key: cord-0812481-ngd1jb6y
authors: Llorens-Rico, V.; Gregory, A. C.; Van Weyenbergh, J.; Jansen, S.; Van Buyten, T.; Qian, J.; Braz, M.; Menezes, S. M.; Van Mol, P.; Vanderbeke, L.; Dooms, C.; Gunst, J.; Hermans, G.; Meersseman, P.; CONTAGIOUS collaborators,; Wauters, E.; Neyts, J.; Lambrechts, D.; Wauters, J.; Raes, J.
title: Mechanical ventilation affects respiratory microbiome of COVID-19 patients and its interactions with the host
date: 2020-12-26
journal: nan
DOI: 10.1101/2020.12.23.20248425
sha: 90411c34a87e0dbf067a6154d0f40dc3f095ca3b
doc_id: 812481
cord_uid: ngd1jb6y

Understanding the pathology of COVID-19 is a global research priority. Early evidence suggests that the microbiome may be playing a role in disease progression, yet current studies report contradictory results. Here, we examine potential confounders in COVID-19 microbiome studies by analyzing the upper (n=58) and lower (n=35) respiratory tract microbiome in well-phenotyped COVID-19 patients and controls combining microbiome sequencing, viral load determination, and immunoprofiling. We found that time in the intensive care unit and the type of oxygen support explained the most variation within the upper respiratory tract microbiome, dwarfing (non-significant) effects from viral load, disease severity, and immune status. Specifically, mechanical ventilation was linked to altered community structure, lower species- and higher strain-level diversity, and significant shifts in oral taxa previously associated with COVID-19. Single-cell transcriptomic analysis of the lower respiratory tract of ventilated COVID-19 patients identified increased oral microbiota compared to controls. These oral microbiota were found physically associated with proinflammatory immune cells, which showed higher levels of inflammatory markers. Overall, our findings suggest confounders are driving contradictory results in current COVID-19 microbiome studies and careful attention needs to be paid to ICU stay and type of oxygen support, as bacteria favored in these conditions may contribute to the inflammatory phenotypes observed in severe COVID-19 patients.

To identify potential associations between COVID-19 severity and evolution and the 95 upper and lower respiratory tract microbiota, we used nasopharyngeal swabs and 96 bronchoalveolar lavage (BAL) samples, respectively. For the upper respiratory tract, we 97 longitudinally profiled the nasopharyngeal microbiome of 58 COVID-19 patients during 98 intensive care unit (ICU) treatment and after discharge to a classical hospital ward 99

following clinical improvement, in conjunction with viral load determination and 100 nCounter immune profiling. For the lower respiratory tract, we analyzed microbial 101 immune cell populations of the host using nCounter (Methods). Of the 112 samples 127 processed and sequenced, 101 yielded over 10,000 amplicon reads that could be 128 assigned to bacteria at the genus level (Figure 1b ; Methods). The microbiome of the 129 entire cohort was dominated by the gram-positive genera Staphylococcus and 130

Corynebacterium, typical from the nasal cavity and nasopharynx 19 . 131 132 Bacterial alpha diversity is strongly associated with ICU stay length 133 134 First, we determined genus-level alpha-diversity for the 101 samples with more than 135 10,000 assigned reads, using Shannon Diversity index (see Methods; Supplementary 136 Table 1 ). We observed that alpha diversity was not significantly correlated to SARS-137

CoV-2 viral load in nasopharyngeal swabs ( Figure 1c ). In contrast, we found the 138 Shannon index to be significantly affected by the sampling moment (Kruskal-Wallis 139 test, p-value = 0.0076; Figure 1d ), with significant differences between swabs procured 140 upon patient admission and later timepoints, suggesting an important effect of disease 141 progression and/or treatment (i.e. due to antibiotics administered throughout ICU 142 stay). We explored these differences further, and observed that Shannon Diversity Table 2) . 177

