RNASE6 is a novel modifier of APOE-ε4 effects on cognition

Apolipoprotein E4 (APOE-ε4), the strongest common genetic risk factor for Alzheimer's disease (AD), contributes to worse cognition in older adults. However, many APOE-ε4 carriers remain cognitively normal throughout life, suggesting that neuroprotective factors may be present in these individuals. In this study, we leverage whole-blood RNA sequencing (RNAseq) from 324 older adults to identify genetic modifiers of APOE-ε4 effects on cognition. Expression of RNASE6 interacted with APOE-ε4 status (p = 4.35 × 10-8) whereby higher RNASE6 expression was associated with worse memory at baseline among APOE-ε4 carriers. This interaction was replicated using RNAseq data from the prefrontal cortex in an independent dataset (N = 535; p = 0.002), suggesting the peripheral effect of RNASE6 is also present in brain tissue. RNASE6 encodes an antimicrobial peptide involved in innate immune response and has been previously observed in a gene co-expression network module with other AD-related inflammatory genes, including TREM2 and MS4A. Together, these data implicate neuroinflammation in cognitive decline, and suggest that innate immune signaling may be detectable in blood and confer differential susceptibility to AD depending on APOE-ε4.


Introduction
Alzheimer's disease (AD) is a debilitating neurodegenerative disorder that is the 6th leading cause of death among all adults in the United States ( Association, 2021 ). The most common form of Alzheimer's disease is sporadic late-onset AD (LOAD), which is complex in etiology and heterogeneous in clinical presentation ( Bekris et al., 2010 ;Reitz et al., 2020 ). Sporadic LOAD is polygenic, and to date, over 40 risk loci for AD have been identified via large genome-wide association studies ( Jansen et al., 2019 ;Kunkle et al., 2019 ;Lambert et al., 2013 ;Wightman et al., 2021 ). One key genetic driver is APOE (apolipoprotein E), which has three common polymorphic alleles: ε 2, ε 3, and ε 4. The APOE -ε 4 allele is the strongest common genetic risk factor for AD ( Corder et al., 1993 ;Saunders et al., 1993 ). A single allele of APOE -ε4 can increase AD risk by up to 3 times compared to APOE -ε3, and 2 APOE -ε4 alleles can increase risk by up to 15-fold ( Reiman et al., 2020 ;Wu and Zhao, 2016 ). In addition to increased AD risk, the APOE -ε4 allele is associated with increased brain amyloid and tau burden ( Baek et al., 2020 ), subsequently leading to downstream neurodegeneration and cognitive impairment ( Jack et al., 2013 ;Jack et al., 2010 ). However, many APOE -ε4 carriers remain cognitively normal throughout life despite the increased AD risk ( Emrani et al., 2020 ), suggesting that there may be neuroprotective molecular modifiers of APOE effects. For example, mutations within the caspase 7 ( CASP7 ) and Klotho ( KL ) genes were suggested to have protective effects (e.g. , reduced AD risk, slower cognitive decline) in APOE -ε4 carriers in comparison to non-carriers ( Seto et al., 2021 ). In the brain, epigenetic modifiers of APOE -ε4 have been observed, such as the recently described epigenomic factor of activated microglia (EFAM; Ma et al., 2022 ). Indeed, identifying and describing APOE modifiers may provide critical insight into the pathophysiology of AD and provide novel targets for therapeutic intervention.
The primary function of the APOE protein is lipid transport and signaling, which plays important roles in the brain, innate immune system, and vascular system ( Husain et al., 2021 ;Safieh et al., 2019 ). Given the roles of APOE in both the peripheral and central nervous system (CNS), whole blood transcriptomics may provide an opportunity to identify novel genes and pathways that contribute to neuroprotection by modifying the effect of APOE . Blood transcriptomics provide some important advantages over brain transcriptomics alone, particularly when seeking for modifiers of APOE effects. While transcriptomic signatures in blood do not perfectly mimic the brain ( Sullivan et al., 2006 ;Tylee et al., 2013 ), many of the gene networks and molecular pathways that change over the course of AD are measurable in the blood and provide a window into relevant biological cascades such as inflammation ( Cullen et al., 2021 ). Moreover, peripheral inflammation changes very early in AD and contributes to AD progression ( King et al., 2018 ;Tao et al., 2018 ), with emerging evidence ( Kloske and Wilcock, 2020 ;Krasemann et al., 2017 ) suggesting that non-CNS inflammation is particularly relevant among APOE-ε4 carriers.
In the present study, we leverage whole blood RNA sequencing (RNAseq) data from the Vanderbilt Memory and Aging Project (VMAP) to identify genes that modify the association between APOE -ε4 and cognitive performance. We then extend our analyses to the brain to characterize whether genes identified in blood show comparable modifying effects in the brain. By leveraging blood and brain transcriptomics, we aim to better characterize the molecular modifiers of APOE on cognition and potentially uncover novel blood-based biomarkers and targets for future drug discovery effort s.

