Human in vivo neuroimaging to detect reprogramming of the cerebral immune response following repeated systemic inflammation

Despite increasing evidence that immune training within the brain may affect the clinical course of neuropsychiatric diseases, data on cerebral immune tolerance are scarce. This study in healthy volunteers examined the trajectory of the immune response systemically and within the brain following repeated lipopolysaccharide (LPS) challenges. Five young males underwent experimental human endotoxemia (intravenous administration of 2 ng/kg LPS) twice with a 7-day interval. The systemic immune response was assessed by measuring plasma cytokine levels. Four positron emission tomography (PET) examinations, using the translocator protein (TSPO) ligand 18F-DPA-714, were performed in each participant, to assess brain immune cell activation prior to and 5 hours after both LPS challenges. The first LPS challenge caused a profound systemic inflammatory response and resulted in a 53% [95%CI 36-71%] increase in global cerebral 18F-DPA-714 binding (p<0.0001). Six days after the first challenge, 18F-DPA-714 binding had returned to baseline levels (p=0.399). While the second LPS challenge resulted in a less pronounced systemic inflammatory response (i.e. 77±14% decrease in IL-6 compared to the first challenge), cerebral inflammation was not attenuated, but decreased below baseline, illustrated by a diffuse reduction of cerebral 18F-DPA-714 binding (-38% [95%CI -47 to -28%], p<0.0001). Our findings constitute evidence for in vivo immunological reprogramming in the brain following a second inflammatory insult in healthy volunteers, which could represent a neuroprotective mechanism. These results pave the way for further studies on immunotolerance in the brain in patients with systemic inflammation-induced cerebral dysfunction.


Introduction
The discovery of innate immune cell memory shifted our understanding of innate immunity profoundly (Kurtz, 2005;Medzhitov and Janeway, 2000;Netea et al., 2011). On one end of the spectrum, innate immune cells exposed to microbial components exhibit an enhanced pro-inflammatory phenotype following subsequent restimulation with unrelated stimuli. This phenomenon was coined 'trained immunity' (Netea et al., 2011). On the other end of the spectrum, an initial challenge of innate immune cells with potent immunostimulants such as bacterial lipopolysaccharide (LPS) results in a severely blunted inflammatory response after restimulation with the same or a different compound. This response is known as 'immunotolerance' or 'immunoparalysis' (Netea et al., 2016). Both trained immunity and immunotolerance may exhibit beneficial as well as detrimental consequences for the host.
In the central nervous system (CNS), neuroinflammation has been shown to play an intricate role in the pathogenesis of neuropsychiatric disorders such as major depression, psychotic disorders, Alzheimer's disease (AD), Parkinson's disease, and amyotrophic lateral sclerosis (Brites and Fernandes, 2015;Heneka et al., 2015;Mondelli et al., 2017;Philips and Robberecht, 2011;Vivekanantham et al., 2015). Up to now, innate immune memory of the brain has almost solely been studied within the trained immunity field. Several studies showed that systemic inflammation may induce an exaggerated pro-inflammatory response in the brain when CNS innate immune cells were previously chronically activated (Combrinck et al., 2002;Cunningham et al., 2009Cunningham et al., , 2005Godbout et al., 2005;Hughes et al., 2010;Lopez-Rodriguez et al., 2018;Palin et al., 2008;Raj et al., 2014;Ramaglia et al., 2012;Yin et al., 2017). This hyperreactive cell phenotype is referred to as priming of CNS immune cells, in which pro-inflammatory cytokines produced by glial cells promote sustained tissue inflammation and a neurotoxic microenvironment (Kettenmann et al., 2011), ultimately resulting in exacerbation of brain pathology (Perry et al., 2007;Wendeln et al., 2018).
In contrast, the concept of immunotolerance of the brain has been sparsely investigated. In vitro and animal experiments have shown that LPS-preconditioned microglia acquire an immune-suppressed phenotype (Schaafsma et al., 2015). In addition, single peripheral administration of LPS in mice induced immune training, whereas repeated challenges induced immune tolerance (Wendeln et al., 2018). Interestingly, immunotolerance alleviated cerebral β-amyloidosis load in a murine AD model and led to less severe stroke pathology (Wendeln et al., 2018). These animal data combined with observational data in humans (Ehlenbach et al., 2010;Honarmand et al., 2020;Moller et al., 1998;Monk et al., 2008;Pandharipande et al., 2013;Shah et al., 2013;Wolters et al., 2013) suggest that cerebral innate immune memory plays an important role in neurodegeneration. However, in vivo data on the trajectory of innate immune memory within the human brain is nonexisting.
Experimental human endotoxemia, in which LPS is intravenously administered to healthy volunteers, induces a standardized, reproducible, and safe, transient systemic inflammatory response (van Lier et al., 2019). LPS administration also results in the development of endotoxin tolerance, which is exemplified by a profoundly attenuated immune response after a second challenge with LPS (Leentjens et al., 2012;Leijte et al., 2018;van Lier et al., 2019). This refractory state of the immune system captures many hallmarks of the immunotolerant state observed after sepsis or major surgery (Leentjens et al., 2012;Leijte et al., 2018;van Lier et al., 2019). Previous work has shown that cerebral positron emission tomography (PET) imaging using radiotracers targeting the mitochondrial translocator protein (TSPO) can be employed to measure microglial and astrocytic activation in vivo during systemic inflammation (Forsberg et al., 2017;Nettis et al., 2020;Sandiego et al., 2015;Woodcock et al., 2020). TSPO binding is minimal in healthy brains, but markedly increased after activation of microglia and astrocytes (Lavisse et al., 2012;Rupprecht et al., 2010).
In the present study, we employed this imaging paradigm combined with repeated endotoxemia to study the in vivo immune response trajectories within the systemic and brain compartment.

