key: cord-0317710-5uolpxzp
authors: Qiu, X.; Li, J.; Bonenfant, J.; Klein, W.; Jaroszewski, L.; Godzik, A.; Nair, M. G.
title: Dynamic changes in human single cell transcriptional signatures during fatal sepsis
date: 2021-03-02
journal: nan
DOI: 10.1101/2021.03.01.21252411
sha: 12a76b34dd26d74391d63614363b54b2e3de19ea
doc_id: 317710
cord_uid: 5uolpxzp

Systemic infections, especially in patients with chronic diseases, result in sepsis: an explosive, uncoordinated immune response that can lead to multisystem organ failure with a high mortality rate. Sepsis survivors and non-survivors oftentimes have similar clinical phenotypes or sepsis biomarker expression upon diagnosis, suggesting that the dynamics of sepsis in the critical early stage may have an impact on these opposite outcomes. To investigate this, we designed a within-subject study of patients with systemic gram-negative bacterial sepsis with surviving and fatal outcomes and performed single-cell transcriptomic analyses of peripheral blood mononuclear cells (PBMC) collected during the critical period between sepsis recognition and 6 hours. We observed that the largest sepsis-induced expression changes over time in surviving versus fatal sepsis were in CD14+ monocytes, including gene signatures previously reported for sepsis outcomes. We further identify changes in the metabolic pathways of both monocytes and platelets, the emergence of erythroid precursors, and T cell exhaustion signatures, with the most extreme differences occurring between the non-sepsis control and the sepsis non-survivor. Our single-cell observations are consistent with trends from public datasets but also reveal specific effects in individual immune cell populations, which change within hours. In conclusion, this pilot study provides the first single-cell results with a repeated measures design in sepsis to analyze the temporal changes in the immune cell population behavior in surviving or fatal sepsis. These findings indicate that tracking temporal expression changes in specific cell-types could lead to more accurate predictions of sepsis outcomes. We also identify molecular pathways that could be therapeutically controlled to improve the sepsis trajectory toward better outcomes.

Sepsis is an inflammatory syndrome caused by a systemic infection that can lead to multi-system organ 2 failure and death. Sepsis is responsible for a significant percentage of in-hospital healthcare both in the US and 3 worldwide and is associated with a high mortality rate 1, 2 . Despite many efforts, no targeted therapeutic agents 4 against sepsis have been developed to assist in the treatment of sepsis. One acknowledged challenge is the 5 complexity of the disease involving the competing interplay between rampant inflammation (cytokine storm) and 6 paradoxically, the almost simultaneous shutdown of the immune system (immunoparalysis) 3, 4 . Another sepsis 7 paradox is that some patients with nearly identical clinical phenotypes as quantified by qSOFA and APACHE 8 scores, die at every stage of the disease while others survive 5 . The host response to sepsis has been studied in 9 several blood and PBMC profiling studies with gene-expression or proteomics methods 6 . These identify several 10 prognostic biomarkers, such as lactate, procalcitonin, C-reactive protein (CRP), ferritin, and erythrocyte 11 sedimentation rate (ESR), which, along with clinical scores, are standardly utilized to evaluate sepsis patients and 12 determine their care 5 . 13 However, connecting these high-level observations to accurate clinical outcomes presents an unresolved 14 challenge, likely due to the complexity and heterogeneity of this disease. To gain molecular insight into this 15 heterogeneity, many studies have been conducted to identify a potential sepsis molecular signature, which could 16 aid in diagnosis or treatment 7 . Recently, the first single-cell analysis of the status of immune cells in sepsis was 17 reported, which identified abnormal monocyte states associated with immune dysregulation 8 . Here, we apply the 18 same approach to focus on the additional question of immune cell trajectory in sepsis survivor and non-survivor 19 outcomes. We performed single cell transcriptomics analyses in fatal or surviving sepsis using a within-subject 20 study design of peripheral blood mononuclear cells (PBMC) collected from septic patients in the Intensive Care 21 Unit (ICU) at 0 and 6 hours from sepsis recognition. Although a limitation in this study is the small sample size 22

of two sepsis patients, we validated our analyses by comparison with similarly processed PBMC from non-sepsis 23 volunteers and corroborated our findings with available public domain expression datasets. Our timed analyses 24 further reveal the emergence of abnormal immune cells, including new types of cells typical of sepsis but not 25 present in healthy controls, as well as classical cell types present in both sepsis and healthy conditions but with 26 an abnormal gene expression profile. Specifically, we observed that fatal sepsis was associated with expansion of 27 platelets and erythrocyte precursors, and the overall expression of genes related to hypoxic stress and 28 inflammation that were increased over 6 hours, especially in the monocytes. Additionally, the lymphocyte subsets 29 in fatal outcomes expressed genes related to exhaustion. On the other hand, sepsis survival over time involved 30 expression of genes related to recovery from cellular stress, including the regain of cytotoxic effector function for 31 lymphocyte subsets, and an increase in genes related to monocyte migration and chemotaxis. Last, we observed 32 a switch in metabolic state at the cellular level, from oxidative phosphorylation to glycolysis in fatal outcomes. 33

