key: cord-0764947-qzg5mep4
authors: Legebeke, J.; Lord, J.; Penrice-Randal, R.; Vallejo, A. F.; Poole, S.; Brendish, N. J.; Dong, X.; Hartley, C.; Holloway, J. W.; Lucas, J. S.; Williams, A. P.; Wheway, G.; Strazzari, F.; Gardner, A.; Schofield, J. P. R.; Skipp, P. J.; Hiscox, J. A.; Polak, M. E.; Clark, T. W.; Baralle, D.
title: Distinct immune responses in patients infected with influenza or SARS-CoV-2, and in COVID-19 survivors, characterised by transcriptomic and cellular abundance differences in blood.
date: 2021-05-15
journal: nan
DOI: 10.1101/2021.05.12.21257086
sha: 830bf558d3664fe90bb38ad579c3eef636c56049
doc_id: 764947
cord_uid: qzg5mep4

Background The worldwide pandemic caused by SARS-CoV-2 has claimed millions of lives and has had a profound effect on global life. Understanding the pathogenicity of the virus and the body's response to infection is crucial in improving patient management, prognosis, and therapeutic strategies. To address this, we performed functional transcriptomic profiling to better understand the generic and specific effects of SARS-CoV-2 infection. Methods Whole blood RNA sequencing was used to profile a well characterised cohort of patients hospitalised with COVID-19, during the first wave of the pandemic prior to the availability of approved COVID-19 treatments and who went on to survive or die of COVID-19, and patients hospitalised with influenza virus infection between 2017 and 2019. Clinical parameters between patient groups were compared, and several bioinformatic tools were used to assess differences in transcript abundances and cellular composition. Results The analyses revealed contrasting innate and adaptive immune programmes, with transcripts and cell subsets associated with the innate immune response elevated in patients with influenza, and those involved in the adaptive immune response elevated in patients with COVID-19. Topological analysis identified additional gene signatures that differentiated patients with COVID-19 from patients with influenza, including insulin resistance, mitochondrial oxidative stress and interferon signalling. An efficient adaptive immune response was furthermore associated with patient survival, while an inflammatory response predicted death in patients with COVID-19. A potential prognostic signature was found based on a selection of transcript abundances, associated with circulating immunoglobulins, nucleosome assembly, cytokine production and T cell activation, in the blood transcriptome of COVID-19 patients, upon admission to hospital, which can be used to stratify patients likely to survive or die. Conclusions The results identified distinct immunological signatures between SARS-CoV-2 and influenza, prognostic of disease progression and indicative of different targeted therapies. The altered transcript abundances associated with COVID-19 survivors can be used to predict more severe outcomes in patients with COVID-19.

The analyses revealed contrasting innate and adaptive immune programmes, with transcripts and cell 43 subsets associated with the innate immune response elevated in patients with influenza, and those 44 involved in the adaptive immune response elevated in patients with COVID-19. Topological analysis 45 identified additional gene signatures that differentiated patients with COVID-19 from patients with 46 influenza, including insulin resistance, mitochondrial oxidative stress and interferon signalling. An 47 efficient adaptive immune response was furthermore associated with patient survival, while an 48 inflammatory response predicted death in patients with COVID-19. A potential prognostic signature 49 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Introduction influenza are able to express type I and type II IFNs at a significantly higher concentration (3) which 95 correlates with quicker recovery and decreased disease severity and mortality (10, 11) . Consistent with 96 this observation, early administration of inhaled recombinant IFN-beta for COVID-19 patients was 97 associated with a lowered in-hospital mortality and quicker recovery (12, 13) . Despite the reduced IFN 98 response in patients with COVID-19 the expression of pro-inflammatory cytokines occurs for a 99 prolonged time at similar levels with influenza patients (3), and interleukin (IL) -6 and IL-10 (14-16) 100 and CCL3 (3) were associated with increased disease severity for COVID-19. The presence of CD4+ and 101 CD8+ T cells, and antibodies were correlated with a positive patient outcome in the case of . This puts elderly patients at a higher risk due a smaller naïve T cell pool (18) (19) (20) and an absence 103 of a pre-existing adaptive immunity (21) resulting in a potential delayed T cell response to a novel virus 104 like SARS- . Delaying an adaptive immune response which, when combined with a high viral 105 load, could lead to a poor outcome (23). As discussed by Sette and Crotty (24) an absent T cell response 106 may cause an increased innate response attempting to control the virus resulting in an excessive lung 107 immunopathology. 108 109 To investigate unique molecular features associated with COVID-19, a cohort of patients was identified 110 from hospitalised individuals that were positive for SARS-CoV-2. As a comparator an equivalent group 111 of patients hospitalised with influenza virus were identified. An extensive record of clinical parameters 112 and peripheral blood, used for RNA-seq to obtain a global blood transcriptome overview, were taken 113 at point of care and could therefore be correlated with any molecular signatures of disease. Through 114 these side-by-side comparisons, we aim to identify distinct patterns of blood transcript abundances 115 and cellular composition related to specific antiviral immune responses. Furthermore, we aim to 116 identify a promising prognostic signature indicative of COVID-19 outcome. 117 118 119 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 

