Gene-Environment Interaction Promotes Alzheimer’s Risk as Revealed by Synergy of Repeated Mild Traumatic Brain Injury and Mouse App Knock-In

Abstract

There is a strong unmet need for translational progress towards Alzheimer’s disease (AD) modifying therapy. Unfortunately, preclinical modeling of the disease has been disappointing, relying primarily on transgenic mouse overexpression of rare dominant mutations. Clinical manifestation of AD symptoms is known to reflect interaction between environmental and genetic risks. Mild traumatic brain injury (mTBI) is an environmental risk for dementia, including Alzheimer’s, but there has been limited mechanistic analysis of mTBI contribution to AD. Here, we investigate the interplay between mTBI and Aβ precursor protein gene mutation in AD pathogenesis. We employed a knock-in (KI) model of AD that expresses the Aß-containing exons from human APP bearing the Swedish and Iberian mutations, namely App^NL-F/NL-F mice. Without environmental risk, this genetic variation yields minimal mouse symptomatology. Anesthetized 4-month-old KI mice and their age-matched wild type (WT) controls were subjected to repeated mild closed head injury (rmCHI), once daily for 14 days. Anesthetized, uninjured genotype- and age-matched mice were used as sham controls. At 3- and 8-months post-injury, amyloid-β, phospho-tau and Iba1 expression in the injured KI cortices were assessed. Our data reveal that rmCHI enhances accumulation of amyloid-β and hyperphosphorylated tau inclusions, as well as neuroinflammation in App^NL-F/NL-F mice. Furthermore, novel object recognition and Morris water maze tests demonstrated that rmCHI greatly exacerbates persistent cognitive deficits in APP^NL-F/NL-F mice. Therefore, study of gene-environment interaction demonstrates that combining risk factors provides a more robust model for AD, and that repeated mTBI substantially accelerates AD pathology in a genetically susceptible situation.

Introduction

Alzheimer’s disease (AD) is the leading cause of dementia and is characterized by the accumulation in the brain of neurofibrillary tangles composed of hyperphosphorylated tau protein and senile plaques made of amyloid beta (Aβ). Despite the advances made in our understanding of AD pathogenesis, there is still no therapy available to prevent, slow, halt or reverse the disease. This failure has been ascribed in part to the pitfalls of the preclinical models currently used to study the disease. Indeed, most of these experimental models rely on promoters endowed with strong constitutive activities to drive artificial overexpression and to create certain pathological hallmarks of the disease (Cheng et al., 2004; Jankowsky et al., 2004; Oakley et al., 2006; Oddo et al., 2003). Such animals exhibit a gain-of-function phenotype due to supraphysiological mutant protein expression that may not accurately model AD. Hence, there is a need for better experimental models to investigate AD pathogenesis. Unfortunately, the contribution of modifiable environmental risk factors is often overlooked when studying AD. Therefore, understanding the action of modifiable risk factors in triggering AD may help decipher the complex interplay between the environment and genetic polymorphisms in the induction and progression of the disease. One such risk factor is traumatic brain injury (TBI) (Breunig et al., 2013; Sivanandam and Thakur, 2012; Washington et al., 2016).

TBI can be defined as any mechanical injury to the brain that disrupts normal CNS function, whether caused by a direct impact to the skull or an indirect acceleration/deceleration mechanism (Faul and Coronado, 2015). TBI is a serious public health concern and a leading cause of both acute and long-term disability worldwide, with more than 50 million new victims each year (Maas et al., 2017). The severity of TBI varies widely, and symptoms are as diverse as loss of consciousness, seizures, temporary amnesia, coma and even death. Although the acute effects of TBI are well characterized, it is now recognized that many consequences of injury take years to develop and can result in profound and long-lasting neurological defects. A silent epidemic has been attributed to delayed effects of TBI (Goldstein, 1990; Langlois et al., 2005). There are several studies showing that a previous history of moderate to severe TBI not only increases the likelihood of developing AD (Barnes et al., 2014; Plassman et al., 2000), but is also associated with an earlier onset of the disease (Alosco et al., 2018; LoBue et al., 2016; LoBue et al., 2017). Furthermore, severe TBI has been shown to increase the formation of Aβ plaques (Ikonomovic et al., 2004; Roberts et al., 1994), including in long term survivors (Johnson et al., 2012). Similarly, tau deposits have been described in the brain after TBI (Goldstein et al., 2012; McKee et al., 2013; Omalu et al., 2011). Altogether, these studies demonstrate that TBI is indeed an important risk factor for dementia (Al-Dahhak et al., 2018; LoBue et al., 2018; Washington et al., 2016), including both AD (Sivanandam and Thakur, 2012) and Chronic Traumatic Encephalopathy (CTE) (Goldstein et al., 2012; McKee et al., 2013).

Mild TBI (mTBI), defined as traumatically induced brain dysfunction with lost consciousness of less than 30 minutes, a Glasgow coma score ≤ 13 of 15 at 30 minutes, and less than 24 hours of posttraumatic amnesia (Rehabilitation), is by far the most common form of the injury, representing 75 to 90% of cases (Control., 2003; Kay and Teasdale, 2001; Kraus and Nourjah, 1988). Whereas in the past the effects of mTBI were thought to be self-limited, it is now well recognized that mTBI may cause symptoms that persist for months after the insult (de Koning et al., 2016; Paul J. McMahon, 2014; Silverberg et al., 2019). Furthermore, in contact sports and military service, repeated mTBI (rmTBI) is a common occurrence that has garnered substantial public attention recently because of its association with dementia (McKee et al., 2009; Smith et al., 2013; Stewart et al., 2016). Despite growing attention to rmTBI, early exposure to mild head traumas has not been considered of importance to AD pathogenesis (Gardner et al., 2014; Plassman et al., 2000), with most studies of the TBI-AD literature focused on moderate to severe head injuries (Barnes et al., 2014; Johnson et al., 2012; Lou et al., 2018). Thus, data regarding a mechanistic relationship between repetitive concussions and AD remains limited. Interestingly, recent in vivo studies have suggested a common link between AD and rmTBI, including enhanced Aβ deposition (Grant et al., 2018), brain phospholipid dysregulation (Ojo et al., 2019), alteration of blood based biomarkers, upregulation of inflammatory agents, and production of reactive oxygen species (Ojo et al., 2018). However, the use of overexpression models may limit the clinical relevance of these observations. As a potential molecular connection, cellular prion protein has been implicated in both Aß-mediated AD phenotypes and injury-triggered tau-associated phenotypes (Chung et al., 2010; Corbett et al., 2019; Gimbel et al., 2010; Lauren et al., 2010; Rubenstein et al., 2017).

Here, we investigate the interplay between rmTBI and genetic risk in the induction of AD pathology using a preclinical knock in (KI) mouse model that expresses physiological levels of mutated human APP. Our data demonstrate that the occurrence of rmTBI in young asymptomatic APP-KI mice synergistically accelerates Aβ deposition, somatic tau accumulation and neuroinflammation, culminating in enhanced cognitive deficits.

Materials and Methods

Animal model

App^NL-F/WT heterozygous KI mice were imported from the RIKEN Institute to Yale University. In these mice, one allele of the APP gene was humanized by substituting murine Aβ exons with its human homolog bearing the Swedish (KM670/671NL) and Iberian (I716F) mutations (Saito et al., 2014). The mice were backcrossed for more than 9 generations to C57Bl6 strain and then expanded to generate homozygous App^NL-F/NL-F mice (NLF) and App^WT/WT wild type (WT) controls. All cohorts consisted of 45–55% female mice. All procedures were performed in agreement with the Yale Institutional Animal Care and Use Committee.

