Hippocampal connectivity with retrosplenial cortex drives neocortical tau accumulation and memory function

: The mechanisms underlying accumulation of Alzheimer’s disease (AD)-related tau 1 pathology outside of the medial temporal lobe (MTL) in older adults are unknown but crucial to 2 understanding cognitive decline. Neural connectivity has recently been implicated in the 3 propagation of tau in humans, consistent with data from animal studies. Using resting state 4 functional connectivity and tau PET imaging, we examined MTL structures involved in medial 5 parietal tau deposition in cognitively normal older adults. Functional connectivity between 6 retrosplenial cortex and hippocampus, but not entorhinal cortex, correlated with tau in medial 7 parietal lobe. Further, hippocampal-retrosplenial connectivity strength modulated the correlation 8 between MTL and medial parietal lobe tau, as well as between medial parietal tau and episodic 9 memory. Medial parietal tau spread thus reflects patterns of neural connectivity that represent a 10 critical step in the evolution of cognitive dysfunction in aging and AD. 11 to the spread of AD-related tau the aging brain, greater functional between is related to greater tau accumulation and correspondence of tau between these regions 11,12 Further, models predicting the spread of tau using functional closely resemble the observed pattern of tau deposition in the brain 7,8 , and functional connectivity may even be a better predictor of tau propagation than traditional Braak staging 13 . Our finding that connectivity between retrosplenial cortex and hippocampus, but not alEC or pmEC, was associated with medial parietal tau suggests that tau originating in MTL accumulates in hippocampus before later spreading directly to medial parietal lobe via with retrosplenial cortex. In addition, the finding that pmEC-RsC connectivity was weaker than HC-RsC connectivity indicates that although pmEC is functionally connected with several posterior medial areas 11,33 , it is not likely to be the primary structure involved in the propagation of tau to the medial parietal lobe. Instead, tau likely spreads to alEC-connected neocortex and hippocampus at a comparable rate, and later accumulates in posterior medial areas with connections to the hippocampus. This is consistent with cross-sectional histopathological data indicating that although the earliest cortical region to exhibit tau pathology is the area 14,34 , tau is typically observed in hippocampus prior to limbic areas such as the retrosplenial region 15 These findings also corroborate recent work that found that structural connectivity between hippocampus and posterior cingulate cortex (PCC) was with tau pathology in PCC 10


Introduction 13
The microtubule-associated protein tau forms neurofibrillary tangles (NFTs) in its 14 hyperphosphorylated state, which together with amyloid-b (Ab) plaques are the hallmark 15 neuropathologies of Alzheimer's disease (AD). Early histopathology studies of these NFTs 16 described tau pathology as first appearing in medial temporal lobe (MTL) structures such as the 17 entorhinal cortex and hippocampus before later accumulating in limbic areas and association 18 cortex 1 . In recent years, the use of positron emission tomography (PET) imaging has allowed the 19 in vivo characterization of the distribution and extent of pathological tau burden in patients with 20 AD as well as cognitively healthy older adults 2,3 . A developing body of work has used other 21 neuroimaging modalities in concert with PET to investigate what factors and characteristics of 22 the aging brain lead to stereotypical patterns of tau pathology. 23 Though the mechanisms underlying patterns of tau spread are not yet fully understood, 1 there is converging evidence of transsynaptic propagation of tau through coactive neurons. 2 Studies in vitro and in animal models suggest that tau pathology can be transferred between 3 synaptic connections, and that enhanced neuronal activity stimulates the release of pathological 4 tau and increases downstream accumulation [4][5][6] . In humans, transneuronal tau spread has been 5 investigated by comparing the topography of tau accumulation with measures of structural and 6 functional connectivity [7][8][9][10][11][12][13] . Graph theoretic measures of resting state functional magnetic 7 resonance imaging (fMRI) indicate that connected nodes in the brain tend to exhibit greater tau 8 accumulation 7 , and computational modeling of tau spread based on both structural and 9 functional connectivity is highly predictive of the observed pattern of tau accumulation 8 . 10 Further, the degree of between-region functional connectivity is associated with the covariance 11 of change in tau PET signal over time 9 . Given that the degree of connectivity across the cortex 12 has been shown to relate to the pattern of tau accumulation, connectivity measures in brain 13 regions that first exhibit tau pathology may provide insights into the structures and processes 14 involved in the spread of tau in aging and AD. 15 NFTs are thought to originate in the transentorhinal region, an area spanning the lateral 16 entorhinal cortex and medial perirhinal cortex in humans 14,15 . Tau pathology is in fact observed 17 in MTL structures such as the entorhinal cortex and hippocampus in virtually all older adults 16 . 18 However, its accumulation in cortical areas outside of the MTL, perhaps facilitated by Ab 17 , is 19 often a feature of the earliest stages of AD, with medial parietal cortex being a particularly 20 common area of deposition 18,19 . A recent study from our group found that functional 21 connectivity of the anterolateral entorhinal cortex (alEC), the entorhinal area most proximal to 22 the transentorhinal region, showed the strongest association with neocortical tau PET signal in 23 Table 1. Participant characteristics. Demographic information for sample of 97 cognitively normal older adults including age, years of education, mini mental state examination (MMSE) score, global beta amyloid positron emission tomography (Aβ PET) signal, sex, beta amyloid positivity status (Aβ+), and apolipoprotein E positivity status ( 4+).

