Liquid Biopsy for Invasive Mold Infections in Hematopoietic Cell Transplant Recipients With Pneumonia Through Next-Generation Sequencing of Microbial Cell-Free DNA in Plasma

Abstract

Background

Noninvasive diagnostic options are limited for invasive mold infections (IMIs). We evaluated the performance of a plasma microbial cell-free DNA sequencing (mcfDNA-Seq) test for diagnosing pulmonary IMI after hematopoietic cell transplant (HCT).

Methods

We retrospectively assessed the diagnostic performance of plasma mcfDNA-Seq next-generation sequencing in 114 HCT recipients with pneumonia after HCT who had stored plasma obtained within 14 days of diagnosis of proven/probable Aspergillus IMI (n = 51), proven/probable non-Aspergillus IMI (n = 24), possible IMI (n = 20), and non-IMI controls (n = 19). Sequences were aligned to a database including >400 fungi. Organisms above a fixed significance threshold were reported.

Results

Among 75 patients with proven/probable pulmonary IMI, mcfDNA-Seq detected ≥1 pathogenic mold in 38 patients (sensitivity, 51% [95% confidence interval {CI}, 39%–62%]). When restricted to samples obtained within 3 days of diagnosis, sensitivity increased to 61%. McfDNA-Seq had higher sensitivity for proven/probable non-Aspergillus IMI (sensitivity, 79% [95% CI, 56%–93%]) compared with Aspergillus IMI (sensitivity, 31% [95% CI, 19%–46%]). McfDNA-Seq also identified non-Aspergillus molds in an additional 7 patients in the Aspergillus subgroup and Aspergillus in 1 patient with possible IMI. Among 19 non-IMI pneumonia controls, mcfDNA-Seq was negative in all samples, suggesting a high specificity (95% CI, 82%–100%) and up to 100% positive predictive value (PPV) with estimated negative predictive values (NPVs) of 81%–99%. The mcfDNA-Seq assay was complementary to serum galactomannan index testing; in combination, they were positive in 84% of individuals with proven/probable pulmonary IMI.

Conclusions

Noninvasive mcfDNA-Seq had moderate sensitivity and high specificity, NPV, and PPV for pulmonary IMI after HCT, particularly for non-Aspergillus species.

In this retrospective study of 114 hematopoietic cell transplant recipients with pneumonia, microbial cell-free DNA sequencing (mcfDNA-Seq) had moderate sensitivity and high specificity for detecting pathogenic molds in plasma, particularly for non-Aspergillus species. McfDNA-Seq may be a useful test to reduce invasive procedures and guide treatment.

Pulmonary invasive mold infections (IMIs) remain a significant cause of morbidity and mortality after hematopoietic cell transplant (HCT) [1]. While pulmonary aspergillosis is more common, the incidence of infections with non-Aspergillus molds is increasing [2–4]. Diagnosis of pulmonary IMI remains challenging since the majority of filamentous molds cannot be isolated in conventional blood cultures and given the relative insensitivity of bronchoalveolar lavage (BAL) fluid testing [5]. Noninvasive diagnostic options are limited, particularly for non-Aspergillus molds [6]. Given differences in the spectrum of activity and toxicities associated with mold-active therapies, early, specific diagnosis of IMI is important to enable targeted treatment.

Noninvasive diagnostic tests for IMI lack in desired performance. Serum testing for Aspergillus galactomannan is frequently negative, prompting use of empiric therapies or invasive tests. Serum testing for Aspergillus using polymerase chain reaction (PCR) was recently recommended for immunocompromised patients with suspected IMI [7], although performance may be worse in patients receiving antimold therapy [8]. Meta-analyses of serum PCR tests for Aspergillus report a mean sensitivity of 80.5% (95% confidence interval [CI], 73.0%–86.3%) and specificity of 78.5% (95% CI, 67.8%–86.4%) with a single positive PCR test; 2 positives improve specificity at the expense of sensitivity [9–11]. There are no guideline-recommended noninvasive tests for non-Aspergillus IMI. PCR tests for Mucorales species based on 18S ribosomal RNA have been developed [12, 13], but none are incorporated into guidelines given limited supporting data [14, 15]. Other noninvasive tests such as serum (1→3)-β-D-glucan (BDG), a cell wall component of many fungi, are nonspecific.

