Abnormal CD8 T cells induce and track Alzheimer’s-like neurodegeneration


 Sporadic Alzheimer’s disease, the most common neurodegenerative disorder of aging, is characterized by cerebral plaques and neurofibrillary tangles. Experimental rodents develop plaques but neither tangles nor substantial neurodegeneration under conditions that guarantee Alzheimer’s in humans, suggesting rodents lack critical co-initiation factors. Accumulation of antigen-reactive memory CD8 T cells increases with aging, and was recently revealed as a hallmark of human Alzheimer’s. The impact of this process on disease initiation, however, has not been established because age-related T cell changes are muted in rodents. We developed a mouse model of human-like CD8 T cell aging that promotes antigen-reactive memory CD8 T cell accumulation. Here we show that these “hiT” mice develop all major hallmarks of Alzheimer’s with aging, including tangle-like inclusions and substantial neurodegeneration. Antigen-reactive CD8 T cells analogous to those in hiT mice increased in Alzheimer’s brain, but decreased earlier in blood, where their loss effectively distinguished the Alzheimer’s continuum from aging controls. Our findings establish a clinically relevant mouse model for sporadic Alzheimer’s and show that age-related immune dysfunction critically contributes to its initiation. They also identify useful immune-based targets to track and potentially treat human Alzheimer’s, while validating a model system to examine age-related disease immuno-biology more generally.


Generation of " hi T" cells in nude mice
We previously demonstrated that young B6.Foxn1 (B6.nude) mice injected with purified donor 56 B6 CD8 T cells rapidly develop a T cell compartment dominated by homeostatically induced, self-57 reactive CD8 T cells with a resident memory phenotype ( hi TRM) identical to age-related CD8 T 58 cells accumulating in affected aged mice. This mirrors the dominance of circulating age-related 59 memory CD8 T cells in moderate-to old-age humans 8 , and rendered selected APP-reactive 60 memory CD8 T cell levels in mice similar to those in aging humans (Extended Data Fig. E1) 9 . 61 hi TRM recipient mice also exhibited age-related tissue pathology, including neuroinflammation and 62 increased memory CD8 T cells in brain, along with other factors associated with AD. We therefore 63 examined whether B6.Foxn1 hi TRM recipients exhibited additional neuropathological features of 64 AD various times after T cell injection (Extended Data Fig. E2a). To ensure our observations were 65 due to functional rather than purely physical aspects of CD8 hi TRM accumulation in the brain, we 66 included B6.Foxn1 cohorts injected with PBS, and with CD8 T cells from wild-type, Perforin 1-67 deficient, or IFN-deficient donors, (PBS,68 respectively). 69 Aβ and neurofibrillary deposition 70 CD8 T cells expanded in circulation of all B6.Foxn1 recipients 7 . By contrast, Amyloid Precursor 71 Protein (APP) and its cleavage products including A were dramatically increased only in brains 72 of wt-CD8 group mice 3 weeks after injection (Fig. 1). Detergent-soluble A1-40, but not A1-73 42, was also elevated 10 weeks post-injection in ELISA analysis, and Western blot confirmed 74 5 prominent involvement of hippocampus at this time point (Fig. 1a,b;Extended Data Fig. E2b). 75 By 6 months post-injection, increased A deposition in brain vasculature was evident in wt-CD8 76 group mice, consistent with selective elevation of A1-40 (Extended Data Fig. E2c). At 15 months 77 post-injection, A1-40 was still significantly elevated in wt-CD8 brain (Fig. 1b), and A plaques 78 were evident in entorhinal cortex, hippocampus, and cingulate cortex of wt-CD8 and IfnKO-CD8 79 groups (Fig. 1c,d). Unlike mice expressing familial gene mutations found in human AD, A 80 plaques in both these groups were mainly diffuse and detergent-soluble (Extended Data Fig. E3a,81 b), with little co-staining by curcumin or ThioS (Fig. 1c;Extended Data Fig. E4a,b). The discrete 82 amyloid pathology in the wt-CD8 group encouraged examination of additional AD-associated 83 features such as tau phosphorylation and aggregation. 84 Detergent-soluble phospho-tau (pTau) was slightly (30%) but significantly increased by 10 weeks 85 post-injection in wt-CD8 group forebrain, while pTau paired helical filaments (PHFs, which 86 mature to form NFTs in AD) were increased nearly 5-fold (Fig. 1e). Soluble pTau did not remain 87 elevated 15 months post-injection, however, while PHFs remained significantly elevated, but at 88 lower than earlier levels (2.5-fold decreased; Fig. 1f). We speculated this could be due conversion 89 of pTau isoforms to more insoluble aggregated species after 10 weeks. Indeed, fibril-staining 90 reagents, including curcumin and Thio-S, revealed cellular inclusions within wt-CD8 group 91 hippocampus 6 months post-injection (Extended Data Fig. E4a, b). These inclusions were apparent 92 before larger amyloid deposits appeared, although small plaques were occasionally associated with 93 them (Extended Data Fig. E4a, b), and were absent from aged AD-transgenic mice, as well as from 94 AD-transgenic rats that reportedly accumulate PHFs (Extended Data Fig. E4a, b) 10 . 95 6 Gallyas staining also revealed discrete silver-stained cellular structures in wt-CD8 and IfnKO-96 CD8 hippocampus and cortex 15 months post-injection, that were not seen in AD-transgenic mice 97 (Fig. 1g,h). These structures appeared similar to NFTs from human AD patients (Extended Data 98 Fig. E5a,b). In addition, sequential staining revealed that Gallyas-stained structures in wt-CD8 99 and IfnKO-CD8 groups were derived from pTau + neurons with intact nuclei (Fig. 1g), and that 100 Gallyas and pTau staining was superimposable and distinct from that of A (Fig. 1g; Extended 101 Data Fig. E5c). "Ghost tangles", NFTs in dead neurons that are often present in human AD 11,12 , 102 were not observed. These data suggest that hi TRM promote the coordinated deposition of 103 parenchymal Aβ40, diffuse plaques, and fibrillar pTau inclusions in live neurons, either directly 104 or by indirectly promoting neuroinflammation.