Mean (SD) or n (%) Range
Age (years) 76.4 (6.1) 60 -93 Education (years) 16. 8  We first investigated resting state functional connectivity between MTL and medial 2 parietal lobe in our sample. We selected three regions of interest within MTL: the anterolateral 3 (alEC) and posteromedial (pmEC) entorhinal cortices as well as the hippocampus (Figure 1a). 4 Within the medial parietal lobe, we identified the retrosplenial cortex as a region where tau is 5 thought to spread to the neocortex via the MTL 15 and with known structural and functional 6 connectivity with MTL 21,24,26 . Hippocampus and retrosplenial cortex ("isthmus cingulate", see 7 Methods) regions of interest were derived using the FreeSurfer segmentation of each 8 participant's native space MRI, and template space alEC and pmEC were defined from a 9 previous high-resolution MRI study 27 . We performed bilateral region-to-region functional 10 connectivity analyses with semipartial correlations adjusting for age and sex. 11 Next, we carried out one-sample t-tests using β-weights from region-to-region semipartial 12 correlations to see if functional connectivity between each region was significantly different 13 from 0 ( Figure 1b). We found significant connectivity between hippocampus and retrosplenial 14 cortex (b = 0.45, p < 0.001), as well as between pmEC and retrosplenial cortex (b = 0.07, p = 15 0.007). We further compared the connectivity of these two pathways using a paired samples t-16 test and found that hippocampal-retrosplenial (HC-RsC) connectivity was significantly greater 17 than pmEC-RsC connectivity (t(96) = 15.74, p < 0.001). By contrast, no significant connectivity 1 was observed between alEC and retrosplenial cortex (b = 0.00, p = 0.891). Thus, hippocampus 2 and to a lesser extent pmEC, but not alEC, exhibited resting state functional connectivity with 3 retrosplenial cortex in our sample of cognitively normal older adults. 4

6
Hippocampal-retrosplenial connectivity strength is related to medial parietal tau pathology 7 Having observed strongest functional connectivity between hippocampus and 8 retrosplenial cortex, we next sought to investigate the extent to which this connectivity is 9 associated with tau accumulation in medial parietal lobe. We operationalized connectivity 10 strength by again extracting the β-weights from region-to-region semipartial correlations for each 11 participant. Tau pathology was quantified as the proportion of voxels above an a priori threshold 12 Anterolateral entorhinal cortex (alEC), posteromedial entorhinal cortex (pmEC), and hippocampus (HC) were included from the medial temporal lobe, as well as retrosplenial cortex (RsC). (B) Retrosplenial cortex exhibits functional connectivity with hippocampus (β=0.45, p<0.001) and posteromedial entorhinal cortex (b=0.07, p=0.007), but not anterolateral entorhinal cortex (β=0.00, p=0.89). Line color and thickness correspond to the Tstatistic of semipartial correlations of resting state activity between regions.
