1. Introduction
Molecular tools as diagnostic criteria for invasive fungal diseases (IFD) has long been questioned because of a lack of reproducibility and insufficient standardization of protocols. Thanks to initiatives such as FPCRI (www.fpcri.eu [1]) and to the dramatic improvement of the quality assessment of molecular technics, Aspergillus PCR is now included in the new EORTC criteria for classification [2]. Regarding intensive care units (ICU) patients, the classification of IFD mainly refers on criteria adapted from neutropenic patients or relies on single center experiences. One algorithm has emerged as a valuable tool to classify invasive aspergillosis (IA) in ICU patients: the AspICU algorithm [3]. This classification is considered as robust because it has been evaluated in patients for whom autopsy results were available, but it is quite awkward to use in routine practice, particularly in COVID-19 patients with clinical and CT-scan signs hard to interpret [4]. Besides, it does not include molecular markers, which are now used routinely [5].
During COVID-19, patients presenting an acute respiratory distress syndrome (ARDS) shared risk factors and underlying diseases classically reported for IA, such as intubation and mechanical ventilation, corticosteroid therapy, immunological storm with high production of inflammatory cytokines. Warnings following preliminary cohort studies from various countries prompted the monitoring of fungal colonization and co-infections in SARS-CoV-2-infected patients hospitalized in an ICU. However, the entry criterion for putative IA, according to Blot et al., is an Aspergillus-positive culture endotracheal aspirate, which may lack specificity. In the recent review by Arastehfar et al. [6], many COVID-19-associated pulmonary aspergillosis (CAPA) benefited from galactomannan (GM) testing of bronchoalveolar fluid (BALF) or even of tracheal aspirates (not approved by the manufacturer). However, some laboratories, such as ours, have stopped various manipulations of highly SARS-CoV-2-infected samples in order to limit the exposure of laboratory technicians to viral infection. Then, direct examination of respiratory samples or galactomannan (GM) determination in broncho-alveolar lavage have thus been replaced by the systematic use of molecular tools. While performances of blood biomarkers such as GM, (1-3)β-d-glucan (BDG) or Aspergillus DNA detection are well evaluated in neutropenic patients, their clinical value is far less known in other conditions and still need evaluation in an ICU.
Here, our objective was to evaluate the concordance between molecular detection of Aspergillus in respiratory and culture and concordance between blood PCR and serum GM. We also aimed at assessing the ability of Aspergillus PCRs to help categorizing patients in the continuum of colonization to invasive infection in COVID-19 patients. Arguments to complement AspICU criteria are suggested.

2. Materials and Methods

2.1. Population of Patients
Forty-five intubated and mechanically ventilated patients hospitalized in a “COVID-19 ICU” of Rennes teaching hospital were screened for this study and benefited from a systematic monitoring to detect Aspergillus.
The hospital’s ethics committee (N 20-56 obtained the 30 April 2020) approved the study. The presence of SARS-CoV-2 in respiratory specimens (nasal and pharyngeal swabs or sputum) was detected by real time reverse transcription-polymerase chain reaction (RT-PCR) methods.
The following data were recorded: age, patient’s preexisting condition (current smoking, diabetes, hypertension, cardiovascular disease, pulmonary disease, and kidney disease), body mass index, ICU length of stay, duration of mechanical ventilation, ventilator-free days at day 28, need for prone position ventilation, and death in the ICU. Initial clinical laboratory workup included a complete blood count and serum biochemical tests. Chest CT scans were performed during the ICU hospitalization. The Simplified Acute Physiology Score (SAPS II) and the Sepsis-Related Organ Failure Assessment (SOFA) score at admission in ICU, at day 7 and 14 days after admission were used to assess severity [7,8].

