Background
Traumatic brain injury (TBI) is a debilitating disorder that affects more than 1.5 million individuals annually in the United States [
1]. Obesity is among the most prevalent pre-existing conditions that can negatively impact TBI outcomes. Currently, approximately 13% of the world population is considered clinically obese [body mass index BMI greater than 30 kg/m
2] [
2] placing a large burden on healthcare systems [
3]. Emerging evidence indicates that obese patients have increased TBI complications and higher mortality rates (36%) [
4‐
7].
Activation of brain and peripheral inflammation pathways is a key pathophysiological feature in both TBI [
8‐
12] and obesity [
12‐
20] and has been implicated in the development of associated neurological dysfunction [
8,
21,
22]. Notably, evidence suggests that pre-existing diet-induced obesity is associated with an amplification of post-TBI pro-inflammatory responses and increased microglia-altered states in several brain regions [
23‐
25]. Additional studies report diet-induced exacerbations in TBI-induced cognitive decline [
24,
25] and central (brain) insulin resistance [
26].
Microglia play a key role in mediating the inflammatory responses after TBI. In both clinical and experimental TBI, microglia undergo chronic reactive changes that may contribute to long-term neurodegenerative processes and cognitive decline [
10,
11]. In animal models, pharmacological depletion of microglia or other interventions that reduce chronic microglial activation decrease TBI-induced neuroinflammation and associated neurological deficits [
8,
27,
28]. Microglia also mediate hypothalamic inflammation and neuronal stress responses in diet-induced obesity [
19]. Selective pharmacological inhibition of microglial phagocytosis limits diet-induced obesity hyperphagia and weight gain, decreases dendritic spine loss, and reduces cognitive dysfunction [
21].
Obesity triggers peripheral inflammation that includes reactive changes in adipose tissue macrophages (ATM), which account for > 50% of the cells in the increased visceral adipose tissue (VAT) [
29]. VAT is an immunogenic tissue and the resulting state of obesity-induced chronic low-grade inflammation and increased secretion of pro-inflammatory factors into the circulation contribute to brain neuroinflammatory responses [
29,
30]. Obesity is also associated with increased cognitive decline and dementia [
31‐
33]. Inflammation may play a role in driving the negative pathological consequences of diet-induced obesity, including behavioral deficits [
34‐
37]. NLR family, pyrin domain-containing 3 (NLRP3) contributes to obesity-induced inflammation [
36,
38] through NLRP3-induced activation of microglial IL-1 receptor 1 (IL-1R1) [
36]. Targeted knockout of NLRP3 at the level of the VAT attenuates diet-induced obesity cognitive deficits, thus implicating a potential role for NLRP3/IL-1β signaling in brain-visceral adipose tissue interactions related to obesity [
36]. Furthermore, the depletion of myeloid NOX-2, decreases VAT inflammation, VAT macrophage infiltration, and cognitive dysfunction [
37].
HFD-induced obesity may promote the development of VAT macrophage reactive states that prime the development of deleterious brain microglia phenotypes, increasing vulnerability to secondary insults including TBI. Thus, the amplification of chronic microglial maladaptive states in combined TBI-HFD may contribute to the observed exacerbation of cognitive dysfunction [
13,
23]. Although the individual effects of TBI on the brain and of diet on the adipose tissue and brain have been well studied, the present study examines several important issues that have not been addressed. These include: the effect of isolated TBI on the adipose tissue inflammatory environment; and how interactive effects of diet-induced obesity and TBI affect transcriptomic changes associated with microglia/macrophage states centrally and systemically to impact neurological function.
Materials and methods
Animals
Studies were performed using adult male C57Bl/6 J mice (8–10 weeks old; Jackson Laboratories). Mice were housed in shoebox cages (5 mice in each cage) at least 1 week prior to any procedures in a room (22–23 °C) with a 12-h/12-h light–dark cycle. Food and water were provided ad libitum. The mice were handled briefly before use. Procedures were conducted from 10:00 to 17:00 in a quiet room. All protocols involving the use of animals complied with the Guide for the Care and Use of Laboratory Animals published by the National Institutes of Health (NIH) (DHEW publication NIH 85-23-2985) and were approved by the University of Maryland Animal Use Committee.