Subjects and ethical approval
Six healthy male volunteers (aged 18-35 years) were recruited. Exclusion criteria included smoking, use of medication, signs of acute illness in the three weeks prior to start of the study, previous participation in experimental human endotoxemia, previous neurotrauma, past psychiatric or substance dependence diagnosis, and contraindications to undergo a MRI-scan. Written informed consent was obtained, and medical history, physical examination, laboratory tests, and a 12-leads electrocardiogram did not reveal abnormalities. Prior to inclusion, subjects were genotyped for the rs6971 polymorphism on the TSPO gene to exclude low-affinity TSPO binders (homozygous T/T). For this, DNA was extracted from blood samples, and the rs6971 polymorphism was genotyped using Taqman analysis on a 7500 Fast Real-Time PCR System (ThermoFisher Scientific, Waltham, USA). Four participants were high-affinity binders, and two were mixed-affinity binders. Due to the small sample size and the within-subject design, these were analyzed as one group, however, sensitivity analysis did not reveal different outcomes if TSPO genotype was incorporated in the analyses. This study was part of a human endotoxemia project that was approved by the Medical Research Ethics Committee (MREC) region Arnhem-Nijmegen (NL61136.091.17 and 2017-3337). The study procedures were performed in accordance with the declaration of Helsinki.

Study design
In this study, subjects received two subsequent LPS challenges, with an interval of 7 days. Subjects underwent 18 F-DPA-714 dynamic PET/CT imaging of the brain to study the neuroinflammatory response following the LPS challenges at four timepoints: before the first LPS challenge (first baseline), 5 hours after the first challenge, 6 days later (second baseline) and 5 hours after the second challenge at day 7. Furthermore, a structural T1-weighted MRI of the brain was obtained on a 3 T TIM TRIO scanner (Siemens, Erlangen, Germany) at baseline for anatomical coregistration of the PET/CT images. Blood was collected before and at numerous timepoints after the LPS challenges as well as during the PET/ CT scans to characterize the systemic immune response and for pharmacokinetic analysis of the PET-data, respectively. This study design is illustrated in Fig. 1A.