for tau PET (SUVR > 1. 4), following previous work that demonstrated this to be a reliable 1 marker of AD-related tau pathology 23 . We computed this suprathreshold tau in a medial parietal 2 lobe composite region comprising the retrosplenial cortex, precuneus, and posterior cingulate 3 cortex such that for each participant, the number of voxels above threshold was divided by the 4 total number of voxels in this region. To visualize this tau signal, we also computed the 5 proportion of participants above threshold in each voxel of the composite region. (Figure 2a). 6 Using multivariate linear regression, we then examined the association of suprathreshold 7 medial parietal lobe tau with connectivity strength between each MTL subregion and 8 retrosplenial cortex. Adjusting for age, sex, and global Ab PET signal, we found that HC-RsC 9 connectivity strength was associated with suprathreshold tau in the medial parietal lobe (b = 10 0.145, p = 0.004; Figure 2b). The global Ab term from this model was also associated with 11 suprathreshold medial parietal tau (b = 0.113, p = 0.014). In contrast with HC-RsC, linear 12 regression models adjusting for age, sex, and global Ab revealed that neither alEC-RsC (b = 13 0.120, p = 0.145) nor pmEC-RsC connectivity strength (b = 0.015, p = 0.842) were associated 14 with medial parietal lobe tau (Figure 2c-d). In a separate model, we further examined whether 15 individuals with more global Ab exhibited a stronger relationship between HC-RsC and medial 16 parietal tau. Adjusting for age and sex, we did not observe a significant interaction between HC 17 RsC connectivity and global Ab (b = 0.442, p = 0.141). 18 1 2 To confirm that the relationship between HC-RsC connectivity and medial parietal tau 3 was specific to retrosplenial cortex and not a general effect of strong resting state functional 4 connectivity, we identified a control region within the superior frontal gyrus (SFG) analogous to We further wanted to verify that connectivity between hippocampus and retrosplenial 9 cortex was associated specifically with tau in medial parietal lobe and not also in other early tau-10 accumulating regions. To this end, we examined tau within entorhinal cortex and inferior 11 temporal cortex, a region often used as a marker for early tau accumulation in aging 29,30 . Again 12 using linear regression adjusting for age, sex, and global Ab, we found that HC-RsC connectivity   Hippocampal-retrosplenial connectivity modulates the relationship between medial temporal 1 and medial parietal tau 2 To further examine the role of functional connectivity in the spread of tau pathology from 3 MTL, we tested whether HC-RsC connectivity strength modulated how closely MTL tau 4 corresponded with medial parietal lobe tau. Measuring tau in MTL, particularly in hippocampus, 5 is challenging given the confound of signal contamination from off-target FTP binding in 6 choroid plexus. To address this, we quantified MTL tau pathology with hippocampal FTP SUVR 7 using Rousset geometric transfer matrix partial volume correction, which minimizes choroid 8 plexus spillover 31 . As an additional precaution, we adjusted for choroid plexus FTP signal in our 9 linear regression model, in addition to age, sex, and global Ab (Table 2).   Figure 3.

Hippocampal-retrosplenial connectivity modulates the relationship between medial temporal 3
and medial parietal tau 4 Having observed that increased functional connectivity between hippocampus and 5 retrosplenial cortex modulates tau pathology burden in medial parietal lobe, we sought to test if 6 connectivity and tau in these areas might also interact to predict cognitive function. We 7

Figure 3. Hippocampal-retrosplenial connectivity modulates the relationship between medial temporal and medial parietal tau.