Of the 19 significant covariates, only 2 accounted for 48.7% non-redundant variation in 179 this dataset, with the rest holding redundant information. These were the patient ID, 180 included due to the longitudinal sampling of patients, and confirming that intra-181 individual variation over time is smaller than patient inter-individual variation 20 , and 182 the type of oxygen support received at the time of sampling (Figure 2a,b) . Notably, the 183 type of oxygen support discriminated samples based on ventilation type, with non-184 invasive ventilation samples (groups 1, 2 and 3) separating from samples from 185 intubated patients (groups 4 to 7; PERMANOVA test, R 2 =0.0642, p-value=0.001). 186

To determine if oxygen support also impacted the microbiome at finer taxonomic 188 resolution, we revisited alpha-diversity at species-and strain-level. We defined species 189 as 97% identity 16S OTUs and strains per species as the clustered 16S sequences 190 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint within each OTU. The species and strain-level diversity per sample were calculated as 191 the number of OTUs and as the mean of the number of strains from five randomly 192 sampled OTU species sampled 1,000 times, respectively. Our analyses revealed both 193 species-and strain-level diversity change with ventilation, even with non-invasive 194 ventilation (e.g. BIPAP, CPAP). Across all samples we observed high species-and low 195 strain-level diversity pre-ventilation, which reversed following any form of ventilation 196 By extracting the amplicon sequence variants (ASVs) corresponding to these 217 differentially abundant genera (see Methods), some of these taxa could be narrowed 218 down to the species level, confirming their origin as typically oral bacteria: for 219 instance, Prevotella species included P. oris, P. salivae, P. denticola, P. buccalis and P. 220 oralis. Within the Mycoplasma genus, ASVs were assigned to Mycoplasma salivarium 221 among other species, an oral bacterium which has been previously associated to the 222 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Methods) but due to the low sample number, none was significant after multiple-test 252 correction. Additionally, as hospital stay (ICU or ward), type of oxygen support 253 (invasive or non-invasive ventilation) and disease (COVID-19 or controls) were highly 254 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Out of the total 29,886 microbial barcodes, only 2,108 were also assigned to host cells, 270

suggesting that the bulk of bacteria found in BAL samples exist as free-living organisms 271 or in bacterial biofilms. However, for those associated to host cells, the distribution 272 across disease types was not random. We found that while 2.3% of the non-COVID-19 273 patient cells were associated to bacterial cells, almost the double (4%) could be 274 observed in COVID-19 patients (Figure 3a ; Chi-squared test; p-value < 2.2·10 -16 ). 275

However, because COVID-19 diagnosis is highly correlated with intubation in this 276 cohort, this effect could be due to higher intubation rates in COVID-19 patients. Within 277 COVID-19 patients, we also evaluated the overlap between bacteria-associated host 278 cells and cells with detected SARS-CoV-2 reads (Supplementary Table 5 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint

We also explored whether host-associated bacterial reads would preferentially be 286 linked with specific cell types, taking into account the varying frequencies of cell types 287 in COVID-19 patients and controls (see Methods). Such a preferential association 288 would suggest that these observations are biologically relevant and not an artifact of 289 the single-cell sample and library preparation. Among control patients, cell types were 290 similarly distributed in both groups (i.e. with and without bacteria), with only a 291 preferential association of microbial cells with neutrophils (p-value = 3.61·10 -12 ; Figure  292 suggesting that the same cell type responsible for defense against the virus would be 309 responding to potentially invasive bacteria in the lung. This subgroup is characterized 310 by the expression of the calprotectin subunits S100A8 and S100A9. It is known that 311 S100A8/A9 heterodimer secretion is increased in infection-induced inflammation and 312 has some antibacterial effects mediated by secretion of pro-inflammatory cytokines, 313 release of reactive oxygen species and recruitment of other inflammatory cells, as well 314 as chelation of Zn 2+ necessary for bacterial enzymatic activity 24 . These mechanisms are 315 mediated by binding of the S100A8/A9 dimer to TLR4 receptors to trigger the release 316 of pro-inflammatory cytokines such as IL-6 and TNF-a, and thus may contribute to 317 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint sustain or exacerbate inflammation 25 . Therefore, the association with bacteria could, 318 at least in part, explain the inflammatory phenotype of this neutrophil subset. To 319 further examine this hypothesis, we explored differential gene expression between 320 bacteria-associated and non-associated S100A12 hi neutrophils (Supplementary Table  321 6). Because association of these cells with SARS-CoV-2 and with bacteria was mutually 322 exclusive, we also compared these changes with the ones triggered by the virus in 323 neutrophils 22 . Within this subset, neutrophils with co-occurring bacteria showed 324 significantly higher expression (Bonferroni-corrected p-value < 0.05) of pro-325 inflammatory genes, including the cytokine IL1B and some of its target genes (PTSG2), 326 the transcription factors FOS and JUN, and several genes involved in degranulation 327 (S100A9, FOLR3, HSPA1A, HSP90AA1, FCGR3B), (Supplementary Table 6 ). Among 328 these, FOLR3, a gene encoding for a folate receptor, is found in neutrophil secretory 329 granules and has antibacterial functions, by binding folates and thus depriving bacteria 330 of these essential metabolites 26 . This response differed to that of virus-engulfing 331 neutrophils in that IFN response genes are not distinctively upregulated by bacteria. Table 6 ). A similar increase was also observed in monocytes, yet not significant 342 (Supplementary Table 6 ), possibly due to the lower monocyte abundances in this 343 dataset. Additionally, bacteria-associated macrophages express significantly higher 344 levels of the calprotectin subunits S100A8/A9, similarly to neutrophils, as well as pro-345