Experimental design
To induce our model of DIO, C57Bl/6 mice (n = 8–12/group) were placed on either a high-fat diet (HFD) (60 kcal% Fat; D12492, Research Diets) or a standard diet (SD) (10 kcal% Fat; D12450B, Research Diets) for a period of 12 weeks prior to exposure to controlled cortical impact (CCI) or Sham surgery. Group sizes were determined based on 25 ± 10% (TBI) and 50 ± 10% (diet), anticipated differences between groups on functional recovery and metabolic parameters, respectively, with a α = 0.05, statistical power of 80%. Standard deviation used in sample size calculation was obtained from previous/pilot experiments within our laboratory.
Cohort 1: Following exposure to CCI/Sham surgery, mice were anesthetized (100 mg/kg sodium pentobarbital, I.P.) and transcardially perfused with ice-cold 0.9% saline (100 ml) at 28 days post-injury (dpi). VAT, brain tissue, and blood samples were collected and stored at − 80 °C until processed for RNA/protein analysis.
Cohort 2: To assess more chronic effects of diet and/or TBI, on neurobehavioral, neuroinflammatory and neurodegenerative outcomes, a second cohort of mice was fed a SD or HFD diet for 12 weeks prior to induction of CCI/Sham surgery. Mice underwent a battery of neurobehavioral tasks to assess cognitive behaviors throughout 90 dpi. Mice were anesthetized (100 mg/kg sodium pentobarbital, I.P.) and transcardially perfused with ice-cold 0.9% saline (100 ml) at 90 dpi. Cluster of differentiation 11b (CD11b) positively selected cells were isolated from hippocampal and cortical tissue using Miltenyi MACS Cell separation. RNA was extracted from both the CD11b positively selected cells (primarily microglia but also resident/infiltrating macrophages) as well as flowthrough (the remaining brain cells, including neurons, astrocytes, and oligodendrocytes) and analyzed using a Nanostring Glial Panel. RNA was also extracted from the VAT and analyzed using a Nanostring Neuroinflammation panel.
Controlled cortical impact
Our custom-designed controlled cortical impact (CCI) injury device consists of a microprocessor-controlled pneumatic impactor with a 3.5 mm diameter tip. Mice were anesthetized with isoflurane evaporated in a gas mixture containing 70% N
2O and 30% O
2 and administered through a nose mask (induction at 4% and maintenance at 2%). The depth of anesthesia was assessed by monitoring respiration rate and pedal withdrawal reflexes. Mice were placed on a heated pad, and core body temperature was maintained at 37 °C. The head was mounted on a stereotaxic frame, and the surgical site was clipped and cleaned with Nolvasan and ethanol scrubs. A 10-mm midline incision was made over the skull, the skin and fascia were reflected, and a 5-mm craniotomy was made on the central aspect of the left parietal bone. The impounder tip of the injury device was then extended to its full stroke distance (44 mm), positioned to the surface of the exposed dura, and reset to impact the cortical surface. Moderate-level CCI (n = 12) was induced using an impactor velocity of 6 m/s and deformation depth of 2 mm as previously described [
8,
39‐
42]. After injury, the incision was closed with interrupted 6–0 silk sutures, anesthesia was terminated, and the animal was placed into a heated cage to maintain normal core temperature for 45 min post-injury. Sham animals underwent the same procedure as TBI mice except for the craniotomy and impact. All animals were monitored daily post-injury.
Neurobehavioral testing
Y-maze spontaneous alternation
The Y-maze was carried out at 70 days dpi to access spatial working memory and was essentially performed as previously described [
8,
41]. Briefly, the Y-maze (Stoelting Co., Wood Dale, IL) consisted of three identical arms, each arm 35 cm long, 5 cm wide, and 10 cm high, at an angle of 120° with respect to the other arms. One arm was randomly selected as the “start” arm, and the mouse was placed within and allowed to explore the maze freely for 5 min. Arm entries (arms A–C) were recorded by analyzing mouse activity using ANY-maze software (Stoelting Co., Wood Dale, IL). An arm entry was attributed when all four paws of the mouse entered the arm, and an alternation was designated when the mouse entered three different arms consecutively. The percentage of alternation was calculated as follows: total alternations × 100/ (total arm entries − 2). If a mouse scored > 50% alternations (the chance level for choosing the unfamiliar arm), this was indicative of proper spatial working memory.