In conclusion, this pilot study, which focused on within-subject analyses of PBMC over time, offers the unique 34

perspective of dynamic immune cell changes in fatal sepsis. Specifically, we identify abnormal immune cell 35 subsets, functional pathways and molecular signatures at the single cell resolution that are associated with fatal 36 or surviving outcomes in sepsis. This study provides foundation data and identifies specific cell subsets and 37 molecular pathways that can be further explored for better prediction of sepsis outcomes. 38

To gain molecular understanding of the immune state in surviving or non-surviving sepsis outcomes, we 3 performed retrospective single-cell RNA sequencing on PBMC from two septic shock patients. Both patients 4 belonged to a cohort of 45 patients admitted with severe sepsis to the RUHS ICU and enrolled in the study based 5 on Sepsis-3 clinical parameters (manuscript in preparation). Both patients were females, aged between 65 and 70 6 years old. The first patient expired by 24 hours post-enrollment (NS, non-survivor). The second patient recovered 7 from infection and was discharged from the hospital after 22 days (S, survivor). Both patients suffered from 8

Escherichia coli bacteremia and were treated with broad spectrum antibiotics within the first 24 hours. While 9 clinical parameters (qSOFA and APACHE scores) were similar, both patients exhibited different plasma cytokine 10 levels, although in both patients these levels were dramatically elevated compared to baseline non-sepsis 11 volunteers, (Table 1) . These results are consistent with the known phenomenon of the sepsis-induced cytokine 12 storm 4 . Interestingly, re-stimulation of PBMC from the same sepsis patients with lipopolysaccharide led to 13 reduced TNFa secretion as compared to PBMC from non-sepsis controls (Table 1) , consistent with monocytic 14 deactivation seen in sepsis immunoparalysis 9 . Additionally, flow cytometric analysis of PBMC revealed different 15 immune subset distribution with sepsis including reduced monocyte and lymphocyte subsets especially in the 16 non-survivor ( Figure 1 ). We also observed the emergence of cell subsets that we were unable to define with 17 common PBMC surface antibodies ( Figure 1C ). Together, these data reveal similar and distinct clinical and 18 peripheral immune profiles in sepsis. However, more detailed subsetting of specific immune cells and insights 19 into how temporal changes in their gene expression relate to sepsis outcome were lacking, which we addressed 20 by single cell RNA sequencing. 21

Single-cell transcriptome analysis was performed on a 10X Genomics platform according to a standard 23 pipeline (Figure 2A ). After quality filtering (see Materials and methods), the transcriptome profiles of 27,685 24 cells were collected (15,546 cells from two non-sepsis control samples, 5,758 cells from the two sepsis non-25 survivor samples, and 6,381 cells from the two sepsis survivor samples). Given that PBMC samples were 26 processed identically, and equivalent numbers of viable cells from each sample were subjected to single cell 27 transcriptomics, the ~3-fold decrease in cell number from sepsis patients that passed quality control is likely 28 caused by the sepsis syndrome and consequent cellular stress and apoptosis. Analysis of the single-cell 29 transcriptomes of all cells from all subjects (Seurat v3 canonical pipeline 10 ) followed by the consensus-based The dominant cell types in the healthy controls (HCs) were CD4 T cells (30.45%), followed by B cells 36 (25.04%), and then CD14 monocytes (18.80%). In sepsis survivor (S) and non-survivor (NS), we observed an 37 approximately two-fold decrease in the percentage of CD4 T cells (S, 13.21%; NS, 8.23%) and B cells (S, 10.55%; 38 NS, 6.06%) as compared to the HCs. CD14+ monocytes showed the opposite trend with the highest percentage 39 in sepsis survivor (38.44%) as compared to the HC (18.80%) and the non-survivor (12.82%). These population 40 distributions exhibited similar trends to the flow cytometry analysis (Figure 1) , however, we also identified the 41 expansion of atypical PBMC subsets in sepsis. Specifically, we observed a dramatic increase in erythroid 42 precursors and platelets in S and NS PBMC. Only traces of these cell types are typically present following PBMC 43 isolation by gradient centrifugation, therefore their high levels here may suggest abnormal expansion and 44 activation in sepsis. Specifically, HCs contained 2.88% of platelets and 0.07% of erythroid precursors but both 45 subsets were elevated in sepsis, especially in the NS samples (37.43% platelets, 10.28% erythroid precursors) and, 46 to a much lower extent, in the S samples (5.45% platelets, 1.18% erythroid precursors). In general, the populations 47 of most cell types in the non-survivor were more distant from the healthy controls than the cell populations in the 48

survivor. The results obtained with Seurat v3 canonical pipeline were consistent with additional calculations 49 performed with a slightly different methodology using the MAGIC imputation algorithm 11 and Seurat v4 map to 50 reference method 12 . The final assignments of cell types were based on the consensus approach (see Materials and 51

