectively analyzed the medical records of patients who received allo-HSCT at Ruijin Hospital, the First Affiliated Hospital of Zhejiang University, and Liquan Hospital, GoBroad Medical Institute of Hematology (Shanghai Center) from April 2021 to January 2023. Institutional databases were retrospectively reviewed to extract demographic, clinical and genetic data. All procedures complied with the tenets of the Helsinki Declaration. The requirement for written informed consent was waived, owing to the non-interventional and retrospective nature of the study.

For herpes simplex virus (HSV) prophylaxis, acyclovir was given at the dose of 400 mg twice a day for 12 months after transplantation (Henze et al., 2022). For pneumocystis jirovecii pneumonia (PJP) prophylaxis, sulfamethoxazole was administrated at a dose of 480mg thrice a week from 2 months to 12 months (Maschmeyer et al., 2016). Between April 2021 and August 2022 (before the introduction of letermovir in China), 111 allo-HSCT recipients were treated with ganciclovir with 5mg/kg twice a day for 7 days prior to transplantation as cytomegalovirus (CMV) prophylaxis (Ljungman et al., 2019). After August 2022, 42 allo-HSCT recipients were treated with letermovir at a dose adjusted to calcineurin inhibitors for at least 3 months. For primary fungal prophylaxis, all patients received posaconazole oral suspension at an initial dose of 200mg thrice a day at the start of the conditioning regimens (Maertens et al., 2018). The efficacy of posaconazole treatment was determined based on therapeutic drug monitoring. The optimal serum trough concentration is>0.5mg/ml for prophylaxis (Chen et al., 2018). In case the trough concentration did not reach the prophylactic concentration, we prefer to increment the dosing frequency to 200mg every 4-6 hours, or switching to tablets or intravenous formulation. The duration of fungal prevention is up to 6 months for allo-HSCT with haploidentical donors (HID) or matched/mismatched unrelated donors (MUD/MMUD) and 3 months for allo-HSCT with matched sibling donors (MSD).

The criteria for determining positive mNGS detection varied depending on the type of microbes being detected (Xu et al., 2023). A microbe was considered a positive result for bacteria and fungi when the stringently mapped read number (SMRN) at the species level was ≥3 and relative abundance at the genus level was >30%. A microbe was considered a positive result for viruses when the SMRN was ≥3. A microbe was considered positive for parasites when the SMRN was ≥100.

All patients underwent CMT as indicated by the treating physicians. Different specimens were collected for testing based on the type of suspected infection (i.e., blood samples, puncture fluids, BALF, and CSF). Positive CMT results was verified when bacteria, fungi were detected via smear microscopy, culture, and (1,3)-β-D-glucan test (G test), or galactomannan test (GM test). Viruses including CMV and Epstein-Barr virus (EBV), were verified with plasma twice a week via the quantitative real-time PCR. HSV1 was verified by PCR with oropharyngeal swab samples.”

The diagnosis was made by an independent clinical committee comprising two hematologists, one pulmonologist and one radiologist. We comprehensively evaluated the clinical symptoms, radiological manifestations, CMT, and responses to anti-infection therapy to classify the final diagnosis into four categories: clinically significant infection, infection without clinical significance, immune events, mixed events of clinically significant infection and immune events. Also, the identification of final causative pathogens was made based on comprehensive consideration. The mNGS results was classified as definite, probable, possible, unlikely, or false-negative causes of infection. (1) Definite: microbes detected were consistent with those detected by CMT; (2) Probable: microbes detected were probably causative pathogens; (3) Possible: microbes detected showed potential to cause infection, but not as a common cause based on the consideration of clinical medical records; (4) Unlikely: microbes detected by mNGS were non-causative pathogens; (5) False negative: mNGS result was negative, but the case was diagnosed with infection. On the condition of “Definite”, “Probable” and “Possible”, mNGS results were judged as true positive, while on the condition of “Unlikely” they tended to be false positive. The positive predictive value (PPV) and negative predictive value (NPV) of mNGS were calculated as the ratio of true positive or negative mNGS detection to all practical mNGS detection. Initial empirical antibiotic therapy was performed within 24-48 h of clinical signs (Wang et al., 2020). Treatment success was defined as the resolution or reduction of clinical symptoms. CMV detection is defined as the detection and quantification of CMV DNA by either mNGS or PCR in plasma (Ljungman et al., 2017). The definition of clinically significant CMV infection requires clinical symptoms and CMV DNA viremia, necessitating the initiation of anti-CMV therapy by a treating physician (Chemaly et al., 2020).