Microbial cell-free DNA sequencing (mcfDNA-Seq) using plasma is an alternative to invasive tests for species-level identification of microorganisms [16–19]. Noninvasive tests that avoid aerosol-generating procedures are particularly relevant in the context of the current coronavirus disease 2019 (COVID-19) pandemic. In a prospective study using mcfDNA-Seq to screen for invasive fungal infections in cancer patients, 4 of 6 (67%) clinically proven cases were detected [19]. In a study of 9 participants with proven invasive fungal infections, plasma mcfDNA-Seq identified the same pathogen in 78% (7/9) [17]. While these studies address the potential diagnostic yield of noninvasive mcfDNA-Seq testing, they are limited by small numbers and lack of controls. Here we address limitations of prior studies by testing the performance of the same mcfDNA-Seq assay in a large, well-characterized cohort of HCT recipients with pulmonary IMI due to Aspergillus- and non-Aspergillus molds, in addition to appropriate controls.

METHODS

Patients and Samples

We conducted a retrospective study of 117 HCT recipients at the Fred Hutchinson Cancer Research Center (FHCRC) who were evaluated for pneumonia after HCT between 1999 and 2018 and met criteria for prespecified categories of lower respiratory tract disease (LRTD) consisting of (1) proven or probable Aspergillus IMI (n = 51); (2) proven or probable non-Aspergillus IMI (n = 24); (3) possible IMI (n = 20); and (4) non-IMI pneumonia controls with proven bacterial or viral infections (n = 19). All patients had a stored plasma sample frozen at –20°C within 14 days of diagnosis, and the closest sample to diagnosis was used for testing. The study was approved by the FHCRC Institutional Review Board.

Diagnostic Evaluation and LRTD Categorization

Diagnostic workup included lung imaging in all patients and BAL and/or biopsy in the majority of patients. Most patients had BAL fluid and/or serum tested for Aspergillus with the galactomannan index (GMI) using the Bio-Rad Platelia Assay (Bio-Rad, Hercules, California). A GMI ≥0.5 with a confirmatory test processed separately on the same sample was considered positive for BAL fluid and serum, consistent with the accepted diagnostic cutoffs at the time of the included IA cases. Samples missing clinical test results for the GMI were retrospectively tested if available. Testing of BAL fluid for fungal pathogens with laboratory-developed PCR assays was available at FHCRC after 2003 and was performed at the discretion of treating physicians. The number of patients in each cohort who did not undergo a BAL or did not have GMI testing of BAL fluid or serum are in Supplementary Table 1. The date of diagnosis was based on the date of a positive microbiologic or histopathologic finding for proven or probable IMI cases, the date of a BAL or computed tomographic scan of the chest (if BAL not performed) for possible IMI cases, and the date of the BAL for non-IMI pneumonia cases. Additional details pertaining to standard microbiologic testing are shown in the Supplementary Data.

We categorized proven, probable, and possible IMI pneumonia according to the 2008 revised European Organization for Research and Treatment of Cancer/Invasive Fungal Infections Cooperative Group and the National Institute of Allergy and Infectious Diseases Mycoses Study Group (EORTC/MSG) criteria [20]. Laboratory-developed PCR results are reported but were not used as diagnostic criteria for IMI categorization. Control patients with non-IMI pneumonia (bacterial or viral) were categorized according to consensus guidelines [21, 22]. All patients in the non-IMI pneumonia control category were required to have complete workup for pulmonary IMI with BAL fluid and serum GMI testing, as well as a clearly documented bacterial or viral cause of LRTD. Final determinations of diagnostic categories were made by authors J. A. H. and C. E. F.

mcfDNA-Seq From Plasma Samples

Frozen plasma was thawed and centrifuged to remove cells and debris, and cell-free DNA was extracted from 250 µL, converted to DNA libraries, and sequenced at a College of American Pathologists–accredited and Clinical Laboratory Improvement Amendments–certified laboratory according to previously validated methods (Karius, Redwood City, California) [17]. Human reads were removed and remaining sequences aligned to a curated database including >400 fungi. A research-use-only analytical pipeline, implementing an optimized significance threshold and BLAST parameters tailored for fungal mcfDNA reads, was utilized. Organisms meeting this optimized significance threshold for detection of fungal pathogens were reported. Pipeline optimization and all analyses of sequencing data were blinded to all clinical data and LRTD categorizations.

