To the Editor:
Sepsis commonly leads to both acute and chronic brain dysfunction (1). Numerous mechanisms likely underlie brain dysfunction during and after sepsis, including oxidative stress, microvascular and blood–brain barrier dysfunction, neurotransmitter imbalance, and neuroinflammation (2). Our understanding of these mechanisms is based on rodent studies, but there are important intrinsic differences between humans and animal models in both the response to sepsis and brain gene expression (3). Furthermore, it is likely that the human brain, supported by modern critical care, encounters significantly greater physiologic and metabolic insults during the course of critical illness than is accounted for in animal models.
Postmortem studies of the human brain have provided valuable insights into the neuropathology of sepsis, including ischemia, hemorrhage, neuroaxonal injury, and innate immune activation (4). Microglia, the innate immune cells of the brain, are activated in sepsis, and markers of inflammation are increased in microglia, astrocytes, and endothelial cells (4). Although these studies have highlighted the activation of classical inflammatory mechanisms using immunohistochemical methods, only a single study has undertaken a molecular analysis of signaling in the human brain during systemic infection (5). Moreover, focused analyses cannot reveal whether innate immune activation is the predominant neuropathologic process during sepsis, or whether brain-specific pathways are also highly differentially regulated.
An unbiased postmortem analysis of sepsis-related brain expression in patients requires high-quality frozen brain tissue from patients with a well-characterized cause of death and a spectrum of chronic neuropathology. We identified 89 subjects in the autopsy cohort of the ACT (Adult Changes in Thought) study who died while hospitalized (6). We determined cause of death by reviewing hospital records of the subjects’ terminal hospitalization using a structured instrument, with attention to evidence of infection (7). Patients with an acute structural brain injury at the time of death or with an indeterminate cause of death due to multifactorial critical illness were excluded.
We identified 12 subjects who died of sepsis and 12 who died of a noninfectious critical illness. RNA was isolated from the parietal cortex gray matter. Age, underlying neuropathology as measured by Braak and Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) scores, dementia diagnosis, and RNA integrity were balanced between subjects with sepsis and control subjects without sepsis (Table 1). The University of Michigan Advanced Genomics Core prepared complementary DNA libraries (pico-input rRNA depletion complementary DNA kit; Takara Bio) and sequenced libraries using the Illumina Hi-Seq platform. Fragments per kilobase of transcript per million mapped reads (FPKM) values were normalized using the default setting, “classic fpkm,” within CuffDiff. Details regarding alignment, quality control, and the full differential expression analysis are available online (8), and individual-level gene count data are available at the National Center for Biotechnology Information Gene Expression Omnibus (accession number GSE135838). Some of the results of this study have been previously reported in abstract form (9).
Table 1.
Descriptive Characteristics of 24 Subjects and Samples Included in RNA Sequencing
| Measure | Sepsis | Nonsepsis |
|---|---|---|
| Age, yr, mean ± SD | 91.2 ± 5.6 | 88.7 ± 9.4 |
| Male/female, n | 4/8 | 8/4 |
| RIN, mean ± SD | 5.62 | 6.19 |
| Dementia diagnosis, n | 3 | 3 |
| Braak score, median (Q1–Q3) | 3.6 (3–4.3) | 3.8 (3–5) |
| CERAD, median (Q1–Q3) | 1 (0.8–1) | 1.5 (1–2) |
| Length of stay, d, median (Q1–Q3) | 3 (2–4) | 3 (2–4) |
| Source of sepsis, n | ||
| Pulmonary | 7 | — |
| Urinary | 2 | — |
| Abdominal | 2 | — |
| Unknown | 1 | — |
| Cause of death, n | ||
| Cardiac | — | 5 |
| Hemorrhage | — | 4 |
| Respiratory failure | — | 1 |
| Comfort care due to pain | — | 2 |
Definition of abbreviations: CERAD = Consortium to Establish a Registry for Alzheimer’s Disease score; Q1 = first quartile; Q3 = third quartile; RIN = RNA integrity number.
A total of 176 genes were considered significant or differentially expressed when the fold change was >1.5 and the Benjamini-Hochberg false discovery rate was <0.05. Notably, the most differentially expressed genes were immune related, including damage-associated molecular patterns (DAMPs) (S100A8, S100A9, and members of the HSP family), markers of astrocyte activation (HSPB2, GBP2, and SERPINA3), and macrophage and microglial markers (SOC3, CHI3L2, and CHI3L1).