2.2. Aspergillus Detection
Respiratory samples, either bronchial or endotracheal aspirates or bronchoalveolar lavages, were systematic twice weekly and Aspergillus detection was performed using culture and real-time quantitative PCR, but GM was not performed to avoid any risk of lab contamination.
Briefly, respiratory samples were first digested using v/v digestEUR (Eurobio) for 30 min under shaking. Mycological culture were performed after centrifugation of fluidified samples by inoculation of 100–200 µL of pellet on Sabouraud-Chloramphenicol dextrose Agar plates, and incubated for 7 days at 30 °C and 37 °C. Mold identification at genus or species complex level was performed microscopically, and confirmed at species level using MALDI-ToF mass spectrometry (MALDI Biotyper, Bruker France, Marne-la-Vallée, France), after fungal material extraction [9]. Spectrum profiles were then submitted to the Mass Spectrometry Identification (MSI) online database for definitive identification (https://msi.happy-dev.fr/ [10]).
For molecular detection, 200 µL of plain fluidified respiratory sample underwent immediate SARS-CoV-2 inactivation by heating at 56 °C overnight in ATL Lysis buffer (Qiagen, Saint-Quentin Fallavier, France), before DNA extraction using the EZ1 DSP virus kit (Qiagen) on a EZ1 Advanced XL device (Qiagen). Molecular detection of A. fumigatus was done using a 28S rDNA Aspergillus-targeted PCR, as previously published [11,12].
In case of Aspergillus positive culture and/or positive PCR in respiratory samples, additional tests were performed on serum, i.e., detection of GM (Platelia GM Aspergillus, Biorad, Marnes-la-Coquette, France), Aspergillus PCR and detection of anti-Aspergillus antibody by ELISA (Platelia IgG Aspergillus, Biorad) and in-house immunoelectrophoresis. Briefly, Aspergillus PCR was performed on 1 mL of serum extracted using MagNA Pure 24 Total NA Isolation kit (Roche diagnostics, Meylan, France) on a MagNA Pure 24 device (Roche diagnostics), according to manufacturer recommendations. DNA was eluted in a volume of 50 µL.

2.3. Statistical Analysis
Continuous variables were expressed as median (interquartile range, IQR) and compared using the nonparametric Mann–Whitney U or Kruskal–Wallis test. Dunn’s correction tests were performed if multiple comparisons were requested. Qualitative data were compared using Chi-square test. Tests were two-sided with significance set at α less than 0.05.
Concordance between categorical results from diagnostic tests was performed using the percent agreement coefficient and Cohen’s kappa coefficient (κ). When comparing quantitative data, an ANOVA test was performed. All data were analyzed with GraphPad Prism 8.4 (GraphPad Software, La Jolla, CA, USA).

3. Results

3.1. Patient Aspergillus Status
A cohort of 45 COVID-19 intubated and mechanically ventilated patients for ARDS was followed. Patients benefited from a systematic screening for Aspergillus. Overall, 211 respiratory samples (culture and PCR) and 32 serum samples (GM detection and Aspergillus PCR) were collected. The mean number of respiratory samples until patient discharge from ICU was 3.8 (median = 3).
We categorized these 45 patients according to the AspICU algorithm and propose two alternative classification methods presented in Table 1: the AspICU algorithm associated to PCR results in respiratory and serum samples, and a modified AspICU proposal. Thirty patients did not present any biological criteria of aspergillosis with any of the algorithms. According to the AspICU classification incorporating PCR detection, 15 were classified as having putative aspergillosis because they met all criteria reported by Blot et al., i.e., compatible clinical signs, abnormal thoracic medical imaging on CT scan and positive screening for Aspergillus on respiratory samples. However, in this particular context of COVID-19 with all ARDS patients presenting compatible clinical signs and abnormal chest CT imaging in all likelihood lacking specificity, we decided to use a modified AspICU algorithm taking into account blood markers; we classified eight patients as colonized and seven patients with a putative/probable IA (Table 1 and Table 2).