Ethics and consent 121 The study was approved by the South Central -Hampshire A Research Ethics Committee (REC): REC 122 reference 20/SC/0138 (March 16 th , 2020) for the COVID-19 point of care trial; and REC 123 reference 17/SC/0368 (September 7 th , 2017) for the FluPOC trial. For full inclusion and exclusion 124 criteria details see (25) and (26). Patients gave written informed consent or consultee assent was 125 obtained where patients were unable to give consent. Demographic and clinical data were collected 126 at enrolment and outcome data from case note and electronic systems. ALEA and BC data 127 management platforms were used for data capture and management. 128 129 Study design and participants 130 All participants were recruited within the first 24 hours of admission to two large studies of molecular 131 point-of-care testing (mPOCT) for respiratory viruses . Blood samples 132 including whole blood in PAXgene Blood RNA tubes (BRT) (Preanalytix) were collected from SARS-CoV-133 2 positive patients and influenza positive patients, within 24 hours of enrolment, and stored at -80°C. 134

The studies were prospectively registered with the ISRCTN trial registry: ISRCTN14966673 (COV-135 19POC) (March 18 th , 2020), and ISRCTN17197293 (FluPOC) (November 13 th , 2017). The study was a non-randomised interventional trial evaluating the clinical impact of mPOCT for SARS-137 CoV-2 in adult patients presenting to hospital with suspected COVID-19, using the QIAGEN QIAstat-Dx 138 PCR testing platform with the QIAstat-Dx Respiratory SARS-CoV-2 Panel (27). The trial took place 139 during the first wave of the pandemic, from 20th March to 29th April 2020, and prior to the availability 140 of approved COVID-19 treatments. All patients were recruited from the Acute Medical Unit (AMU), 141

Emergency Department (ED) or other acute areas of Southampton General Hospital. The FluPOC study 142 was a multicentre randomised controlled trial evaluating the clinical impact of mPOCT for influenza in 143 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) generate the RNA libraries, followed by 11 or 13 cycles of amplification and purification using AMPure 156 XP beads. Each library was quantified using Qubit and the size distribution assessed using the Agilent 157 2100 Bioanalyser and the final libraries were pooled in equimolar ratios. Libraries were sequenced 158 using 150 bp paired-end reads on by an Illumina® NovaSeq 6000 (Illumina®, San Diego, USA). Raw 159 fastq files were trimmed to remove Illumina adapter sequences using Cutadapt v1.2.1 (29). The option 160 "−O 3" was set, so that the 3' end of any reads which matched the adapter sequence with greater than 161 3 bp was trimmed off. The reads were further trimmed to remove low quality bases, using Sickle 162 v1.200 (30) with a minimum window quality score of 20. After trimming, reads shorter than 10 bp 163 were removed. 164 165 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Data QC and alignment 166 Quality control (QC) of read data was performed using FastQC (31) (v0.11.9) and compiled and 167 visualised with MultiQC (32) (v1.5). Samples with <20 million total reads were excluded from further 168 analysis. The STAR index was created with STAR's (33) (v2.7.6a) genomeGenerate function using 169

GRCh38.primary_assembly.genome.fa and gencode.v34.annotation.gtf (34) Differential gene expression analysis between patient groups 203 HTSeq (46) (v0.11.2) count was used to assign counts to RNA-seq reads in the Samtools sorted BAM 204 file using GENCODE v34 annotation. Parameters used for HTSeq were --format=bam, --order=pos, --205 stranded=reverse, --type=exon and the other options were kept at default. EdgeR (47) (v3.30 .3) was 206 used for differential gene expression analysis with R (v4.0.2) in RStudio (v1.3.959). Genes with low 207 counts across all libraries were filtered out using the filterByExpr command. Filtered gene counts were 208 normalised using the trimmed mean of M-values (TMM) method. Differentially expressed genes were 209 identified, after fitting the negative binomial models and obtaining dispersion estimates, using the 210 exact test and using a threshold criteria of FDR p-value < 0.05 and log2 fold change < -1 and > 1. Genes 211 which were within the threshold criteria were used for ToppGene gene list enrichment analysis. A 212 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 principal component analysis (PCA) graph was constructed based on all differentially expressed genes 213 to assess sample clustering. 214 215 Assessment of difference in adaptive immune response related gene expression 216 A higher abundance of transcripts from 83 immunoglobulin genes, overlapping with the genes in the 217 Gene Ontology (GO) (48,49) biological process term 'adaptive immune response' (Additional file 1), 218 was found in patients with COVID-19 compared to influenza. To assess gene transcript abundance 219 differences for these 83 genes in each patient a heatmap was generated and Z-scores were summed 220 to give an overall positive (high) or negative (low) total Z-score. Patient baseline clinical characteristics 221 were explored, as above, for any explanatory factors for the involvement of a high or low total Z-score 222 between patients with COVID-19 or influenza, and those that survived COVID-19 versus those that 223 died within 30 days of hospital admission. Metadata comparison plots were made with the R package 224 CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101/2021.05.12.21257086 doi: medRxiv preprint Assessment of differential splicing between patient groups 237 Three different tools were used to assess differential gene splicing between patients with COVID-19 238 or influenza, and COVID-19 survivors or non-survivors after 30 days of hospital admission. rMATs (35) 239 (v4.1.0) was run using BAM files with soft clipping suppressed, generated with STAR and GENCODE 240 v34 gene annotation. Additional settings used were -t paired, --readLength 150 and --libType fr-241 firststrand. Results were filtered for FDR p-value < 0.05. LeafCutter (54) (v0.2.9) was run in stages 242 following the Differential Splicing protocol (55) (bam2junc.sh generated junction files from BAMs, 243 leafcutter_cluster.py grouped junctions into clusters, leafcutter_ds.R tested for differential splicing, 244 all with default settings, except -min_samples_per_intron was set to be approximately 60% of the 245 smaller group size for each comparison (46 for COVID-19 vs influenza, 9 for COVID-19 survivors vs non-246 survivors), and results were filtered to exclude events with delta PSI <10%, based on 247 recommendations (56). The LeafViz script (57), prepare_results.R was used to generate a data table 248 from which gene names for significant events were extracted, while the map_clusters_to_genes R 249 function was used to assign genes to non-significant tested events. Overlap between LeafCutter 250 differentially spliced and EdgeR differentially expressed genes was tested for significance using 251