Novel object recognition task
The Novel object recognition (NOR) was carried out to assess non-spatial hippocampal-mediated memory as previously described with slight modifications, on 77–78 dpi [
8]. Mice were placed in the NOR chamber where two identical objects were placed near the left and right corners of the open field for training (familiar phase) and allowed to freely explore until they spent a total of 20 s exploring the identical objects (exploration was recorded when the front paws or nose contacted the object). After 24 h, object recognition was tested by substituting a novel object for a familiar training object (the novel object location was counterbalanced across mice). The time spent with each object was recorded using the computed Any-Maze automated software. Because mice inherently prefer to explore novel objects, a preference for the novel object (more time than chance [15 s] spent with the novel object) indicates intact memory for the familiar object. Mice that failed to explore and remained immobile throughout the NOR test were excluded from the analysis.
Morris water maze (MWM)
Spatial learning and memory were assessed using the Morris Water Maze (MWM) task carried out at 80–84 dpi, as previously described [
8]. The MWM protocol included two phases: (1) standard hidden platform training (acquisition) and (2) the twenty-four-hour probe test. Briefly, a circular tank (100 cm in diameter) was filled with water (23 ± 2 °C) and was surrounded by various extra-maze cues on the wall of the testing area. A transparent platform (10 cm in diameter) was submerged 0.5 cm below the surface of the water. Starting at 80 dpi, the mice were trained to find the hidden submerged platform located in the northeast (NE) quadrant of the tank for 4 consecutive days (80-83dpi). The mice underwent four trials per day, starting from a randomly selected release point (east, south, west, and north). Each mouse was allowed a maximum of 90 s to find the hidden submerged platform. The latency to the platform was recorded by the Any-Maze automated video tracking system. Reference memory was assessed by a probe test carried out at 24 h following the last acquisition day, on 84 dpi. The platform was removed, and the mice were released from the southwest (SW) position, and the time in the target quadrant was recorded. Additionally, search strategy analysis was performed as previously described [
8]. Mice that failed to swim and remained immobile throughout the 90-s trial in the MWM test were excluded from the analysis.
Isolation of CD11b-positive cells
A magnetic bead-conjugated anti-cluster of differentiation 11b (CD11b) was used to isolate microglia/macrophages from ipsilateral (injured hemisphere) cortical and hippocampal tissue using Miltenyi MACS Separation Technology (Miltenyi Biotec, Auburn, CA) as per manufacturer’s instructions. Briefly, ipsilateral perilesional cortex and hippocampus from Sham and CCI mice were rapidly microdissected and a single cell suspension was prepared from the combined tissues (pooled tissue from two mice) using enzymatic digestion (Neural Tissue Dissociation Kit; Miltenyi Biotec) in combination with a gentleMACS Dissociator. Myelin was removed using Debris Remval Solution step (Miltenyi Biotec) and cells were incubated with anti-CD11b MicroBeads (Miltenyi Biotec) and loaded onto an LS column (Miltenyi Biotec) placed in the magnetic field of a MACS separator. The negative fraction (flow-through) was collected, and the column was washed 3 times with MACS buffer (Miltenyi Biotech). CD11b-positive cells were eluted by removing the magnetic field, resulting in the isolation of viable CD11b-positve cells (microglia/macrophages) from Sham and CCI mice. Cells were snap-frozen on liquid N2 for RNA extraction performed using Direct-zol RNA MicroPrep kit (Zymo Research).
Real-time PCR
Quantitative gene expression analysis in the adipose tissue as well as ipsilateral hippocampus and peri-lesional cortex of Sham and CCI mice was performed using Taqman technology as previously described [
8,
41]. Real-time PCR for target mRNAs was performed using TaqMan gene expression assays (NADPH oxidase 2 (NOX2), Mm01287743_m1; human neutrophil cytochrome blight chain (p22
phox), Mm00514478_m1; Interleukin-1beta (IL-1β), Mm01336189_m1; Tumor necrosis factor-alpha (TNF-α), Mm00443258_m1; Interleukin-10 (IL-10), Mm01288386_m1; NLR family pyrin domain containing 3 (NLRP3), Mm00840904_m1; glial fibrillary acidic protein (GFAP), Mm01253033_m1; Integrin Subunit Alpha M (ITGAM; CD11b). Mm00434455_m1; and GAPDH, Mm99999915_g1; Applied Biosystems, Carlsbad, CA) on a QUANTSTUDIO 5 Real Time PCR machine (Applied Biosystems). Samples were assayed in duplicate in one run (40 cycles), which was composed of 3 stages, 50 °C for 2 min, 95 °C for 10 s for each cycle (denaturation) and finally the transcription step at 60 °C for 1 min. Gene expression was normalized by GAPDH and compared to the control sample to determine relative expression values by 2
−ΔΔCt method.