Experimental human endotoxemia
Subjects were admitted twice to the intensive care research unit of the Radboud university medical center in Nijmegen, The Netherlands, to receive both LPS challenges. All subjects underwent the same procedures on both hospitalization days which were 7 days apart. One day prior to both admissions, subjects needed to refrain from alcohol and caffeine, and no food or drinks were allowed from 10 hours before start of the challenge days. After admission, an intravenous cannula was inserted in the antebrachial vein to administrate fluids and LPS. A radial artery catheter was placed to withdraw blood and monitor blood pressure continuously. One hour before the challenge, 1.5 L prehydration fluid (2.5% Glucose/0.45% NaCl) was administered. Subsequently, a bolus of 2 ng/kg LPS (E. coli type O113, Lot no. 94332B1; List Biological Laboratories, Campbell, USA) was administered intravenously and similar hydration fluid was continued at an infusion rate of 150 mL/h for 8 hours. Heart rate was monitored using a 4-lead electrocardiogram (M50 Monitor, Philips, Eindhoven, The Netherlands) and intra-arterial blood pressure was measured with a pressure transducer (Edward Lifesciences, Irvine, USA). Every 30 minutes, we measured core temperature with a tympanic thermometer (FirstTemp Genius 2, Covidien, Dublin, Ireland) and LPS-induced symptoms (headache, nausea, cold shivers, muscle-and back pain) were scored using a numeric six-point scale (0 = no symptoms, 5 = worst symptoms experienced ever). The sum of these scores in addition of 3 points in case of vomiting, resulted in a total symptom score ranging from 0 to 28. Blood was withdrawn before prehydration (T = -60 minutes), just before LPS administration (T = 0 minutes), and respectively 60, 90, 120, 180, 240, 360 and 480 minutes after the LPS bolus. Plasma cytokine concentrations (tumour necrosis factor-alpha (TNF-α), interleukin (IL)-6, IL-8, IL-10, macrophage inflammatory protein (MIP)-1α, MIP-1β, monocyte chemoattractant protein (MCP)-1 and IL-1RA (receptor antagonist)) were determined batchwise using a simultaneous luminex assay (Milliplex, Millipore, Billerica, USA).

TSPO neuroimaging -PET acquisition and image reconstruction
Each subject participated in four 18 F-DPA-714 PET scan sessions prior to and approximately five hours after each LPS challenge. This timing was chosen since we expected increased TSPO expression at this moment, and to reduce the risk on movement-induced artefacts due to LPS-induced symptoms. The interval between LPS administration and the post-LPS scan varied slightly between subjects (5.00 h -5.45 h) but within-subject the interval was exactly the same for both post-LPS scans. 18 F-DPA-714 was synthesized at the Radionuclide Centre of VUmc, Amsterdam, The Netherlands, as previously described (Golla et al., 2015). The specific activity at the end of synthesis was 40.3 ± 11.0 GBq/ μmol and the average radiochemical purity was greater than 99%. Three batches did not pass quality control and since these concerned post-LPS scans, and LPS had to be administered to the volunteers before the radiotracer synthesis was concluded, we could not reschedule these missed scans. 18 F-DPA-714 (123 ± 5 MBq) was injected intravenously as a bolus over 40 seconds by a computer-controlled pump (OptiVantage® Multi-use, Guerbet, Paris, FR). Sixty minutes dynamic PET scans were acquired on a Biograph 40mCT scanner (Siemens Medical Solutions, Erlangen, Germany) of the brain region (15 cm field of view (FOV)), starting at time of injection. Before each radiotracer administration, an ultra-low dose brain CT-scan (80 kV, 30 mA) was acquired for attenuation correction of the PET emission data. Dynamic list-mode scan data were reconstructed into 32 frames (18 × 10 s, 4 × 30 s, 5 × 1 min, and 5 × 10 min) using the Ordered Subset Expectation Maximization (21 subsets and 3 iterations) with inclusion of time of flight, attenuation correction, and in-plane gaussian post smoothing (Full Width at Half Maximum kernel of 3 mm).