inflammatory chemokines (such as CCL4, CXCL10 and CXCL1). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint Altogether, our results suggest that the bacteria detected in these cell subsets via 348 scRNA-seq analyses may be contributing to the inflammatory response observed in the 349 host. We further found that between patient microbiome variation (as measured by genus-378 level microbial beta-diversity) was also affected by different severity indicators such as 379 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ;

the clinical status of the patient, or more importantly the type of oxygen support 380 received, with intubated patients harboring a different microbiota than non-intubated 381 patients. The impact of oxygen support was also reflected at the species-and strain-382 levels, with intubation causing a significant decrease and increase, respectively, in 383 diversity. We hypothesize that the introduction of forced oxygen may drive the fast 384 extinction of certain microbial species enabling the diversification of existing or newly 385 colonizing species into new strains. Our results suggest that non-invasive ventilation 386 patients diagnosed of community-acquired pneumonia found no differences in 396 respiratory microbiome composition between both groups of patients, but both 397 groups did differ from healthy controls 29 . Together, these results indicate that patient 398 intubation or even non-invasive ventilation are to be considered as important 399 confounders when studying the upper respiratory microbiome, and we strongly 400 suggest future COVID-19 microbiome studies should foresee and include strategies to 401 account for this covariate. A recent study found a single ASV corresponding to the 402 genus Rothia that was specific for SARS-CoV-2 patients after controlling for ICU-related 403

To better understand the potential functional consequences of these procedures and 406

linked microbial shifts, we also profiled the microbiome of the lower respiratory tract 407 using single-cell data obtained from a cross-sectional cohort of patients derived from 408 the same hospital. Our results show that 'standard' single-cell RNA-seq, even though 409 not optimized for microbial detection and profiling, can identify bacteria alone or in 410 association with specific human cells. Unfortunately, the low numbers of microbial 411 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint reads obtained in this cohort, together with the fact that ICU stay, COVID-19 diagnosis 412 and intubation are highly correlated in this set of patients, only allow for a descriptive 413 analysis of the results. In this cohort, we identified different oral commensals and 414 opportunistic pathogens previously linked to COVID-19 patients in both groups of 415 samples, thus pointing again at a potential ventilation-linked origin. More interestingly, 416

we identified a subset of bacteria associated with host cells, more specifically with 417 neutrophils, monocytes and macrophages. This enrichment shows that these bacteria 418 are likely not random contaminants, from which an even distribution across cell types 419 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. To increase bacterial cell lysis efficiency, glass beads and DX reagent (Pathogen Lysis 504

Tubes, QIAGEN, catnr. 19091) were added to the lysis buffer, and samples were 505 disrupted in a FastPrep-24 TM instrument with the following program: 1-minute beating 506 at 6.5m/sec, 1-minute incubation at 4°C, 1-minute beating at 6.5m/sec, 1-minute 507 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint incubation at 4°C. After lysis, the remaining extraction steps followed the 508 recommended protocol from the manufacturer. DNA was eluted in 50µL EB buffer. 509