We next identified differentially expressed genes (DEGs), and associated pathways (whose down-and 53 upregulations were evaluated using modules' scores) by comparisons of sepsis samples and healthy controls 54 (Sepsis vs. HCs) and sepsis non-survivor versus sepsis survivor samples (NS vs. S) ( Table 2) . We also performed 55 temporal comparisons to find DEGs between hour 6 (T6) and hour 0 (T0) (with time counted from sepsis 56 recognition). This temporal analysis was done separately for sepsis non-survivor (NS T6 vs. T0), and sepsis 57 survivor (S T6 vs. T0). Unsurprisingly, the greatest changes were observed between HCs and sepsis patients. in sepsis, especially in non-survivors. In platelets, almost two times more genes were downregulated in sepsis 62 compared to controls than upregulated, suggesting transcription shutdown in sepsis. This trend was also observed 63 when comparing non-survivor to survivor, suggesting that aberrant gene expression and likely transcription 64 shutdown in platelets was indicative of sepsis severity. When comparing gene expression changes within 6 hours 65 within the same patient, there were less DEGs. However, we observed a stark contrast in DEGs for lymphocyte 66 subsets (B cells, T cells, and NK cells); NS exhibited low DEG numbers while S had almost 10-fold more DEGs. 67

This suggests that lymphocyte subsets undergo more transcriptional changes in surviving outcomes, while in fatal 68 sepsis, these cells may be exhausted. The opposite trend was observed in platelets, with more dynamic changes 69 (up and down) occurring in the non-survivor after 6 hours, while there were almost no changes in platelet gene 70 expression in survivors. Taken together, analysis of major changes in gene expression with regards to immune 71 subsets identifies that sepsis leads to a loss in lymphocyte subsets and the expansion of platelets and erythroid 72 precursors, which is more dramatic in fatal sepsis. Additionally, investigation of gene expression over time within 73 the same sepsis patient suggests that fatal sepsis is associated with transcriptional shutdown especially in T 74 lymphocytes, while transcriptional recovery of these immune subsets is observed in surviving sepsis. 75 MHC class II-related and translation initiation-related genes exhibit distinct sepsis-specific expression 76 patterns 77

Studies previously reported the decrease of expression of MHC class II-related genes in immune cells in 78 sepsis [13] [14] [15] [16] . We further explored these genes' expression changes in sepsis, with specific focus on individual cell 79 types, temporal changes within subjects, and association with survival outcomes. We confirmed that MHC class Figure 2A) , which provided a more robust measure than individual DEGs and, thus, 88 more suitable for the smaller number of datapoints. Across most cell types, the MHC class II module score was 89 decreased in sepsis in comparison with HCs, especially in the non-survivor. 90

Average expression of genes related to translation initiation (i.e., genes encoding ribosomal proteins) 91 exhibited a distinctive pattern in sepsis and control samples. In the sepsis survivor (S), these genes, including 92 those encoding L ribosomal, S ribosomal, and mitochondrial ribosomal proteins 17 , were expressed at a higher 93 level, than in healthy controls and their expression also increased with time (Supplementary Figure 3) . The NS 94 samples had the lowest expression score of the module representing these genes across almost all cell types and, 95 consequently, the deviation of this module score from healthy controls was also most noticeable in NS samples 96 (Bonferroni adjusted p-values below 0.01) (Supplementary Figure 2B) . The biggest decreases in this module in 97 NS were observed in platelets, CD14+ monocytes, and FCGR3A+ monocytes. These cell types had over 60 98 ribosomal proteins encoding genes downregulated in NS as compared to S; other cell types had fewer than 15 99 significantly downregulated genes in the same comparison, ribosome proteins are listed in Supplementary Table  100 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. ;

1. Interestingly in monocytes, B cells and CD4+ T cells, the ribosomal genes module score was higher in S 101 samples than in HC, suggesting a mechanism for re-establishment of immune homeostasis in recovery from sepsis. 102

In contrast, the score was lower in platelets, CD8+ T cells, NK cells, and erythroid precursors. These results are 103 consistent with previous studies which have demonstrated that sepsis halts protein translation and hypothesized 104 that this would potentially lead to organ failure 18, 19 . Our analysis additionally highlights platelets and monocytes 105

as the main cell subsets affected in fatal sepsis outcomes, where they exhibit translation shutdown within hours. 106

Aberrant platelet gene expression is observed in fatal sepsis 107

The role of platelets in the development of sepsis pathophysiology is not well established. Still, increasing 108

number of studies are beginning to recognize that platelets are altered in sepsis, and that transcriptional and 109

translational changes in platelets are related to mortality 20 . Out of all the cell types analyzed, platelets contained 110 the highest number of DEGs in our comparisons of sepsis samples vs. HCs and in comparisons of NS vs. S 111 samples. Specifically, the number of DEGs in platelets from both comparisons was more than two-fold higher 112 than in FCGR3A+ monocytes, which ranked second in the number of DEGs in these comparisons ( Table 2) . 113

While most DEGs in platelets were down-regulated in sepsis, genes contributing to microvascular coagulation 114

were up-regulated (64% in Sepsis vs. HC, and 79% in NS vs. S, see Supplementary Table 1 for the list of genes 115 included in coagulation module). Sepsis non-survivor samples had higher coagulation module scores than HC 116 and S samples in most cell types, but this trend was most pronounced in platelets (Supplementary Figure 2C) . In 117 contrast, in survivor samples, only platelets exhibited higher coagulation module scores than HC. 118

In platelets, significant gene expression changes were observed for the Rho GTPases RAC1 and RHOA, 119

which regulate cell adhesion 21 . RAC1, the main GTPase required for cell barrier maintenance and stabilization 120 was downregulated in NS vs. S comparison (logFC of 0.31 and adjusted p-value < 0.001). RHOA, the GTPase 121 that negatively regulates barrier properties under both resting and inflammatory conditions was up-regulated in 122