Repeated mild closed head injury protocol

Repeated mild closed head injury (rmCHI) was induced in mice aged 4 months using Impact One, an electromagnetic stereotactic Impactor (Leica Biosystem, Illinois, USA). Within the WT and KI groups, mice were assigned randomly to injury and sham protocols. Anesthesia was induced with 4% isoflurane in 1L/min O_2, until corneal and pedal reflexes were absent, upon which they were maintained under 1% isoflurane for the remainder of the procedure. Anesthetized mice were then transferred to the stereotaxic device and the head was shaved to expose the scalp. To induce the head trauma, the impactor was adjusted to produce an impact depth of 1 mm, using a 5 mm diameter impactor tip, at a speed of 5 m/s, with a dwell time of 0.1 s. These parameters have been previously described to produce a mild effect on the brain (Velosky et al., 2017). The impact tip was adjusted to be 5 mm lateral from the sagittal line, 5 mm caudal from the eye, at an angle of 20° from the horizontal. Throughout the procedure, animals were kept on a heating pad and the body temperature kept at 37°C via a homeothermic warming system (Harvard Bioscience, Massachusetts, USA). After the impact, mice were given 0.1 mg/kg of buprenorphine. This procedure was repeated daily for a total of 14 days, alternating between sides of the skull with an inter-injury interval of 24 h. Age-matched sham controls were shaved and anesthetized but did not receive the impact. The injury induced was considered mild as animals met the following criteria: less than 6 min before full recovery of the righting reflex, no skull fracture, and no major haemorrhage at the time of euthanasia (Mouzon et al., 2018b).

Brain collection for pathology

Animals were euthanized with CO₂ and perfused with cold, phosphate buffer saline (PBS), pH 7.4. Brains were dissected, and the two hemispheres separated, with one post-fixed in 4% paraformaldehyde at 4° for 3 days. The second hemisphere was quickly micro-dissected to isolate the cerebral cortex, which was then separated into 3 segments: anterior, middle and posterior. After discarding the anterior and posterior segments, the middle third of the cerebral cortex (containing the impact site) was snap-frozen in liquid nitrogen and stored at −80°C for subsequent use in biochemistry experiments.

Immunohistochemistry

Post-fixed brains were sliced into 40 μm coronal sections using a Leica WT1000S vibratome (Leica, Wetzlar, Germany), and the sections stored in PBS at 4°C. For histology experiments, either 5 (3 mpi) or 6 (8 mpi) mice were randomly selected from each group. For each stain, 6 free floating sections per mouse were collected along the entire antero-posterior axis of the cerebrum, spaced at approximately 960 μm intervals. For immunofluorescence, free-floating sections were incubated in blocking solution (5% bovine serum albumin, 0.3% Triton-X, 7.4 pH PBS) for 1h at room temperature (RT). The following primary antibodies in blocking solution were added to the sections overnight at 4°C: rabbit anti-Aβ (1:1000, D54D2, Cell Signaling Technology, Massachusetts, USA), rabbit anti-tau (1:5000, #A0024, DAKO, Glostrup, Denmark), rabbit anti-phospho-tau Ser 396 (1:500, 44–752G, Invitrogen, California, USA), rabbit anti-Iba1 (1:250, #019–19741, Wako Chemicals, Virginia, USA), mouse AT8 (1:500, MN1020, Invitrogen, California, USA), rat anti-CD68 (1:500, MCA1957GA, Biorad, California, USA), chicken anti-GFAP (1:2000, #ab4674, Abcam, Massachusetts, USA), mouse anti-Olig2 (1:500, #MABN50, Millipore, Massachusetts, USA), mouse anti-oligomeric tau TOMA (1:250, MABN819, Millipore, Massachusetts, USA). Tau and APP null mouse brain tissue was used to confirm specificity of total tau, phospho-S396-tau, TOMA and Aβ antibodies (not shown). Sections were washed 3 times in PBS and incubated 1h at RT in Alexa-Fluor secondary antibodies (donkey anti-rabbit, anti-mouse, anti-rat or anti-chicken, 1:1000, Invitrogen, California, USA) in blocking solution. Sections were washed 3 times and mounted onto Superfrost Plus microscope slides (Thermo-Fisher Scientific, Massachusetts, USA) and coverslipped with Vectashield mounting medium (H-1200, Vector Laboratories, California, USA). Sections were imaged with a Zeiss 800 confocal microscope (Carl Zeiss, New-York, USA) using a 20X objective at fixed exposure time and laser power.

For Nissl staining, sections were dry-mounted onto Superfrost Plus microscope slides and then bathed for 10 min in a prewarmed solution of 0.1% cresyl violet, 0.25% acetic acid. After removing the excess dye with water, sections were destained in 95% ethanol for 10 minutes and bathed twice in 100% ethanol for 5 minutes. Sections were then washed twice in xylene for 5 min and the slides coverslipped with Cytoseal60 (Thermo-Fisher Scientific, Massachusetts, USA). Sections were visualized with a Zeiss AxioImager Z1 microscope (Carl Zeiss, New-York, USA), under a 5X objective and pictures of whole sections acquired with an Axiocam 305 color camera using the tiling mode.

Protein extraction for biochemistry

Cortices were homogenized in two-fold volume of 20 mM Tris buffer saline (TBS) pH 7.5, containing PhosSTOP phosphatase inhibitor (Roche, #04906837001, Mannheim, Germany), and cOmplete-mini protease inhibitor (Roche, #11836170001, Mannheim, Germany) cocktails. Homogenates were ultracentrifuged at 100,000 g for 30 min at 4°C, and the supernatants collected and stored at −80°C as the TBS soluble fraction. The pellets were resuspended in RIPA buffer made of 20 mM TBS, 0.1% SDS, 135 mM NaCl, 1% NP-40, 10% Glycerol, supplemented with phosphatase and protease inhibitor cocktails. The homogenates were ultracentrifuged at 100,000 g for 30 min at 4°C and the supernatants recovered as the RIPA-soluble fraction and stored at −80°C.

Immunoblotting

Tissue extracts (10 μg protein) were boiled for 5 minutes in Laemmli buffer and separated in 4–15% precast polyacrylamide gel (Biorad, #4561086, California, USA). Gels were transferred onto nitrocellulose membranes using the iBlot™ 2 device (Invitrogen, California, USA). Membranes were incubated in a commercial blocking solution (Rockland, MB-070, Pennsylvania, USA) for 1h at room temperature. The following primary antibodies diluted in the blocking solution was then applied overnight at 4°C: mouse anti-APP_Nterm (1:1000, #MAB348, Millipore, Massachusetts, USA), anti-rabbit APP_cterm (1:1000, #512700, Invitrogen, CA, USA), mouse anti-tau (1:500, #05–348, Millipore, MA USA), rabbit anti-phospho-tau Ser 396 (1:500, #44–752G, Invitrogen, California, USA), rabbit anti-phospho-tau Ser199 (1:500, 44768G, Invitrogen, California, USA) mouse anti-β-actin (1:2500, #8H10D10, Cell Signaling, Massachusetts, USA) mouse anti-βIII-tub (1:1000, #G712A, Promega, Wisconsin, USA). After 3 washes with 0.1% TBS-Tween 20 (American Bio, Massachusetts, USA), blots were incubated for 1h at room temperature in secondary antibodies diluted in blocking solution: Odyssey donkey anti-rabbit/donkey anti-mouse IRDye 680 or IRDye 800 (1:10,000, Licor Biosciences, Nebraska USA). Blots were washed 3 times with 0.1% TBS-Tween 20 and revealed with Licor Odyssey Infrared Imaging System. Densitometry analysis of the blots was performed with Image Studio Lite software (Licor Biosciences, Nebraska USA).

Image analysis

Quantitative analysis of histological data was done using the FIJI software (NIH, Washington D.C, USA), with the experimenter blinded to the genotype and injury. Stained regions of confocal images were selected by setting a single common threshold intensity for all images for a particular staining method. The percentage area occupied by the thresholded signal, as well as the mean fluorescent intensity in that area were then measured. In rare cases where the antibody background was high and there was evidence for poor perfusion by observation of blood cells in vessels, those particular mice were excluded from the quantitative analysis for a particular stain. Quantification of phospho- and total tau cellular inclusions was done by manual counting of thresholded images.