Flow cytometry analysis
Immediately following euthanasia, a 30 mg piece of adipose tissue was dissected from each mouse, minced into small pieces, and incubated with 150U/ml Collagenase IV (Worthington Biochemical Corporation Lakewood, NJ) and 10 mg/ml DNase II (Sigma) for 1 h at 37 °C in a rotational shaker. The suspension was passed through a 70 µm filter to mechanically dissociate adipose tissue. For immune cell surface markers, leukocytes were washed with FACS buffer (5% fetal bovine serum in 1 × HBSS) with sodium azide (NaN3) and blocked with 1:50 mouse Fc Block (anti-CD16/32) prior to staining with primary antibody-conjugated fluorophores at 1:50 concentration, including CD45-eF450, and CD11b-APCeF780. All antibodies were commercially purchased from Biolegend. For live/dead discrimination, a fixable viability dye, Zombie Aqua (Biolegend), was diluted at 1:100 in Hank's balanced salt solution (HBSS; Gibco). Cells were briefly fixed in 2% paraformaldehyde (PFA). Data were acquired on a LSRII using FACSDiva 6.0 (BD Biosciences) and analyzed using FlowJo (Treestar Inc.).
To measure reactive oxygen species (ROS) levels, leukocytes were incubated with dihydrorhodamine (DHR)123 (5 mM; 1:500 in RPMI; Ex/Em: 500/536), a cell-permeable fluorogenic probe (Life Technologies/Invitrogen, Waltham, MA). DHR123 passively diffuses into cells and is oxidized by peroxide and peroxynitrite, causing a reaction that produces a green fluorescence that can be measured by flow cytometry. Cells were loaded for 20 min in a 37 °C water bath, washed twice with FACS buffer (without NaAz), and then stained for surface markers including viability dye, and subsequently fixed in PFA.
Leptin, insulin, and monocyte chemoattractant protein (MCP)-1 determination
At either 28 or 90 dpi, serum Monocyte Chemoattractant Protein (MCP)-1, Leptin and Insulin were measured using the Mouse CCL2/JE/MCP-1 DuoSet ELISA (Cat No. DY479), Mouse/Rat Leptin Quantikine ELISA Kit (Cat No. MOB00B, R&D Bio systems, Minneapolis, MN, USA) and the Insulin Mouse ELISA Kit (Cat No. EMINS; Thermo-Fisher, Waltham, MA, USA). The assays were performed according to the manufacturer’s instructions.
Determination of resting blood glucose levels
Prior to euthanasia at 28 dpi, mice were fasted for 12 h. Using the AlphaTRAK Blood Glucose Monitoring System (Zoetis Inc. Parsippany, NJ, USA) resting blood glucose levels were measured via tail snip samples according to the manufacturer’s instructions.
Nanostring gene expression data analysis
The raw Nanostring files were analyzed using Rosalind software to generate the normalized gene expression data (used for all later analyses) and log2 fold changes. The average gene expression across groups as well as the expression Z-scores were calculated in Excel; where indicated, we used Z-score addition to create a composite score (Z-score sum) [
43]. Two-way ANOVA in R was used to study the gene expression variability among groups and identify interaction effects between injury and diet, as well as main effects—genes which are affected only by the injury or only by diet, independent of each other—as previously described [
44]. The resulting p values and log2 fold changes were imported into IPA package to analyze the activation of various molecular pathways including Canonical, Upstream regulators (genes, RNAs, and proteins); abbreviated throughout as Upstream regulators – Genes
, Upstream regulators (drugs and chemicals); abbreviated throughout as Upstream regulators – Drugs
, and Diseases and Bio Functions; abbreviated throughout as Diseases
.
Two-way ANOVA in GraphPad Prism was used to determine the significance of differences between cumulative gene expression Z-scores and the pathway activation Z-scores generated by IPA software. PCA analysis of the Rosalind differentially expressed genes was performed in R using the pcaExplorer software package [
45]. The principal components were derived from the Z-score values of the differentially expressed subset of genes.