Fisher's Exact Test (fisher.test in R (v3.5.1) using a 2x2 contingency table and two.sided alternative 252 hypothesis). MAJIQ (58) (v2.2) was run in two stages (majiq build and majiq deltapsi) with default 253 settings, and results were filtered (delta PSI >20%, probability >0.95) using Voila (58) (v2.0). 254

In silico immune profiling predicting immune cell levels between patient groups 256 Relative abundance of 22 immune cell types and their statistical significance was deconvoluted from 257 whole blood using the reference gene signature matrix (LM22) using CIBERSORTx (59). CIBERSORTx 258 analysis was conducted on the CIBERSORTx website (60) using 100 permutations. Immune cell 259 distribution between the groups were compared by Mann-Whitney test. 260 261 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The baseline clinical characteristics of the patients used in this study for the comparison between 285 influenza and COVID-19 were assessed and no differences in distribution of sex or age were detected 286 between patient groups, however, more patients with influenza were of White British ethnicity (p-287 value 1.12x10 -05 ) and more were current smokers (p-value 9.07x10 -05 ). There were also differences in 288 the proportion of cases with underlying comorbidities, with patients with COVID-19 more commonly 289 having hypertension (p-value 1.42x10 -02 ), liver disease (p-value 3.63x10 -02 ) and diabetes mellitus (p-290 value 6.44x10 -03 ) than those with influenza. However, underlying respiratory disease was more 291 common in patients with influenza (p-value 1.22x10 -03 ). Patients with COVID-19 generally exhibited 292 more severe clinical symptoms. While the National Early Warning Score 2 (NEWS2) was not different 293 between patients with COVID-19 or influenza, patients with COVID-19 had a higher respiratory rate 294 (p-value 2.79x10 -02 ) and a greater proportion of patients with COVID-19 were on supplementary 295 oxygen at hospital admission (p-value 6.81x10 -03 ). Laboratory results indicated higher levels of C-296 reactive protein (p-value 1.73x10 -03 ) and lymphocytes (p-value 2.76x10 -02 ) in patients with Furthermore, COVID-19 patients had a longer duration of symptoms prior to presentation to hospital 298 (p-value 1.17x10 -05 ) and once admitted a longer length of stay (p-value 5.51x10 -10 ). Longer stay time 299 was associated with increased 30 day mortality after hospital admission and patients with COVID-19 300 were more likely to have died compared to patients with influenza (p-value 4.42x10 -05 ) ( Table 1) . 301 302 Between patients with COVID-19 who survived and those who died, a fatal outcome occurred in older 303 patients (p-value 2.58x10 -09 ). COVID-19 non-survivors also had a shorter duration of symptoms before 304 being admitted to hospital (p-value 5.38x10 -03 ). COVID-19 non-survivors more commonly had 305 underlying comorbidities including hypertension (p-value 1.93x10 -03 ), cardiovascular disease (p-value 306 3.97x10 03 ), diabetes mellitus (p-value 2.31x10 -02 ) and underlying respiratory disease (p-value 1.06x10 -307 02 ). While the NEWS2 scores were not different, COVID-19 survivors had a higher heart rates than 308 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101/2021.05.12.21257086 doi: medRxiv preprint COVID-19 non-survivors (p-value 9.27x10 -03 ). Laboratory results showed an increase of white blood 309 cell count (p-value 3.83x10 -02 ), total protein levels (p-value 2.5x10 -03 ), creatinine (p-value 3.87x10 -02 ), 310 alanine aminotransferase levels (p-value 2.85x10 -02 ), troponin levels (p-value 2.37x10 -04 ), tumour 311 necrosis factor α (TNFα) (p-value 1.43x10 -02 ), interleukin (IL)-6 levels (p-value 2.78x10 -03 ), IL-8 (p-value 312 2.24x10 -02 ), IL-1β (p-value 3.78x10 -02 ) and IL-10 (p-value 7.51x10 -02 ) in COVID-19 non-survivors. Patient 313 outcome and length of hospital stay were different due to separation based on patient survival (Table  314 2). 315 316 