Sepsis vs. HC (logFC of 0.6 and adjusted p-value < 0.001) ( Figure 3B ). In NS, the expression changes in platelets 123

suggest both increased microvascular permeability and microvascular coagulation. These phenomena may 124

contribute to the rapid development of multi-organ failure. 125

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. The published studies reporting biomarker genes in sepsis were mostly based on bulk data from all 139 immune cells from the peripheral blood. The extensive study by Sweeney et al. 7 used a community-based 140 approach to establish optimized lists of genes whose expressions have the strongest prognostic value for sepsis 141 mortality and survival. We tested if these lists could be used in single-cell transcriptomic data to provide insights 142 into cell types crucial for sepsis outcomes. Interestingly, genes linked to sepsis survival were mostly expressed in 143 monocytes from HC and S T6 samples ( Figure 3C ). On the other hand, the genes linked to sepsis mortality were 144 more represented in monocytes from NS and S T0 samples ( Figure 3D ). 145

Next, we evaluated monocyte-specific cytokine expression. Compared to HCs, CD14+ monocytes from 146 sepsis patients were characterized by upregulation of chemokines (CCL3 and CCL4) ( Figure 4A ). Pro-147

inflammatory cytokines, chemokines and adipokines including IL6, CCL2, CCL3, CCL4, CCL7, HMGB1, and 148

NAMPT were overexpressed in NS as compared to S samples ( Figure 4B ). Amphiregulin (AREG) was also 149 overexpressed in NS, where it may have a pro-inflammatory role through promoting production of pro-150

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. ;  inflammatory cytokine IL8 (CXCL8) 24 . CXCL8 itself was upregulated in NS as compared to S in FCGR3A+ 151 monocytes ( Figure 4D ). Another up-regulated gene -EBI3, is also reported to promote pro-inflammatory IL6 152

functions by mediating trans-signaling 25 . Of note, PPBP (CXCL7), a potent neutrophil chemoattractant also 153 expressed by platelets 26 , was one of the top genes upregulated in both monocyte subsets and in both comparisons 154

(Sepsis vs HC, and NS vs S). These results suggest that monocytes in sepsis are in hyperinflammatory state, which 155 is more severe in patients with fatal outcomes. 156

Sepsis survival is associated with increased IFN response and reduced exhaustion in cytotoxic lymphocytes. 157

The patients as compared to HC (Supplementary Figure 4A) . Notably, in NK cells the NS samples had a much higher 166 exhaustion score than S samples (Supplementary Figure 4B ). 167

We investigated T cell-mediated cytotoxicity using a list of genes associated with GO term "T cell 168 mediated cytotoxicity" (GO:0001913). Compared to the HCs, sepsis patients had higher expression of genes 169 associated with T-cell cytotoxicity in CD8+ T cells (Supplementary Figure 4C) . However, when comparing NS 170 and S samples, we observed that CD8+ T and NK cells from NS had significantly lower expression of 171 cytotoxicity-related genes than S (Supplementary Figure 4D) , confirming the exhaustion and immunosuppressed 172 state of lymphocytes in NS.

Taken together, these transcriptomic findings indicate that NK and T lymphocytes in the sepsis survivor 174

had a more robust response to IFNs. Although all CD8+ T and NK cells in the sepsis samples exhibited gene 175

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. ;  expression signatures associated with exhaustion, the sepsis survivor had adaptive cytotoxic lymphocytes 176 activated, while the non-survivor exhibited reduced cytotoxic activity in both innate and adaptive lymphocyte 177

subsets. These findings support the predictive potential of lymphocyte exhaustion for detrimental outcomes in 178 sepsis. Consistent with other studies, our findings also support the therapeutic potential of inhibiting lymphocyte 179 apoptosis and restoring lymphocyte function for new treatments to restore immune homeostasis after sepsis 180 episodes [29] [30] [31] [32] [33] [34] . 181

Sepsis drives metabolic shift and hypoxic stress which is exacerbated in fatal outcomes 182

The metabolic state of specific immune cells profoundly determines their function. In the steady-state, 183

immune cells rely on oxidative phosphorylation (OXPHOS) for ATP production as the energy source, maintaining 184

homeostasis. In sepsis, immune cells such as monocytes undergo the process known as Warburg effect, in which 185 the metabolism mechanism shifts from OXPHOS to glycolysis by using lactic acid fermentation as the 186

predominant energy source 35 . The transcription factor hypoxia-inducible factor (HIF1A) is a main driver for this 187 metabolic switch, of relevance to the hypoxemia in severe sepsis 36 . We examined HIF1A expression among cell was observed in multiple cell types from the sepsis non-survivor to significantly exceed values in NS and HC 200 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. ;  samples (CD14+ monocytes, FCGR3A+ monocytes, and platelets, Bonferroni adjusted p-value < 0.0001). None 201 of the cell types from the sepsis survivor appeared in the highest module score in glycolysis across three conditions 202 ( Figure 5B ). These results indicate that in severe sepsis, monocytes and platelets shift their metabolic function 203 from OXPHOS to glycolysis, and this shift is more extreme in fatal outcomes. Interestingly, in sepsis platelets 204

showed increased module scores for both OXPHOS and glycolysis, suggesting metabolic dysfunction and 205 competition for energy sources. In healthy controls, monocytes and platelets had OXPHOS and glycolysis module 206 scores significantly below levels observed in both sepsis samples (Bonferroni adjusted p-value < 0.001) ( Figure  207 5A, B), confirming that monocytes and platelets have dysregulated metabolic activity in sepsis patients. 208