Data Collection and Statistical Considerations

We abstracted patient and laboratory data from medical records and databases. We calculated the sensitivity and specificity of the mcfDNA-Seq assay for pulmonary IMI overall and in the aforementioned subcategories of LRTD. As this was a case-control study, we extrapolated estimates for the prevalence of IMI to approximate negative predictive values (NPVs) [2, 8, 23]. The effect on sensitivity of time between plasma collection and diagnosis, year of sample collection, serum GMI positivity, and receipt of mold-active agents for ≥3 days prior to sample collection was analyzed. Our sample size for patients with any proven or probable IMI was based on a priori calculations to determine the number of patients needed to provide a 95% confidence interval (CI) within 15%–20% assuming a test sensitivity ranging from 70% to 80% (Supplementary Table 2). We preselected a target sample size of 20 patients for the exploratory objective of evaluating possible IMI cases.

RESULTS

Patients and Samples

We identified 117 subjects who were evaluated for pneumonia after HCT, had a plasma sample available within 14 days of diagnostic testing, and met our prespecified infection categories. Three patients were excluded due to samples failing quality metrics. The final patient cohorts consist of cohort 1 (51 subjects with proven [n = 2] or probable [n = 49] pulmonary aspergillosis; cohort 2 (24 subjects with proven [n = 11] or probable [n = 13] non-Aspergillus IMI (5 of whom also had proven or probable pulmonary aspergillosis); cohort 3 (20 subjects with possible pulmonary IMI); and cohort 4 (19 subjects with non-IMI bacterial or viral pneumonia). Plasma samples were obtained within a median of 3 days of the clinical diagnosis (interquartile range [IQR], 1–7 days pre- or postdiagnosis). Antimold active therapy was being administered to 41% of patients at the time of sample testing. Additional demographics, clinical characteristics, and diagnostic test results are shown in Table 1.

Table 1.

Demographic and Clinical Characteristics of the Cohort

Variable	All Subjects (N = 114)	P/P Aspergillosis (n = 51)a	P/P Non-Aspergillus IMI (n = 24)	Possible IMI (n = 20)	Non-IMI Controls (n = 19)
Age, y, median (range)	51 (16–74)	50 (16–74)	53 (21–66)	58 (33–67)	47 (26–72)
Male sex	61 (54)	30 (59)	9 (38)	13 (65)	9 (47)
White race	97 (85)	45 (88)	19 (79)	20 (100)	13 (68)
Year of HCT
1999–2009	75 (66)	46 (90)	12 (50)	4 (20)	13 (68)
2010–2018	39 (34)	5 (10)	12 (50)	16 (80)	6 (32)
HCT type
Allogeneic	111 (97)	50 (98)	22 (92)	20 (100)	19 (100)
Autologous	3 (3)	1 (2)	2 (8)	0 (0)	0 (0)
Stem cell source
BM/PBSC	106 (93)	49 (96)	20 (83)	20 (100)	17 (89)
Cord blood	8 (7)	2 (4)	4 (17)	0 (0)	2 (11)
HLA match
Matched related	29 (25)	9 (18)	9 (37.5)	4 (20)	7 (37)
Matched unrelated	44 (39)	22 (43)	5 (21)	12 (6)	5 (26)
Mismatched related	5 (4)	1 (2)	1 (4)	0 (0)	3 (16)
Mismatched unrelated	36 (32)	19 (37)	9 (37.5)	4 (20)	4 (21)
Underlying disease
Acute leukemias and myelodysplastic syndromes	52 (46)	30 (59)	6 (25)	5 (25)	11 (58)
Other malignancies	51 (45)	20 (39)	11 (46)	15 (75)	5 (26)
Nonmalignant conditions	11 (9)	1 (2)	7 (29)	0 (0)	3 (16)
Days between HCT and diagnosis, median (range)	72 (2–1824)	71 (6–1824)	90 (2–338)	86 (6–1130)	38 (9–89)
Days between plasma and diagnosis, median (IQR)	3 (1–7)	3 (1–7)	2.5 (1–5)	4 (3–6.5)	2 (1–10)
Antimold agent at time of tested plasma
Any	47 (41)	17 (33)	14 (58)	6 (30)	10 (52)
Median duration, d (IQR)	29 (11–80)	23 (12–46)	20 (6–84)	34 (10–94)	80 (19–226)
Voriconazole	37 (32)	16 (31)	7 (29)	6 (30)	8 (42)
Posaconazole, isavuconazole, or amphotericin	10 (9)	1 (2)	7 (29)	0 (0)	2 (10)
Positive diagnostic testb
Histopathology	15 (13)	5 (10)	10 (42)	0 (0)	0 (0)
Culture	30 (26)	14 (3)	15 (63)	0 (0)	0 (0)
BAL GMI	26 (23)	20 (39)	6 (25)	0 (0)	0 (0)
Serum GMI	29 (25)	29 (57)	0 (0)	0 (0)	0 (0)
BAL PCR	18 (16)	4 (8)	13 (54)	1c (5)	0 (0)

