Epigenetic impacts of stress priming of the neuroinflammatory response to sarin surrogate in mice: a model of Gulf War illness

A coalition of 34 countries deployed approximately 956,600 troops during the 1990–1991 Gulf War [1], with the majority, ~ 700,000, from the USA. An estimated 25–32% of these veterans fulfil the requirements of a Gulf War illness (GWI) diagnosis [2]. GWI is an archetypal, medically unexplained, chronic condition characterised by persistent sickness behaviour, with neuroimmune and neuroinflammatory components.

Symptoms of GWI include fatigue, musculoskeletal pain, cognitive dysfunction, chemical sensitivities, loss of memory and sleep disruption, which can be characterised as ‘sickness behaviour’ [3, 4]. ‘Sickness behaviour’ is normally a result of inflammatory response to illness or injury, which usually resolves itself over time after the initial insult is removed. Symptoms were reported within 6 months of the conflict [5‐7].

Although the exact cause of GWI is still unknown, there is a consensus that exposure to environmental toxins is the most likely cause [8]. GWI symptoms are highly heterogeneous, and specific symptoms may be related to specific experiences: for example, Gulf War veterans exposed to nerve agents or oil well fires are at increased risk of brain cancer compared to other Gulf War veterans [9].

A leading hypothesis for the cause of GWI is that the high physical and psychological stress of combat interacted with exposure to acetylcholinesterase (AChE) inhibitors [1, 4, 10‐19]. Military personnel were exposed to a number of AChE inhibitors [1, 20], including pyridostigmine bromide (PB), a reversible AChE inhibitor, as a prophylactic against nerve agents; sarin, soman, and related nerve agents, irreversible AChE inhibitors, which combatants were inadvertently exposed to after demolition of Iraqi supply depots, such as at Khamisiyah; organophosphate pesticides, irreversible AChE inhibitors, which were widely used to prevent pest-borne diseases and irritation [20]; permethrin, an insecticide which may inhibit AChE [10, 21]; and DEET, an insect repellent and a weak AChE inhibitor which may enhance the activity of other AChE inhibitors [22]. For example, an estimated 95,000 deployed personnel were exposed to the plume from the Khamisiyah demolition, and approximately 250,000 may have been exposed to low levels of nerve agents during aerial bombardments earlier in the conflict [23]. Further, the number of nerve agent alarms heard is correlated with risk for GWI [24]. Accumulating research has indicated that deleterious health effects of exposures to psychophysiological stress [25, 26] and environmental toxicants [27, 28] involve epigenetic modifications that affect transcriptional regulation.

The overall objective of this study was to examine genome-wide epigenetic transcriptional modifications in the brain using an established mouse model of GWI [4, 12, 15, 16]. Our previous research demonstrates that effects on neuroinflammatory pathways occur shortly after initial exposures. For example, pre-exposure with the stress hormone corticosterone (CORT) causes an increase in expression of specific chemokines and cytokines in response to diisopropyl fluorophosphate (DFP) [4], an irreversible acetylcholinesterase inhibitor [29] used here as a sarin surrogate. This corresponds well with the work in GWI study participants [30‐37], which have shown immunological abnormalities, and a recent paper [38], which has shown specific immune-related biomarkers for GWI veterans. We hypothesized that epigenetic and transcriptomic changes upon initial exposures would identify gene pathways linked to poor health outcomes in GWI.

To our knowledge, this is the first transcriptome- and epigenome-wide study to examine evidence for transcriptional, chromatin and DNA methylation modifications in a model of GWI. A previous study [12] examined 84 microRNAs (MiRNAs) and global but not gene-specific changes in DNA methylation and hydroxymethylation in rats subjected to restraint stress and a protocol of PB, DEET and permethrin to simulate troops’ chemical exposures. This study found differential expression of two miRNAs and in increase in global methylation in the hippocampus. Although this was an important step forward, our study was able to examine the whole transcriptome, a histone modification associated with chromatin accessibility and DNA methylation at a genome-wide, single cytosine level.