To investigate if HIF1A was the main driver of the observed metabolic changes, and to evaluate which 209 cell types were responsive to HIF1A, we evaluated correlations between HIF1A expression and 210

OXPHOS/glycolysis module scores. Cells with no detected HIF1A expression were filtered out, resulting in 211

14,077 cells suitable for the analysis. Across the three conditions (HC, S and NS), where significant correlation 212 between HIF1A and OXPHOS was observed (correlation p-value < 0.05), the correlation was negative in CD14+ 213

Mono, FCGR3A+ Monocytes, and platelets, which had the absolute correlation coefficient |R| above 0.25 and p-214 value < 0.05 (Supplementary Figure 5B) . The similar analysis of correlation between HIF1A expression and 215

glycolysis showed the opposite effect where CD14+ Mono, FCGR3A+ Monocytes, and platelets across the three 216

conditions with significant correlation (p-value < 0.05) had a positive correlation coefficient. The highest 217 correlation coefficients, exceeding 0.25, were observed CD14+ and FCGR3A+ monocytes from the sepsis non-218

survivor (Supplementary Figure 5C) .

The erythroid precursor expansion observed in the non-survivor sepsis patient was also indicative of 220 severe hypoxemia in fatal sepsis 39 . We compared pathway changes in sepsis vs. HC and NS vs. S. Erythroid 221 precursors in sepsis expressed genes related to hypoxic stress (oxygen transport, erythrocyte differentiation, 222 response to oxidative stress, cofactor catabolic process and cellular oxidant detoxification) ( Figure 5C ). The 223

pathways that were down-regulated in sepsis vs. HC and NS vs. S ( Figure 5C-D) , included translational initiation, 224

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. ; https://doi.org/10.1101/2021.03.01.21252411 doi: medRxiv preprint RNA splicing, protein nitrosylation and regulation of sequence-specific DNA binding transcription factor activity, 225

suggesting that erythroid precursors in sepsis exhibited a halt in protein translation. 226

Together, these data suggest that the peripheral immune cells in sepsis are responsive to the hypoxic 227 environment that leads to HIF1A-mediated shifts in metabolic function towards glycolysis, with fatal outcomes 228 exhibiting more extreme hypoxic stress leading to protein shutdown especially in erythrocyte precursors. These 229 results support the possibility of therapeutically targeting HIF1A to restore metabolic homeostasis in the sepsis 230 immune response. 231

Temporal changes in the transcriptome profile elucidate differences between the survivor and the non-232 survivor in the first hours of sepsis recognition 233

To reveal trends associated with recovery and fatal outcomes we analyzed temporal changes of gene 234 expression in immune cell subsets during the first 6 hours from sepsis recognition. When comparing hour 6 and 235

hour 0 in sepsis non-survivor (NS T6 vs. T0), the cell type that had the most differentially expressed genes (DEGs) 236

and pathway changes were the CD14+ monocytes. For the sepsis survivor (S T6 vs. T0 comparison), the cell type 237

with the highest number of DEGs were the CD4+ T cells, and the cell type with the highest number of pathway 238 changes were the CD14+ monocytes (Table 2) . To examine how CD14+ monocytes contribute to the fatal sepsis 239 outcome, we investigated functional pathways for which expression was increased in non-survivor (NS T6 vs. 240 T0) but decreased in survivor (S T6 vs. T0) ( Figure 6A ). One of the GO pathways associated with fatal outcomes 241 appeared unrelated to sepsis ('female pregnancy'). However, the genes in this pathway were involved in tissue 242 remodeling and fibrosis (eg. CALR/TIMP1/ADM), which are also related to the "inflammatory response to 243 wounding pathway) potentially suggesting vessel disruption and remodeling in fatal sepsis. Additionally, 244 pathways related to metabolic dysfunction, and inflammatory and oxidative stress were also increased over time 245

in fatal sepsis. To examine the role of CD14+ monocytes in sepsis recovery we tested which pathways' expression 246

was decreasing in sepsis non-survivor (NS T6 vs. T0) but increasing in sepsis survivor (S T6 vs. T0). Those 247 functional pathways, likely associated with recovery, included cell migration, and regulation of inflammatory 248 response ( Figure 6E ). The above results suggest that within 6 hours of sepsis progression the CD14+ monocytes 249 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. ; https://doi.org/10.1101/2021.03.01.21252411 doi: medRxiv preprint in the non-survivor undergo increasing response to cellular stress and inflammation and trigger tissue remodeling 250 processes, while exhibiting waning functions of neutrophils, cell migration, and reduced immune cell proliferation. 251 FCGR3A+ monocytes also exhibited multiple pathways with opposite temporal expression trends in non-survivor 252 and survivor (NS T6 vs. T0 and S T6 vs. T0, Figure 6B ). The only significant pathway which increased over time 253

in fatal sepsis was regulation of chemokine (genes up in NS T6 are PYCARD/GSTP1/CSF1R, and genes down 254 in S T6 are HIF1A/CLEC7A/DDX3X), which may suggest that this regulatory monocyte subset in non-survivors 255 was more immunosuppressed and/or refractory to outside chemotactic signals. In the FCGR3A+ monocytes from 256 the sepsis survivor, there was an increase of genes over time related to monocyte migration, cell-cell interaction 257

and metabolic activity, as well as regulation of apoptosis. These suggest a return to homeostasis and normal 258 monocyte function and increased cell survival.