Prior studies of brain gene expression that did not take cause of death into account found that RNA integrity is significantly related to gene expression patterns (6). Likewise, we hypothesize that underlying neuropathology, such as amyloid plaques and neurofibrillary tangles, may also drive gene expression (5). Given the difficulty of obtaining suitable, well-annotated brain specimens from patients with sepsis, our modest sample was not suitable for a multivariate analysis of differential gene expression. We therefore performed dimensional reduction with weighted gene coexpression network analysis (WGCNA), which is robust for small sample sizes, reduces the burden of multiple hypothesis testing, and identifies clusters or networks of genes that are potentially dysregulated (10).
WGCNA generated 35 modules of covarying genes. The association of module expression with sepsis, age, sex, RNA integrity, Braak score for neurofibrillary tangles, CERAD score for amyloid plaques, and dementia diagnosis was tested in multiple univariate analyses. The modules, their associated gene lists, and association with covariates are available online (8). Six modules were significantly correlated (P ≤ 0.05) with sepsis (Table 2). Although three modules were correlated with Braak scores and seven were correlated with dementia, these modules were distinct from those associated with sepsis. Age was significantly correlated with six modules, only one of which was also associated with sepsis.
Table 2.
Summary of Gene Modules That Are Significantly Related to Sepsis in Weighted Gene Coexpression Network Analysis
| Module | Genes (n) | Hub Genes | Ten Most Differentially Expressed Genes | Variables (ρ*,P Value) | Enriched Terms | |
|---|---|---|---|---|---|---|
| grey60 | 225 | C1R, CNN3, CSDA, ICAM1, IFITM2, PXDC1, SHC1, STAT3, TMBIM1, and TNFRSF1A | BCL3, IER3, S100A8, SHC1, IL15RA, TNFRSF1A, ANGPTL4, PNP, S100A9, and IL6 | Sepsis | (0.51, 0.01) | Cytokine-mediated signaling†, inflammatory response†, response to external stimulus†, and cell death† |
| RIN | (0.27, 0.2) | |||||
| Braak | (0.0057, 0.8) | |||||
| CERAD | (0.13, 0.6) | |||||
| Age | (0.43, 0.04) | |||||
| midnightblue | 259 | BAI1, CACNA1I, DUSP8, GLIS1, ITSN1, MLLT1, NEURL, PALM, PITPNM2, and PLEKHA6 | PRB1, YAP1, RHOF, INPP5E, TRIM7, TRAF7, ZC3HAV1, INTU, ASIC4, and PHLDA1 | Sepsis | (0.41, 0.05) | Synapse‡, actin cytoskeleton process‡, dendritic tree§, voltage-gated calcium channel§, and glycerophospholipid synthetic process§ |
| RIN | (0.54, 0.008) | |||||
| Braak | (0.068, 0.8) | |||||
| CERAD | (0.099, 0.7) | |||||
| Age | (0.063, 0.8) | |||||
| darkorange | 63 | AFF1, ASB15, EVC2, HMGB2, IL16, INPP5E, LRP2BP, SERP1, UTS2R, and WFIKKN1 | FOSL1, LOC643733, LOC649330, GJA4, WNT8B, FAM181A-AS1, LRRC43, LOC100505835, HMBG2, and LAG3 | Sepsis | (0.53, 0.009) | DNA binding‖ and regulation of transforming growth factor β signaling‖ |
| RIN | (0.52, 0.01) | |||||
| Braak | (0.051, 0.8) | |||||
| CERAD | (0.26, 0.2) | |||||
| Age | (0.039, 0.9) | |||||
| lightgreen | 213 | ADSS, CDK17, KIAA1244, LANCL1, LPGAT1, MTMR2, PAPSS1, SPTLC1, TBCB, and USP32 | SNX30, MFSD6, ACAT2, HSD17B4, NIPA1, ST3GAL5, PCYT2, SCHIP1, TUBGCP3, and ABHD5 | Sepsis | (0.45, 0.03) | COPI-coated vesicle membrane§, cholesterol biosynthesis‖, and protein glycosylation‖ |
| RIN | (0.66; 5 × 10−7) | |||||
| Braak | (0.069, 0.8) | |||||
| CERAD | (0.12, 0.6) | |||||
| Age | (0.14, 0.5) | |||||
| skyblue | 54 | B4GALNT1, DZIP1, GOLGA7B, GPR162, IQUB, KIAA1522, MAST1, MCF2L2, PSD, and RHOF | FZD4, ATP6V0A4, SRP14, GRHL1, GOLGA7B, LGALS3, ANKRD36BP2, SLC10A1, EPB41L5, and SPRY1 | Sepsis | (0.52, 0.01) | Regulation of adenylate cyclase activity involved in G protein–coupled receptor signaling pathway‖, and regulation of neurofibrillary tangle assembly‖ |
| RIN | (0.32, 0.1) | |||||
| Braak | (0.19, 0.4) | |||||
| CERAD | (0.085, 0.7) | |||||
| Age | (0.078, 0.7) | |||||
| paleturquoise | 49 | AFF1, DTNA, LRRC37B, MCL1, PLIN2, SAT1, SOX9, TGFBR3, UBR5, and ZNF395 | METRNL, GNA14, FNDC3B, GPR4, ZC2HC1C, MCL1, SERP1, IL18R1, SAT1, and SCN9A | Sepsis | (0.46, 0.03) | CD4 receptor binding‡, T-cell cytokine production‖, IL-1 receptor activity‖, and protein ubiquitination‖ |