Amplification of the V4 region of the 16S gene was done with primers 515F and 806R, 510 using single multiplex identifiers and adaptors as previously described 35 . RNA was 511 eluted in 30µL of nuclease-free water and used for SARS-CoV-2 viral load 512 determination in the swabs as well as to measure inflammatory markers and cytokines 513 and to estimate host cell populations via marker gene expression using nCounter. In 514 brief, raw nCounter data were processed using nSolver 4.0 software (Nanostring), is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint All the analyses were performed using R v3.6.0 and the packages vegan 44 , phyloseq 42 , 539

CoDaSeq 45 , DESeq2 46 , Biostrings 47 , rstatix 48 and DECIPHER 49 . 540

To analyze the 16S amplicon data, technical replicates were pooled and counts from 542 technical replicates were added. For all the analyses using genus-level agglomerated 543 data, only samples containing more than 10,000 reads assigned at the genus level 544

were used (101 samples in total). Alpha-diversity was analyzed using Shannon's 545 Diversity Index. Comparison of the alpha diversity values across different groups was 546 performed using Wilcoxon signed-rank tests for 2-group comparisons, and Kruskal-547

Wallis tests for comparisons across multiple groups. In the latter case, pairwise 548 comparisons (when applicable) were performed using Dunn post-hoc tests. To de-549 confound for the effect of the ICU length, we fitted a quadratic model between the 550 days spent at ICU and the Shannon index using the lm function in R. The residuals of 551 this model were used to test the association with the SARS-CoV-2 viral load. 552 553 Beta diversity analyses were performed using distance-based redundancy analyses 554 (RDA), using Euclidean distances on CLR-transformed data. RDA analyses were 555 performed using the capscale function from vegan. Non-redundant contribution to 556 variation was calculated using the ordiR2step function from vegan, using only the 557 variables that were significant individually in the RDA, and a null model without any 558 explanatory variables. For these analyses, taxa with prevalence lower than 10% were 559 excluded. Metadata variables containing dates, as well as non-informative metadata 560

(containing a single non-NA value or unique for only one patient) were also excluded. 561 562 Differential taxa abundance analyses were performed using DESeq2's likelihood ratio 563 tests and controlling for potential confounders when indicated, including them in the 564 null model. All statistical tests are two-sided, and when multiple tests were applied to 565 the different features (e.g. the differential taxa abundances across two conditions) p-566 values were corrected for multiple testing using Benjamini-Hochberg's method. 567

In order to explore species-level and strain-level diversity, 16S sequences were first 569 clustered into 97% nucleotide diversity operational taxonomic units (OTUs) using the R 570 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint packages Biostrings and DECIPHER. These OTUs were used to represents the species-571 level. The number of unique 16S sequences clustered within each OTU were used to 572 represent the number of detectable strains per species. To calculate strain-level 573 diversity per sample, the number of strains of 5 detected OTU species were randomly 574 selected and averaged. This was repeated 1,000x and the average of the all 1,000 575 subsamplings was used as the final strain-level diversity value for each sample, as 576 previously described 50 . To account for uneven sampling assessing diversity differences 577 based on different parameters, we randomly selected and averaged the species-and 578 strain-level diversity of 5 samples per parameter. This was repeated 100x and the 579 subsamplings were used to assess the significant differences between species-and 580 strain-level diversity across the parameters. The average was of all 100 subsamplings 581 was used to as the input for a Pearson's correlation between species-and strain-level 582 diversity. 583 584

Single-cell data was processed with an in-house pipeline to identify microbial reads. 587

Only read 2, containing the information on the cDNA, was used. Trimmomatic 51 (v0.38) 588 was used to remove trim low quality bases and discard short reads. Additionally, 589