Overall, our results represent several interesting findings. First, as expected, there was a large change in the expression of immune-related genes in both the frontal cortex and hippocampus, building upon previous findings in this model [4]. Second, many genes associated with synaptic function are changed in their activity, as shown by our frontal cortex H3K27ac ChIP-seq data and hippocampus RNA-seq. These changes in gene expression seem to be subtler than those found for immune-related genes (lower expression), but differential expression of genes related to synaptic function in mice is associated with impaired memory and cognition, consistent with impairments reported by GWI suffers. Interestingly, long-term potentiation- and depression-related genes are enriched in the ChIP-seq data. Finally, we see evidence of not just a change in gene activity but of a suggested change in cell proportions. This is in line with previous work with CORT [78, 79].

It is possible that microglia are responsible for this large change in immune-related gene expression, but we acknowledge that other cells, such as astrocytes, also express cytokines and chemokines (e.g. Kim et al. [80]). However, this is usually at a much lower level than that in microglia: in genes significantly differentially expressed uniquely with CORT + DFP in both the frontal cortex and hippocampus the six genes with the largest fold change are expressed in microglia (Additional file 26: Table S18). Therefore, although other cell types may be contributing to cytokine and chemokine expression, it is highly unlikely that microglia are not the cell type driving this change. Similarly, we attribute many of the transmembrane transporter-related annotations we see in the frontal cortex ChIP-seq data and hippocampus RNA-seq data to neurons; however, many of these transporters are also expressed in glial cells [54]. For this reason, future work should be carried out to isolate or enrich specific cell populations, allowing these predictions to be tested.

Our findings of altered cholinergic neurotransmitter expression after DFP exposure are in line with previous studies of low dose sarin exposure in rats [81, 82]. Specifically, changes in M1 and M3 acetylcholine receptor expression were seen in the frontal cortex in a dose-dependent manner when the animals were maintained under hyper-thermic conditions (i.e. a stressor; Henderson et al. [81, 82]). In relation to this, we see choline-related annotations at all levels we analysed: our H3K27ac ChIP-seq analyses show changes in cholinergic synapse-related genes and differential binding related to Chrm2, the gene coding for M2; the KEGG annotation, ‘choline metabolism in cancer’, is enriched in our hippocampus RNA-seq genes and our H3K27ac ChIP-seq genes; finally, in our hippocampus RRBS, we see altered methylation of a CpG site within Ache, the gene coding for acetylcholinesterase.

These findings not only link to rat models but also to GWI subjects, where differences in prefrontal cortex working memory have been previously found and have been attributed to the cholinergic system [83]. This again connects with our findings from ChIP-seq where both ‘learning and memory’ and ‘cholinergic synapse’ annotations are significantly enriched. Therefore, pre-exposure with CORT may not only be potentiating the immune response to AchE inhibitors but also causing longer term changes to the cholinergic signalling system in line with previous animal models. Human studies have been more varied, with evidence both for [84, 85] and against [86] long-term changes in the cholinergic system, perhaps related to the heterogeneity of symptoms or to differences in tissues being examined. Recent work by Locker et al. [15] has shown that increased neuroinflammatory response in this model is not directly related to AChE inhibition but instead may result from changes in other aspects of signaling (e.g. epigenetic alterations in gene expression).

In relation to the potential reduction in myelinating oligodendrocytes in the cortex, this may have an effect on some of the phenotypes seen in GWI: reduced oligodendrocytes have been linked to major depressive disorder (MDD), functional consequences in neurons and mood-related symptoms in rats [87]. As this change in cell proportion would affect myelinating cells, it could also contribute to the reported alterations in white matter in GWI veterans [20, 88‐90].

Our RNA-seq data suggests both immune dysfunction and oxidative stress. Previous reports have shown immune dysfunction and oxidative stress in other disorders with similar phenotypes to GWI such as myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) [91]. However, whether the immune response is causing oxidative stress or oxidative stress is causing an immune response is less clear. This is directly relevant, as oxidative stress contributes to the toxicity of AchE inhibitors [29] and has been suggested as a cause of GWI [1, 92].

One of the six genes (Additional file 26: Table S18) we find strongly differentially expressed in both the frontal cortex and hippocampus, Tlr2, has previously been linked to sickness behaviour when expressed in hypothalamic microglia [93] and was reported to show increased expression in a model of GWI [94]. Therefore, it is of great interest in terms of the sickness-like behaviour that defines GWI. TLR activation by immune insult appears to increase Tlr2 expression [95]. We can speculate that the change in Tlr2 expression we see is upstream of the change in Il1b and TNF, as specific activation of TLR2 increases their expression [93, 96]. Further, TLR stimulation increases SLC15A3 expression in dendritic cells, which in turn regulates TNF and IL-1β expression [97]. At the very least, these six, consistently, strongly upregulated genes make up a robust core of immune- and microglia-related genes for further investigation (Additional file 26: Table S18).