| RIN | (0.56, 0.005) | |||||
| Braak | (0.025, 0.9) | |||||
| CERAD | (0.081, 0.7) | |||||
| Age | (0.3, 0.2) | |||||
Definition of abbreviations: CERAD = Consortium to Establish a Registry for Alzheimer’s Disease score; RIN = RNA integrity number.
Hub genes are determined by correlation between the expression of an individual gene and the eigengene of each module and reflect connectivity of the gene to expression of the module as a whole. Top differentially expressed genes were determined by cross-referencing module membership with differential expression calculated from the dataset as a whole. Gene enrichment analysis was performed on modules that were significantly correlated with sepsis using the enrichment analysis tool within the weighted gene coexpression network analysis and with the bioinformatics tool DAVID (14).
Absolute value of biweight-mid correlation between the variable of interest and the module eigengene.
P < 1 × 10−10.
P < 1 × 10−5.
P < 1 × 10−4.
P < 0.01.
WGCNA revealed a strong correlation of sepsis with a large module (grey60) that is enriched in genes that are associated with cytokine signaling and immune function and regulation, and are not correlated with RNA integrity. This module contained many of the most highly differentially expressed genes in the dataset, including DAMPs, primary cytokines, and other genes related to innate immunity. Other modules were enriched in pathways that have received less study than innate immunity in prior studies of sepsis-related brain injury, including synaptic function, ion channel function, neuronal growth (midnightblue), and T-cell signaling and regulation (paleturquoise).
Although this analysis is limited by the small number of samples, it is the only such study to date and is strengthened by the direct analysis of hospital records. In addition, the analysis is limited by our use of whole, unperfused brain tissue, which may allow the response of highly activated cell types, such as microglia and endothelial cells, to mask important transcriptional changes in neurons or astrocytes. It is possible that intravascular or infiltrating leukocytes, such as monocytes, contribute to tissue-level gene expression changes. However, these populations likely are not the sole drivers of gene expression. S100A8 and CHI3L1, two highly differentially expressed genes in this analysis, are expressed by astrocytes in sepsis (5, 11). Our analysis revealed unexpected increases in the expression of pathways that support brain function, such as synaptic and neuronal growth. Preclinical studies have demonstrated synaptic injury during sepsis, and mechanistic studies will be needed to differentiate between injurious responses and compensatory processes (12). In addition, the advanced age of the participants, which could alter the brain response to inflammation, may limit the generalizability of these results (13).
These results demonstrate that sepsis is associated with a specific transcriptional response in the human brain. Gene expression in subjects who died of sepsis was compared with that in subjects who died in-hospital of noninfectious critical illnesses with similar lengths of stay, suggesting that the transcriptional response is specific to sepsis and not agonal factors, which would be common to critically ill patients regardless of underlying illness. In this unbiased analysis, we found that innate immune activation, especially expression of primary cytokines, complement factors, and DAMPs, was robustly associated with sepsis in the brain, substantiating the results of prior focused analyses in both animal models and human tissues (5, 11). WGCNA also revealed differential expression of processes related to synaptic function and neuronal growth, although the fold change of genes in these modules was less than that observed for those related to innate immunity. These results provide both confirmation and new targets for mechanistic explorations of sepsis-associated brain injury.