Visualization of interaction between hippocampus mean tau (partial volume corrected FTP SUVR) and hippocampal-retrosplenial functional connectivity strength (HC-RsC) from linear regression model (see Table 2). Medial parietal lobe (MPL) suprathreshold tau is associated with a hippocampal tauｘHC-RsC interaction. Plot displays the relationship predicted by linear regression at low (10 th percentile), median, and high (90 th percentile) HC-RsC. constructed a composite episodic memory measure consisting of the mean z-score of four 1 episodic memory tasks: the California Verbal Learning Test (CVLT) immediate free recall, 2 CVLT long-delay free recall, Visual Reproduction (VR) immediate recall, and VR delay recall 3 32 . Adjusting for age, years of education, practice effects, and global Ab, we did not observe a 4 significant main effect of either medial parietal lobe tau (b = 1.111, p = 0.318) or HC-RsC 5 connectivity (b = 0.073, p = 0.861) on episodic memory. However, we did find a significant 6 interaction between medial parietal tau and hippocampal-retrosplenial connectivity (b = -9.482, p 7 = 0.018), such that episodic memory performance was poorest when both medial parietal lobe 8 tau and hippocampal-retrosplenial connectivity were greatest (Table 3). To further examine 9 which subdomain of episodic memory might drive this finding, we broke down the episodic 10 memory composite score into a verbal memory component consisting of the two CVLT tasks, 11 and a visuospatial memory component consisting of the two visual reproduction tasks. The 12 interaction between medial parietal lobe tau and hippocampal-retrosplenial connectivity was 13 associated with visuospatial memory performance (b = -12.016, p = 0.006; Figure 4a), but not 14 verbal memory performance (b = -6.933, p = 0.145; Figure 4b). Taken together, these findings 15 indicate that the combination of greater tau in medial parietal lobe and greater connectivity 16 between hippocampus and retrosplenial cortex is associated with poorer episodic memory, an 17 effect that is perhaps driven by a relationship with visuospatial memory in particular. 18 19 In this study of cognitively unimpaired older adults, we measured functional connectivity 3 between the MTL and medial parietal lobe using resting state fMRI, and measured tau pathology 4 burden using tau PET imaging. Retrosplenial cortex exhibited strong functional connectivity 5 with hippocampus, weaker yet significant connectivity with pmEC, and no connectivity with 6 alEC. We found that the strength of HC-RsC connectivity was associated with the degree of tau 7 accumulation in the medial parietal lobe, whereas neither alEC-RsC nor pmEC-RsC connectivity 8 correlated with medial parietal lobe tau. Control analyses demonstrated that this result was 9 specific to HC-RsC connectivity and to tau pathology in the downstream medial parietal lobe. 10 We further observed that the correspondence between tau in hippocampus and medial parietal 11 lobe was greater with stronger HC-RsC connectivity. Finally, we demonstrated that greater 12 medial parietal lobe tau in combination with stronger HC-RsC connectivity was associated with 13 poorer episodic memory performance, particularly in the visuospatial domain. Together, these 14 findings provide strong evidence that AD-related tau pathology disrupts memory processing by 15 spreading first from entorhinal cortex to hippocampus before later accumulating in medial  These results add to a growing literature linking patterns of resting state functional 3 connectivity to the spread of AD-related tau pathology. In the aging brain, greater functional 4 connectivity between regions is related to greater tau accumulation and correspondence of tau 5 between these regions 11,12 . Further, models predicting the spread of tau using functional 6 connectivity closely resemble the observed pattern of tau deposition in the brain 7,8 , and 7 functional connectivity may even be a better predictor of tau propagation than traditional Braak 8 staging 13 . Our finding that connectivity between retrosplenial cortex and hippocampus, but not 9 alEC or pmEC, was associated with medial parietal tau suggests that tau originating in MTL 10 accumulates in hippocampus before later spreading directly to medial parietal lobe via 11 connectivity with retrosplenial cortex. In addition, the finding that pmEC-RsC connectivity was 12 weaker than HC-RsC connectivity indicates that although pmEC is functionally connected with 13 several posterior medial areas 11,33 , it is not likely to be the primary structure involved in the 14 propagation of tau to the medial parietal lobe. Instead, tau likely spreads to alEC-connected 15 neocortex and hippocampus at a comparable rate, and later accumulates in posterior medial areas 16 with connections to the hippocampus. This is consistent with cross-sectional histopathological 17 data indicating that although the earliest cortical region to exhibit tau pathology is the 18 alEC/transentorhinal area 14,34 , tau is typically observed in hippocampus prior to limbic areas 19 such as the retrosplenial region 15 . These findings also corroborate recent work that found that 20 structural connectivity between hippocampus and posterior cingulate cortex (PCC) was 21 associated with tau pathology in PCC 10 . 