105
Immune & neuroinflammatory infiltration 106 Our previous work established that astrogliosis, microgliosis, and CD8 T cell brain infiltration 107 were all increased in wt-CD8 group hi TRM hosts 7 . We therefore examined the relationship of 108 observed neuroinflammatory features to A plaque burden to determine the immune population 109 most directly associated with neuropathology. Aβ plaque burden correlated strongest with 110 hippocampal CD8 T cell numbers compared to astrogliosis or microgliosis, consistent with a more 111 direct impact of adaptive than innate immune cells on amyloid pathology (Extended Data Fig. E6).

112
In this context, it is intriguing that, while PrfKO-CD8 group mice failed to exhibit either CD8 T 113 cells in brain, or any significant AD-like neuropathology, IfnKO-CD8 group mice retained 114 significantly increased plaques and NFT-like structures in hippocampus and entorhinal cortex, but 115 not in cingulate cortex (Fig. 1d,h). This resembles the distribution of protein aggregates early in 116 human AD 13 , and as such suggests that neuroinflammation hastens AD-like neuropathology in 117 7 hi TRM mice, but is not required for its development. Taken together, our data suggest that hi TRM 118 may directly promote pathologic features of AD-like neurodegeneration. 120 Robust neurodegeneration is not present in mouse AD models without addition of transgenes 121 uninvolved in human AD 1,14 . To determine if overt neurodegeneration was present in hi TRM mice, 122 we stained and counted NeuN + neurons in CA1, CA2, and CA3 of hippocampus, assessed brain 123 mass, and quantified Western blots of NeuN and synaptic proteins. Loss of NeuN + cells in wt-CD8 124 group mice was visually apparent in hippocampal immunostains, and was verified by NeuN + cell 125 counts at 15 months post-injection . Loss of brain mass was also evident in wt-CD8 126 group, and progressed from 5% at 6 months, to 10% 15 months post-injection (Fig. 2d), which is 127 comparable to terminal brain atrophy in human AD 15 . Western blots confirmed an approximate 128 10% decrease in NeuN, as well as in the synaptic protein, Drebrin, and a non-significant trend 129 toward lower Synaptophysin protein, 15 months post-injection (Fig. 2e,f). Importantly, loss in 130 brain mass correlated with decreased NeuN across all T cell injection groups, establishing a direct 131 relationship between brain mass and neuronal loss (Fig. 2g). Thus, B6.nude hi TRM recipients 132 exhibited robust and easily discernible neurodegeneration by multiple measures, with related brain 133 atrophy. This revealed the possibility that hi TRM mice might exhibit AD-like dementia as well.