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101/2021.05.12.21257086 doi: medRxiv preprint 324 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101/2021.05.12.21257086 doi: medRxiv preprint that died. A median of 60.4 million reads in patients with COVID-19, and 58.9 million reads in patients 336 with influenza was obtained (Supplementary figure 2A) . In patients who died of COVID-19 a median 337 of 55.7 million reads was obtained and for COVID-19 survivors the median was 62.6 million reads 338 (Supplementary figure 2B) . Clustering analysis between patients with COVID-19 or influenza indicated 339 a homogeneity of blood transcriptome profiles suggesting any variation between groups to be subtle 340 (Supplementary figure 3A) . A partial separation was found between patients who survived or died of 341 COVID-19 based on patient outcome after 30 days of hospital admission, indicative of a larger variation 342 in the blood transcriptome (Supplementary figure 3B) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101/2021.05.12.21257086 doi: medRxiv preprint 

Interestingly, an increased abundance of gene transcripts in patients with COVID-19 are involved in 387 adaptive immunity, pointing to activation/priming of T cells and B cells, including induction of 388 proliferation (cluster 4, FDR p-value 3.97x10 -57 ), Additionally, an increased abundance of gene 389 transcripts encoding neutrophil degranulation (cluster 9) and blood coagulation (cluster 6) clearly 390 differentiated patients with COVID-19 from patients with influenza (FDR p-value 4.33x10 -19 and FDR p-391 value 2.84x10 -12 respectively). In contrast, an decreased abundance of gene transcripts in the blood 392 transcriptome of patients with COVID-19 in comparison to patients with influenza were associated 393 with innate immunity, including biological processes involved with defence response to virus (cluster 394

2) (FDR p-value 1.34x10 -37 ), type 1 helper T cell stimulation (cluster 10) (FDR p-value 4.53x10 -03 ), 395 dendritic cell morphogenesis (cluster 11) (FDR p-value 1.37x10 -02 ), and myeloid cell activation (clusters 396 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101/2021.05.12.21257086 doi: medRxiv preprint 1 and 8) (FDR p-value 5.16x10 -13 and FDR p-value 4.15x10 -04 respectively). Importantly, the largest 397 decrease of transcript abundances in patients with COVID-19 comprised genes expressed in 398 plasmacytoid dendritic cells (pDC) (FDR p-value 4.17x10 -22 ), indicating impaired immune responses to 399 viruses (FDR p-value 1.34x10 -37 ) and impaired IFN signalling (FDR p-value 5.56x10 -30 ). This was 400 suggestive of contrasting innate and adaptive immune programmes between the different infections 401 and these were further investigated. Complimentary to the findings from gene co-expression analysis, the transcripts with increased 410 abundance in patients with COVID-19 were found to be involved with humoral immune response, 411 complement activation and B cell mediated immunity ( Figure 3B) . 412 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 than those with influenza (p-value < 2.22x10 -16 , Wilcoxon test) (Figure 3C and Supplementary figure  424   4 ) and by using a total Z-score, patients with COVID-19 or influenza were classified as having either a 425 high or low abundance of these 83 immunoglobulin genes. A high abundance was associated with a 426 total positive Z-score (1.46 to 175.46) which was identified in 59 patients with COVID-19 and 21 427 patients with influenza indicating a higher than average abundance of these 83 adaptive immune 428 response related immunoglobulin genes. While a low abundance was associated with a total negative 429 Z-score (-0.12 to -154.93) identified in 19 patients with COVID-19 and 62 patients with influenza 430 indicating a lower than average abundance of adaptive immune response related immunoglobulin 431 genes. COVID-19 patients with lower abundance of adaptive immune response related 432 immunoglobulin genes, a total negative Z-score, were found to be significantly older (p-value 6.32x10 -433 3 , T-test) and had a shorter duration of symptoms before being admitted into hospital (p-value 5.9x10 -434 04 , Wilcoxon test). Additionally, COVID-19 patients with high abundance of adaptive immune response 435 related immunoglobulin genes, a total positive Z-score, were significantly more likely to be still alive 436 30 days after admitted into hospital (x 2 13.39 and p-value 2.52x10 -04 , Chi-square test) (Figure 4) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Topological mapping of global gene patterns 456 Topological analysis allows the measurement of the global profiles of transcript abundances relative 457 to gene pathways without data reduction and this was used to define a global map of differentially 458 activated pathways between COVID-19 and influenza. The first differentially activated TopMD 459 pathway was enriched for ribosomal and insulin related pathways, with peak gene UBA52: named by 460 GO analysis as cytoplasmic ribosomal proteins (adjusted p-value 1.55x10 -146 ). This pathway was also 461 found to be enriched for genes expressed by transcription factor Myc (adjusted p-value 7.07x10 -53 ) 462 against the ChEA 2016 transcription factor database and of dendritic cells in the ARCHS4 transcription 463 factors' co-expression database (adjusted p-value 1.34x10 -36 ). Activated Myc represses interferon 464 regulatory factor 7 (IRF7) and a significant lower abundance of IRF7 was found in patients with COVID-465 19 compared to influenza (Supplementary figure 5) . The second differentially activated TopMD 466 pathway had peak gene NDUFAB1; named by GO analysis as mitochondrial complex I assembly model 