Prinseq++ 52 (v1.2) was used to discard reads with low-complexity stretches. Following 590 quality control, reads from human and potential sequencing artifacts (phage phiX174) 591

were mapped with STAR 53 (v2.7.1) and discarded. The remaining reads were mapped 592 against bacterial genomes using a 2-step approach: first, we scanned the reads using 593 mash screen 54 (v2.0) against a custom database of 11685 microbial reference genomes 594 including bacteria, archaea, fungi and viruses. Genomes likely to be present in the 595 analyzed sample (selected using a threshold of at least two shared hashes from mash 596 screen) were selected and reads were pseudoaligned to this subset of genomes using 597 kallisto 55 (v0.44.0). To remove potential artifacts, two additional filters were applied: 598 first, if less than 10 different functions were expressed from a given species, the 599 species was discarded. Second, if one function accounted for more than 95% of the 600 mapped reads of a given species, it was also discarded. These filters were aimed at 601 . CC-BY-NC-ND 4.0 International license It is made available under a perpetuity.

is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint removing potential artifacts caused by contamination or errors in the genome 602 assemblies used in our database. 603 604 Differences in lower respiratory tract microbial taxa between COVID-19 patients and 605 controls, ICU and ward patients, and invasive and non-invasive ventilation types were 606 calculated using Wilcoxon rank-sum tests on centered-log-ratio (CLR)-transformed 607 data. Prior to CLR data transformation, we filtered the data using the CoDaSeq.filter 608 function, to keep samples with more than 1,000 reads and taxa with a relative 609 abundance above 0.1%. Zeros were imputed using the minimum proportional 610 abundance detected for each taxon. This more lenient approach than the one used for 611

16S data was chosen due to the low number of samples available and the reduced 612 number of bacterial reads identified per sample. 613

Bacterial reads were assigned their specific barcodes and UMIs as follows: IDs from the 615 mapped microbial reads were retrieved from the kallisto pseudobam output, and used 616 to retrieve their specific barcodes and UMIs using the raw data files from read 1, 617 assigning each barcode and UMI univocally to a microbial species and function. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; determine enrichment of bacteria-associated cell subtypes. Previous annotations of 634 cell subtypes 22 were used to generate new clusters manually and identify marker 635 genes for these subtypes, using the function findAllMarkers from Seurat. This function 636 was also used to find differentially expressed genes between bacteria-associated and 637 not-bacteria-associated host cells of each subtype. When using this function, reported 638 is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; of the sample. For samples taken after discharge to ward, the total number of days 730 spent in ICU was used. The shaded area surrounding the trend line represents the 95% 731 confidence interval. e) Correlation between SARS-CoV-2 viral load and Shannon 732 diversity index, after controlling for the time spent in ICU. The residuals of a quadratic 733 fit between the Shannon diversity and the days in ICU were correlated to the SARS- is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

The copyright holder for this this version posted December 26, 2020. ; https://doi.org/10.1101/2020.12.23.20248425 doi: medRxiv preprint

COVID-19 in Wuhan, China: a retrospective cohort study

Risk Factors Associated with Mortality among Patients with 797 COVID-19 in Intensive Care Units in Lombardy

Risk Factors for Mortality in Patients with COVID-19 in New 800

Diabetes is a risk factor for the progression and prognosis of 802 <scp>COVID</scp> -19

Obesity in Patients Younger Than 60 Years Is a Risk Factor for 804 COVID-19 Hospital Admission

SARS-CoV-2 viral load in sputum correlates with risk of COVID-19 807 progression

Impact of Severe Acute Respiratory Syndrome Coronavirus

Viral Load on Risk of Intubation and Mortality Among Hospitalized Patients With 810

Coronavirus Disease

SARS-CoV-2 Viral Load Predicts Mortality in Patients with 812 and without Cancer Who Are Hospitalized with COVID-19

The cytokine 815 storm in COVID-19: An overview of the involvement of the 816 chemokine/chemokine-receptor system

On the Alert for Cytokine Storm: Immunopathology in 819 COVID-19

COVID-19: consider cytokine storm syndromes and 821 immunosuppression

Lung microbiome and coronavirus disease 2019 823 (COVID-19): Possible link and implications