22 The specificity of this key finding in our study is striking. The lack of association 1 between HC-RsC connectivity and tau in entorhinal cortex suggests this connectivity is 2 specifically related to tau in the downstream medial parietal region of this pathway. In addition 3 to the medial parietal lobe, the inferior temporal cortex is one of the first areas outside of the 4 MTL where tau pathology begins to accumulate, and tau burden in this region is often used as a 5 marker of AD disease progression 29,30 . That the strength of HC-RsC connectivity was only 6 associated at trend level with tau in this region suggests that this connectivity is specific to tau 7 pathology in the medial parietal lobe and not in other early-accumulating neocortical areas. In 8 addition, though hippocampus showed strong functional connectivity with the medial portion of 9 Brodmann area 10, a region of the default mode network in the superior frontal gyrus, there was 10 no association between the strength of this connectivity and medial parietal tau. Thus, it appears 11 that connectivity specifically between hippocampus and retrosplenial cortex is associated with 12 medial parietal tau, not simply connectivity between hippocampus and other highly-connected 13 cortical area. Finally, we failed to find an association between HC-RsC and Ab burden within 14 medial parietal lobe. This is not surprising given that Ab does not originate in the MTL and is 15 not thought to spread in the same transneuronal manner as tau, lending further support to the 16 notion that functional connectivity is uniquely useful in predicting tau accumulation. Taken 17 together, these results support a narrative of tau pathology in hippocampus spreading to medial 18 parietal lobe in cognitively unimpaired older adults, reflected by resting state functional 19 connectivity between these regions. 20

Functional connectivity may modulate tau propagation at different time scales 22
In addition to the association between functional connectivity and tau pathology, our 1 finding of greater correspondence between hippocampal and medial parietal tau with greater HC-2 RsC connectivity strength lends further support to the notion that neural connectivity may 3 influence tau to spread from early-accumulating regions to connected downstream areas. In 4 particular, individuals with greater connectivity strength in our sample were more likely to have 5 tau pathology burden in hippocampus align with medial parietal lobe tau than individuals with 6 weaker connectivity. It is notable that although replicating this analysis using entorhinal tau PET 7 demonstrated this same relationship at trend level, the strongest relationship was evident with 8 hippocampal tau. Because tau in entorhinal cortex and hippocampus tend to be highly correlated 9 with one another, it is not surprising that using entorhinal tau yielded a similar result, and further 10 suggests that contamination of hippocampal signal from non-specific choroid plexus FTP 11 binding is not likely to be driving this finding. 12 An alternative explanation for these findings is that tau accumulation influences the 13 degree of functional connectivity between these regions. Indeed, some studies have found a 14 negative association between tau pathology and functional connectivity across the cortex 29 connectivity. These distinct short-and long-term effects may help explain regional differences in 23 the tau-connectivity relationship such that pathways of early tau propagation are first to show 1 local degeneration, whereas later pathways, such as the MTL-medial parietal lobe, may 2 concurrently demonstrate increased connectivity leading to further tau spread. In the context of a 3 cascading network failure model of AD 37 , functional connectivity has proven to be a powerful 4 tool in examining the effect of tau accumulation on functional isolation of brain regions 32 and in 5 predicting inter-individual variability in the pattern of tau spread and disease progression 13 . 6 7 Greater connectivity and tau together lead to worse episodic memory 8 In this study, functional connectivity and tau were also found to have consequences for 9 cognitive function. Episodic memory performance, particularly visuospatial memory, was related 10 to the combination of greater medial parietal lobe tau accumulation and greater HC-RsC 11 strength. Though neither connectivity strength nor medial parietal tau alone predicted memory 12 performance in our sample, it is notable that the interaction between them was associated with 13 poorer episodic memory in individuals without clinical cognitive impairment, even after 14 controlling for global Ab burden. Existing work has yet to establish a clear relationship between 15 memory performance and connectivity changes in aging. Greater functional connectivity 16 between MTL and posterior medial areas is related to poorer memory performance both cross-17 sectionally and over time 36 . By contrast, a positive association has been reported between MTL-18 medial parietal connectivity and memory performance 38,39 , though these studies did not measure 19 tau pathology with PET imaging. Abnormal