Statistical analysis
Blinding was achieved by ensuring that the individual who carried out behavioral and stereological analyses were blinded to injury or diet groups. Quantitative data were expressed as mean ± SEM or mean ± STDEV with individual data points as indicated. Normality testing was performed; as datasets met normality requirements (D'Agostino and Pearson omnibus normality test), parametric analyses were used. Statistical analysis was performed using a two-way ANOVA with Tukey post-hoc tests with post-hoc statistics shown in the representative figures. When comparisons were made between two conditions, an unpaired Student's t test was performed. Statistical analyses utilized Prism v8 for Windows (GraphPad Software) or in R. Significance level was set at p < 0.05.
The consistent application of two-way ANOVA model permits rigorous determination of the effects of each of the two factors (diet and injury), as indicated by the separate factor significance as well as the direction, and magnitude of their combined effects as evidenced by the interaction and
post-hoc test significance [
46]. A non-significant interaction denotes a combined effect that is simply additive whereas, in a significant interaction, the effects of one of the factors of change depend on the condition of the other factor, and the combined effect reflects either a super additive amplification (synergy) or the opposite. Only significant comparisons evident in
post-hoc tests are indicated in respective figures.
We used two-way ANOVA significant post-hoc p values to select genes for IPA analysis rather that t-test p-value (or multiple comparisons-adjusted p values). Although this approach is more stringent and may exclude some biologically significant changes, it is also more appropriate considering the multiple groups evaluated. Furthermore, the application of adjusted p values is not appropriate here in which we focus on Nanostring panels that include genes selected based on their contribution to similar pathways (e.g., inflammation) and may therefore be modulated in a similar fashion. The analysis of IPA data employed repeated two-way ANOVA to account for changes in specific pathways across groups. To perform the analyses, we introduced the following conventions: a baseline Sham-SD vs Sham-SD comparison with a z-score “0” across all pathways and replaced all IPA-generated N/A classifications (not able to detect activation level) by a score of “0”.
The absence of additive effects can also result in a significant interaction factor and may reflect a “ceiling effect” in which the relatively severe TBI model used in our studies prevents detection of the additive impairments after combined TBI and HFD insults, as was observed for cognitive deficits and lesion volume (an indicator of post-traumatic neurodegeneration, data not shown).
Discussion
The present study examined the individual and combined effects of HFD and TBI on VAT, systemic circulation, and brain, with a specific focus on cellular transcriptomic programs and cognitive changes. At 28 dpi (cohort 1), HFD is the major driver of time-dependent changes in body weight, metabolic markers, VAT weight, and VAT immune cell activation. Although HFD alone was not associated with a significant increase in ATM/myeloid cell numbers, HFD in the presence of TBI modified the VAT immune environment with significantly increased numbers of CD45+ and CD11b+ ATM/myeloid cells at 28 dpi; effects of which were associated with increased ATM/myeloid cell phagocytic activity. Furthermore, HFD increased VAT expression of the pro-inflammatory mediators TNF-a, NLRP3, NOX-2, and p22phox as well as the anti-inflammatory mediator IL-10. Notably, HFD in the presence of TBI showed significant diet*injury interactions with exacerbation of HFD-dependent increases in VAT pro-inflammatory molecules IL-1β and NLRP3, indicating a brain trauma-dependent amplification of diet-induced adipose tissue inflammation. Systemically, HFD resulted in evident metabolic dysfunction including increased leptin, insulin, and resting glucose levels; effects of which were independent of TBI, except for evident TBI-induced increases in circulating leptin. Notably, circulating levels of the inflammatory mediator, MCP-1, was significantly increased in the presence of -TBI-HFD. At the level of the brain, although TBI was shown to be the primary driver of microglial inflammatory changes; HFD exacerbated, at least in part, TBI-induced deficits in cognitive function. Overall, this study demonstrates a potential bi-directional neuroimmune relationship between the adipose tissue and brain in the presence of co-morbid diet-induced obesity and TBI; that may be targetable.