Quantitative assessment of microglia morphology was performed as previously described with a few modifications (Morrison and Filosa, 2013). After thresholding, images of Iba1 labelled sections were skeletonized with the Analyze Skeleton plugin of FIJI (http://imagejdocu.tudor.lu/) to render an arborisation of the microglia processes. The number of trees (processes emerging from the soma) and branches (processes emerging from trees) were counted by the program and expressed as trees/cell and branches/cell respectively, with the number of microglial cells counted manually.

Behavioral testing

Before the behavioral essays, mice were handled for 5 minutes each for 5 days to habituate the animals to the experimenter and reduce anxiety. Mice were randomized and assigned to different genotype and injury groups to which the experimenter was blinded.

Novel object recognition (NOR) was performed as described previously with minor modifications (Salazar et al., 2019; Salazar et al., 2017). First in the habituation phase, mice were kept for 1 h in a clean, rectangular rat cage in a quiet, dimly lit room. During the acquisition phase, mice were taken out of the cage and two identical objects chosen pseudorandomly were introduced in the cage, perpendicularly to the long axis of the cage, ~ 2 cm from the sides. The cage was oriented so that the experimenter was facing its long axis. The mice were returned to the cage and allowed to explore the objects for 10 min with the time required to accumulate 30 s of orofacial exploration (whisking or sniffing) recorded by the experimenter. At the end of these 10 min, the two familiar objects were discarded, and the animal left in the cage for 1 h. Finally, in the trial phase mice were removed from the cage, and two different objects were placed in the cage at pseudorandom corners, one object being identical to the previous familiar objects, and one being novel. The animals were then reintroduced in the cage and the exploration of each object timed. The observer was blind as to which of the two objects was novel. After each trial the objects were discarded, and the cage cleaned to remove olfactory cues. Because the test was repeated at several time points, different pairs of objects were used in NOR: a 15 mL Falcon tube and an unwrapped 5 mL plastic syringe in the 1^st test, a large binder clip and a plastic glue stick in the 2^nd test and a large binder clip and 15 mL Falcon tubes in the 3^rd test. Mice were provided 10 min to accumulate 30 s of orofacial exploration of the objects, and a failure of reach 30 s led to exclusion from the analysis.

Morris water maze (MWM) was performed as described with minor modifications (Salazar et al., 2019; Salazar et al., 2017). A tank of 1 m diameter with visual cues was filled with water and maintained at 21–23°C. A transparent platform was placed at the center of the target quadrant, hidden 2.5 cm below the water surface. Trials were conducted daily, each consisting of 4 attempts, in which animals were put in the opposite quadrant of the tank, facing the wall and the time it took them to find the escape platform recorded. The test was considered a success if the mouse found the platform within 60 s and stayed there for 1 s, upon which it was removed from the tank. If the mouse failed to reach the escape platform after 60 s, in the first 2 trials the mouse was guided to the platform and allowed to stay there for 10 s before being taken out of the tank; for the remaining trials it was just guided to the platform. The probe trial was conducted 24 h after the last session. The hidden platform was removed, and the mouse was placed in the tank facing the wall in the quadrant opposite the target quadrant, and allowed to swim for 60 s. For the entirety of the assay, visual cues and ambient lightning were kept constant. Repeat tests at 3 or 8 months placed the target in different quadrants from earlier tests. The latency to platform and the probe trials were recorded on a JVC Everio G-series camcorder (JVCKenwood, Yokohama, Japan) and tracked by Panlab’s Smart software (Harvard Bioscience, Massachusetts, USA).

Statistical analysis

Data were plotted as mean +/− sem. All statistical analyses were made using one-way or two-way ANOVA analysis followed by Tukey’s multi-comparison post-hoc test as indicated in the figure legends. Where appropriate, one sample or two-tailed unpaired Student’s t-test was used. GraphPad Prism 8 (GraphPad Software Inc., San Diego, CA, USA) was used for most statistical analyses. Repeated one way and two-way ANOVA analyses where done using SPSS Statistics (IBM, New-York, USA). Statistical significance was set at p <0.05.

Data Availability

The data that support the findings of this study are available from the corresponding authors, upon reasonable request.

Results

Repeated mild closed head injury does not causes macroscopic brain damages

To study the role of multiples concussions in dementia induction, we used a protocol of rmCHI that subjected 4-month-old mice to multiple mild head impacts. In contact sport, such as football or hockey head impact can occur multiple times to a player within the span of a single season (Brainard et al., 2012; Crisco et al., 2010; McAllister et al., 2012; Montenigro et al., 2017). Therefore, to better simulate repetitive concussions, a total of 14 mild controlled cortical impacts were administered to animals, one per day over 2 weeks. The first behavioral essays were conducted 1.5 months after the induction of rmTBI. The same tests were repeated at 3 and 8 months after the injury protocol after which some of animals were sacrificed and their brain collected for histology and biochemistry (Fig 1A). To assess how AD genes might modulate the effects of rmCHI on the brain, we used NLF mice genetically modified to express the Aß-encoding exons from human APP bearing the Swedish and Iberian mutations, two mutations known to increase the production of aggregation prone Aβ42, within the endogenous mouse App locus (Fig 1B) (Citron et al., 1992; Guardia-Laguarta et al., 2010).

Figure 1. Repeated mild closed head injury does not cause macroscopic brain damage.

A)Mice aged 4 months received a daily repeated mild closed head injury (rmCHI) for 14 days, or only anesthesia (sham) and were allowed to recover for either 3 or 8 months followed by behavioral, histological and biochemical analysis. The distribution of mice to each procedure is shown. NT, not tested.

B)Mice either expressed the murine App gene (WT), or were knock-in for the human sequence at App exons encoding Aß and bearing the KM670/671NL and the I716F mutations (NLF).

c)Image of a perfused mouse brain showing that the impact site in the temporal cortex (dashed lines) presents no macroscopic lesions.

D-G) Nissl staining of coronal sections of sham and injured WT and NLF brain revealing the presence of microlesions at the impact site of injured mice (red arrow). For immunohistochemical analysis we used the perilesional area and the distal lesional area corresponding respectively to cortical layers I to IV (red square) and layers V to VI (blue square), respectively. Scale bar: 1000 μm.

H-K) Higher magnification view of the Nissl stained rmCHI brain showing microlesions at the impact site. Scale bar: 200 μm.

L-O) Immunolabelling of NeuN in brain sections revealing at most a slight decrease of neuronal density at the impact site irrespective of genotype. Scale bar: 40 μm.

This closed head impact was quite mild, as evidenced by the fact that within 6 minutes of injury, and within the time that the anesthetic effects had resolved in the control Sham injury mice, there was no obvious deficit in the level of alertness or the righting response for injured mice. Furthermore, close observation of the perfused brain shows that rmCHI caused neither macroscopic brain injury nor substantial hemorrhage in the impacted area (Fig 1C). However, Nissl staining of brain sections revealed that, in comparison to sham controls (Fig 1D, E), the injured brains showed a microlesion below the impact site (Fig 1F, G). Higher magnification showed that the microlesion was restricted to the impact site and was limited to more superficial cortical layers, with the perilesional and distal sites remaining unaltered by routine histological stains (Fig. 1H–K). In addition, staining of the brain tissue with the neuronal marker NeuN revealed that near the lesion site there was limited or no decline of neuronal density relative to sham controls (Fig 1L–O). Therefore, our injury model produced minimal effects in mice, consistent with rmTBI features described previously (Kane et al., 2012; Velosky et al., 2017).

rmTBI accelerates the formation of Aβ deposits in the brain of humanized AD model

Amyloid pathology in NLF mice develops slowly, with detectable plaque formation not being apparent before 12 months of age (Saito et al., 2014). On the other hand, TBI has been shown to increase APP processing and to promote Aβ accumulation (Gentleman et al., 1993; Ikonomovic et al., 2004; Roberts et al., 1994; Stein et al., 2015). Thus, we sought to determine whether multiple episodes of mild head trauma could precipitate Aβ accumulation in young, otherwise plaque-free AD mice. We probed Aβ expression in the brain by immunostaining, using an antibody capable of recognizing all Aβ species. As expected, we found that at 3 months post-injury (mpi) there was no change in Aβ expression between WT and NLF sham mice (Fig 2A, B). For injured NLF mice, but not injured WT mice, there was a qualitative modest increase of Aβ at the perilesional site relative to sham controls (Fig 2C, D). Cortical areas more than 2 mm distal from the injury site showed no changes in Aβ accumulation between injured and intact controls. By densitometric analysis, neither Aβ immunoreactive area nor intensity were significantly altered in the perilesional or distal sites (Fig 2I–L).