Among lymphocytes, CD4+ T was the cell type that had the highest number of pathways showing opposite 260 temporal trends in survivor and non-survivor (NS T6 vs. T0 and S T6 vs. T0, Figure 6C ). Interestingly, all these 261 pathways were decreasing in the CD4+ T cells from non-survivor and increasing in the survivor ( Figure 6G ). The 262

functions of these pathways, possibly associated with recovery, are related to T cell cytotoxic function including 263 and response to pathogens. This suggests that the rapid recovery of T cells (in particular, their reversal from 264 exhaustion) may be crucial for positive outcomes in sepsis.

Other cell types where pathways showed opposite temporal trends in non-survivor and survivor (NS T6 266

vs. T0 and S T6 vs. T0) were the CD8+ T cells, NK cells, and platelets (Supplementary Figures 6A-C) . CD8+ T 267 cells and NK exhibited trends similar to CD4+ T cells, where all pathways with opposite trends were decreasing 268 in non-survivor (NS T6 vs. T0) but increasing in survivor (S T6 vs. T0). The functions of these pathways were 269 mainly related to protein synthesis (Supplementary Figures 6D, E) . In platelets, the pathways increasing non-270

survivor (NS T6 vs. T0) but decreasing in survivor (S T6 vs. T0) involved apoptotic, metabolic, and protein 271

folding processes (Supplementary Figure 6F) .

In summary, the analysis of single-cell transcriptomics in sepsis with different outcomes revealed distinct 273 dynamic trends in expression in immune cell subsets, highlighting the benefits of tracking temporal changes by 274 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. ; https://doi.org/10.1101/2021.03.01.21252411 doi: medRxiv preprint single cell analysis, which could be specifically targeted to improve outcomes. In monocytes, the fatal outcome 275 appears to be associated with increasing expression of pro-inflammatory pathways but, at the same time, with the 276 inability to respond to external stimuli. Similarly, in CD4+T cells, increasing exhaustion and hyporesponsiveness 277

were associated a fatal outcome. For the other PBMC subsets, pathways correlated with a fatal outcome were not 278

immune-specific, but rather associated with protein production shutdown, that is reflective of dysfunction and 279 exhaustion, as observed in the monocytes and T cells, respectively. 280

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Sepsis is a dysregulated systemic inflammatory response, which alters the innate and adaptive immune 2 responses to microbial invasion and results in organ injury with mortality rates of 15-25% 40, 41 . Given significant 3 sepsis disease heterogeneity, single cell transcriptomics offers an valuable approach to provide a better 4 understanding of the molecular mechanisms of sepsis, however, from over 1000 single-cell transcriptomics 5 studies have been published to date 42 , only two studied sepsis 8, 43 . In those two studies, the authors focused on 6 monocytes and myeloid-derived suppressor cells by sorting on specific surface markers. In comparison, our study 7 used the centrifuge gradient-based approach before performing the single-cell RNA-seq, which expanded the cell 8 subsets investigated in the study. We additionally collected samples at different time points from a survivor and 9

non-survivor of sepsis, which provided temporal details of the immune response in severe sepsis. Limitations of 10 this study include the small sample size, therefore, we focused on two dramatically different clinical outcomes 11 (fatal versus recovery), while using a within-subject study design where patients had similar clinical syndrome 12 and sepsis etiology. In these focused analyses were able to identify specific immune cell subsets and gene 13 expression patterns over time that correlated with beneficial or fatal outcomes. 14 Our results are consistent with the previous studies both in single-cell and bulk sepsis transcriptomic 15 studies. The peripheral blood cell composition of non-survivors is much more "distant" from healthy controls 16 than the blood cells of survivors ( Figure 2B-D) . In addition, we identify cell types that were not studied by the 17 previous single-cell studies, which include platelets and erythroid precursors. These were found to be expanded 18 in sepsis patients, especially in the sepsis non-survivor. Examination of platelets from the non-survivor sepsis 19 patient revealed increased expression of genes related to microvascular permeability and microvascular 20 coagulation, which in the literature were reported to lead to the development of organ failure 21 . The erythroid 21 precursors, that were dramatically expanded in non-survivors, showed upregulation of genes related to hypoxic 22 stress and apoptosis, reflective of the hypoxic environment in severe sepsis that leads to emergency erythropoiesis. 23

Interestingly, according to a longitudinal COVID-19 study 44 , erythroid cells were identified as the hallmark of 24 severe disease and had defined molecular signatures linked to a fatal COVID-19 disease outcome. Here we also 25 All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. ; https://doi.org/10.1101/2021.03.01.21252411 doi: medRxiv preprint observed that erythrocyte expansion and expression of genes related to hypoxic stress was a major predictor of 26 fatal outcomes. 27