ATM are a diverse population of cells with distinct developmental lineages including yolk-sac derived and self-maintaining tissue resident macrophages as well as macrophages resulting from recruitment of bone-marrow derived monocytes, a process that is accelerated in the adipose tissue of obese animals [
51]. Monocytes infiltrate the adipose tissue [
52,
53] and differentiate in macrophages beginning at 8 weeks following initiation of diet-induced obesity [
54,
55] and these bone-marrow (myeloid) derived monocytes/macrophages may be primarily involved in adipose tissue inflammation [
51]. Infiltrating ATM secrete pro-inflammatory mediators including TNF-α, IL-1β, MCP-1, and IL-6 which are suggested to play an important role in driving a pro-inflammatory state within the VAT microenvironment and subsequent low grade chronic inflammation in the systemic circulation [
56‐
59]. Other myeloid immune cells, namely neutrophils, have been shown to transiently infiltrate the adipose tissue as early as 3 days following initiation of HFD [
60] and are identified as key drivers of obesity-induced increases in IL-1β [
61]. In the present study, we demonstrate that HFD has a significant factor effect to elevate CD45
+ and CD11b
+ cell numbers in the VAT microenvironment and that HFD in the presence of TBI, results in a significant increase in the numbers of these cell populations at 28 dpi; these effects are associated with increased phagocytic activity. While we cannot distinguish the specific functions of the various lineages of the CD45
+/CD11b
+ cells including the contribution of yolk-sac vs. myeloid infiltrating ATM/monocytes or that of other infiltrating myeloid cells such as neutrophils, our data suggest that overall, the ATM/myeloid populations contribute to driving the reported inflammatory changes in the adipose tissue microenvironment.
To further explore the VAT immune cell microenvironment, we performed nanostring analysis on the VAT at 90 dpi. The data show an ordered clustering of differential gene expression, which establishes diet as the primary factor driving changes. Disease associated macrophages (DAM) molecules, as well as markers of various pro-inflammatory states, likely expressed in ATM, are among the most numerous HFD/TBI upregulated genes, whereas other immune and epigenetic regulators are downregulated. Importantly, the VAT shows a significant effect of both TBI and HFD, and/or the presence of significant interactions driving super-additive pro-inflammatory effects. The TBI/HFD-dependent elevation was evident not only for the inflammatory states but also for the homeostatic state, suggesting a complex response in which HFD increases some macrophage sub-populations, whereas the combined TBI/HFD increases all macrophage sub-populations, typically at higher levels than HFD alone. IPA pathway analysis extends these findings, identifying the inflammatory pathways as the largest component of upregulated pathways, with both diet and injury as significant factors (although diet remains the dominant factor), and a significant diet-injury interaction driving a synergistic increase in the pathway activation levels after combined TBI-HFD compared to Sham-HFD.
A recent report provides evidence that ATM accumulate lipids in obese mice and become polarized toward the lipid-associated macrophages (LAM) state [
62]. LAM development is driven by Trem2 and characterized by upregulation of components of an enzymatic machinery that recognizes, scavenges, and catabolizes lipids such as fatty acids transporter Cd36, fatty acid binding proteins 5 (Fabp5) and lipoprotein lipase (Lpl). Notably, these and other molecules defining LAM establish a profile that is virtually identical with the conserved signature for DAM [
63]. The LAM/DAM signature in VAT samples is elevated by HFD and appears to be a Trem2-dependent uniform response and responsible, at least initially, for the clearance of injured cells and damaged tissue, resolution of inflammation, and improvement of metabolic changes in obesity [
64,
65]. However, ATM also transform to pro-inflammatory phenotypes, including pro-inflammatory associated genes, changes that may underlie the disease-driving role played by ATM in metabolic conditions and obesity-associated inflammation [
66].
Circulating levels of MCP-1 are increased in animal models of diet-induced obesity [
67,
68] where it appears to play a role in HFD-induced insulin resistance, glucose intolerance [
52,
69] and ATM infiltration [
52,
69,
70]. In the present study, we show that whereas HFD alone does not significantly increase serum levels of MCP-1, HFD in the presence of TBI causes a significant increase in serum levels of MCP-1 at 90 dpi. This observation may have significant implications, given that modulating circulating MCP-1 signaling reduces microglial activation and improves neurological recovery in hepatic encephalopathy [
71]. Furthermore, limiting MCP-1 signaling suppresses LPS-induced seizures [
72]. The adipokine leptin is another factor reported to play a key role in the communication between peripheral immune responses and the and brain [
73,
74]. Notably, both HFD and TBI, albeit to a lesser extent, increase circulating levels of leptin at 28 dpi, suggesting a potential role for HFD-induced increases in circulating leptin in driving brain changes.