Figure 2. rmTBI accelerates the formation of Aβ deposits in the brain of APP-KI mice.

A-D) Aβ immunostaining (green) of the brain at 3 months post-injury shows that compared to sham and WT-CHI controls there is a moderate accumulation of Aβ at the perilesional site of the NLF-CHI mice.

E-H) No changes in Aβ levels between animals was detected at the distal lesional site.

I-L)Quantitative analysis of Aβ signal showed a slight increase of Aβ labelled area in the injured transgenic subject in regions proximal but not distal to the impact site with the intensity remaining unaltered throughout the cortex.

M-Q) Aβ immunolabelling (green) of the brain at 8 months post-injury demonstrates that contrary to the WT control, rmCHI induced a marked increase in the formation Aβ plaques in the NLF brain at the perilesional site. Double stain of NLF-CHI sections with CD68 (Q, red) shows that Aß plaques are surrounded by activated microglia.

R-V) Similarly in the area distal to the lesion site and in opposition to WT controls, rmCHI triggered the formation of Aβ deposits in NLF mice (green). Double stain with CD68 (V, red) reveals activated microglia in the vicinity of Aß plaque.

W-Z) Statistical analysis confirmed a significant increase of Aβ area and intensity at the perilesional site at 8 months post-injury in the knock-in mice. In contrast, within the distal site, and relative to controls, there were no significant changes of Aβ-labelled area and intensity in the injured NLF brain.

lotted in each graph are the mean +/− S.E.M, each circle represents a mouse. I-L) WT-Sham, n=5; NLF-Sham, n= 5; WT-CHI, n=4, NLF-CHI, n=5. U-X) WT-Sham, n=4; NLF-Sham, n=5; WT-CHI, n=6, NLF-CHI, n=5. One-way ANOVA, Tukey’s post-hoc multi-comparison test, *: p<0.05, **p<0.01, ***p<0.001, Scale bars: 40 μm.

Aβ accumulation in the injured brain was also assessed at 8 mpi, i.e. at 12 months of age. Consistent with previous results (Saito et al., 2014), Aβ plaques were detected in the superficial cortical layer of sham-injured NLF brain when compared to WT sham controls (Fig 2M, N). There was also a moderate Aβ build-up in the injured WT brain (Fig 2O). Strikingly, the injured NLF mice exhibited substantially greater formation of Aß plaques when compared to transgenic sham control (Fig 2P), and these plaques were surrounded by activated microglia as demonstrated by the presence of CD68-positive cells in the vicinity of the deposits (Fig 2Q). In areas distal to the impact site and in stark contrast to all 3 control groups, there was a pronounced deposition of Aβ in the injured KI mice, albeit to a lesser extent than in the perilesional area (Fig 2R–U). Here too, Aß plaques were juxtaposed to activated microglia (Fig 2V). These results were confirmed by quantitative analysis showing that at the perilesional site, total Aβ-positive area was greater for NLF-sham mice than for WT-sham controls. rmCHI caused a significant increase in Aβ area in the transgenic brain relative to its genotype-matched intact control (Fig 2W) (one-way ANOVA, F(3, 16) = 12.66, p=2×10⁻⁴). All injured brains exhibited a significantly higher Aβ intensity at the perilesional site when compared to intact controls (Fig 2X) (one-way ANOVA, F(3, 16) = 6.39, p=4.7×10⁻³). In cortical zones distal to the impact site, Aβ area and mean intensity were not significantly altered (Fig 2Y, Z).

Since acutely increased expression and processing of APP is reported as a feature of TBI (Gentleman et al., 1993; Wang et al., 2018) we assessed whether there were chronic changes in APP level with rmCHI brain. When compared to genotype-matched controls at 3 mpi, injured cortices did not present changes in APP expression in either the TBS or RIPA fractions. In addition, βIII-tubulin level was similar between all groups indicating that the injury protocol did induce overt neuronal loss, thus further confirming the mild nature of the injury (Sup Fig S1A–D). At 8 mpi, we observed in the TBS fraction an augmentation of APP expression in injured cortices relative to intact controls, although this was significant only for WT. In the RIPA fraction, head trauma caused a significant elevation of cortical APP level in the NLF brain only (Sup Fig S1E–H). In relationship to these APP measurements, the incremental Aβ accumulation in the injured NLF brain might relate to increased APP levels or increased amyloidogenic processing or impaired clearance.

Taken together, these results demonstrate that the early occurrence of multiple concussions in young APP-KI brain causes a time-dependent mild increase of Aβ pathology.

The APP-KI genotype exacerbates the intracellular accumulation of tau after chronic head injury

In AD, the formation of tau tangles is a cardinal pathophysiological manifestation (Alonso et al., 1996; Braak and Braak, 1995). Similarly, the aggregation of misfolded tau is an essential feature of CTE, a disease caused by rmTBI (McKee et al., 2009; McKee et al., 2013). In contrast, the uninjured App^NL-F/NL-F KI mouse does not to exhibit tau pathology, even with advanced age (Saito et al., 2014). We considered whether the genotype might modulate tauopathy triggered by concussive injuries.

First, we performed immunohistochemical analysis of brain sections labelled with antibody against total tau. Staining of total tau showed that by the 3^rd month post-lesion, and relative to sham controls, there was an accumulation of tau in the cells of the App^NL-F/NL-F superficial cerebral cortex (Fig 3A–D). In deeper cortical layers more distant from the impact site of NLF animals, there was a marked accumulation of tau in cell bodies (Fig 3E–H). Quantitatively, total tau area was not significantly altered in the perilesional area; while tau staining intensity was increased in concussed brains relative to sham controls (one-way ANOVA, F(3, 12) = 10.68, p=1.1×10⁻³) (Fig 3I, J). At the more distal site, there was no change in tau immunoreactive area, but the density of tau-positive cell bodies somata was increased by injury, with the NLF KI mice showing the most dramatic effect (one-way ANOVA, F(3, 12) = 5.25, p=1.52×10⁻²) (Fig 3K, L). Critically, by 8 months after rmCHI, the somatic tau deposits were more abundant in the injured KI animals, and were now found in zones both proximal and distal to the injury site (Fig 3M–P, s–V). As might be expected, the majority (58 ± 2%, mean ± S.E.M., n = 3 mice) of somatic tau inclusions where neuronal, showing colocalization with NeuN (Fig 3Q, W; Sup Fig 2A–C). Interestingly, co-staining of tau with the oligodendroglial marker, Olig2, revealed that the formation of tau deposits occurred to a lesser extent (32 ± 9%, mean ± S.E.M., n = 3 mice) in oligodendrocytes as well (Fig 3R, X; Sup Fig 2D–F). Thus, chronic head injury and APP gene polymorphism act in tandem to promote the accumulation of total tau inside both neuronal and oligodendroglial cells.

Figure 3. The APP-KI genotype exacerbates the intracellular accumulation of tau after chronic head injury.

A-H) Anti-tau labelling (red) of cerebral cortex 3 months after repeated mild traumatic head injury from the indicated groups of mice. The perilesional site of injured KI animals shows an increase of tau accumulation relative to other groups. In the distal area, the NLF-CHI brain displayed marked somatic tau deposits.

I-L) Quantification of images similar to those in A-H. Total tau positive area was not significantly altered, but there was an increase in tau signal intensity as well as the somatic tau accumulation for the KI animals subjected to head trauma compared to other groups, as indicated.

M-R) Brain sections stained for tau immunoreactivity (red) 8 months post-rmTBI clearly demonstrated tau inclusions for the impacted NLF brain near the microlesion site. Tau inclusions were in both neurons and oligodendrocytes as shown by colocalization with NeuN (Q, green) and Olig2 (R, green, arrowheads).