Supplementary Material
Footnotes
Supported by NIH grants T32HL00774921 (A.C.B.), R01HL123515 (T.J.S.), K08NS101054 and UL1TR002240 (B.H.S.), K23AG038352 (W.J.E.), and U01AG006781 (E.B.L. and P.K.C.), and the Nancy and Buster Alvord Endowment (C.D.K.).
Originally Published in Press as DOI: 10.1164/rccm.201909-1713LE on January 15, 2020
Author disclosures are available with the text of this letter at www.atsjournals.org.
References
- 1.Widmann CN, Heneka MT. Long-term cerebral consequences of sepsis. Lancet Neurol. 2014;13:630–636. doi: 10.1016/S1474-4422(14)70017-1. [DOI] [PubMed] [Google Scholar]
- 2.Mazeraud A, Pascal Q, Verdonk F, Heming N, Chrétien F, Sharshar T. Neuroanatomy and physiology of brain dysfunction in sepsis. Clin Chest Med. 2016;37:333–345. doi: 10.1016/j.ccm.2016.01.013. [DOI] [PubMed] [Google Scholar]
- 3.Hawrylycz M, Miller JA, Menon V, Feng D, Dolbeare T, Guillozet-Bongaarts AL, et al. Canonical genetic signatures of the adult human brain. Nat Neurosci. 2015;18:1832–1844. doi: 10.1038/nn.4171. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Sharshar T, Annane D, de la Grandmaison GL, Brouland JP, Hopkinson NS, Françoise G. The neuropathology of septic shock. Brain Pathol. 2004;14:21–33. doi: 10.1111/j.1750-3639.2004.tb00494.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Rakic S, Hung YMA, Smith M, So D, Tayler HM, Varney W, et al. Systemic infection modifies the neuroinflammatory response in late stage Alzheimer’s disease. Acta Neuropathol Commun. 2018;6:88. doi: 10.1186/s40478-018-0592-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Miller JA, Guillozet-Bongaarts A, Gibbons LE, Postupna N, Renz A, Beller AE, et al. Neuropathological and transcriptomic characteristics of the aged brain. Elife. 2017;6:e31126. doi: 10.7554/eLife.31126. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Iwashyna TJ, Odden A, Rohde J, Bonham C, Kuhn L, Malani P, et al. Identifying patients with severe sepsis using administrative claims: patient-level validation of the Angus implementation of the international consensus conference definition of severe sepsis. Med Care. 2014;52:e39–e43. doi: 10.1097/MLR.0b013e318268ac86. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Bustamante A, Opron K, Ehlenbach W, Crane P, Keene D, Standiford T, et al. doi: 10.7302/6fv8-bv53. Transcriptomic profiles of sepsis in the human brain [data set]. University of Michigan - Deep Blue; 2019. Available from: [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Bustamante AC, Opron K, Standiford TJ, Singer BH. Transcriptomic profiles of sepsis in the human brain [abstract] Am J Respir Crit Care Med. 2019;199:A2387. doi: 10.1164/rccm.201909-1713LE. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Langfelder P, Horvath S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics. 2008;9:559. doi: 10.1186/1471-2105-9-559. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Denstaedt SJ, Spencer-Segal JL, Newstead MW, Laborc K, Zhao AP, Hjelmaas A, et al. S100A8/A9 drives neuroinflammatory priming and protects against anxiety-like behavior after sepsis. J Immunol. 2018;200:3188–3200. doi: 10.4049/jimmunol.1700834. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Moraes CA, Santos G, Spohr TCL de SE, D’Avila JC, Lima FRS, Benjamim CF, et al. Activated microglia-induced deficits in excitatory synapses through IL-1β: implications for cognitive impairment in sepsis. Mol Neurobiol. 2015;52:653–663. doi: 10.1007/s12035-014-8868-5. [DOI] [PubMed] [Google Scholar]
- 13.Tanaka N, Cortese GP, Barrientos RM, Maier SF, Patterson SL. Aging and an immune challenge interact to produce prolonged, but not permanent, reductions in hippocampal L-LTP and mBDNF in a rodent model with features of delirium. eNeuro. 2018;5 doi: 10.1523/ENEURO.0009-18.2018. ENEURO.0009-18.2018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Huang DW, Sherman BT, Tan Q, Collins JR, Alvord WG, Roayaei J, et al. The DAVID gene functional classification tool: a novel biological module-centric algorithm to functionally analyze large gene lists. Genome Biol. 2007;8:R183. doi: 10.1186/gb-2007-8-9-r183. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.