Diet-induced obesity is associated with brain neuroinflammation [
16], including in the hypothalamus [
75,
76]. There are conflicting reports regarding the effect of HFD and obesity on extra-hypothalamic inflammation [
21,
37,
77‐
79], with some studies reporting no change [
78,
79] and others identifying neuroinflammation in the hippocampus and cortex [
37,
77]. Isolated cortical microglia from mice fed on a HFD regime release more TNF-α compared to SD-fed counterparts [
80]. Diet-induced obesity, unlike short-term HFD feeding, leads to altered hippocampal microglial states, reductions in dendritic spines at sites of excitatory synapses and memory impairments; the cognitive deficits were reversed by partial knockdown of microglia and associated reduction in microglial phagocytosis [
21]. In TBI patients and experimental models, microglia undergo a chronic transformation to reactive phenotypes that are associated with neurodegenerative processes and cognitive decline [
10,
11]. Delayed transient removal of microglia after experimental TBI reduced the inflammatory lesion microenvironment and improved long-term neurological recovery [
8]. Although previous studies have reported increased microglial altered states in the hypothalamus and the prefrontal cortex in the context of co-morbid diet-induced obesity and TBI [
23,
26], the present study is the first to investigate hippocampal and cortical inflammatory responses at 28dpi (
cohort 1). Unlike TBI, HFD significantly altered only two pro-inflammatory genes (
p22phox and
CD11b) in the hippocampus. However, combined TBI-HFD displayed a significant diet*injury interaction, including an elevation in hippocampal
p22phox compared to TBI-SD mice. These results suggest that HFD can prime development of TBI-induced, pro-inflammatory responses.
To better characterize the interaction between HFD and TBI on secondary posttraumatic injury, we examined cognitive (memory/learning) functions as well as large-scale transcriptomic changes in specific cellular compartments after chronic brain trauma as late as 90 dpi (cohort 2). Y-Maze and NOR tests showed significant effects for both TBI and HFD, although no significant diet*injury interaction was observed, suggesting a simple additive association. In contrast, HFD and TBI had significant factor effects in MWM but not an additive one. However, combined TBI-HFD did show additive deficits in the more sensitive search strategy measurement in the MWM test.
Transcriptome signatures have been proposed to define the complex and dynamic microglial states [
81,
82]. To examine the impact of diet and injury on the molecular pathways potentially related to neurodegeneration and neurological dysfunction, we isolated and analyzed transcriptomic profiles of microglial populations from the perilesional cortex and hippocampus. The heatmap of microglia gene differential expression at 90 dpi displayed a consistent and ordered clustering of the experimental groups: Sham-SD, Sham-HFD, TBI-SD and TBI-HFD. The fact that Sham-HFD diet is consistently distinct from Sham-SD suggests that, at least at a global level, HFD influences microglia responses. The secondary grouping of Sham-SD with Sham-HFD and TBI-SD with TBI-HFD demonstrate that TBI is the dominant factor responsible for change, with HFD having a secondary, albeit persistent, impact. Major components of differential expression in response to TBI and/or HFD include downregulation of homeostatic microglia that occur in response to pathological conditions (injury and/or disease) [
81,
83] and may serve an adaptive role when acting to promote repair, but may also represent a maladaptive response when associated with dysfunctional proinflammatory states that perpetuate tissue damage [
83]. The transcriptomic signature of the DAM and related microglia neurodegenerative phenotype (MGnD) state represents one of the most consistent differentially expressed profiles. However, as with other microglia pro-inflammatory signatures it reflected TBI, but not HFD, as a significant factor. The collective magnitude for overexpression of virtually all DAM/MGnD markers- including ApoE, Spp1, Trem2, Axl, Clec7a and Lpl [
81,
83]- is paralleled by opposite directional changes for homeostatic markers, indicating a strong shift of microglial populations toward the DAM reactive phenotype after TBI/HFD. Downregulation of the homeostatic signature is one of the most common microglial changes shared across various neurodegenerative conditions [
81]. The DAM/MGnD state, driven by the Trem2-APOE signaling pathway [
84], is found in many neurodegenerative disorders, but its function in this regard remains to be elucidated [
83].
Trem2, which along with
APOE is elevated in microglia after TBI in our study, plays a key role in microglial responses to pathological signals such as amyloid accumulation, promoting their exit from the homeostatic state and shift to the DAM reactive state [
85], especially DAM stage 2 [
83]. Trem2 may be commonly required for microglia to switch from the homeostatic state to all reactive states [
83]. Conversely, Trem2 absence impairs phagocytosis of key substrates, including APOE, resulting in an impaired response to beta-amyloid; thus, in certain circumstances the shift of Trem2-dependent microglia to the DAM state may reflect an adaptive response [
85]. In contrast to the differential expression approach, which identified only TBI as a significant factor, IPA analysis using a pathway-centric approach detected a significant effect for TBI and HFD, as well as the significant interaction underlying the super-additive activation of multiple microglial reactive inflammatory pathways in TBI-HFD compared to TBI-SD.