S-X) The deeper cortical layers also showed a marked increase of tau accumulation selectively in the NLF-CHI brain. Tau accumulations colocalized to NeuN+ (W, green) and Olig2+ (R, green, arrowheads) cells.

Y-B’) Quantification of stained images as in M-X revealed that at 8 mpi the total tau-positive area was unaltered, but the number of tau-accumulating cells was higher in the NLF-CHI brain.

Plotted are the mean +/−S.E.M, each circle represents a mouse. I-L) WT-Sham, n=4; NLF-Sham, n=4; WT-CHI, n=4; NLF-CHI, n=4. U-X) WT-Sham, n=5; NLF-Sham, n=5;and WT-CHI, n=5; NLF-CHI, n=6. One-way ANOVA, Tukey’s post-hoc multi-comparison test, *: p<0.05, **p<0.01. Scale bars: 40 μm.

Alzheimer’s pathology promotes cell-specific accumulation of phosphorylated tau after rmTBI

In tauopathies, hyperphosphorylation of tau renders the protein insoluble leading to its accumulation and aggregation in the brain (Alonso et al., 1996). Therefore, we investigated whether the accumulation of total tau in the injured brain is correlated with an increase in phosphorylation. We first examined the phospho-tau epitope at Serine 396 (pS396) since its hyperphosphorylation contributes to tau dysfunction, accumulation, and spreading in AD (Bramblett et al., 1993; Regalado-Reyes et al., 2019; Rosenqvist et al., 2018). Interestingly, three months after the mild head trauma, we found that in KI mice there was the accumulation of pS396 in cortical areas near the lesion site (Sup Fig 3A–D). Cortical areas distal to the injury sites also showed somatic pS396 inclusions, which were more pronounced in NLF mice than WT controls, and were not observed in tissue from sham injured animals (Sup Fig 3E–H). These findings were supported by quantitative analysis (Sup Fig 3I–L). We also performed a biochemical assessment of the effects of trauma on tau expression and phosphorylation in the NLF cerebral cortex at 3 mpi. In addition to pS396, we probed total tau and tau phosphorylated on Ser199 (pS199) another epitope described to correlate with disease severity (Augustinack et al., 2002). In opposition to histology data, immunoblot densitometry of the TBS-soluble and RIPA-soluble fractions did not reveal consistent changes in total tau expression or tau phosphorylation between sham and injured groups (Sup Fig 3N–S). Therefore, at 3 months after the injury, the marked local increase in tau phosphorylation observed histologically is not widespread enough to be detected by whole cortex immunoblot.

We continued to assess tau pathology in 8 mpi samples. We found that contrary to sham controls and injured WT subjects, there was a pronounced formation of somatic phospho-tau inclusions in the injured KI brain both at the perilesional site (Fig 4A–D) and at distal sites (Fig 4E–H). While pS396 immunoreactive area was not altered by injury (Fig 4M, O), head trauma caused a significant increase in the frequency of pS396-positive cell inclusions selectively in the NLF brain, in areas both perilesional areas (Fig 4N) (one-way ANOVA, F (3, 17) = 8.13, p=1.4×10⁻³) and distal to the impact zone (Fig 4P) (one-way ANOVA, F (3, 17) = 5.26, p=9.5×10⁻³). Intriguingly, we found that within injured KI mouse brain the pS396+ cells did not colocalize with NeuN (Fig 4E, K; Sup Fig 4A–C) or GFAP (Sup Fig 4D–F), ruling out neurons and astrocytes as the main site of these somatic deposits. Surprisingly, most of the pS396+ cells instead colocalized with the oligodendrocytic marker, Olig2 (Fig 3F, L; Sup Fig 4G–I). This was confirmed by quantitative analysis showing that in the perilesional tissue of injured KI animals 84 ± 5% (mean ± S.E.M., n = 3 mice) of oligodendrocytes where pS396 positive.

Figure 4. Alzheimer’s pathology promotes the cell-specific accumulation of phosphorylated tau after rmTBI.

A-L) Immunolabelling of the brain at 8 months post-injury using an antibody against tau phosphorylated on S396 (pS396, green) demonstrated the increase of phosphorylated tau in the injured AD brain both in the perilesional and distal cortical areas. By double immunofluorescence, pS396-tau deposits did not colocalize with neuronal NeuN (E, K, red), but did colocalize with oligodendrocyte Olig2 (F, L, red).

M-P) Quantitative analysis demonstrated that total pS396 area remained unchanged, whereas there was a stark increase in the number of somatic inclusions of phospho-tau in the injured AD brain relative to both injured WT and sham controls.

Q) Immunoblot analysis of RIPA soluble cortical extracts probed with antibodies raised against total and phospho-tau showed that compared to controls, at 8 months post-injury NLF mice presented an apparent increase in both total tau and tau phosphorylated on S396 and S199.

R-T) Densitometry analysis confirmed that relative to both WT-CHI and sham controls, NLF-CHI cortices presented a significant increase of total and phospho-S396 tau while the S199 epitope instead exhibited a reduced phosphorylation on the 50kDa isoform.

U-W) Statistical analysis of immunoblots from TBS soluble cortical extracts depicting an increase in total tau level in NLF-sham mice relative to WT-sham subjects; the 60 kDa tau isoform showed an increased phosphorylation on S396 in NLF-CHI compared to WT-CHI.

Plotted are the mean +/− S.E.M, each circle is one mouse. M-P) WT-Sham, n=5; NLF-Sham, n=5; WT-CHI, n=5, NLF-CHI, n=6. R-W) WT-Sham=5, NLF-Sham=5, WT-CHI=5, NLF-CHI=5. One-way ANOVA, Tukey’s post-hoc multi-comparison test, *: p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, Scale bar: 40 μm.

In order to assess whether the neuronal total tau accumulation we observed in the injured NLF brain (Fig 3) was associated with alternate phosphorylation sites, we performed AT8 co-staining with NeuN. Detection of tau phosphorylation on Serine 202 and Threonine 205 by AT8 antibody (Goedert et al., 1995) was chosen as a key biomarker of tau pathology in AD (Braak and Braak, 1995; Goedert et al., 1993). Near the lesional site in the NLF brain, there were numerous AT8+ cellular inclusions, and all AT8+ cells were also NeuN+ (n=3 mice), consistent with a neuronal identity (Sup Fig 4J–L). Furthermore, immunoprobing of the impacted NLF brain with the TOMA antibody (Castillo-Carranza et al., 2014) revealed the presence of tau oligomers inside neurons, as indicated by TOMA/NeuN double-labelled cells (Sup Fig 4M–O). Thus, specific pathological tau conformations and phosphorylations are associated with different cell types in the injured NLF mice.

By 8 months after injury, immunoblot of the RIPA-soluble cortical extracts revealed that total tau, pS396 and pS199 levels all increased in the injured NLF brain (Fig 4Q). Densitometry confirmed that in the NLF brain, rmCHI caused a significant increase of total tau, both for the 50 kDa (one-way ANOVA, F (3, 16) = 9,09, p=1×10⁻³) and 60 kDa isoforms (one-way ANOVA, F (3, 16) = 16.69, p<1×10⁻⁴) (Fig 4R). Consistent with histology results, pS396 level was increased in the injured AD brain, for both the 50 kDa (one-way ANOVA, F (3, 16) = 6.42, p=4.6×10⁻³) and 60 kDa isoforms (one-way ANOVA, F (3, 16) = 5.46, p=8.9×10⁻³) (Fig 4S). In contrast, immunoblot measurements for Ser199 phosphorylation of the 50 kDa isoform showed a reduction in the concussed APP-KI brain relative to its genotype-matched control (one-way ANOVA, F (3, 16) = 8.49, p=1×10⁻³) (Fig 4T). Quantitative analysis of total and phospho-tau expression was also completed with the TBS-soluble fraction. Here, data showed that relative to WT sham, only the intact NLF brain presented a significant increase of both 50 kDa (one-way ANOVA, F (3, 16) = 6.72, p=3.8×10⁻³) and 60 kDa tau isoforms (one-way ANOVA, F (3, 16) = 3.38, p=0.044) (Fig 4U). For the Ser396 epitope, the injured KI cortices showed a significantly increased phosphorylation of 60 kDa TBS-soluble tau (one-way ANOVA, F (3, 16) = 3.96, p=0.027) (Fig 4V), while phosphorylation on Ser199 remained unaltered (Fig 4W). Overall, the combination of trauma and KI yielded a time-dependent and consistent accumulation of pS396-tau.