Data are presented as No. (%) unless otherwise indicated.

Abbreviations: BAL, bronchoalveolar lavage; BM, bone marrow; GMI, galactomannan index; HCT, hematopoietic cell transplant; HLA, human leukocyte antigen; IMI, invasive mold infection; IQR, interquartile range; PBSC, peripheral blood stem cell; PCR, polymerase chain reaction assay; P/P, proven or probable.

^aFive subjects had proven disease and 46 subjects had probable disease.

^bCategories are not mutually exclusive. The number of patients who did not undergo BAL or have Aspergillus GMI testing in BAL fluid or serum are detailed in Supplementary Table 1. BAL PCR testing was performed in 24 patients overall.

^cIn this patient, the serum GMI, BAL GMI, and BAL cultures were negative.

Sensitivity and Additive Diagnostic Value of Plasma mcfDNA-Seq for Pulmonary IMI

In 75 participants with any proven or probable pulmonary IMI (cohorts 1 and 2 combined), plasma mcfDNA-Seq detected ≥1 pathogenic mold from 38 patients, yielding a sensitivity of 51% (95% CI, 39%–62%; Figure 1A). The genus and species of all detected molds are depicted in Figure 2.

Figure 1.

Sensitivity of plasma microbial cell-free DNA sequencing for identifying pulmonary invasive mold infections. ^aFour subjects with Aspergillus spp detection also had detection of a non-Aspergillus spp mold, and an additional 3 subjects without Aspergillus spp detected had detection of a non-Aspergillus spp mold. ^bAll samples with detected Aspergillus spp DNA were collected within 7 days of the clinical diagnosis. ^cFor this group, we excluded patients only receiving voriconazole. Abbreviations: GMI, galactomannan index; IMI, invasive mold infection; P/P, proven or probable.

Figure 2.

Pathogenic molds detected using plasma microbial cell-free DNA sequencing in patients with clinically proven or probable invasive mold infection (IMI). Numbers indicate the unique number of individuals.

Among 51 participants in cohort 1 with proven or probable pulmonary aspergillosis, we detected Aspergillus DNA in plasma from 16 patients, yielding a sensitivity of 31% (95% CI, 19%–46%). Both clinically proven cases were identified by mcfDNA-Seq; in 1 of these cases, both the serum test and BAL GMI test were negative. The mcfDNA-seq test identified additional non-Aspergillus molds that were not detected by clinical testing in 7 patients, 4 of whom also had Aspergillus detected. All of these patients except for 1 had findings potentially consistent with non-Aspergillus pulmonary IMI based on nonspecific histopathologic findings of hyphal elements in 2 patients, death due to progressive respiratory failure in 1 patient only receiving voriconazole, and breakthrough IMI in 3 patients who were receiving voriconazole prophylaxis. Including the detection of all molds, the sensitivity increased to 37% (95% CI, 24%–52%). A summary of the diagnostic yield of the plasma mcfDNA assay in patients with proven or probable pulmonary aspergillosis is depicted in Figure 3A.

Figure 3.

Summary of the diagnostic yield of plasma microbial cell-free DNA sequencing (mcfDNA-Seq) in patients with proven or probable (P/P) invasive mold infection (IMI). A, Among 51 participants in cohort 1 with P/P pulmonary aspergillosis, Aspergillus (Asp) DNA was detected in plasma from 16 patients (31%). Non-Aspergillus molds were detected in an additional 7 patients (14%, 4 of whom included those with Aspergillus detection), resulting in an overall sensitivity for detection of IMI of 37%. B, Among 24 participants in cohort 2 with P/P pulmonary IMI with non-Aspergillus spp, non-Aspergillus molds were detected in 19 patients (79%). C, Among 62 patients with P/P IMI who had a serum galactomannan index (GMI) test, serum GMI was positive in 29 (47%) individuals, mcfDNA-Seq detected a mold in 31 (50%) individuals, and both tests were positive in 8 individuals, yielding a composite sensitivity of 84% (52/62).