TSPO neuroimaging -Arterial input function and metabolite correction
Metabolite-corrected arterial input functions were collected and used for pharmacokinetic modelling to estimate 18 F-DPA-714 binding. Arterial blood samples were taken manually immediately before 18 F-DPA-714 injection and during the scan ~ every 2 seconds for the first 2 minutes, and at 2.5, 3.0, 4.0, 6.0, 8.0, 10.0, 20.0, 30.0, 40.0, 50.0 and 59.5 minutes after injection. Radioactivity in a sub-selection of whole blood samples was measured using an automated gamma-counter (Wizard 2 2480, PerkinElmer, USA). All blood samples were centrifuged and the amount of radioactivity in plasma was additionally phenyl]-5,7-dimethylpyrazolo[1,5-a]pyrimidine-3-acetamide; bpm: beats per minute; TNFα: tumour necrosis factor alpha; IL: interleukin; MCP-1: monocyte chemoattractant protein 1; MIP: macrophage inflammatory protein; IL-1RA: interleukin-1 receptor antagonist; p-values in B-D were tested with repeated measures twoway ANOVA. * p < 0.05; ** p < 0.01, *** p < 0.001 and NS = not significant in E, according to paired t-tests on AUC. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) measured. Then the ratio between whole blood and plasma was calculated to generate a population based whole blood/plasma ratio. This ratio was applied to the individual plasma curves to generate a whole blood curve. In three samples per subject, obtained at 3, 10, and 30 minutes after tracer injection, during post-LPS scans from two participants, the 18 F-DPA-714 parent fraction curve was determined, using solid phase extraction as described previously (Peyronneau et al., 2013). The metabolite-corrected arterial input function was calculated as the product of the plasma time-activity curve and parent fraction curve. The arterial input function in two scans had technical failures and were excluded from the analysis.

TSPO neuroimaging -Image processing and analysis
PET data analysis was performed using PMOD 4.0 software (PMOD Technologies LLC, Zürich, Switzerland). First, the MR images were segmented using the 6 Probability Maps and the PET images were matched to the MR images. Then, the Hammers Atlas was spatially normalized to the PET images. Volume-of-interests (VOI) of individual brain regions were automatically derived to obtain time-activity curves (TACs, in kBq/ml). These included the frontal, parietal, temporal and occipital lobes, thalamus, hippocampus, amygdala, brainstem, cerebellum and whole brain. The metabolite-corrected arterial plasma activity curve was used as input function to estimate the binding potential (BP ND , k 3 /k 4 ) in each VOI. BP ND is the product of receptor density and affinity of ligand binding and dependent on the fraction of nondisplaceable binding in the tissue (Innis et al., 2007;Mintun et al., 1984). BP ND may be directly related to TSPO receptor expression and thus glial activation. In this study, the 2TCM K 1 /k 2 with K 1 /k 2 constrained for gray matter VOIs, showed more reliable (smaller SE) values for BP ND , compared to the 2TCM, 2TCM K 1 /k 2 and to the 2TCM with vascular trapping (2TCM-1 K) (Rizzo et al., 2014), and was therefore reported. The change in BP ND (%ΔBP ND ) across brain regions from the first baseline (PET 1) was computed as (BP ND PET2/3/4 -BP ND PET1) / BP ND PET 1 *100%. For visualization of global PET tracer uptake, standardized uptake value (SUV) images were calculated in PMOD with averaged time frames.

TSPO mRNA expression and 18 F-DPA-714 binding on peripheral blood cells
The methodology of assessing TSPO mRNA expression on peripheral blood cells and the in vitro 18 F-DPA-714 binding assay is described in the Supplementary Information.