Taken together, these results show that in the young APP-KI brain, rmTBI triggered time-dependent and cell-specific changes in tau phosphorylation which ultimately promoted the formation of intracellular tau deposits. In oligodendrocytes, the pathology consisted chiefly in tau hyperphosphorylation on S396, whereas accumulation in neurons was associated with tau hyperphosphorylation on S202/T205 and with oligomerization.

Persistent activation of microglia in the injured AD brain

We then asked whether the chronic neuroinflammation known to occur with rmTBI (Aungst et al., 2014; Luo et al., 2014) was altered in the presence of Alzheimer’s pathology. For that purpose, we studied microglia activation through labelling of tissue sections with antibodies directed against CD68, a marker of activated microglia. We found that at 3 mpi CD68+ cells were detected only in the perilesional site of impacted animals, with the injured NLF brain showing the strongest effect (Sup Fig 5A–L).

We continued the analysis of microglia activation 8 months after concussions, this time using Iba1, a pan-microglial marker. We found that in the perilesional zone, microglial cells in both WT and NLF sham mice extended numerous cell processes that were highly complex and ramified (Fig 5A, B), a morphology that is consistent with a baseline quiescent state (Kettenmann et al., 2011). However, in injured WT mice microglia were less ramified suggesting that they had become reactive (Fig 5C). For impacted APP-KI mice these cells were amoeboid in shape, a sign consistent with the adoption of a phagocytic phenotype (Fig 5D)(Stence et al., 2001). For distal areas, and in comparison to intact WT controls, microglia were less ramified in all injured brains. Interestingly this also applied to sham KI animals, suggesting that cells have been activated by genetically driven plaque formation in this case (Fig 5E–H). Quantification of morphological changes demonstrated that in the perilesional area the number of trees/cell was significantly lower in injured NLF animals relative to intact WT controls (one-way ANOVA, F (3, 15) = 5.46, p=9.7×10⁻³) (Fig 5H). The number of branches/cell was also starkly reduced in injured NLF animals comparatively to intact WT controls with intermediates changes in head-injured WT and intact NLF-sham subjects (one-way ANOVA, F (3, 15) = 8.30, p=1.7×10⁻³) (Fig 5I). In the distal area, the number of trees/cell was significantly higher in WT sham controls relative to the other groups (one-way ANOVA, F (3, 15) = 10.14, p = 7×10⁻⁴) (Fig 5K). A similar result was observed when quantifying the number of branches/cell far from the concussion site (one-way ANOVA, F (3, 15) = 10.18, p=7×10⁻⁴) (Fig 5L). These results demonstrate that persistent microgliosis is exacerbated by the combination of rmCHI and genetic predisposition due to NLF knock-in.

Figure 5. Persistent activation of microglia in the injured APP-KI brain.

A,B) Labelling of Iba1 in the brain to visualize microglia shows that 8 months after rmCHI, microglial cells from both WT and NLF sham controls are highly ramified which indicates a quiescent state.

(C) At the perilesional site microglial cells from WT-CHI mice are markedly less ramified, which indicates that they are in a reactive state.

D) in NLF-CHI mice there is a strong microgliosis around the impact site as demonstrated by the amoeboid shape of the cells.

E-H) In cortical zones distal to the impact site, microglial cells in WT-sham controls remain in a quiescent state whereas in both NLT-Sham, WT-CHI and NLF-CHI, they are all reactive.

I,J) Quantification of the number of processes (trees and branches) per cell via morphometric analysis demonstrates that near lesional site there is a marked and significant transition from a quiescent state in WT-Sham animals to a fully reactive state in NLF-CHI mice.

K,L) Morphometric analysis of microglial cells in cortical areas distal to the injury site reveals that they are also in a reactive state in both NLF-Sham, WT-CHI and NLF-CHI mice relatively to WT-sham controls.

Plotted in each graph is the mean +/− S.E.M, each circle represents a mouse; WT-Sham, n=4; NLF-Sham, n=4; WT-CHI, n=5, NLF-CHI, n=6. t, one-way ANOVA, Tukey’s post-hoc multi-comparison test, *: p<0.05, **p<0.01, Scale bar: 40 μm.

We also assessed reactive astrogliosis, through histological staining of GFAP. We found that in areas both proximal and distal to the impact site, there was an enhancement of GFAP expression in the injured NLF brain at 3 mpi (Sup Fig 6A–L). This was more pronounced at 8 mpi (Sup Fig 6M–X). Taken together, these results demonstrate a time-dependent enhancement of the injury-induced neuroinflammatory response in the KI brain.

AD persistently aggravates the cognitive defects triggered by repetitive concussions

Mutations in the APP gene (Scheuner et al., 1996; Selkoe and Hardy, 2016) and repeated concussive injuries (Montenigro et al., 2017; Nolan et al., 2018), detrimentally affect cognition. We sought to determine whether the interaction between these two factors might enhance cognitive deficits. A behavioral analysis of injured NLF mice and control groups was performed at various intervals after the injury protocol. Of note, previous study of APP^NL-F/NL-F mice without injury have shown no deficit in learning at 6 and 12 months of age (Saito et al., 2014; Sakakibara et al., 2019).

Novel object recognition (NOR) testing demonstrated that at 1.5 mpi and in opposition to WT animals, repeated concussions significantly altered the ability of NLF mice to discriminate between a novel and a familiar object (p_WT-Sham=1.21×10⁻², p_NLF-Sham=4.6×10⁻³, p_WT-CHI=2.49×10⁻², p_NLF-CHI=0.669) (Fig 6A). In contrast, at the 3 mpi time point, all animals successfully distinguished between novel and familiar objects (p_WT-Sham=3×10⁻⁴, p_NLF-Sham=1.03×10⁻², p_WT-CHI=4×10⁻³, p_NLF-CHI=2.99×10⁻²), though injured NLF mice spent the least time with the novel object (Fig 6B). By 8 mpi the effect of trauma had resolved in the AD mice, as they spent considerably more time with the novel object than the familial one, similar to sham controls (p_WT-Sham= 0.001, p_NLF-Sham=0.167, p_WT-CHI=0.204, p_NLF-CHI=0.011) (Fig 6C). Thus, repeated head trauma caused a slowly resolving impairment of object memory retention for young APP KI mice.

Figure 6. Combination of repetitive concussions and APP-KI genotype persistently impairs cognitive performance.

A)NOR of mice at 1.5 months after rmCHI shows that WT-sham significantly spent more time with the novel object compared to chance performance (dotted line), the same was true of NLF-sham and WT-CHI mice. On the contrary, NLF-CHI subjects could not discriminate between familiar and novel objects.

B)At 3 months post-injury all animals spent significantly more time with the novel object. The injured transgenic group spent the least amount time with the novel object.

c)By 8 months post-injury all impacted mice had fully recovered their ability to identify the novel object as they performed similarly to their genotype-matched controls.

D-F) During the acquisition phase of Morris water maze trial performed 1.5, 3 and 8 months after rmCHI, all injured animals had a higher escape latency than their genotype-matched sham control. However, this injury effect was more severe in transgenic mice relative to WT control, especially at 1.5 months post-injury where the difference was significant.

G-I) Probe trial performed at 1.5, 3- and 8-months post-injury demonstrated that relative to genotype-matched sham controls, there was a decline in the time spent in the target quadrant by all head-injured animals. The decline was significantly more drastic in transgenic mice at 1.5 and 3 months after injury, with the 8-month time point showing a trend towards significance.