Consistent with previous studies showing that CD14+ monocytes play a significant role in the sepsis 28 pathogenesis 45, 46 , we observed aberrant gene expression and pathway changes in the monocytes of sepsis patients, 29

suggesting they were in a hyperinflammatory state (Table 2, Figure 3C-D) . However, in a novel observation, 30 monocytes from sepsis patients, especially the non-survivor, exhibited signs of being refractory to the external 31 environment, with reduced expression of interferon responsive genes and the inability to produce TNFa in 32 response to LPS treatment (Table 1) focused on the metabolic pathway changes, we found that monocytes and platelets from the sepsis non-survivor 36 exhibited a switch in their energy utilization, from the OXPHOS to partial OXPHOS and glycolysis ( Figure 5A By exploring the temporal changes in the transcriptome profile from the sepsis non-survivor and the 40 survivor, we found dramatic differences predominantly involving CD14+ monocytes ( Figure 6A ). 6 hours 41 following sepsis recognition, the non-survivor's CD14+ monocytes demonstrated an intense response to 42 stimulation, stress, and inflammatory behaviors. In contrast, these pathways were downregulated in the sepsis 43 survivor during this time ( Figure 6D ). Further, the survivor's CD14+ monocytes expressed pathways related to 44 cell migration, neutrophil function, lymphocyte proliferation, while these were downregulated in the non-survivor 45 ( Figure 6E ). Overall, our study suggests that not only the initial status of the sepsis patient but also the dynamic 46 changes in cell behavior during the critical period following diagnosis significantly effect sepsis outcome. Future 47 focus on these changes, specifically addressing immune cell metabolic dysfunction and identifying mechanisms 48 to promote their recovery from exhaustion may provide therapeutic and prognostic insight into sepsis. with big variation between cell types in the reference data set, then compares each cell's scRNA-seq data with 45 each sample from the reference data set and, lastly, performs iterative fine-tuning to select the most likely cell 46 type of each cell. The microarray dataset from Human Primary Cell Atlas Data with assigned labels was used as 47 the reference. Finally, each cluster was assigned a cell type with the highest percentage of cells assigned to that 48 type by SingleR. The third applied method of cell type assignment was scCATCH, where cell types are assigned 49 using the tissue-specific cellular taxonomy reference databases [49] [50] [51] and the evidence-based scoring protocol. Our 50 final assignment of cell types to clusters was based on the consensus of the three aforementioned methods as 51 follows: First, each cluster was assigned a cell type selected by most methods if possible. If each method gave a 52 different result, then the priority was given to the assignment based on canonical markers. If the markers-based 53 assignment was inconclusive, the consensus assignment was based on the results from SingleR method. 54

Independently from the main computational pipeline described above, we applied two alternative pipelines: DEGs with FDR-adjusted p-value < 0.05. The results of differential gene expression and functional enrichment 72 are summarized in Table 2 . Each bar shows an average percentage of a given cell-type across all samples from that condition (only cell types with fractions over 1% across all samples in a condition are shown). Error bars represent ± SEM for 2 HCs and 2 samples from each patient (T0 and T6). HC (n = 2), NS (n = 2) and S (n = 2). The significance was evaluated using two-sample t-tests. The differences with p-values below 0.05 are indicated as *, "ns" -not significant. A) The comparison of expression of MHC Class II -related genes in the analyzed samples. Heatmap coloring represents z-scored, log-normalized mean gene expression counts averaged across all cells from a given sample.

The percent of cells with non-zero expression of RHOA and RAC1 genes and average expression of these genes in platelets across three conditions. The color saturation indicates the average expression level, and the circle's size indicates the percentage of cells expressing a given gene. C), D) Expression of genes from C) survival All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. monocytes from sepsis vs. HC; D) FCGR3A+ monocytes from NS vs. S; Each dot represents one gene, red color indicates significantly overexpressed genes (LogFC > 0.5, and FDR adjusted p-value < 0.05); blue indicates significantly under expressed genes (0 < LogFC < 0.5, and FDR adjusted p-value < 0.05); green indicates genes with LogFC > 0.5, but FDR adjusted p-value > 0.05; gray indicates genes without significant expression differences. Cytokines (i.e. genes included in Gene Ontology term "cytokine activity" -GO:0005125) are labeled.

Volcano plots were prepared with R package EnhancedVolcano 54 . Venn diagrams describing temporal changes in pathways from each sepsis patient. Set labeled 'NS T0' contains pathways decreasing in NS between T0 and T6, set labeled 'NS T6' contains pathways increasing between T0 and T6 in NS, etc. A) CD14+ monocytes, B) FCGR3A+ monocytes, C) CD4+ T cells. Sets of pathways increasing in NS (up-regulated in NS T6 as compared to NS T0) and decreasing in S (down-regulated in S T6 as compared to S T0) is colored in orange. Set of pathways decreasing from NS T0 to T6 and increasing from S T0 to T6 re colored in green. D) Heatmap illustration of temporal changes of pathways which are changing in opposite All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) 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 March 2, 2021. ; https://doi.org/10.1101/2021.03.01.21252411 doi: medRxiv preprint direction in NS and in S D),E) in CD14+ monocytes, F),G) in FCGR3A+ monocytes, H) in CD4+ T cells. The sets of overlapping GO terms were reduced to representative ones using Revigo 55 (the cutoffs were, more than 10 overlapping GO terms and similarity > 0.4).