Statistical analysis
Statistical analyses were performed with R software packages, version 3.6.2 (http://www.R-project.org) and GraphPad version 6.0 (Sandiego, USA). Results are reported as mean + SEM or median with interquartile range (IQR), depending on the distribution. Area under the curves (AUC) were calculated and systemic responses following the first and second LPS challenge were compared with paired t-tests. To examine differences in 18 F-DPA-714 scan parameters between baseline and follow-up scans, log-transformed BP ND were analyzed using a linear mixed model with time of the PET-scan as within-subjects fixed effect, and subject ID and brain region as random effects. In addition we analyzed the data with time and cortical region as fixed effects to assess differences in the neuroinflammatory response in cortical and other brain regions. Pearson correlations were performed between the AUC of cytokine and leukocyte responses, and vital signs, and the percentage delta in frontal BP ND (%ΔBP ND ) after both LPS challenges, to assess whether the systemic inflammatory response correlates with the immune response of the brain.

Results
Fig . 1A illustrates the study procedures. Six healthy male volunteers aged 24 ± 4 years, with a body mass index (BMI) of 24.7 ± 1.5 were included. From one subject, the baseline PET results deviated more than 2 SD from the mean and was regarded an outlier. Since the second baseline of this subject was missing due to issues with PET tracer synthesis, no baseline measurement was available and therefore this subject was excluded from the analysis. eTable 1 illustrates the homogeneity of the cohort.

Characterization of the systemic immune response after repeated endotoxemia
The first LPS challenge induced a profound systemic inflammatory response, characterized by fever, flu-like symptoms (including headache and shivering), and haemodynamic changes (Fig. 1B). Kinetics in leukocyte differentiation are illustrated in Fig. 1C. Plasma concentrations of both pro-inflammatory and anti-inflammatory cytokines increased substantially following the first LPS challenge and most returned to baseline levels within 8 hours ( Fig. 1D and eFigure 1). Seven days later, all inflammatory parameters had returned to baseline values. The second LPS challenge resulted in an attenuated inflammatory response (Fig. 1B-D and eFigure 1), with an approximately 60% less pronounced cytokine response compared to the first (Fig. 1E), illustrative of systemic immunotolerance in vivo.

The first but not the second LPS challenge significantly increases glial activity
There were no significant differences before or after LPS challenges with regard to injected radiotracer volumes (8.0 ± 0.1 mL vs. 8.0 ± 0.1 mL, respectively, p = 0.49) or dose of radioactivity (122.8 ± 4.4 MBq vs. 123.1 ± 6.2 MBq, respectively, p = 0.79). Three post-LPS scans were cancelled due to radiotracer synthesis batches not passing quality control and two scans (one pre-LPS and one post-LPS) had to be excluded from the analysis due to technical failures. Therefore fifteen scans were included in the final analysis (n = 9 and n = 6 PET scans prior to and after LPS challenges, respectively). Sensitivity analyses between subjects with complete follow-up and subjects with missing scans did not reveal case-mix differences. Fig. 2 shows diffuse uptake of 18 F-DPA-714 throughout the brain. The 18 F-DPA-714 binding potential (BP ND ), which is related to TSPO receptor density and thus a marker for glial activity, was measured in regions of interest for baseline and LPS challenge scans. Table 1 presents the trajectory of regional BP ND values. The first LPS administration significantly increased whole brain BP ND (+53% [95%CI 36-71%], p < 0.0001). This increase was most pronounced in cortical brain regions (+89% [95%CI 33-167%], p = 0.0002), and less in subcortical regions (+28% [95%CI 12-48%], p = 0.0005), which is illustrated in Fig. 3. Importantly, analysis of the kinetic parameter representing influx of the radiotracer between the blood and brain compartment did not suggest a change in blood brain barrier (BBB) permeability or blood flow after the LPS challenges. Six days after the first LPS challenge, BP ND had returned to levels similar to those observed before the first challenge (Table 1 and Fig. 3). Although the second endotoxemia at day 7 induced a less pronounced systemic inflammatory response compared to the first, reflecting systemic immunotolerance, clear increases in all systemic inflammatory parameters were still present as described before. In contrast to these systemic findings, whole brain BP ND was not attenuated, but significantly decreased below baseline levels following the second challenge (-38% [95%CI -47 to -28%], p < 0.0001, Table 1 and Fig. 3). This decrease was similar across all studied brain regions (Table 1 and Fig. 3).