Plotted is the mean +/− S.E.M; each circle is one mouse. A) WT-Sham, n=18; NLF-Sham, n=16; WT-CHI, n=22; NLF-CHI, n=22. B) WT-Sham, n=18; NLF-Sham, n=14; WT-CHI, n=17; NLF-CHI, n=14. C) WT-Sham, n=7; NLF-Sham, n=6; WT-CHI, n=3; NLF-Sham, n=9. D, G) WT-Sham, n=30; NLF-Sham, n=32; WT-CHI, n=30; NLF-CHI, n=30. E, H) WT-Sham, n=19; NLF-Sham, n=20; WT-CHI, n=17; NLF-CHI, n=16. F, I) WT-Sham, n=14; NLF-Sham, n=15; WT-CHI, n=12; NLF-CHI, n=11. A-C: One sample t-test of novel object exploration time compared to chance time of 15 seconds (represented by the dotted lines in the graphs). D-F: Asterisks to the right of the curves represent repeated measures one-way ANOVA over the last 3 trial blocks with pairwise Tukey’s post-hoc multi-comparison test. The stars above the blocks represents one-way ANOVA of each trial session for WT-Sham versus NLF-CHI after Tukey’s post-hoc multi-comparison test. *: p<0.05, **p<0.01, ***:p<0.001, ****:p<0.0001. G-I: one-way ANOVA, Tukey’s post-hoc multi-comparison test, *: p<0.05, **p<0.01, ***:p<0.001, ****:p<0.0001.

We extended the analysis to assess spatial memory in the Morris water maze (MWM). Here, we found that at 1.5 mpi, and in contrast to WT and KI sham controls, the escape latency was significantly longer in injured WT mice. Importantly, the delayed escape latency was significantly more pronounced in injured KI mice than in impacted WT animals (repeated measures one-way ANOVA, F=35.73, p< 1×10⁻⁴) (Fig 6D). A similar pattern was observed regardless of sex (data not shown). At 3 mpi, we observed that the learning deficits caused by repeated concussions remained more severe in the NLF mice (repeated measures one-way ANOVA, F=23.09, p< 1×10⁻⁴) (Fig 6E). This differential effect of rmCHI on the learning abilities of WT versus NLF mice persisted at 8 mpi (repeated measures one-way ANOVA, F= 17.44, p< 1×10⁻⁴) (Fig 6F). It should be emphasized that there was an interaction between APP-KI gene status and repeated concussions on the escape latency that was statistically significant at 1.5 mpi (repeated measures two-way ANOVA F_interaction (1, 118) = 11.08, p_interaction = 0.01) and 3 mpi (repeated measures two-way ANOVA, F_interaction (1, 68) = 5.96, p_interaction = 0.01).

To test whether the location of the escape platform was remembered over one day, we carried out a probe trial 24 h after the final swim at each time point. At 1.5 mpi, impacted mice spent significantly less time in the target quadrant than did intact animals, with the injury effect being significantly more severe in NLF mice than WT (one-way ANOVA, F(3, 118) = 11.90, p < 1×10⁻⁴) (Fig 5G). Similarly, at 3 mpi, the injured KI mice continued to spend significantly less time in the target quadrant relative to the other groups (one-way ANOVA, F(3, 68) = 2.71, p = 0.05) (Fig 5H). The same pattern was observed at 8 mpi (one-way ANOVA, F(3, 48) = 3.05, p = 0.03) (Fig 5I).

Due to the prominent role of the hippocampus in spatial memory tasks, we assessed whether the synergistic effect of rmTBI and NLF genotype on the accumulation of Aß, pS396-tau and total tau observed in the cerebral cortex extended to this brain region. However, the accumulation of these proteins remained unaltered in the hippocampus at 8 mpi (Sup Fig 6A–L), suggesting that the memory performance was either due to aggregation-independent hippocampal dysfunction or to contributions of other regions to the behavior. Despite the lack of hippocampal protein aggregation, we observed that multiple brain regions in addition to the medial cerebral cortex displayed various degrees of tau, p396-tau and Aß protein deposition in concussed NLF mice (Table 1), with potential contribution to the impaired behavior.

Table 1:

Distribution of Aß, pS396-tau and total tau in the NLF-CHI brain.

	Aß	pS396-tau	Tau
Medial motor cortex	+++	+++	+++
Lateral motor cortex	+	+	+
Corpus callosum	0	++	+
Striatum	0	0	0
Hippocampus	0	0	0
Thalamus	0	0	0
Dorsal lateral geniculate	0	+	0
Amygdala	0	+	++
Optic tract	0	++	++

NLF-CHI brains at 8 m.p.i. were examined immunohisotologically, and protein accumulation scored qualitatively as follows: 0 (no protein deposits), + (mild), ++ (moderate), +++ (pronounced). n = 3 mice.

Overall, we conclude that the APP-KI genetic status severely aggravates chronic cognitive deficits triggered by traumatic head injury even though the App^NL-F/NL-F genotype has minimal or no effect alone.

Discussion

The major finding of the current work is that rmTBI and genetic predisposition from the APP KI synergize to create profound deficits in learning and memory, with accompanying Aß plus tau accumulation and inflammation. At the ages studied here, the APP-KI status alone produced minimal to no deficit, and the repeated mildTBIs yielded only limited cognitive deficits and minor pathological changes. Specifically, the combination of environmental trauma and genetic knock-in resulted in a robust and persistent deficit.

The aim of this study was to investigate the effects of mild TBI, an environmental risk factor for AD, in the induction of the disease. To avoid the pitfall of overexpression systems, we opted for a KI model in which mice produce physiological levels of APP from the mouse locus with the human Aß sequence and with the Swedish and Iberian mutations known to promote the production of aggregation prone Aβ42 (Saito et al., 2014). Thus, we were able to demonstrate for the first time in this model that the interaction between multiple mild head injuries and physiological levels of mutant APP expression dramatically enhances the initiation and progression of a process exhibiting features of AD. An exacerbation of Aß accumulation and behavioral deficits by rmCHI had been reported in an APP overexpression model, though effects on tau and inflammation were not described (Uryu et al., 2002). Another study using single TBI in an 8 month old APP/PS1 double KI model reached conclusions similar to ours (Webster et al., 2015). However, the novelty of our work stems from the fact that we used a model of repeated mild concussive injuries, and that we intervened at a younger age, as is typically the case for contact sports and military exposure. We also conducted our analyses over an extended follow-up period to assess long-term consequences of repetitive brain trauma. Arguably the most important contribution of this work is that we demonstrate aggregation of endogenous tau, as discussed below.

In addition to a very late onset and minimal functional deficit, the uninjured APP-KI model is limited as an AD model in that it does not develop tau pathology (Saito et al., 2014). To overcome this issue, we took advantage of the ability of rmTBI to sporadically trigger tau pathology in the brain (McKee et al., 2009; Montenigro et al., 2017). Thus, using a combined environment plus genetic strategy, we were able to demonstrate that rmTBI and NLF status synergized to generate tau pathology in NLF mice with altered phosphorylation and somatic accumulation. Of significant importance is the fact that the enhanced tauopathy caused by head injury in a context of Aβ deposition did not require a mutation in the tau gene. Nevertheless, we noted that at 3 mpi, when histology revealed initial accumulation of total and phospho-tau in the cortex of impacted NLF brain, immunoblots of cortical homogenates failed to show a global difference relative to controls. We believe that this is because the degenerative process was still in a focal early stage and tau pathology was not robust or widespread enough to be detectable in bulk cerebral cortex. By 8 mpi, we were able detect considerable accumulation of hyperphosphorylated tau by both histologically and biochemically methods, confirming that the tau dysfunction induced by head injury in the NLF brain was time-dependent. At this stage, the injured NLF brain showed profound alterations of total and phosphorylated tau primarily in RIPA-soluble fraction, not TBS-soluble fractions. Thus, under the influence of repeated head trauma, mouse tau in the APP-KI brain exhibits a gradual reduction in extractability from TBS-soluble to RIPA-soluble. These observations are consistent with our understanding of clinical tau pathology where hyperphosphorylation causes the protein to become insoluble and accumulate in the cell soma (Alonso et al., 1996; Augustinack et al., 2002).