134
Severe cognitive impairment 135 Prior to cognitive testing in hi TRM mice, we established that spontaneous locomotor activity was 136 not significantly different between treatment and control groups in Open Field testing 3, 6, and 13 137 months after T cell injection (Fig,3a;Extended Data Fig. E7a), ruling out motor deficits such as 138 those in multiple sclerosis. Nevertheless, all groups exhibited motor deficits that increased with 139 8 age, but this was unrelated to treatment. In contrast to motor activity, Fear Conditioning (FC) 140 response to contextual but not cued learning was reduced in wt-CD8 group mice 6 months after T 141 cell injection, with both contextual and cued learning impaired in the same mice tested 5 months 142 later (Fig 3b). The hippocampus is required for contextual FC responses, whereas both 143 hippocampus and amygdala are required for cued FC responses. Hence, the FC results suggest that 144 hi TRM mediate damage to hippocampus alone early on, and cause further damage to amygdala later, 145 a pattern commonly seen in human AD 16 . Contextual performance at 6 and 11 months also 146 correlated with brain mass (Extended Data Fig. E7b), further underscoring the relationship of 147 cognitive decline to physical neurodegeneration. 148 Spontaneous Alternation Behaviour (SAB) 12 months post-injection independently confirmed 149 behavioral abnormalities in wt-CD8 group mice exclusively. This test is based on the preference 150 of mice to alternately explore two alleys, which requires working memory of the alley previously 151 entered. The lowest possible score of 50% indicates random alley choice, reflecting either no 152 working memory, or complete lack of preference. The 55-56% SAB score in the PBS group was 153 comparable to published wild-type values 17 , but the wt-CD8 group SAB was significantly lower 154 at 50% (Fig. 3c). To verify whether this reflected loss of working memory or lack of preference, 155 we employed the Barnes Maze test at 14 months, a more focused measure of hippocampus-156 dependent learning and memory. In contrast to all other groups, wt-CD8 mice showed no ability 157 to learn the maze over the initial 4-day training period, indicating a profound learning and memory 158 deficit (Fig. 3d). Given this inability, wt-CD8 mice were uniquely impaired on subsequent memory 159 retention and reversal phases of the maze as well . As with Fear-Conditioning, latency 160 to solve the Barnes Maze correlated inversely with brain mass (Extended Data Fig. E7c). Taken   161   9 together, these tests suggest that fully functional hi TRM mediate severe, progressive impairment of 162 hippocampus-dependent learning and memory, but not locomotor activity. 163 Because cognitive impairment is differentially associated with amyloid and tau pathology in AD, 164 we further addressed whether cognitive loss was associated with A and/or pTau metrics in hi TRM 165 mice. Poor performance on Barnes Maze (total latency below median = BM lo ) exhibited significant 166 association only with increased pTau PHFs on Western blots (Extended Data To examine possible involvement of T cells analogous to those in hi TRM mice in AD, we quantified 173 CD8 T cells in blood, their relationship to cognitive decline, and their presence in brain, using 174 three human cohorts (Fig. 4a). We first examined KLRG1 + and KLRG1 -CD8 T cell 175 subpopulations in blood from aging control subjects (CTRL), MCI patients with an AD-typical 176 CSF biomarker profile (MCI-AD), MCI patients without an AD-typical CSF biomarker profile 177 (MCI), and sporadic AD patients (AD) (Cohort 1). KLRG1 + CD8 T cells were not significantly 178 increased in AD blood (Fig. 4b), but increased in rough correlation with age, while KLRG1 -CD8 179 T cells did not (Extended Data Fig. E8). In contrast, KLRG1 + CD8 T cells co-stained with pHLA-180 A2 multimers to a human T cell epitope analogous to that recognized by T cells in hi TRM mice 181 [APP(471-479)] were dramatically decreased in the blood of MCI, MCI-AD, and AD patients (Fig. 182 4c, d). While this differs from other (i.e., EBV-specific) CD8 T cells that increase in AD 3 , 183 10 segregation of CTRL and MCI patients by cognitive performance score did reveal that KLRG1 + 184 CD8 T cells increased during age-related cognitive decline (Fig. 4d). Moreover, APP(471-479)-185 specific KLRG1 + CD8 T cell levels correlated significantly with cognitive decline but not age 186 itself (Fig. 4e). The parental KLRG1 + CD8 T cell population thus appears to expand in blood 187 earlier than other memory CD8 T cells 3 , and appears to contract as cognitive decline is clinically  197 We next examined CD8 and Perforin-1 content in AD brain. Importantly, Western blots rendered 198 the expected antibody specificities (68-75 kDa Prf1; 33-35 kDa CD8α), with anti-Prf1 staining 11 2.3 + 0.55, P = 0.31; not shown). Nevertheless, APP-specific CD8 T cells were increased in 207 perivasculature and cortical regions of AD hippocampus (Fig. 4i), similar to hi TRM mice and as 208 predicted by that model.