3.2. Demographic, Clinical and Biological Characteristics
Demographic, clinical and biological baseline characteristics at admission are detailed in Table 3 and Table S1. Basic demographic characteristics were well-balanced between the three groups of patients (no aspergillosis, Aspergillus colonization, putative/probable aspergillosis). Of note, we observed a high proportion (71.1%) of male patients in the study population. Clinical and biological baseline data did not differ among the three groups, except C-reactive protein which was higher in the “no aspergillosis” group. Regarding the severity scores at admission, no differences were observed either, among the groups of patients, but SAPS II and SOFA scores at day one tended to be higher in patients with putative invasive aspergillosis.

3.3. Concordance of Diagnostic Tools
Table 4 gathers the results of the techniques used for Aspergillus detection. DNA detection by PCR showed the highest sensitivity, with a number of positive respiratory samples near twice higher, compared to the culture. Only one sample grew in culture, whereas PCR was negative, but the species obtained in culture was A. tubingensis (Nigri complex species), which is theoretically not amplified when using the 28S-targeted PCR specific for A. fumigatus. Interestingly, the correlation between cultural and molecular quantification showed a significant difference between the two techniques, with a mean Cq threshold of 32.6 ± 0.7 when cultures were negative, highlighting the higher sensitivity of PCR (Figure 1).
Overall, the concordance coefficient between PCR and culture on respiratory samples was 90.52% with a Cohen’s Kappa coefficient of 0.588. Regarding blood samples, three patients had a positive detection of a systemic biomarker: 3/3 had a positive PCR and 2/3 had a positive GM (Table 5). All three patients had a simultaneous detection of Aspergillus in respiratory samples by culture (n = 2) and/or PCR (n = 3). Overall, the concordance coefficient between PCR and culture on respiratory samples was 93.75% with a Cohen’s Kappa coefficient of 0.632.

3.4. Relevance of Various Tests and Categorization of Patients and Outcome
Table 6 presents the classification of the 45 patients using original or modified AspICU algorithms. It appears that using an AspICU algorithm, nine patients were considered as having a putative IA (22% of the cohort). When including PCR, the number of patients with putative IA would increase from 9 to 15 (33%) patients, while most patients might be only colonized because all presented compatible clinical signs and abnormal chest CT scan (Table 5). Regarding Aspergillus detection, eight patients had a single detection of fungi using culture and/or PCR in respiratory samples and thus were classified as colonized. One of these patients had a concomitant GM detection in serum (index = 0.551), was not treated and is still alive, thus was considered as a false positive result. Finally, seven (16%) patients presented a heavy burden of Aspergillus in the respiratory tract with repeated positive cultures and/or PCR. In order to rule out a chronic colonization before the episode, an anti-Aspergillus antibody testing was performed and showed negative results. These patients were classified as putative IA, and three of them could even be considered as probable IA because of a positive biomarker of angioinvasion (serum PCR and/or GM) in agreement with EORTC/MSG classification.
Interestingly, following these classification criteria, CT scan abnormalities showed a gradation according to patient group. Diffuse reticular or alveolar opacities were observed in patients classified as probable IA (Figure 2), nodules in half of putative IA, and in colonized patients, only non-specific and hard to interpret signs in the context of COVID-19 infection could be described.
In addition, putative/probable aspergillosis patients appeared more severely ill than patients without aspergillosis, since SOFA score at day seven was significantly higher in this group (p = 0.01) with a continuum between no infection, colonization and IA (Table 5). Similarly, the mean ICU length of stay increased significantly from 12 days in patients without aspergillosis to 23 days in colonized patients, and 27 days in putative/probable invasive aspergillosis (p = 0.02). All patients with a putative/probable IA were treated either with voriconazole or isavuconazole. Only one colonized patient was treated with voriconazole. Six patients died; there was a trend towards higher mortality in the group of putative/probable IA compared to uninfected patients, although not significant (2/7; 28.6%) versus 4/30 (13.3%), respectively (Table 7).