Previous studies using WT as well as human tau transgenic mice have reported variable CHI effects on tau, with expression changes ranging from nil (Mouzon et al., 2014; Xu et al., 2016), to transient (Mouzon et al., 2019; Mouzon et al., 2018a) to chronic (Kondo et al., 2015; Ojo et al., 2016; Petraglia et al., 2014; Zhang et al., 2015). Previous repeated mild head injury studies of APP transgenic mice failed to elicit tau pathology (Cheng et al., 2019; Uryu et al., 2002), or relied on tau overexpression to achieve this such pathology (Hu et al., 2018; Tran et al., 2011; Wu et al., 2020). Our rmCHI injury triggered a mild tau pathology in WT mice, which was substantially enhanced in the APP KI mice. Our observation of a time-dependent accumulation of total tau in injured animals is likely due to the chronicity of the injury paradigm and follow up period (Ojo et al., 2016; Petraglia et al., 2014). Most critically, the enhanced chronic deposition of tau in injured APP KI mice demonstrates a compounding action of APP mutation and trauma.

Interestingly, we found pathological tau deposits in both neurons and oligodendrocytes of concussed APP KI animals. This was the case not only in the injured cortex, but also in the corpus callosum and amygdala. This finding is consistent with a study reporting the presence of tau tangles in oligodendrocytes in the brain of AD patients (Nishimura et al., 1995). Furthermore, it has been reported that intracerebral inoculation in WT mice of sarkosyl-insoluble extracts derived from brains afflicted with AD and other tauopathies leads to the accumulation of hyperphosphorylation tau in oligodendrocytes which colocalize with activated tau-kinases (Ferrer et al., 2019). Our data therefore raise the possibility that in addition to neurons, oligodendrocytes may also play an active role in the enhancement of trauma-induced neuropathology in AD subjects. We also observed a cell-specific pattern of phospho-tau inclusions with phosphorylation at S396 being relatively oligodendrocyte-specific and phosphorylation at S202/T205 residues being detected mainly in neurons. This may reflect cell-specific dysregulation of tau kinases and phosphatases, but requires future study.

Although TBI promotes brain Aβ deposition, the association between head trauma and amyloid pathology has not been clear since only 30–40% of post-mortem cases of severe TBI (Johnson et al., 2012; Roberts et al., 1994) or CTE form Aβ deposits (Stein et al., 2015). This partially penetrant effect of head injury on Aβ accumulation could be explained by interindividual differences in genetic predisposition to amyloid pathology. In support of this notion, we found that at 12 months of age while concussions caused the formation of a few Aβ inclusions around the impact site in the WT cortex, in the APP-KI brain there was a considerable augmentation in the number of dense Aβ plaques. Importantly, the relative difference in Aβ immunoreactivity between intact and injured brains was higher for APP-KI mice than WT control. Thus, repeated concussion accelerates Aβ accumulation of in the presence of predisposing APP mutation.

Our behavioral data show that the spatial memory deficits exhibited by impacted NLF animals in the MWM persisted from 1.5 to 8 mpi. Given the time-dependence of our neuropathological findings, we initially expected that the cognitive deficit caused by rmCHI in KI mice might increase gradually from a moderate level at 1.5 mpi. Instead, the MWM studies showed that the injured KI mice already exhibited pronounced learning deficits at 1.5 mpi, and that this remained maximal for the remainder of the study. This temporal pattern suggests that the underlying pathological processes triggered by injury quickly interact with the NLF genotype to trigger early and persistent cognitive deficits. Specifically, it appears that the modest neuropathological changes seen at 3 mpi correspond to advanced functional deficits. Distinct from spatial memory testing, NOR performance demonstrated defects in injured KI mice that were transient, with no deficits at 3 and 8 mpi. The behavioral differences for impacted NLF mice between MWM and NOR may be ascribed to the different neural pathways needed for each test. MWM tests spatial learning and long-term memory, which require complex interactions between multiple brain regions including the neocortex and the hippocampus (D’Hooge and De Deyn, 2001), the NOR paradigm for its part evaluates episodic memory and relies more specifically on networks connecting the prefrontal cortex to hippocampus (Chao et al., 2016; Chao et al., 2017). Thus, on one hand, our observation of the most severe Aβ and tau pathologies in the neocortex near the impact site, might explain why the injured transgenic mice showed profound and persistent defects in MWM. Other regions with protein aggregation in NLF-CHI mice (Table 1) might also contribute. Alternatively, the MWM deficits may result from hippocampal dysfunction not driven by protein aggregation, such as injury-induced calcium dyshomeostasis (Sun et al., 2008). Overall, relative sparing of prefrontal cortex and hippocampus from Aβ and tau aggregation might explain why the deficits in NOR were moderate and transient, with injured animals recovering completely by 8 mpi.

Several cellular and molecular mechanisms are likely to contributor to the synergistic interaction between concussion and APP mutations. One potential pathway derives from the neuroinflammatory response. With a single concussion, there is only minor neuroinflammation. However, with subsequent impacts occurring while the brain is still in an active response phase, cellular changes may accrue, yielding full glial activation (Weil et al., 2014). The inflammatory mediators released by the reactive cells may in turn stimulate Aβ production and aggregation (Buxbaum et al., 1992; Yamamoto et al., 2007). The consequence could then be a reinforced activation of microglia and chronic neuroinflammation. This is supported by our observation that microglial cells remained chronically activated in the injured KI brain months after the impact. In addition, inflammatory mediators can cause the activation of tau kinases such as GSK3ß or Cdk5 (Kinney et al., 2018). Alternatively, head injury has been shown to increase β-secretase activity, which would stimulate Aβ production (Lou et al., 2018), an effect enhanced by the presence of the Swedish and Iberian mutations. The result will be an upregulation of Aβ release, oligomerization and plaque deposition in the AD brain. In parallel, our group has previously shown that Aβ oligomers, by forming a complex with cellular Prion protein, can allosterically hyperactivate mGluR5 (Haas et al., 2017; Salazar et al., 2017; Um et al., 2013), an event known to lead to Fyn and Pyk2 phosphorylation (Kaufman et al., 2015; Salazar et al., 2019). The net result of this cascade is increased tau phosphorylation, either directly, or through enhance activation of GSK3ß, a well-known tau kinase. (Brody and Strittmatter, 2018). Therefore, the potential exists for a feedforward loop with concussions causing inflammation and enhancing Aβ production coupled with accrued amyloidogenic processing of APP leading to persistent gliosis and synaptotoxicity, with both converging to accelerate tau hyperphosphorylation. Future studies will further delineate the mechanistic events driving the synergism between concussion and APP mutation in AD pathogenesis.

Concerning the limitations of this study, first is the fact that the current APP KI model does not express human tau. Recent AD brain extract innoculation studies performed using double mutant mice containing both the APP-KI allele and human WT tau inserted into the mouse locus show markedly accelerated tau spreading when compared APP-KI controls (Hashimoto et al., 2019). Therefore, rmCHI studies in human APP/tau double KI mice will be of interest to extend further our understanding of head injury in AD induction. Secondly, because of the limited number of mice used for histology, we could not assess whether there were sex differences in protein accumulation for injured mice. Another limitation of the study is that during the 14-day injury protocol, only CHI mice received buprenorphine for analgesia. Therefore, we cannot rule out drug administration confounding differences between Sham and CHI groups.

In conclusion, we have demonstrated that the exposure of mice carrying mutations associated with familial AD to multiple mild concussions results in an acceleration of amyloid and tau brain pathologies in concert with prolonged neuroinflammation. The synergistic brain pathology culminates in long-lasting spatial memory deficits. Combining environmental and genetic risks provides a novel approach for modeling AD, and allows investigation of the pathophysiological mechanisms driving AD in the context of mild head trauma history.

Highlights.

• Mild repeated closed head injury (rmCHI) created in APP knock-in (KI) mice

• Synergistic deficits in learning and memory for combined APP KI plus rmCHI group

• Greater phospho-Tau accumulation and inflammation in combination group

• Combining genetic and environmental factors yields more robust preclinical AD model