diffusivity of the hippocampal cingulum bundle has 20 also been shown to be related to greater decline in memory performance in older individuals with 21 high PCC tau and high Ab 10 . It may be that propagation of tau, not tau burden per se, is 22 associated with the earliest deficits in cognitive function, representing a state of mild circuit 23 disruption that has not yet reached the stage of widespread neurodegeneration and cascading 1 network failure. This view comports with a model of AD stemming from peptide-dependent 2 circuit dysfunction, as tau propagation leads to changes in circuit excitability associated with the 3 earliest-detectable cognitive changes 40 . Indeed, the correlation between tau spread to the medial 4 parietal lobe and worse episodic memory performance may be an indicator of consequences for 5 the PM memory system, which while not affected by tau as early as the AT system may begin to 6 be disrupted in MCI and AD 18 . 7 It is also striking that our cognitive findings are driven by an association with visuospatial 8 memory in particular. Medial parietal areas have long been implicated in processing of spatial 9 information 41-43 , and the representation of visuospatial context is thought to be a key function of 10 the PM memory system 24 . Spatial information processing may also be one of the earliest-11 affected domains in cognitive aging and Alzheimer's disease 44 , but the relationship between 12 functional connectivity and spatial memory performance has not been extensively studied in 13 humans. In rodents, RsC-PCC resting state functional connectivity has been shown to be 14 associated with impaired spatial memory performance 45 . Inferring a causal link between 15 propagation of tau to the medial parietal lobe and domain-specific cognitive decline is beyond 16 the scope of this study, but these processes may underlie the beginnings of cognitive decline in 17 healthy older adults. 18 19 Limitations 20 It should be noted that the cross-sectional nature of these data means that we can only 21 infer the spread of tau rather than directly observing it. Longitudinal studies of tau accumulation 22 are needed to fully validate the notion of tau propagating to different regions over time via 23 connectivity. In addition, though we adjusted for Ab PET signal throughout this study, we did 1 not find that the relationships described here were stronger in individuals with greater global Ab 2 burden. Though Ab was associated with medial parietal lobe tau independent of HC-RsC 3 connectivity, there was no significant interaction with Ab in any of our analyses. This was a 4 somewhat surprising finding given a number of studies that have found a stronger association 5 between tau and connectivity for those with greater Ab pathology [9][10][11] . It is possible that our 6 sample did not provide us with enough statistical power to observe this interaction, though it was 7 enriched to include nearly half Ab-positive individuals. Another explanation is the processes 8 examined here represent features of normal aging independent of AD-specific pathologies. 9 Further study is needed to assess the role of Ab in the association between medial parietal lobe 10 tau and functional connectivity in cognitively normal older adults. 11 12

Conclusions 13
The findings described here support the view that tau pathology spreads from its origin in 14 the entorhinal cortex to the hippocampus in cognitively normal older individuals, before later 15 depositing in medial parietal lobe via direct connectivity between hippocampus and retrosplenial 16 cortex. Though the accumulation of tau pathology in MTL and even some areas of the AT 17 network has been observed in older adults without cognitive impairment, tau spread into the PM 18 network and subsequent domain-specific memory decline may reflect a significant transition 19 between normal aging and the processes involved in Alzheimer's disease. Future work with 20 longitudinal data can help establish tau propagation into the medial parietal lobe as a crucial 21 marker of the beginnings of Alzheimer's disease. 22

Study Design 2
The main objective of this study was to determine what medial temporal lobe structure is 3 primarily involved in the propagation of tau into the medial parietal lobe in cognitively normal 4 older adults. We hypothesized that hippocampus would show the strongest functional 5 connectivity with medial parietal lobe, and that the strength of this connectivity would be 6 associated with the degree of medial parietal tau. After initial analysis, we further hypothesized 7 that the strength of this connectivity would modulate the correlation between medial temporal 8 and medial parietal tau, as well as between tau and episodic memory performance. To test these 9 hypotheses, we included data from 97 cognitively normal older adults from the Berkeley Aging 10 Cohort Study. All participants underwent 3T structural and resting state functional MRI, 3T 11 structural MRI, and a standard neuropsychological assessment. These participants also received 12 tau PET imaging using 18 F-Flortaucipir (FTP) and Ab PET imaging using 11 C-Pittsburgh 13 Compound B (PiB). There were 3 individuals who did not have PiB PET data available for 14 analysis, and so were excluded from all analyses that adjusted for global Ab signal. We included 15 only participants whose resting state fMRI data was collected within 146 days of their 16 corresponding tau PET scan (M = 42.5, SD = 37.9). 