Correlations between systemic and cerebral immune responses
To assess whether systemic inflammatory response parameters were related to the brain immune response, we performed correlation analyses between the area under the curve (AUC) cytokine and leukocyte responses, vital parameters, and the percentage change in frontal BP ND (%ΔBP ND ) after both LPS challenges. Fig. 4 shows strong positive correlations between the endotoxemia-induced increase in heart rate (R 2 = 0.80, p = 0.017) and body temperature (R 2 = 0.73, p = 0.031), as well as the pro-inflammatory cytokines IL-6 (R 2 = 0.79, p = 0.017) and IL-8 (R 2 = 0.79, p = 0.018) with %ΔBP ND .

TSPO mRNA expression on peripheral monocytes
Single-cell RNA sequencing analysis revealed that approximately 55% of circulating classical monocytes do not express TSPO mRNA, and this fraction does not change following a LPS challenge. In the subset of monocytes expressing TSPO mRNA, slightly but significantly increased   [-41 to -34] %ΔBP nd : percentage change in binding potential (BP ND ) from baseline ( 1 :post-LPS challenge 1 vs. 1 st baseline and 2 : post-LPS challenge 2 vs. 2 nd baseline) for subjects with paired measurements. expression was observed 4 hours after the first LPS challenge, which reverted to baseline after 7 days (eFigure 2).

18 F-DPA-714 binding to peripheral blood cells
In addition to TSPO mRNA levels in peripheral monocytes, we evaluated whether this resulted in functional changes of 18 F-DPA-714 binding of peripheral blood cells, to allow for comparison with the observed changes in binding potential of cerebral immune cells. To this end, we performed an in vitro binding assay using whole blood obtained before and 4 hours after LPS administration. eFigure 3 shows a significant 70 ± 7% decrease of plasma/pellet ratio 4 hours after LPS administration (p = 0.012), indicating increased 18 F-DPA-714 binding to peripheral blood cells during systemic inflammation.