All rights reserved. No reuse allowed without permission.

(which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Table 1 . Characteristics of enrolled non-sepsis volunteers and sepsis patients at sepsis recognition (T0). Clinical parameters, cytokine levels in the plasma, and supernatant following LPS stimulation (10 ng/mL) of PBMCs. APACHE II: Acute physiology and chronic health evaluation II, SOFA: Sequential organ failure assessment, N.D.:

not detected.

All rights reserved. No reuse allowed without permission.

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

The Economic and Humanistic Burden of Severe Sepsis

Global, regional, and national sepsis incidence and mortality, 1990-2017: analysis for the Global Burden of Disease Study

Current trends in inflammatory and immunomodulatory mediators in sepsis

The immunopathology of sepsis and potential therapeutic targets

Biomarkers of sepsis: time for a reappraisal

Genomic landscape of the individual host response and outcomes in sepsis: a prospective cohort study

A community approach to mortality prediction in sepsis via gene expression analysis

An immune-cell signature of bacterial sepsis

Immunomodulatory therapies in sepsis

Recovering Gene Interactions from Single-Cell Data Using Data Diffusion

Integrated analysis of multimodal single-cell data. bioRxiv

Alteration of mononuclear cell immune-associated antigen expression, interleukin-1 expression, and antigen presentation during intra-abdominal infection

T Cell-and Monocyte-Specific RNA-Sequencing Analysis in Septic and Nonseptic Critically Ill Patients and in Patients with Cancer

Sepsis-induced changes in macrophage co-stimulatory molecule expression: CD86 as a regulator of anti-inflammatory IL-10 response

Coincidence of pro-and anti-inflammatory responses in the early phase of severe sepsis: Longitudinal study of mononuclear histocompatibility leukocyte antigen-DR expression, procalcitonin, C-reactive protein, and changes in T-cell subsets in septic and postoperative patients

Genenames.org: the HGNC and VGNC resources in 2019

Sepsis halts renal protein translation

Bacterial sepsis triggers an antiviral response that causes translation shutdown

Sepsis alters the transcriptional and translational landscape of human and murine platelets

Role of GTPases in control of microvascular permeability

Are Platelets Cells? And if Yes, are They Immune Cells? Front Immunol

Platelets present antigen in the context of MHC class I

Proinflammatory role of amphiregulin, an epidermal growth factor family member whose expression is augmented in rheumatoid arthritis patients

Epstein-Barr virus-induced gene 3 (EBI3) can mediate IL-6 trans-signaling

Platelet-Derived Chemokine CXCL7 Dimer Preferentially Exists in the Glycosaminoglycan-Bound Form: Implications for Neutrophil-Platelet Crosstalk. Front Immunol

Exhausted CD8(+) T cells exhibit low and strongly inhibited TCR signaling during chronic LCMV infection

Defining 'T cell exhaustion

Prevention of lymphocyte apoptosis--a potential treatment of sepsis?

IL-7 promotes T cell viability, trafficking, and functionality and improves survival in sepsis

Delayed administration of anti-PD-1 antibody reverses immune dysfunction and improves survival during sepsis

A shift toward inhibitory receptors and impaired effector functions on NK cells contribute to immunosuppression during sepsis

STAT5 phosphorylation in T cell subsets from septic patients in response to recombinant human interleukin-7: a pilot study

Frontline Science: Defects in immune function in patients with sepsis are associated with PD-1 or PD-L1 expression and can be restored by antibodies targeting PD-1 or PD-L1

Cellular metabolism of myeloid cells in sepsis

The immune system's role in sepsis progression, resolution, and long-term outcome

High expression of oxidative phosphorylation genes predicts improved survival in squamous cell carcinomas of the head and neck and lung

A prospective analysis of lymphocyte phenotype and function over the course of acute sepsis

The pathways of mitophagy for quality control and clearance of mitochondria

A randomized trial of protocol-based care for early septic shock

A curated database reveals trends in single-cell transcriptomics

Single Cell RNA-SEQ of Human Myeloid Derived Suppressor Cells in Late Sepsis Reveals Multiple Subsets with Unique Transcriptional Responses: A Pilot Study

Longitudinal Multi-omics Analyses Identify Responses of Megakaryocytes, Erythroid Cells, and Plasmablasts as Hallmarks of Severe COVID-19

Immunosuppression in sepsis: a novel understanding of the disorder and a new therapeutic approach

Immunological characterization of compensatory anti-inflammatory response syndrome in patients with severe sepsis: a longitudinal study*

Reference-based analysis of lung single-cell sequencing reveals a transitional profibrotic macrophage

scCATCH: Automatic Annotation on Cell Types of Clusters from Single-Cell RNA Sequencing Data. iScience

CellMarker: a manually curated resource of cell markers in human and mouse

Mapping the Mouse Cell Atlas by Microwell-Seq

CancerSEA: a cancer single-cell state atlas

MAST: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single-cell RNA sequencing data

clusterProfiler: an R package for comparing biological themes among gene clusters

All rights reserved. No reuse allowed without permission. (which was not certified by peer review) is the author/funder