Data analysis and visual representation of the results were performed using IBM SPSS Statistics for Windows, version 25.0 (IBM Corp., Armonk, N.Y., USA) and R 4.2.0 software (R Foundation for Statistical Computing, Vienna, Austria). Continuous and categorical variables were presented as medians and rang variables, as well as counts and percentages, respectively. Comparative analysis was conducted by Pearson χ2 test and Fisher exact test for discrete variables where appropriate. P-value <0.05 was considered to be statistically significant.

Based on the inclusion or exclusion criteria ( Figure 1 ), we included 271 samples with parallel results for mNGS and CMT from 153 participants. General characteristics of 153 patients were summarized in Table 1 . The median age was 47 (range, 15-69) years. We divided these samples into two cohorts based on neutropenic status at the time of sampling: cohort A (neutropenia [n=92]) and cohort B (non-neutropenia [n=179]) ( Table 2 ). The most frequent mNGS sampling specimen was peripheral blood (78.2%), followed by BALF (10.7%), CSF (8.5%), and others (2.6%). Peripheral blood specimen was more common in cohort A than cohort B due to the non-invasive and easily accessible nature.

We excluded 13 paired mNGS samples collected from different tissues in the same febrile event, and a comparison of the diagnostic results of mNGS with those of CMT methods was made for all 153 participants with 258 samples. The positive detection rate of mNGS was 76.4%, significantly higher than 43.8% of CMT (P<0.001). The positive detection rate was much higher with mNGS when compared with CMT in both cohort A (70.7% vs. 39.1% P<0.001) and cohort B (79.5% vs. 46.4% P<0.001). mNGS and CMT were concordant for 50 of 92 (54.3%) patients (kappa=0.128) in cohort A, and 91 of 166 (54.8%) patients (kappa=0.130) in cohort B ( Table 3 ).

The detection rate of mNGS in peripheral blood specimens was 75.0%, higher than 41.5% with CMT methods (P<0.001). In non-blood specimens, the detection rates with mNGS and CMT were 79.7% and 54.2%, respectively (P=0.003).

Regarding the clinical causative pathogens, bacteria accounted for 57.4% of infection in cohort A, followed by viruses (29.4%) and fungi (13.2%). In cohort B, viruses accounted for 61.1%, followed by bacteria (24.8%) and fungi (14.1%).

For bacteria, mNGS reached a 22.9% positivity rate compared with CMT (16.7%, P=0.077). There was no significant difference in the detection rates of bacteria using mNGS and CMT in cohorts A and B ( Figure 2A ). Similar results were observed when peripheral blood specimens were analyzed ( Figure 2B ). Among the 31 suspected bacterial infection cases in cohort A, mNGS results for 16 cases were consistent with CMT methods, with a concordance rate of 51.6% (16/31), and the concordance rate was 35.7% (10/28) in cohort B ( Figure 3 ). Distribution of microbes detected by mNGS and CMT is shown in Supplemental Figure S1 .

For suspected fungi, the detection rate with mNGS was similar to that of CMT (9.3% vs. 6.2% P=0.188) in all cases, but was higher than that of CMT in cohort A (9.8% vs. 2.2% P=0.03, Figure 2A ). In cohort A and cohort B, 77.8% and 53.3% of the fungal infection verified by mNGS were not confirmed by CMT, respectively ( Figure 3 ). Of the 14 cases of Aspergillus infection, thirteen cases (92.9%) were detected via mNGS, and only one case (A7, Supplemental Table S1 ) was detected via CMT method. A1 patient had respiratory symptoms and radiologic manifestation one month after mNGS-positive detection with peripheral blood in the neutropenic stage ( Supplemental Figure S2 ).