Overall, our data indicate that in the more chronic phase (90 dpi), microglia remain strongly shifted toward heterogenous reactive states/responses primarily driven by TBI, but with HFD having an effect. Thus microglia, due to inability to promote effective repair, especially in the presence of an additional insult (HFD), enter persistent altered states that are maladaptive and may propagate secondary injury.
The gene expression changes in the non-microglia populations also show TBI-dependent down-regulation of neuronal responses and neurogenesis, as well as upregulation of astrocyte and oligodendrocyte signatures, that are consistent with neurodegeneration [
86,
87]. Notably, HFD is also able to decrease neuronal signature. Pro-inflammatory changes were also found in non-microglia populations, suggesting that microglia are not the only cell population involved in post-traumatic inflammatory responses. Overall, principal component analysis (PCA) illustrates the predominant importance of injury for brain changes and diet for VAT changes, but also shows clear separation of responses across the various experimental groups when a composite score is generated that includes both brain and VAT changes.
Evidence demonstrates that pre-existing obesity is associated with worsened outcomes following TBI [
23‐
26]. A prior study implicated a pivotal role for VAT NLRP3-induced signaling in driving obesity-induced cognitive deficits; these effects reflected IL-1-mediated microglial activation [
36]. Furthermore, an important role for the IL-1/IL-1R1 signaling axis in driving TBI-induced cognitive deficits has been reported [
88,
89]. Thus, supporting the notion that targeting the NLRP3/IL-1 signaling axis may be a feasible target in the context of co-morbid TBI-HFD. Notably, the present study demonstrates an interaction effect of TBI-HFD on VAT expression of both NLRP3 and IL-1β mRNA expression and a combined HFD-TBI effect on levels of systemic inflammatory mediators. These findings support the conclusion that HFD-TBI induces a reinforcing loop in which pro-inflammatory molecules released by VAT into the circulation promote a ‘primed’ state within the brain, wherein a subsequent insult (i.e. TBI) serves to exacerbate neuroinflammatory responses and related cognitive decline.
However, there are important limitations in the present studies that should be recognized:
1.
HFD was initiated prior to TBI and continued after injury. Thus, it is not possible to distinguish between the pre- and post-injury effects of diet. Furthermore, HFD and resultant obesity are not equivalent because HFD is maintained throughout the experiment, even after obesity was established. Therefore, we cannot distinguish the effects of continuous HFD on the VAT (and indirectly on the brain) versus more direct actions on the brain.
2.
The Nanostring transcriptomic analysis, although including more than 750 genes, is less comprehensive than RNAseq. More importantly, the panels used are highly enriched in inflammatory molecules, and as such it is not possible to apply gene set enrichment analysis, which restricted our IPA work to pathway activation score analysis.
3.
Sampling limitations include the probable presence of brain-resident macrophages and infiltrating monocytes as a smaller component in the CD11b positively selected fraction, the inclusion of combined injured hippocampus and cortex regions for cell isolation, and the absence of cell-specific isolation (macrophages) in the VAT. In addition, the presence of heterogeneous cellular populations in samples such as brain flowthrough (neuronal, astrocytic, oligodendrocytes, etc.) and adipose tissue (adipocytes, macrophages, etc.) complicate the analysis and may represent an important confounding factor when interpreting reported molecular/cellular changes in this data set. Future studies should also examine specific cell populations in the adipose tissue, including adipocytes, ATM, and neutrophils.
4.
This study focused on high throughput transcriptomic analysis at level of the adipose tissue and brain. Detailed proteomic analysis was not performed; therefore, the biological relevance of identified gene expression changes remain speculative. Future studies should include protein changes related to key identified pathways in both the brain and adipose tissue, effects of modulating such pathways, and elucidating more detailed region-specific brain changes.
In summary, our findings further support the concept of bi-directional communications between the adipose tissue and brain, which may contribute to brain injury and be potentially targetable. In-depth transcriptomic analysis at both the tissue and cellular level revealed that HFD enhances TBI-induced brain pro-inflammatory responses and that TBI augments HFD-induced VAT inflammatory pathway changes. We hypothesize that this interaction creates a self-propagating secondary inflammatory injury loop with an exacerbated microglial response, which contributes to chronic posttraumatic disability including cognitive decline.