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 Cell subsets supporting innate and adaptive immune response differences 482 Analysis of the blood transcriptome can be used to predict the immune cells present (64). Levels of 483 different predicted cell types were assessed to determine whether there were differences in immune 484 system associated cells between patients with COVID-19 or influenza (Figure 6) . Statistical testing was 485 done on cell type levels identified with CIBERSORTx. M0 macrophages (p-value 3.63x10 -06 ), plasma 486 cells (p-value 5.05x10 -04 ), cytotoxic CD8+ T cells (p-value 4.58x10 -03 ), regulatory T cells (p-value 487 7.30x10 -03 ) and resting natural killer cell (p-value 8.90x10 -03 ) were found to be significantly higher in 488 COVID-19 patients, while in influenza patients activated dendritic cells (p-value 2.23x10 -02 ) were 489 significantly higher. 490 491 CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101/2021.05.12.21257086 doi: medRxiv preprint neutrophils (p-value 2.84x10 -04 ) in patients who died of COVID-19. In contrast, an increase of naïve 498 CD4+ T cells (p-value 1.92x10 -03 ), M0 macrophages (p-value 1.20x10 -02 ), M2 macrophages (p-value 499 1.48x10 -02 ), naïve B cells (p-value 1.57x10 -02 ) and naïve cytotoxic CD8+ T cells (p-value 2.31x10 -02 ), were 500 identified in patients who went on to survive COVID-19 (Figure 7) . 501 502 503 As already noted, a high abundance of the GO biological process 'adaptive immune response' related 510 transcripts, mostly immunoglobulin genes, was associated with COVID-19 survival (Figure 4) . Here a 511 direct assessment was done of the blood transcriptome differences between patients who, at 30 days 512 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 after hospital admission, survived or who died of COVID-19. A total of 23,850 abundance measures of 513 gene transcripts were obtained after filtering out transcripts with low counts, of which 6.645 514 transcripts were found to be significant (FDR p-value < 0.05) of which, 537 transcripts exceeded a log2 515 fold change of < -1 or > 1, with CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 in annotation is the ratio of the input query genes overlapping with the genes in the pathway database. D) Increase of T cell 526 and B cell proliferation in .

In patients who died of COVID-19 an enrichment for biological processes involved with an 529 inflammatory response including interleukin signalling and neutrophil activation and degranulation 530 was detected (Figure 8B) . While in COVID-19 survivors biological processes involved with the adaptive 531 immune system including complement activation, B cell mediated immunity and a humoral immune 532 response mediated by circulating immunoglobulins was found to be enriched ( Figure 8C) . Additionally, 533 transcript abundances associated with T cell and B cell proliferation were significantly higher in COVID-534 19 survivors (p-value < 1.0x10 -04 , paired non-parametric T-test) ( Figure 8D) . 535 536 Immune signatures as predictors of COVID-19 outcome 537 Distinct immune signature genes were selected and assessed for their prediction accuracy in 538 stratifying patients with COVID-19 for disease outcome, fatality or survival. A signature of 47 genes 539 was identified (Figure 9A) , representative of the four biggest clusters of genes associated with either 540 patients with COVID-19 who survived or died. The associated GO biological process terms were 541 'humoral immune response mediated by circulating immunoglobulin' (FDR p-value 2.23x10 -46 ), 542 'nucleosome assembly' ), 'regulation of T-helper 1 cell cytokine production' 543 (FDR p-value 4.24x10 -03 ) and 'regulation of T cell activation' (FDR p-value 4.51x10 -04 ) (Supplementary  544   figure 6 ). This was highly predictive for outcome, with a maximum specificity of 75% and sensitivity of 545 93% ( Figure 9B and Table 4 ). 546 547 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 variables that can provide prognostic information on COVID-19 severity have previously been reported 574 (67). We compared these known COVID-19 prognostic variables (67) between patients with COVID-19 575 or influenza and found more active smokers among influenza patients. High C-reactive protein (CRP), 576 which previously has been reported to be similar upon admission to hospital between patients with 577 COVID-19 or influenza (3), hypertension and diabetes were more common among patients with 578 COVID-19. We also found an increase of liver disease, which has been classified as a low or very low 579 certainty predictor (67), in patients with COVID-19. In our cohort more patients with influenza had an 580 underlying respiratory disease. Similar to what has been previously reported (3) upon admission to 581 hospital both patients with COVID-19 or influenza presented with similar WBC and neutrophil counts, 582 and although we detected a difference in lymphocytes between patients with COVID-19 or influenza, 583 there was no difference the N/L ratio. Similar to Piroth et al. (4) we found that the average length of 584 stay was higher for patients with COVID-19 compared to influenza, and more patients with COVID-19 585 needed supplementary oxygen, and finally while Piroth et al. (4) report a roughly three times higher 586 relative risk of death for COVID-19, in our cohort no influenza patients died whilst admitted to hospital 587 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 and so this could not be assessed. In addition, we compared COVID-19 survivors and non-survivors, 588 and as reported the high certainty prognostic variables for mortality and/or severity of increased age, 589 hypertension, cardiovascular disease, diabetes, underlying respiratory disease (including COPD) and a 590 high WBC (67) were increased in non-survivors. While it has previously been reported that CRP and 591 N/L ratio were elevated in patients with COVID-19 who became critically ill (3), in our study we saw 592 no difference in CRP, neutrophil count and lymphocyte count between COVID-19 survivors and non-593 survivors. However, we found that urea, creatinine, alanine aminotransferase, troponin and several 594 cytokines, including IL-1β, IL-6, IL-8, IL-10 and TNFα, to be higher in patients who died of COVID-19. 595