210
In conclusion, we show for the first time that all major hallmarks of human AD can be elicited by 211 a single inductive event in mice. This pathology is dependent on expansion in blood and entry into 212 brain of age-related resident memory CD8 T cells (CD8 TRM), and exhibits gross similarity to 213 human AD with important distinctions. Notably, amyloidopathy in hi TRM mice was limited to A1-214 40 and mainly diffuse plaques, unlike the A1-42 and mature neuritic plaques that predominate in 215 many AD patients. Moreover, while hi TRM mice exhibited fibrillar NFT-like structures in live cells, 216 they did not harbor the ghost tangles in dead neurons often seen in human AD. While A 217 differences could reflect deficiencies in A1-42 clearance and/or fibril formation characteristic of 218 rodent brains 20,21 , the lack of ghost tangles is likely due to decreased expression in adult mice of 219 the ghost tangle-promoting isoforms of MAPT, the gene encoding tau proteins 11,12 . 220 221 Despite these distinctions, hi TRM mice exhibited unique similarities to sporadic AD in humans 222 beyond amyloidosis, fibrillar tauopathy, and robust neurodegeneration. These included cognitive 223 decline that initiated with hippocampus-dependent tasks with later progression to amygdala-224 dependent tasks 16 ; significant association of cognitive loss exclusively with fibrillar tau pathology 225 18 ; neuroinflammation that exacerbated neuropathology 22 ; and accumulation of antigen-specific 226 memory CD8 T cells in brain 7 , the most recently characterized hallmark of human AD 3 . The 227 pattern of ample NFT-like structures, vascular amyloidosis, A1-40 and diffuse plaque 228 12 predominance seen in hi TRM mice also resembled that of at least one subpopulation of human AD 229 patients in carriers of the APP Iowa mutation 23,24 . 230 CD8 T cells reactive to an antigenic epitope nearly identical to that recognized by brain-localized 231 T cells in hi TRM mice, were increased in AD brain but decreased in AD and MCI blood, suggesting 232 that their movement from blood to brain is involved in neuropathology. Consistent with this 233 notion, decreased levels of KLRG1 + APP(471-479)/HLA-A2 multimer-binding CD8 T cells in blood 234 correlated with cognitive impairment in MCI patients. While somewhat reminiscent of a distinct 235 subpopulation of CD8 effector-memory T cells, TEMRA, hi TRM analogues are distinct in that they 236 were weakly age-related, their decrease rather than increase correlated with cognitive loss, and 237 were prominently reactive to self antigen 3 . TEMRA are also absent from mice 3,25,26 . Further    were blinded to both group definition and anticipated outcomes. injection. The BM test is a hippocampus-dependent, spatial-learning task that allows subjects to 498 use spatial cues to locate a means of escape from a mildly aversive environment (i.e. the mice are 499 required to use spatial cues to find an escape location). Mice were assessed for their ability to learn 500 the location of an escape box over the course of 9 days in the BM apparatus 28,29 . The escape hole 501 is constant for each mouse over the five training days. Each mouse was tested three times per day 502 (3 trials) for 4 days, followed by no testing for 2 days, and re-testing on day 7. A 35-60 min inter-503 trial interval separates each trial. Each trial began by placing one mouse inside a start box with a 504 bottomless cube positioned centrally on the maze. After 30 seconds, the start box was lifted and 505 the mouse was released from the start box to find the one hole with access to the escape box. Two 506 fluorescent lights located approximately 4 feet above illuminated the testing room. Each trial lasted 507 up to 4 min or until the mouse entered the escape box. The experimenter guided mice that failed 508 to find the escape hole within 4 min, to the correct hole after each training test. Once the mouse 509 entered the escape box, it was allowed to remain in the box for 1 min. Following the 7 th day of 510 testing, and never on the same day, mice were tested an additional two-days, in which the escape 511 box was placed in the reverse position on day 8, and replaced in the original position on day 9.

512
The same exact testing procedure was applied to all mice in all groups. The maze and all 513 compartments were cleaned thoroughly with isopropyl alcohol to remove any olfactory cues after 514 each trial, and prior to each day of testing. (1.5 mL) and 5000-0050 (4.5 mL)), immediately frozen in liquid nitrogen, and subsequently stored 560 at -80°C until analysis.