Among 24 patients in cohort 2 with proven or probable non-Aspergillus pulmonary IMI, mcfDNA-Seq detected non-Aspergillus molds in 19 patients, yielding a sensitivity of 79% (95% CI, 56%–93%). The sensitivity of mcfDNA-Seq was superior to BAL fluid culture and similar to BAL fluid PCR for fungal pathogens: 11 of 21 (52%) patients who had a BAL had positive BAL fluid fungal cultures, and 13 of 17 (76%) patients who had BAL fluid tested by targeted mold PCR assays were positive. A summary of the diagnostic yield of the plasma mcfDNA-Seq test in patients with proven or probable non-Aspergillus IMI is depicted in Figure 3B.

Among 20 patients in cohort 3 with possible IMI pneumonia, mcfDNA-Seq identified Aspergillus fumigatus in 1 patient.

Specificity, Positive Predictive Value, and Negative Predictive Value of Plasma mcfDNA-Seq for Pulmonary IMI

Among 19 control patients from cohort 4 with non-IMI pneumonia, mcfDNA-Seq was negative in all patients, suggesting a high specificity (95% CI, 82%–100%) and up to 100% positive predictive value (PPV) for IMI pneumonia. Additionally, among the 30 individuals with clinical identification of a mold based on histopathology, culture, or PCR and who had detection of mold DNA by the mcfDNA-Seq assay, the reported genus and/or species were the same in all instances; greater resolution of the genus or species was achieved by mcfDNA-Seq compared to clinical testing in 47% of these cases. Based on estimated prevalences of pulmonary IMI among HCT recipients evaluated for pneumonia, the estimated NPV of the test ranged from 80% to 99% in each cohort (Table 2), but will vary depending on the true prevalence in the tested patient population.

Table 2.

Estimated Performance Characteristics of Plasma Microbial Cell-Free DNA Sequencing for Identifying Pulmonary Invasive Mold Infections

Clinical Diagnosis	Sensitivity	Specificitya	NPVb	PPVa
Any IMI, proven or probable (n = 75)	51% (39%–62%)	100% (82%–100%)	89%–92% (87%–94%)	100%
Cohort 1: Aspergillus IMI, proven or probable (n = 51)	31% (19%–46%)	100% (82%–100%)	89%–93% (87%–94%)	100%
Cohort 2: Non-Aspergillus IMI, proven or probable (n = 24)	79% (56%–93%)	100% (82%–100%)	99%–99% (98%–100%)	100%
Cohort 3: Possible IMI (n = 20)	5% (0%–25%)	100% (82%–100%)	81%–86% (81%–86%)	100%

Data in parentheses indicate 95% confidence intervals.

Abbreviations: IMI, invasive mold infection; NPV, negative predictive value; PPV, positive predictive value.

^aThe specificity and PPV of the assay are estimated to be 100% based on the finding of no false positives in 19 negative controls. The 95% confidence interval for the PPV could not be calculated.

^bThe NPV was calculated as the number of true negatives divided by the sum of the number of true negatives and false negatives. To determine the number of true negatives, we used category-specific estimates of population prevalence and a specificity of 100% based on cohort 4. For any proven or probable IMI, the NPV was based on estimated population prevalences of 15%–20%. For cohort 1, the NPV was based on estimated population prevalences of 10%–15%. For cohort 2, the NPV was based on estimated population prevalences of 2.5%–5%. For cohort 3, the NPV was based on estimated population prevalences of 15%–20% [2, 8, 23].