Our initial global analysis of blood transcriptomic differences between patients with COVID-19 or 597 influenza detected contrasting innate and adaptive immune programmes. An impaired immune 598 response to viruses and interferon signalling in patients with COVID-19 was found, as described 599 previously (6-9), compared to patients with influenza, which are known to produce an IFN response 600

(3). Furthermore, in accordance with accumulating evidence of aberrant blood clotting in patients with 601 COVID-19 (68,69), transcripts expressed by megakaryocytes and platelets associated with blood 602 coagulation were in a higher abundance in COVID-19 patients. Gene clusters associated with an innate 603 immune response were found to be associated with influenza. While, in contrast, gene clusters 604 associated with an adaptive immune response and an increase of predicted plasma cells and CD8+ T 605 cells with COVID-19, pointing to T cell and B cell activation / priming. 606 607 Further analysis revealed various immunoglobulin genes had increased transcript abundance in 608 patients with COVID-19 compared to patients with influenza. This significant over representation of a 609 wide range of heavy chain and light chain V genes in patients with COVID-19 has been described before 610 (70) and the implementation of antibody analysis in plasma samples has been used as an additional 611 tool in diagnosing . We found that the 86.8% (53/61) of patients who survived COVID-612 19 had a higher than average transcript abundance of 83 immunoglobulin genes, which overlap with 613 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101/2021.05.12.21257086 doi: medRxiv preprint the GO biological term 'adaptive immune response', while this was 37.5% among the patients who 614 died of COVID-19. Further analysis revealed that the aforementioned higher than average transcript 615 abundance is associated with a younger age of the patient, a longer symptom duration before 616 admittance into hospital and a positive survival outcome 30 days after hospital admission. A lower 617 than average transcript abundance of 83 immunoglobulin genes was detected in 62.5% (10/16) of 618 patients who died of COVID-19, compared to 14.8% (9/61) of patients who survived COVID-19. 619 620 We subsequently detected an increased transcript abundance from genes associated with T cell and 621 B cell proliferation, an enrichment for gene pathways involved with an adaptive immune response, 622 and an increase in predicted CD4+ and CD8+ T cells and naïve B cells in patients who survived COVID-623 19, highlighting the importance of an efficient adaptive immune response as previously reported (17). 624

The predicted cell fraction of naïve CD4+ T cell was found to be higher compared to CD8+ T cells 625 indicating a higher CD4+ T cell response to SARS-CoV-2 than a CD8+ T cell response, supporting 626 previous observations (17,72), which has been found to control primary SARS-CoV-2 infection (22). 627

We note that the CD8+ T cells were mostly seen in COVID-19 survivors, compared to COVID-19 non-628 survivors, which has been associated with a positive COVID-19 outcome (22, 73) . 629

In contrast, we detected in COVID-19 non-survivors an enrichment of pathways involved with the 631 negative regulation of lymphocyte activation and increased neutrophil activation and degranulation, 632 supported by a significant decrease in predicted cell fraction of naïve B cells and naïve CD4+ and CD8+ 633 T cells and an increase of the neutrophil cell fraction. This is consistent with previous studies finding 634 elevated levels of neutrophils in blood (74) and lungs (75-78) in severe COVID-19. Furthermore, gene 635 pathways involved with an inflammatory response and cytokine signalling were enriched in COVID-19 636 non-survivors and we detected that a higher transcript abundance of several IL genes (IL1-RAP, IL-10, 637 IL1-R1, IL1-R2, IL18-R1 and IL18-RAP) and laboratory results indicated a increase of TNFα, IL-1β, 638 and IL-33 with the largest increase for IL-6 and IL-10. This is consistent with the previously reported 639 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 

Undiagnosed Pneumonia -China (Hubei): Request for information

COVID-19 revealed by temporal type I/III interferon patterns and flu comparison

Comparison of the characteristics, 780 morbidity, and mortality of COVID-19 and seasonal influenza: a nationwide, population-based retrospective cohort 781 study

Immune responses to influenza virus infection

Systems biological assessment of 785 immunity to mild versus severe COVID-19 infection in humans

CoV-2 Drives Development of COVID-19

A dynamic COVID-19 immune signature

Autoantibodies against type I IFNs in 791 patients with life-threatening COVID-19

Dysregulated Type I Interferon and 793

Inflammatory Monocyte-Macrophage Responses Cause Lethal Pneumonia in SARS-CoV-Infected Mice

Microbe

Non-redundant Front-Line Antiviral Protection against Influenza Virus Infection without Compromising Host Fitness