Discussion
In the current study, we demonstrate that an initial LPS challenge induced robust glial activation, reflected by diffusely increased TSPO binding in the brain, in particular in cortical brain regions. This response was relatively short-lived, as TSPO binding returned to baseline 6 days later. Following the second LPS challenge on day 7, we observed not increased but rather reduced cerebral TSPO binding, which suggests reprogramming of the brain immune response following LPS-induced systemic inflammation in healthy young male volunteers. Our data also revealed that cerebral and systemic immune responses are strongly correlated. These are important findings, as they indicate the development of cerebral immunotolerance in humans in vivo as a possible brainprotective mechanism.
The Toll-like receptor (TLR) 4 agonist LPS activates innate immune cells to release pro-inflammatory cytokines, which, among others, results in the recruitment of other immune cells. LPS administration has clear effects on brain functioning as reflected clinically by sickness behaviour, headache, nausea and hyperthermia (Schedlowski et al., 2014). The human endotoxemia model enabled us to study the cerebral immune response following a standardized systemic inflammatory stimulus. The profound increase in glial activity observed following the first LPS challenge is in line with previous work (Sandiego et al., 2015;Woodcock et al., 2020), but our study significantly extends these findings. First, our data indicate that the pro-inflammatory state of glial cells following a LPS challenge is relatively short-lived, confirming that the experimental human endotoxemia model can be safely used to study neuroinflammation in healthy volunteers. Second, following the second LPS challenge, cerebral TSPO binding potential was decreased, suggestive of reprogramming of the cerebral immune response. One could argue that this was due to a blunted systemic immune response, which in turn may be less sufficient to activate microglia and astrocytes. However, the second LPS challenge still elicited a clear systemic inflammatory response, including clinical features such as tachycardia and hyperthermia, as well as significant increases in plasma cytokine levels, albeit less pronounced compared to the first challenge. A previous study also demonstrated increased TSPO binding after a single LPS administration, using a 50% lower dose than we used (Sandiego et al., 2015). If the brain were to follow the systemic compartment, we would have expected an increase in TSPO binding following both LPS challenges with a milder increase during the second endotoxemia, rather than the observed decrease following the second LPS challenge.
Thus far, research on immune memory within the neuroinflammation field mainly focused on trained immunity, while cerebral immunotolerance is less well studied. Recent studies demonstrated that increased TSPO binding is specific for pro-inflammatory activated glial cells in the brain in response to inflammatory stimuli (Beckers et al., 2017;Pannell et al., 2020). Since TSPO binding actively decreased during the second endotoxemia, our findings indicate that our healthy volunteers developed reprogramming towards less cerebral proinflammation, which may indicate cerebral immunotolerance. In view of the associations between neuroinflammation and neurodegenerative diseases (Cunningham et al., 2009;Heneka et al., 2015;Perry et al., 2007;Vivekanantham et al., 2015), this immunotolerant phenotype within the brain might serve as a neuroprotective response to reduce inflammation-associated neurotoxicity (Boche et al., 2006;Thackray et al., 2004). Since peripheral immunotolerance is defined as a less pronounced, but still present, pro-inflammatory response upon subsequent stimuli, the decrease in TSPO binding below baseline may not reflect a tolerant state but rather a counterregulatory response. A decrease in TSPO binding from baseline suggests a less proinflammatory phenotype, possibly mediated by a more antiinflammatory environment in the brain. This remains speculation as an in vivo biomarker for anti-inflammatory microglial activity is currently not available. Our exploratory data justify future studies to confirm reprogramming of the cerebral immune response and to explore to what extent cerebral immunotolerance leads to an anti-inflammatory, neuroprotective phenotype in the brain.
In a first, and to date only, clinical TSPO neuroimaging study in patients undergoing surgery, a global decrease of TSPO binding was found 3-4 days after prostatectomy (Forsberg et al., 2017). Interestingly, animal disease models with repeated LPS and concomitant cerebral immunotolerance demonstrated decreased β-amyloid plaque load and modulated stroke pathology (Wendeln et al., 2018). Currently, the effects of cerebral immunotolerance on tau phosphorylation, an important inductor of neurodegeneration, are unknown, and this forms an interesting aim for future studies. A post-mortem study in Alzheimer patients demonstrated an immunosuppressed microglial phenotype in patients who died during infections relative to patients who died without infection (Rakic et al., 2018), indicating the modification of neuroinflammatory processes by systemic inflammation. Cerebral immunotolerance could be a beneficial response to prevent neuroaxonal damage due to exaggerated neuroinflammation. One could argue that in those patients with brain dysfunction, e.g. delirium or long-term cognitive decline following a systemic inflammatory insult, the balance between pro-and anti-inflammatory responses is skewed to excessive pro-inflammation. A brain-specific intervention aimed at polarization towards a more immunosuppressive state (Rinaldi et al., 2016;Song and Suk, 2017) could therefore have therapeutic potential in certain patients. Future studies, including markers for cognitive and mental functioning, are however warranted to validate these hypotheses.