Sensitivity of Plasma mcfDNA-Seq for Proven or Probable Pulmonary IMI in Subgroups Defined by Clinical Characteristics

Due to the retrospective nature of this study, there was variability in patient characteristics and the timing of plasma available for testing. To explore the impact that specific variables may have on the performance of the plasma mcfDNA-Seq for the diagnosis of proven or probable IMI, we evaluated the sensitivity of the assay in subgroups defined by samples obtained within ±3 days of the clinical diagnosis, within the last 10 years, from patients with a positive serum GMI, and from patients not receiving mold-active drugs (Figure 1B). When considering test results obtained within ±3 days of the clinical diagnosis, the sensitivity of mcfDNA-Seq increased by 10% for any proven or probable IMI pneumonia. Additionally, all samples with Aspergillus spp DNA detection were collected within 7 days of the clinical diagnosis. There was less of an apparent effect of the other variables on test performance, including the receipt of mold-active therapy.

Noninvasive Detection of Proven or Probable Pulmonary IMI by Combining mcfDNA-Seq and Serum GMI

To explore the potential complementarity of serum GMI testing with mcfDNA-Seq, we assessed to what extent combining these tests augmented the noninvasive diagnosis of pulmonary IMI. Among the 75 participants with proven or probable IMI, serum GMI testing was positive in 29 of 62 (47%) tested individuals. In this subgroup of 62 patients, mcfDNA-Seq detected a mold in 31 (50%) individuals, only 8 of whom also had a positive serum GMI. Together, these 2 tests had 84% sensitivity in this subgroup of patients with proven or probable IMI (Figure 3C).

DISCUSSION

The limited armamentarium of noninvasive biomarkers to diagnose IMI in high-risk patients is a critical barrier to providing targeted and timely interventions. In this large case-control study of HCT recipients with proven, probable, and possible pulmonary IMI and non-IMI pneumonia controls, we found moderate sensitivity and high specificity, NPV, and PPV of a plasma mcfDNA sequencing assay for pulmonary IMI. The mcfDNA-Seq assay was complementary to serum GMI; combined, at least 1 of these noninvasive tests was positive in 84% of individuals with proven or probable pulmonary IMI. Thus, plasma mcfDNA-seq has great potential to improve timely delivery of appropriate antifungal therapy and reduce the need for invasive procedures in HCT recipients with LRTD.

In HCT recipients with clinical diagnoses of proven or probable pulmonary IMI, we demonstrated that plasma mcfDNA-Seq had good sensitivity for non-Aspergillus IMI (79%), exceeding positive results obtained by BAL fluid culture and similar to results from BAL fluid PCR. Plasma mcfDNA-Seq had lower sensitivity for Aspergillus IMI (31%) but identified other pathogenic molds in a number of cases that were not identified clinically. This could be related in part to misclassification of probable aspergillosis with currently available diagnostics. Although plasma mcfDNA-Seq had lower sensitivity for detection of Aspergillus compared to studies using serum GMI, BDG, and Aspergillus-targeted PCRs [9, 24], mcfDNA-Seq had sensitivities and specificities for non-Aspergillus IMI that were similar to these noninvasive Aspergillus biomarkers and provided species-level resolution. Thus, this assay may provide a major advance to the field given that no approved noninvasive tests for non-Aspergillus IMI currently exist. Additionally, the sensitivity of mcfDNA-Seq increased by 10% when restricted to plasma collected within 3 days of clinical diagnoses. Interestingly, there was minimum overlap between patients with a positive serum GMI and mold detection by plasma mcfDNA-seq, highlighting the utility of these noninvasive tests to provide a positive result in up to 84% of patients with proven or probable disease when used in combination. The mcfDNA-Seq assay also had excellent PPV and NPV across all categories of pulmonary IMI based on prevalence estimates within the context of current transplantation practices. Notably, the test had limited additional diagnostic value in patients with possible IMI, which continues to be a challenging population in which to make a microbiologic diagnosis.

The availability of a noninvasive diagnostic test for pulmonary IMI has clear potential to improve patient outcomes. In randomized controlled trials comparing standard diagnostic strategies with and without PCR-based testing of blood for Aspergillus in high-risk patients, the addition of PCR resulted in reduced use of empiric antifungal treatment, earlier diagnosis, and improved fungal-free survival [15, 25, 26]. However, there are fewer tools for blood-based biomarkers to diagnose non-Aspergillus IMI [27], which are increasing in the era of mold-active prophylactic or empiric therapy [4, 28, 29]. Metagenomic next-generation sequencing assays have emerged as a promising technology to facilitate pathogen identification, improve turnaround time, streamline testing, and improve clinical outcomes [30, 31]. These tools are particularly relevant in the context of reducing invasive procedures to establish diagnoses [32].