4. Discussion
In France, the global burden of severe fungal infection is estimated at approximately 1,000,000 (1.47%) cases each year [13] and IFD account for a higher risk of mortality in patients with co-morbidities from 9 to 40% [14]. During the COVID-19 pandemic, warning messages considering similarities between Sars-CoV-2 and influenza infections stressed the importance of vigilance towards IFD [15,16]. Local experiences are now published and show high numbers of putative IA [17,18,19,20,21,22].
The diagnosis of IA still remains challenging because of a wide diversity of underlying conditions and growing number of criteria, particularly biological tools [6]. In deeply immunosuppressed patients, such as neutropenic patients, patients under antineoplastic and prolonged corticosteroid therapy or solid organ transplantation, criteria for classification of IFD and notably IA have recently been revised incorporating Aspergillus molecular detection [2]. In ICU, the AspICU algorithm published by Blot et al., [3] is a robust and helpful tool for aspergillosis classification but needs to be more evaluated and even updated. In order to address limitations of the various classification definitions for ICU patients, the ongoing FUNgal infections Definitions in ICU patients (FUNDICU) project aims to develop a standard set of definitions for IFD in critically ill patients [5].
The breaking news of SARS-CoV-2 co-infection urges the need for a critical analysis of the criteria of AspICU algorithm. Indeed, COVID-19 patients, particularly ARDS patients with mechanical ventilation, present with compatible clinical signs as depicted by the algorithm (refractory fever, pleuritic chest pain and rub, dyspnea, hemoptysis and worsening respiratory insufficiency, see [3] for full description) and CT-scan signs are hard to interpret because of COVID-19 CT-scan presentation, leading to absence or very poor discrimination between Aspergillus colonization and infection [19,23]. As a result, IA during COVID-19 has been reported with a possible overestimated high prevalence (until 30%), as favorable outcomes have been described in patients who did not receive any antifungal treatment.
In order to have a well-balanced patient management, limiting unnecessary and costly antifungal treatments while not neglecting the life-threatening feature of IA, we included A. fumigatus PCR as a monitoring tool for fungal detection in both respiratory and blood samples in addition to classical culture and GM approaches but with some restrictions. As expected, PCR allowed detecting Aspergillus in much more respiratory samples. We previously showed that PCR improved the detection of Aspergillus in BAL, with a particular added value in ICU patients compared to hematology patients [11]. Furthermore, PCR using in-house but also marketed kits is also capable of identifying specific gene mutations associated with azole resistance [11,24]. Besides, the sensitivity of GM detection in blood is less sensitive in ICU than for patients with hematological malignancies [5]. Here, the higher sensitivity of Aspergillus detection also incites us to adopt modified criteria for case definition to gain in specificity. Two major changes were introduced to modify the granularity of the classification: (i) the first one is to combine Aspergillus detection in respiratory samples and anti-Aspergillus antibody testing, to distinguish chronic colonization (positive serology) from acute massive colonization (negative serology) and (ii) the second is to introduce of obvious biomarkers of angioinvasion (serum GM and blood PCR), similar to those of the EORTC/MSG classification [2]. Of note, the combination of positive culture, positive anti-Aspergillus antibody testing and positive GM in the context of chronic respiratory diseases characterized a transition step from chronic pulmonary aspergillosis to probable IA [25,26].
Using this refined classification, we were able to categorize our patients in five classes: no infection, colonization, putative IA, probable IA and proven IA (no case of proven IA in the cohort), with a better relevance than the initial AspICU classification, and better specificity than the AspICU + PCR classification. The decision of antifungal treatment onset was taken according to this modified AspICU classification and the outcome observed gives confidence in this patient management. Of course, the limitation of this work is the relatively small number of patients and should be evaluated on larger cohorts in order to correctly analyze the performance of this alternative. A remaining question is also to determine the place of the serum biomarker (1,3)-β-d-glucan in ICU patients, a question that has recently been raised by Honoré et al. [27]
In conclusion, molecular techniques are now key tools for monitoring IFD, particularly IA as recently updated in the EORTC/MSG definitions, but also Pneumocystis jirovecii or mucorales infections. Here, we suggest some adaptations of the AspICU clinical algorithm to gain in sensitivity and specificity. Large multicentric data are needed to confirm this proof of concept study.