The present study is the first to study neuroinflammatory responses, using longitudinal PET examinations, after repeated inflammatory challenges, with the aim to assess the immune response within the brain. Four previous human TSPO neuroimaging studies (Forsberg et al., 2017;Nettis et al., 2020;Sandiego et al., 2015;Woodcock et al., 2020), three in healthy volunteers and one in postoperative patients, examined the neuroinflammatory response after a single inflammatory stimulus. Our study differs on several topics from this previous work. First, inflammatory stimuli differed between studies in type and dosage, affecting the severity of the inflammatory insult. A recent study used interferon-α (IFN-α) as inflammatory stimulus (Nettis et al., 2020), which resulted in a very mild inflammatory response. Second, the timing of PET-scans ranged from 3 hours (Sandiego et al., 2015;Woodcock et al., 2020), 5 hours (present study), 24 hours (Nettis et al., 2020) and 3-4 days (Forsberg et al., 2017) after the inflammatory stimulus. Since in vivo data on the neuroinflammatory response during systemic inflammation are scarce, it is unknown to what extent timing of the scans affected the results. Our study made use of 18 F-DPA-714, which has the great advantage of having a longer half-life of 18 F, making it feasible to use in centres without an on-site cyclotron facility. With sufficient betweenscan intervals, as in this study (≥24 hours), there is no risk of carryover effects during repeat scanning.
This study has several limitations. First, we only studied healthy young males. Biological characteristics such as age, sex, and body mass index (BMI) contribute significantly to TSPO binding estimates (Tuisku et al., 2019) and may therefore be confounders in clinical studies. Second, the statistical power of the study was affected by its limited size and missing scans, especially for the final post-LPS scan. Furthermore, the relatively small sample size likely resulted in overinflated correlation coefficients for the relationship between endotoxemia-induced clinical and inflammatory parameters and changes in binding potential, which should therefore be interpreted with caution. Since these longitudinal PET studies with human endotoxemia in healthy volunteers are quite invasive, challenging to conduct, as well as expensive, we could not replace subjects to increase the study power. Replication of our findings in another (larger) cohort is warranted. Fortunately, we found no casemix differences and interindividual differences were small. Third, with this dataset we could not rule out inflammation-induced changes in 18 F-DPA-714 metabolism. However since the measured 18 F-DPA-714 metabolism post-LPS in our study was similar to previously published results in healthy volunteers without systemic inflammation (Arlicot et al., 2012), we made the assumption that the radiotracer metabolism did not change after an inflammatory stimulus. Moreover it should be mentioned that dynamic PET neuroimaging is designed with the assumption that the BBB remains intact. With increased BBB permeability, radiotracer metabolites that cannot bind to TSPO could cross the BBB and complicate the analysis. Inflammation is a well-established cause of BBB disruption, but since it has been shown that the BBB is relatively resistant to LPS-induced disruption (Banks et al., 2015), and we observed no change in the kinetic influx parameter after LPS administration, we expect the impact on our results negligible. Finally, TSPO expression is not specific to microglia and astrocytes and the measured PET signal can be driven by other factors such as recruitment of peripheral monocytes into the brain tissue, adherence of circulating leukocytes to the vascular epithelium, and expression of TSPO in neurons and vascular endothelial cells. Although we tried to correct for the endothelial component using the 2TCM-K1 model (Rizzo et al., 2014), this did not result in reliable 18 F-DPA-714 binding estimates.
Strengths of this study include that this is a human study with a longitudinal set-up and repeated strong LPS challenges, resulting in a wide range of inflammatory activation enabling us to detect strong correlations between the systemic and cerebral immune responses. In addition, we have not only shown neuroinflammatory responses following endotoxemia, but also revealed increases in TSPO on a transcriptional and functional level in peripheral blood cells, which are more easily accessible than tissue-resident immune cells. Our study includes extensive characterization of the systemic immune response. Finally, our setup presents a human experimental model that can be used to study cerebral immunotolerance in a very standardized, controlled experimental set-up.
In conclusion, we demonstrated a profound increase in cerebral immune activity after a first inflammatory stimulus with LPS, which was most pronounced in cortical brain regions. While systemic endotoxin tolerance following the second inflammatory stimulus occurred, immune activity within the cerebral compartment was not attenuated but reduced. This may suggest cerebral reprogramming towards a more antiinflammatory phenotype in humans, which may serve as a neuroprotective mechanism. Future studies are warranted to assess whether the presence or absence of differentiation towards an immunotolerant state is related to longer-term cerebral dysfunction in patients that suffered from systemic inflammation.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.