Safety and efficacy of inhaled nebulised 799 interferon beta-1a (SNG001) for treatment of SARS-CoV-2 infection: a randomised, double-blind, placebo-controlled, 800 phase 2 trial

Therapy Is Associated with Favorable Clinical Responses in COVID-19 Patients

Clinical and immunological features of severe and moderate 805 coronavirus disease 2019

Clinical features of patients infected with

The role of interleukin-6 in monitoring severe case of coronavirus 809 disease 2019

Targets of T Cell Responses to SARS-CoV-811

Coronavirus in Humans with COVID-19 Disease and Unexposed Individuals

Reduced naïve CD8+ T-cell priming efficacy in 813 elderly adults

Diversity and clonal selection in the human T-cell repertoire

Aging and cytomegalovirus infection 817 differentially and jointly affect distinct circulating T cell subsets in humans

Emerging Pandemic Diseases: How We Got to COVID-19

SARS-CoV-2 in Acute COVID-19 and Associations with Age and Disease Severity

Coronavirus 2 Viral Load on Risk of Intubation and Mortality Among Hospitalized Patients With Coronavirus Disease 824 2019

Adaptive immunity to SARS-CoV-2 and COVID-19

Evaluating the clinical impact of routine molecular point-of-care testing for COVID-19 in adults presenting to 827 hospital: A prospective, interventional, non-randomised, controlled study (CoV19POC)

Pragmatic multicentre randomised controlled trial 27

Clinical impact of molecular point-of-833 care testing for suspected COVID-19 in hospital (COV-19POC): a prospective, interventional, non-randomised, 834 controlled study

Clinical impact of a routine, molecular, point-of-836 care, test-and-treat strategy for influenza in adults admitted to hospital (FluPOC): a multicentre, open-label, 837 randomised controlled trial

Cutadapt removes adapter sequences from high-throughput sequencing reads

Sickle: A sliding-window, adaptive, quality-based trimming tool for FastQ files

FastQC: a quality control tool for high throughput sequence data

MultiQC: summarize analysis results for multiple tools and samples in a 845 single report

Mapping RNA-seq Reads with STAR

GENCODE: the reference human 848 genome annotation for The ENCODE Project

Robust and flexible detection of differential alternative 850 splicing from replicate RNA-Seq data

The Sequence Alignment/Map format and SAMtools

RStudio Team. RStudio: Integrated Development Environment for R

Pipe-Friendly Framework for Basic Statistical Tests

Rich B. table1: Tables of Descriptive Statistics in HTML

Molecular signatures of antibody 862 responses derived from a systems biology study of five human vaccines

limma powers differential expression analyses for RNA-865 sequencing and microarray studies

Network visualization and analysis of gene expression data 867 using BioLayout Express 3D

ToppGene Suite for gene list enrichment analysis and candidate gene 869 prioritization

HTSeq-a Python framework to work with high-throughput sequencing data

edgeR: a Bioconductor package for differential expression analysis of digital 873 gene expression data

Gene Ontology: tool for the unification of biology

The Gene Ontology resource: enriching a GOld mine

Elegant Graphics for Data Analysis

ggplot2" Based Publication Ready Plots

STRING v9.1: protein-protein interaction 885 networks, with increased coverage and integration

Annotation-free quantification of RNA splicing 887 using LeafCutter

Differential Splicing protocol

Re: Leafcutter results

A new view of 895 transcriptome complexity and regulation through the lens of local splicing variations

Determining cell type abundance and 898 expression from bulk tissues with digital cytometry

Open-Source Knowledge Discovery 902 Platform. Patterns

Docker: lightweight Linux containers for consistent development and deployment

Tissue-Specific Immunopathology 906 in Fatal COVID-19

Transcriptomic signatures 908 differentiate survival from fatal outcomes in humans infected with Ebola virus

COVID-19 or influenza

Analysis of an Ebola virus disease survivor 67

Prognostic factors for severity 915 and mortality in patients infected with COVID-19: A systematic review

COVID-19, immunothrombosis and venous thromboembolism: biological mechanisms

Coagulation abnormalities in SARS-CoV-2 infection: 919 overexpression tissue factor

Convergent antibody responses to SARS-CoV-921 2 in convalescent individuals

Profiling Early Humoral Response to Diagnose Novel Coronavirus 923

Disease (COVID-19)

Convalescent Individuals with Asymptomatic or Mild COVID-19

Broad and strong memory CD4 + and CD8 + T cells induced by 927 SARS-CoV-2 in UK convalescent individuals following COVID-19

Comprehensive mapping of 929 immune perturbations associated with severe COVID-19. Sci Immunol

Clinical and pathological investigation of patients with severe COVID-19. JCI 932 Insight

Single-cell landscape of bronchoalveolar immune cells in patients with 934 COVID-19

Viral presence and immunopathology in 936 patients with lethal COVID-19: a prospective autopsy cohort study

Neutrophil extracellular traps infiltrate 938 the lung airway, interstitial, and vascular compartments in severe COVID-19

COVID-19 and the human innate immune system

Centers for Disease Control and Prevention. COVID Data Tracker [Internet]. Centers for Disease Control and 943 Prevention