Our results correspond very well with those of Broderick et al. [98] in blood samples from GWI study participants, who found changes to NF-KappaB-related genes (which are enriched in our hippocampus RNA-seq data and in the genes significant in both the cortex and hippocampus RNA-seq), and in pathways under the broad theme of neuronal development and migration, which we see in our hippocampus RNA-seq and our cortex ChIP-seq. They also saw ligand–receptor interactions supporting neurotransmission, which again we see strongly in our cortex ChIP-seq data. Further, our KEGG pathway analysis of the transcriptomics data highlighted rheumatoid arthritis-related gene enrichment. Interestingly, a study in veterans with GWI also highlighted the possibility of medication used to treat rheumatoid arthritis also being used to treat GWI [99]. The same study identified the TNF-alpha pathway (which we see in our transcriptomics) and the estrogen pathway (which we saw in our ChIP-seq) as potential drug targets [99]. Therefore, data from our model corresponds to data identified from veterans with GWI and could be used as a model to test these compounds as potential therapeutics.

A recent paper described eight potential blood biomarkers for GWI, the strongest being a 9.27-fold increase in CaMKII protein in the blood of these veterans [100]. In our study, Camk2b was found to be differentially methylated in the hippocampus and have differential H3K27ac in the cortex.

Further linking our study with human data was the finding that plasma CRP levels are increased in the blood of GWI veterans [38]. CRP is often used as a biomarker of IL-6-mediated inflammation [38, 101], and IL-6 related annotations were enriched in the genes differentially expressed in both the cortex and hippocampus (Additional file 15: Table S10). This annotation was due to four genes, Il1b, Tlr2, Il1a and Tnf, four of our six most strongly and consistently differentially expressed genes.

We see minimal overlap in genes with significant differential methylation between the hippocampus and frontal cortex (two genes, Bcar3 and Tmem242). This is likely due, at least partly, to the two factors outlined above: cellular heterozygosity and a short time point after treatment. Of the two consistent genes, Bcar3 is involved in cell proliferation, which when overexpressed in breast cancer conveys estrogen resistance, and Tmem242 is a transmembrane protein with very little current annotation. We cannot detect any coordinated methylation changes (denoted by enriched annotations in our significant genes) in the frontal cortex and very few in the hippocampus. As mentioned above, we detect differential methylation of one CpG site within Ache, the gene coding for acetylcholinesterase. A number of the genes found to be differentially methylated can be linked to GWI. Examples include Lims1, which is known to be regulated by TNF and is involved in cell growth and survival [102]; Sesn1, Aplnr, Pxn and Actn1, which have been linked to ME/CFS [103, 104]; Col5a3, which was identified in a rat model of GWI in the hippocampus [12]; and Slc1a2, which has been linked to ALS, a disorder GW veterans are at increased risk of [105, 106]. This may indicate that networks of genes linked to GWI are only just beginning to be methylated, or that different groups of genes are methylated in different cell types. However, until these coordinated networks are elucidated, investigation of individual genes could lead to spurious associations.

We are able to show a range of changes in the transcriptome of this well-established mouse model, many of which reflect gene expression seen in veterans with GWI. Further, we see alteration in H3K27ac, showing potential chromatin configuration changes, which could lead to epigenetic effects with long-lasting implications. We also find differences in DNA methylation, although these are less easily interpretable than the transcriptome and H3K27ac changes.

Additional research is needed to assess whether effects of these epigenetic and transcriptional modifications on long-term health outcomes are cumulative and/or are potentiated by later exposures (e.g. infection). Notably, however, our findings reveal gene pathways known to be involved in long-term adverse health effects in GWI veterans. These results suggest that epigenetic and transcriptional regulation during the initial exposure period likely contribute to pathological outcomes in GWI. It would be important to examine these modifications in peripheral tissues from GWI veterans to ascertain whether biomarkers could be developed to predict future health outcomes.