More viruses had been identified via mNGS than via CMT because of its high sensitivity (cohort A: 62.0% vs. 14.1% P<0.001; cohort B: 72.3% vs. 36.1% P<0.001) ( Figure 2 ). The concordance rates between mNGS and CMT for viral detection were 22.8% (13/57) in cohort A and 45.8% (55/120) in cohort B ( Figure 3 ). CMV was the most frequently detected virus in both cohorts (cohort A: 33.7%; cohort B: 45.2%). As shown in Figure 4 , the peak CMV detection rate using mNGS and the clinically significant CMV infection rate were concentrated between 30 and 60 days after allo-HSCT.

In BALF specimens (29 cases, Supplemental Table S2 ), Pneumocystis jirovecii (P. jirovecii) and viruses constituted 79.3% of the microbes detected ( Supplemental Figure S3A ). Notably, P. jirovecii was detected only by mNGS. In patients with confirmed infectious pneumonia, the mNGS-positive detection rates were 89.3% (25/28 cases) and 64.2% for CMT (18/28 cases). In cases with PJP, the positive rate of the G test in the BALF was 83.3% (5/6 cases), and 12.5% in the peripheral blood (1/8). In 11 cases with positive BALF G/GM results, fungi were detected by mNGS in 6 cases (54.5%), including Candida in 16.7% (1 case) and P. jirovecii in 83.3% (5 cases). In 13 cases with Aspergillus infection verified by mNGS, five of which (38.5%, 5/13) was detected in BALF with mNGS test.

Viruses accounted for 69.6% of causative pathogens in suspected encephalitis ( Supplemental Table S3 and Figure S3B ). Four patients (27.7%, C2, C5, C6, C18, C23) had simultaneous viremia when diagnosed with viral encephalitis, indicating that CSF testing is pivotal when encephalitis is suspected.

A total of 168 patients (65.1%) were eventually diagnosed with clinically significant infection, 12 (4.7%) with infection but without clinical significance, 28 (10.9%) with mixed events, and 50 (19.4%) with immune events. The diagnostic capabilities of mNGS, CMT, and the combined protocols for different samples are shown in Table 4 . The mNGS test yielded a higher sensitivity (77.4% vs. 52.9%, P<0.001) than the CMT for diagnosing clinically significant infection in peripheral blood samples, and the effect was more prominent in cohort B. In non-peripheral blood specimens, mNGS tended towards higher sensitivity than in peripheral blood specimens. (88.7% vs. 77.4%, P=0.075).

Among the febrile events with negative mNGS results (61 samples), 35 patients (57.4%) were eventually diagnosed with immune events. Among the patients with negative CMT results (143 samples), 48 (33.5%) were diagnosed with immune events (P=0.002). The detailed immune events types are displayed in Supplemental Table S4 .

The spectrum of microorganisms in CMT negative cases were displayed in Supplemental Figure S5 . In cohort A, 30 cases with negative CMT results were clinically diagnosed with an infection. Among the 30 cases, mNGS detected pathogens matched the final causative pathogens in 20 cases, including 10 cases of bacterial infection, 2 of fungal infection, 5 of viral infection and 3 of polymicrobial infection. In cohort B, mNGS verified 7 cases of bacterial infection, 3 of fungal infection, 26 of viral infection, and 5 of polymicrobial infection in infected participants with negative CMT results.

Lastly, we retrospectively reviewed the antibiotic regimen and clinical outcomes of the 86 patients. The overall treatment success rate (TSR) was 85% (34/40) for patients treated based on the mNGS results, and 56.5% for those receiving empirical therapy (26/46, P=0.004, Supplemental Figure S4 ).