Participants
The Vanderbilt Memory and Aging Project (VMAP) is a longitudinal aging cohort that was established in 2012 to investigate the relationship between vascular health and brain aging. Three hundred thirty-five individuals were enrolled; the study preferentially recruited participants with mild cognitive impairment (MCI, N = 168) aged 60 and above along with matched counterparts who had normal cognition (N = 167). Individuals with cognitive diagnoses other than MCI or normal cognition (NC), history of neurological disease, MRI contraindications, heart failure, major psychiatric illness, or systemic or terminal illness were excluded. At enrollment, participants underwent a comprehensive evaluation including, but not limited to, APOE genotyping, neuroimaging, cognitive assessment, blood draw, and optional lumbar puncture ( Jefferson et al., 2016 ).
Independent data for replication were acquired from the Religious Orders Study (ROS) and the Rush Memory and Aging Project (MAP), known as ROS/MAP collectively. The ROS began in 1994 and enrolls priests and nuns from across the United States. The MAP cohort began in 1997 and enrolls lay persons from northeastern Illinois. Both longitudinal aging studies were launched to better understand risk factors for and the neurobiology of cognitive decline and dementia. Both studies were approved by an Institutional Review Board of Rush University Medical Center. All participants were without known dementia at enrollment, agree to comprehensive neuropsychological evaluations, and sign an Anatomic Gift Act and Repository Consent to allow their data to be shared . Additional ROS/MAP data can be requested via the Accelerating Medicines Partnership -Alzheimer's Disease (AMP-AD) Knowledge Portal ( https://adknowledgeportal.synapse. org/ ) as well as the Rush Alzheimer's Disease Center Resource Sharing Hub ( https://www.radc.rush.edu/ ).

Neuropsychological Assessment
Composite measures for memory and executive function in VMAP were generated following previously described procedures ( Crane et al., 2012 ;Kresge et al., 2018 ). Briefly, the memory composite leveraged data from the California Verbal Learning Test (2nd edition) and the Biber Figure Learning Test. The executive function composite score in VMAP was derived from the following tasks: Digit Span from the Wechsler Adult Intelligence Scale (third edition), Trail Making Test, Stroop Color Word Inhibition, and Controlled Oral Word Association.
The global cognition variable in ROS/MAP was generated by averaging the Z-scores of 17 neuropsychological tests across five domains of cognition (i.e., episodic, semantic, and working memory, perceptual orientation, and perceptual speed). This composite measurement has been described fully elsewhere ( Wilson et al., 2015 ).

Vanderbilt Memory and Aging Project
Blood draws were performed in the morning under fasting conditions. Approximately 2.5 mL of whole blood were kept frozen at −80 °C in a PAXgene tube (QIAGEN, 761115) until processing ( Jefferson et al., 2016 ). RNA extraction, library preparation, and RNA sequencing were performed by the VANTAGE Core (Vanderbilt University, TN). Total RNA was extracted from whole blood using the QIASymphony RNA Kit (QIAGEN, 931636), and both ribosomal RNA and hemoglobin were depleted with the NEBNext Globin and rRNA Depletion Kit (New England BioLabs, Inc., E7750). Library preparation was completed using the NEBNext Ultra Directional Library Prep Kit (New England BioLabs, Inc., E7420) before sequencing was performed using 150 base pair (bp) paired end reads on an Illumina NovaSeq 60 0 0 (Illumina), targeting an average of 50 million reads per sample.

Religious Orders Study and Memory and Aging Project
Fifty milligrams of frozen brain tissue were dissected and homogenized in DNA/RNA shield buffer (Zymo Research, R1100). RNA was extracted from the dorsolateral prefrontal cortex (DLPFC), posterior cingulate cortex (PCC), and head of the caudate nucleus (CN) using the Chemagic RNA tissue kit (PerkinElmer, Inc. CMG-1212) on a Chemagic 360 instrument. Five hundred nanograms of total RNA was used as input for sequencing library generation and rRNA was depleted with RiboGold (Illumina, 20020599). A Zephyr G3 NGS workstation (PerkinElmer, Inc.) was utilized to generate TruSeq stranded sequencing libraries (Illumina, 20020599). Libraries were normalized for molarity and sequenced using 2 × 150 bp paired end reads on a NovaSeq 60 0 0 (Illumina) targeting a total of 40-50 million reads. Additional details are previously described ( De Jager et al., 2018 ;Lee et al., 2021 ;Mostafavi et al., 2018 ). These data are available on the AMP-AD Knowledge Portal ( https://www.synapse.org/#!Synapse:syn3219045 ).