While prior studies evaluating molecular-based tests to diagnose IMI provide important proof-of-concept data [12, 13, 15, 17, 19, 24–26], most have significant limitations including small patient numbers, inclusion of only proven cases, lack of controls, use of tests that detect a limited number of fungi, and/or use of nonhuman samples. Our study design addressed many of the limitations of prior studies. First, we leveraged a unique sample biorepository to include a large number of patients within well-defined categories of pulmonary IMI, particularly with the less common non-Aspergillus molds. We also included a control group that was comprised of HCT recipients with non-IMI pulmonary infections, making the comparison more reflective of the target patient population. We used a comprehensive sequencing approach optimized for detection of IMI beyond that reported previously [17] by tailoring the computational pipeline to better detect DNA reads generated by invasive molds from a panel of >400 species. Additionally, we analyzed several factors that could plausibly affect the performance of the assay given that our study relied on previously obtained samples over a period of 20 years. The overall findings are notable given that 41% of patients were receiving mold-active medications at the time of sample collection, and prior studies of molecular diagnostics for Aspergillus demonstrate a substantial reduction in sensitivity in this context [8]. Similar to PCR-based strategies to detect IMI in this patient population, the performance of plasma mcfDNA-Seq may improve with serial sampling as demonstrated for other PCR-based strategies [9, 24].

This study had several limitations. This was a single-center, retrospective study. Due to the relative rarity of non-Aspergillus mold infections, the study spanned a large time period to improve sample size, and it is possible that older samples could affect the performance of the test due to prolonged sample storage or differences in clinical practice. In the present study, there was no difference noted in sensitivity between earlier and later samples, and controls spanned the same timeframe. The varied timing of sample collection relative to clinical diagnosis did appear to affect our results, with decreased sensitivity in samples collected >3 days from the time of diagnosis, suggesting real-time clinical use may result in better performance characteristics. We only tested 1 sample per patient, and serial sampling may improve test performance. Additionally, BDG is not routinely tested at our institution, so these results were not available. Twelve of 49 patients with probable aspergillosis would not meet updated diagnostic criteria [33], although mcfDNA-seq detected Aspergillus in 3 of these patients and excluding them does not substantively change the sensitivity (33% vs 31%). Finally, this assay is unable to provide antifungal susceptibility information.

In conclusion, this study highlights the potential utility of noninvasive mcfDNA-Seq in HCT recipients with suspected IMI pneumonia. This test may serve as a useful adjunctive diagnostic assay to facilitate timely and targeted antimicrobial therapy and reduce invasive procedures. Prospective studies are required to further support the utility of mcfDNA-Seq as an additional tool for noninvasive diagnosis of pneumonia in HCT recipients and to assess its performance in different patient populations.

Supplementary Data

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Author contributions. J. A. H., C. E. F., M. B., and D. K. H. designed the study. J. A. H., C. E. F., J. M., J. K.-C., and T. S.-A. collected data. D. K. H., S. C. D., C. H., D. H., L. B., A. A. A., and T. A. B. performed the testing. J. A. H., C. E. F., S. C. D., and T. A. B. analyzed the data. J. A. H. wrote the first draft. All authors contributed to the writing and revision of the manuscript and approved the final version.

Acknowledgments. The authors acknowledge the Infectious Disease Sciences Biorepository at the Fred Hutchinson Cancer Research Center for providing samples for this study.

Financial support. This study was supported by an investigator-initiated award from Karius, Inc (to J. A. H. and C. E. F.) and by the National Institute of Allergy and Infectious Diseases (grant number K23 AI119133 to J. A. H.). Additional resources were provided by the National Institutes of Health (grant numbers HL143050 [to C. E. F.], HL088021, CA78902, CA18029, CA015074, and HL122173).

Potential conflicts of interest. J. A. H. received research support from Karius for this study; serves as a consultant for Amplyx, Allovir, and Gilead; and has received research support/personal fees from Takeda, outside the submitted work. D. K. H., S. C. D., A. A. A., C. H., D. H., L. B., and T. A. B. are employees of Karius. M. B. reports grants and personal fees from Merck and Co, Astellas, Takeda, and Gilead Sciences, all outside the submitted work. C. E. F. received research support from Karius for this study and has received research support outside the submitted work from Gilead Sciences. All other authors report no potential conflicts of interest.

All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.