The role of the microbiome in 826 exacerbations of chronic lung diseases

The respiratory tract microbiome 828 and lung inflammation: A two-way street

Metatranscriptomic Characterization of COVID-19 Identified A 831

Temporal dynamics of human respiratory and gut microbiomes 834 during the course of

The active lung microbiota landscape of 837 COVID-19 patients

The microbiota of the 843 respiratory tract: gatekeeper to respiratory health

Longitudinal sampling of the lung microbiota in individuals 846 with cystic fibrosis

Low-pathogenicity Mycoplasma spp. alter human monocyte 848 and macrophage function and are highly prevalent among patients with 849 ventilator-acquired pneumonia

Discriminating Mild from Critical COVID-19 by Innate and 851

Adaptive Immune Single-cell Profiling of Bronchoalveolar Lavages. Patrick 852 Matthys

Pneumocystis jirovecii pneumonia: 854 Current knowledge and outstanding public health issues

S100A8/A9 in inflammation

Myeloid-related protein 8 induces self-tolerance and cross-859 tolerance to bacterial infection via TLR4-and TLR2-mediated signal pathways

Characterization of soluble folate receptors (folate 862 binding proteins) in humans. Biological roles and clinical potentials in infection 863 and malignancy

The dynamics of the pulmonary microbiome during 866 mechanical ventilation in the intensive care unit and the association with 867 occurrence of pneumonia

Pulmonary microbiome patterns correlate with the course 869 of disease in patients with sepsis-induced ARDS following major abdominal 870 surgery

Genomic Diversity of Severe Acute Respiratory Syndrome-872

Clinical infectious 873 diseases : an official publication of the Infectious Diseases Society of America 71

Title: Microbial context predicts SARS-CoV-2 prevalence in 876 patients and the hospital built environment

Analysis of the upper respiratory tract microbiotas as the 879 source of the lung and gastric microbiotas in healthy individuals

The role of oral microbial colonization in ventilator-881 associated pneumonia. Oral Surgery, Oral Med. Oral Pathol

Ventilator-Associated Pneumonia: The Importance of Oral Care 884 in Intubated Adults

Enrichment of the lung microbiome with oral taxa is associated 886 with lung inflammation of a Th17 phenotype

CC-BY-NC-ND 4dual-index sequencing strategy and curation pipeline for 889 analyzing amplicon sequence data on the miseq illumina sequencing platform

Simultaneous RNA quantification of human and retroviral 892 genomes reveals intact interferon signaling in HTLV-1-infected CD4+ T cell lines

Pathogen transcriptional profile in nasopharyngeal aspirates 895 of children with acute respiratory tract infection

Diagnostic accuracy of digital RNA quantification versus real-898 time PCR for the detection of respiratory syncytial virus in nasopharyngeal 899 aspirates from children with acute respiratory infection

In situ immune signatures and microbial load at the 902 nasopharyngeal interface in children with acute respiratory infection

LotuS: An efficient and 905 user-friendly OTU processing pipeline

DADA2: High-resolution sample inference from Illumina 907 amplicon data

phyloseq: An R Package for Reproducible 909 Interactive Analysis and Graphics of Microbiome Census Data

Simple 912 statistical identification and removal of contaminant sequences in marker-gene 913 and metagenomics data

It's all relative: 916 analyzing microbiome data as compositions

Moderated estimation of fold change and 919 dispersion for RNA-seq data with DESeq2

Efficient 921 manipulation of biological strings

Pipe-Friendly Framework for Basic Statistical Tests

Using DECIPHER v2.0 to Analyze Big Biological Sequence Data in R

Marine DNA Viral Macro-and Microdiversity from Pole to 926 Pole

Trimmomatic: a flexible trimmer for 928 Illumina sequence data

PRINSEQ++, a multi-threaded tool for fast 930 1 and efficient quality control and 2 preprocessing of sequencing datasets

STAR: Ultrafast universal RNA-seq aligner

Mash Screen: high-throughput sequence containment 935

association (i.e. depletion) of bacteria for a given cell type. Asterisks denote 762 significance as follows: * = p-value ≤ 0.05; ** = p-value ≤ 0.01; *** = p-value ≤ 0.001; 763 **** = p-value ≤ 0.0001.