Coronavirus and Obesity: Could Insulin Resistance Mediate the Severity of Covid-19 946 Infection? Front Public Health

Receptor and Regulator of the Renin-Angiotensin System: Celebrating the 20th Anniversary of the Discovery of ACE2

Loss of ACE2 exaggerates high-calorie 951 diet-induced insulin resistance by reduction of GLUT4 in mice

IRF-7 is the master regulator of type-I interferon-953 dependent immune responses

Decoding SARS-CoV-2 hijacking of host mitochondria in COVID-19

positive regulation of genes encoding the activation of innate immune system, viral and IFN response 640(3), and increase of proinflammatory macrophages (79) and elevated IL-6 and IL-10 in severe . 642 643 When comparing the immune response between patients who either survived or died of COVID-19 it 644 appears that, as Sette and Crotty (24) summarised, that COVID-19 severity is largely due to an early 645 virus-driven evasion of innate immune recognition leading to a subsequent delayed adaptive immune 646 response with a fatal COVID-19 outcome, as shown by Lucas et al. (80) , where the innate immune 647 response is ever-expanding due to an absence of a quick T cell response. A delayed adaptive immune 648 response to COVID-19 can occur in the elderly due to their reduced ability to mount a successful 649 adaptive immune response leading to an increased risk of death (22) . This reduced ability to mount 650 an adaptive immune response in the elderly is due to a scarcity of naïve T cells caused by aging (18-651 20) and the association of age and severe or fatal COVID-19 is already known, for example, as of April 652 15 th 2021 in the United States 95.4% of COVID-19 deaths have occurred in 50-year-olds and older, and 653 59.3% in 75-year-olds and older (81). Similarly, we found that patients who survived COVID-19 were 654 younger, had a higher predicted naïve CD4+ T cell and naïve B cell fraction, and had an increased heart 655 rate compared to non-survivors. Further research is needed to assess the causality of these factors, 656 for example the relationship between increased age and heart rate in non-survivors. 657 658 Topological analysis was performed to identify the global map of gene pathways differentially 659 activated between COVID-19 and influenza. The first differentially activated pathway was enriched for 660 genes related to ribosomal and insulin pathways indicating differences in effects on translational 661 machinery and supporting the reported roles of insulin resistance linked to COVID-19 severity (82). 662Although highly speculative, insulin signalling differences may reflect the role of angiotensin 663 converting enzyme 2 (ACE2), the binding site for SARS-CoV-2, which degrades angiotensin 2, 664 protecting against oxidative stress and insulin resistance driven by the renin-angiotensin-aldosterone 665 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101/2021.05.12.21257086 doi: medRxiv preprint system (83). Additionally, ACE2 expression has been found to be increased in rats given a high sucrose 666 diet or insulin sensitisers (84). Furthermore, the first pathway was also found to be enriched for genes 667 transcribed by Myc. Activated Myc represses IRF7 which regulates type I IFN production (85), and 668 correspondingly we found a significant lower IRF7 expression and a lower induction of IFN in patients 669 with COVID-19 compared to influenza. This low IFN induction in COVID-19 may be due to the virus 670 avoiding or delaying an intracellular innate immune response to type I and type III IFNs (6-9). The 671 second most differentially activated pathway, peak gene NDUFAB1, involved with the mitochondrial 672 complex I assembly model OXPHOS system supports reported increased COVID-19 disease severity 673 linked to SARS-CoV-2 being able to highjack mitochondria of immune cells, replicate and disrupt 674 mitochondrial dynamics (86). The third differentially activated pathway was associated with the 675 cellular ubiquitin-proteasome pathways which are known to play important roles in coronavirus 676 infection cycles (87). The protein synthesis and ubiquitination-related pathways might reflect 677 mechanisms of increased viral replication and suppression of host interferon signalling pathways, 678including increased degradation of IκBα which suppresses the IFN-induced NF-κB activation pathway. 679Also, in SARS-CoV, accessory protein P6, whose sequence is conserved in SARS-CoV-2 (88), promotes 680 the ubiquitin-dependent proteasomal degradation of N-Myc interactor, thus limiting IFN signalling 681 (89). However, the peak marker of this pathway PSMD14 which prevents interferon regulatory factor 682 3 (IRF3) autophagic degradation and therefore, permits IRF3-mediated type I IFN activation (90); 683 shedding light on the complex mechanistic differences regulating interferon production between 684 COVID-19 and influenza. 685 686

In this study, we have compared side-by-side SARS-CoV-2 and a stereotypical respiratory viral infection 688 (influenza), and COVID-19 survivors and non-survivors. Distinct patterns of transcript abundances and 689 cellular composition were found in whole blood that can differentiate the infection source, furthering 690 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Randox laboratories Ltd and Cidara therapeutics. TWC has been a member of advisory boards for 718Roche and Janssen and has received reimbursement for these. TWC is member of two independent 719 data monitoring committees for trials sponsored by Roche. TWC has previously acted as the UK chief 720 CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021 

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)The copyright holder for this preprint this version posted May 15, 2021. ; https://doi.org/10.1101 https://doi.org/10. /2021