17 Additional inclusion criteria for this study were 60+ years of age, cognitively normal 18 status (Mini Mental State Examination score ³ 25 and normal neuropsychological examination, 19 defined as within 1.5 SDs of age, education, and sex adjusted norms), no serious neurological, 20 psychiatric, or medical illness, no major contraindications found on MRI or PET, and 21 independent community living status. This study was approved by the Institutional Review 22 Boards of the University of California, Berkeley, and the Lawrence Berkeley National 1 Laboratory (LBNL). All participants provided written informed consent. 2 3 3T MRI acquisition 4 Structural and functional MRI data were acquired on a 3T TIM/Trio scanner (Siemens 5 Medical System, software version B17A) using a 32-channel head coil. A T1-weighted whole 6 brain magnetization prepared rapid gradient echo (MPRAGE) image was acquired for each 7 subject (voxel size = 1mm isotropic, TR = 2300ms, TE = 2.98ms, matrix = 256´240´160, FOV 8 = 256´240´160mm 3 , sagittal plane, 160 slices, 5 min acquisition time). Resting state functional 9 MRI was then acquired using T2*-weighted echo planar imaging (EPI, voxel size = 2.6mm 10 isotropic, TR = 1.067ms, TE = 31.2ms, FA = 45, matrix 80´80, FOV = 210mm, sagittal plane, 11 300 volumes, anterior to posterior phase encoding, ascending acquisition, 5 min acquisition 12 time). During resting state acquisition, participants were told to remain awake with eyes open 13 and focused on a white asterisk displayed on a black background. 14 15

Structural MRI preprocessing 16
Structural T1-weighted images were processing using Statistical Parametric Mapping 17 (SPM12). Images were first segmented into gray matter, white matter, and CSF components in 18 native space. DARTEL-imported tissue segments for all individuals in the sample were used to 19 create a study-specific template, which was then used to warp native space T1 images and tissue 20 segments to MNI space at 2mm isotropic resolution. Finally, native space T1 images were 21 segmented with Freesurfer v.5.3.0 using the Desikan-Killany atlas parcellation. 22 Resting state fMRI preprocessing 1 Resting state fMRI images were preprocessed using a standard SPM12 pipeline. Slice 2 time correction was first applied to adjust for differences in acquisition time for each brain 3 volume. Then, all EPIs were realigned to the first acquired EPI, and translation and rotation 4 realignment parameters were output. Each EPI was next coregistered to each individual's native 5 space T1 image. Next all resting state EPIs and structural images were warped to the study-6 specific DARTEL template in 2mm isotropic MNI space from structural preprocessing. 7 Unsmoothed fMRI data in MNI space was used to extract the time series correlation of all ROI 8 seeds used in these analyses. 9 Functional connectivity analyses for these preprocessed resting states images were 10 carried out using the CONN functional connectivity toolbox (version 17e) 46  Anterolateral entorhinal cortex (alEC) and posteromedial entorhinal cortex (pmEC) were 12 defined in a previous study with high-resolution 7T MRI 27 . In brief, anatomical borders of the 13 entire entorhinal cortex were manually defined on a high-resolution T1-group template. 14 Multivariate classification in a group of young adults was used to then identify clusters of voxels 15 within this mask that showed preferential functional connectivity with perirhinal cortex, 16 comprising the alEC ROI, or with the parahippocampal gyrus, comprising the pmEC ROI. These 17 alEC and pmEC ROIs were then warped to a 2mm isotropic MNI template and made publicly 18 available. In this study, we used these bilateral MNI space ROIs in our functional connectivity 19 analyses. Because these regions are in close spatial proximity to one another, we extracted time 20 series from the unsmoothed, denoised MNI space resting state data to avoid smoothing signal 21 from each seed into each other. 22 To address signal dropout in these and every ROI in these analyses, we derived an 1 explicit mask to remove regions of low signal across the whole brain. This mask was defined by 2 calculating the mean functional MNI space image across all individuals, restricted to a group 3 level grey matter mask. We then excluded voxels with less than 40% of the mean signal intensity 4 of the image. Using this mean signal intensity threshold mask, a mean of 15.8% of voxels (SD = 5 11.9%) were removed across all ROIs with the highest proportion of voxels being removed from 6 the A10m region (34.8%). 7 8 Functional connectivity analysis 9 Seed-to-seed functional connectivity analysis was carried out with the CONN toolbox 46 10 using the MNI space resting state fMRI data. We used semipartial correlations for all first-level 11 analyses to compute the time series correlation between each seed, controlling for the variance of 12 all other seed regions entered into the same model. We first constructed a model of functional 13 connectivity between alEC, pmEC, HC, and RsC using bilateral ROIs. Semipartial correlations 14 and unsmoothed data were used to minimize spillover of signal between adjacent MTL regions. 15 Statistical significance of functional connectivity was determined using one-sample t-tests of β-16 weights from region-to-region semipartial correlations to see if connectivity was significantly 17 different from 0. All analyses were performed using an explicit mask to remove areas of high 18 signal dropout across the whole brain, as described above. 19 20 PET acquisition and processing 21 PET was acquired for all participants at LBNL. Tau accumulation was assessed with 18 F-22 Flortaucipir (FTP) synthesized at the Biomedical Isotope Facility at LBNL as previously 23 described 3 . Data were collected on a Biograph TruePoint 6 scanner (Siemens, Inc) 75-115 min 1 post-injection in listmode. Data were then binned into 4 x 5 min frames from 80-100 min post-2 injection. CT scans were performed before the start of each emission acquisition. Ab burden was 3 assessed using 11 C-Pittsburgh Compound B (PiB), also synthesized at the Biomedical Isotope 4 Facility at LBNL 48 . Data were collected on the Biograph scanner across 35 dynamic frames for 5 90 min post-injection and subsequently binned into 35 frames (4 x 15, 8 x 30, 9 x 60, 2 x 180, 10 6 x 300, and 2 x 600s), and a CT scan was performed. All PET images were reconstructed using an 7 ordered subset expectation maximization algorithm, with attenuation correction, scatter 8 correction, and smoothing with a 4mm Gaussian kernel. 9 Processing of FTP images was carried out in SPM12. Images were realigned, averaged, 10 and coregistered to 3T structural MRIs. Standardized uptake value ratio (SUVR) images were 11 calculated by averaging mean tracer uptake over the 80-100 min data and normalized with an 12 inferior cerebellar gray reference region 31 . The mean SUVR of each ROI (structural MRI 13 FreeSurfer segmentation) was extracted from the native space images. This ROI data was partial 14 volume corrected using a modified Geometric Transfer Matrix approach 49 as previously 15 described 31 . SUVR images were then warped to 2mm MNI space for voxelwise analyses using 16 the study-specific DARTEL template produced from structural data (see above). No additional 17 spatial smoothing was applied. 18 Using SPM12, PiB images were realigned. An average of frames within the first 20 min 19 was used to calculate the transformation matrix to coregister the PiB images to the participants' 20 3T structural MRI; this transformation matrix was then applied to all PiB frames. Distribution 21 volume ratio (DVR) images were calculated with Logan graphical analysis over 35-90 min data 22 and normalized to a whole cerebellar gray reference region 50,51 . Global PiB was calculated 23 across cortical FreeSurfer ROIs as previously described 52 , and a threshold of DVR > 1.065 was 1 used categorize participants as Ab-positive or Ab-negative. In addition, mean DVR within each 2 FreeSurfer ROI was extracted from coregistered, MNI space PiB images. 3 4 Suprathreshold tau quantification 5 To quantify tau deposition, we used the proportion of voxels above an a priori threshold 6 of SUVR > 1.4 for FTP PET signal. This suprathreshold tau measure has previously been shown 7 to be a reliable marker of AD-related tau pathology 23 , and has been used in previous studies 8 investigating functional connectivity and tau 11 . One distinct advantage of using suprathreshold 9 FTP over mean SUVR is that it is not confounded by different number of voxels within ROIs. 10 For each individual, we computed the number of suprathreshold FTP voxels within each ROI 11 and divided by the total number of voxels in the region. 12

Statistical analysis 14
All statistical analysis was carried out in R version 3.6.3, with a two-sided significance 15 level of a=0.05 throughout. We assessed the relationship between tau, resting state functional 16 connectivity, and cognitive function in our sample using linear regression models carried out 17 with the lm() function in the {stats} package. All analyses were adjusted for age at time of tau 18 scan, sex, and mean global Ab burden. Analyses involving cognitive test performance were 19 additionally adjusted for years of education as well as practice effects quantified as the square 20 root of the number of prior testing occasions 53 . 21

Cognitive measures 23
To assess episodic memory in our sample of cognitively normal older adults, we used 1 neuropsychological assessment data from closest in time to each individual's tau scan. There was 2 a mean of 84.2 days (SD = 56.9) between each individual's cognitive assessment and tau PET 3 scan. We computed an episodic memory composite measure by averaging the z-transformed 4 individual test scores using mean and SD from the sample 32,54 for four different tasks. These 5 tasks were the California Verbal Learning Test (CVLT) immediate free recall, CVLT long-delay 6 free recall, Visual Reproduction I (immediate recall), and Visual Reproduction II (delay recall). 7 We analyzed distinct verbal and visuospatial episodic memory performance by considering 8 performance in CVLT and Visual Reproduction tasks separately.