RNAseq Alignment and Quality Control
RNAseq alignment and quality control (QC) for both VMAP and ROS/MAP samples largely followed a previously reported procedure used by the AMP-AD Consortium ( Logsdon et al., 2019 ). Alignment was performed using STAR (version 2.5.2b) with twopassMode set to basic ( Dobin et al., 2013 ). Reads were aligned to the Ensembl (GRCh38, version 99) reference genome ( Howe et al., 2021 ), and gene counts were computed using the featureCounts ( Liao et al., 2014 ) command from the Subread package (version 2.0.0). Summary metrics were calculated using Picard (version 2.18.27, http: //broadinstitute.github.io/picard/ ) to evaluate sample quality and for later use as covariates ( Institute, 2019 ).
Before QC of the VMAP whole blood RNAseq, samples with RNA integrity number (RIN) less than 3.0 were excluded. In addition, genes with missing gene length or GC-content were removed, after which all gene counts were quantile normalized using the cqn R package (version 1.30.0) to remove technical variability due to gene length and GC-content ( Hansen et al., 2012 ). At this time, gene expression values greater than three standard deviations from the mean expression for each gene were removed. Additional samples were removed if deemed principal component outliers or if missing RIN, age, sex, other demographic information, or cognition data prior to batch correction. Expression values were adjusted for batch effects using the R package limma (version 3.40.6; Law et al., 2014 ;Ritchie et al., 2015 ). This left 60,669 genes and 324 samples in VMAP for discovery analyses.
QC of the bulk brain RNAseq from ROS/MAP followed the aforementioned pipeline. From these data, samples with RIN less than 4.0 or with postmortem interval (PMI) greater than 24 hours were excluded. Additional samples were removed if missing covariates or cognitive data resulting in a final dataset of 535 samples.
Sensitivity analyses leveraged RNAseq data from VMAP that was additionally adjusted (i.e., along with quantile normalization and controlling for batch) for the following covariates using limma: sex, race, APOE -ε4 allele count, RIN, age, education, percentage passfilter reads aligned, and percent coding, intergenic, intronic, and ribosomal bases.
A β and phosphorylated tau were identified via immunohistochemistry and quantified via image analysis. The overall amyloid level is defined as the mean percent of cortex occupied by A β across 8 brain regions (hippocampus, angular gyrus, and entorhinal, midfrontal, inferior temporal, calcarine, anterior cingulate, and superior frontal cortices). The overall tangle density is defined as the mean cortical density per mm 2 of the same 8 brain regions mentioned above. Neuritic plaque and NFT burden were determined by microscopic examination of silver-stained slides across 5 brain regions (hippocampus and entorhinal, midtemporal, inferior parietal, and midfrontal cortices). These neuropathological measures were previously characterized by ROS/MAP . CERAD scores (Definite and/or Probable AD, 1 and 2) and Braak staging (Braak < 4) were used to determine amyloid and tau positivity ( Bennett et al., 2006 ;Braak and Braak, 1991 ;Mirra et al., 1991 ).

Statistical Analyses
All statistical analyses were performed using R (version 3.6.3, https://www.r-project.org/ ). False-discovery rate ( Benjamini and Hochberg, 1995 ) was used to correct for multiple comparisons in all analyses, with family-wise α set a priori to 0.05. Both baseline and longitudinal memory and executive function scores were used as continuous outcome variables. Linear regression was used to assess the interaction between APOE -ε4 positivity (i.e., presence of at least one ε4 allele) and gene expression measured by RNAseq on cross-sectional memory performance. Linear-mixed effects regression tested the APOE interaction with gene expression on longitudinal memory, where the intercept and interval from baseline were entered as both fixed and random effects. Covariates in both models included baseline age and sex.
Replication analyses were performed using ROS/MAP samples. Specifically, linear regression models were used to examine the interaction between APOE -ε4 positivity and bulk brain gene expression in the DLPFC, CN, and PCC (Supplementary Table 2) on the last global cognition composite score before death. Covariates included PMI, sex, and age at death.

Sensitivity Analyses
Sensitivity analyses included models with more stringent QC for VMAP whole blood RNAseq data (i.e., adjusted for batch, sex, age, RIN, percentage of coding, intronic, and intergenic bases) to determine whether the observed results were due to technical effects. Additional sensitivity analyses included stratifying by cognitive diagnosis and covarying for education because of their effects on cognitive performance.

Post-hoc Blood to Brain RNASE6 Expression Correlation
To investigate the correlation between peripheral RNASE6 expression and brain RNASE6 expression, gene expression data from the NIH Genotype-Tissue Expression (GTEx) Project was leveraged ( Lonsdale et al., 2013 ). Linear regression and Pearson correlation were used to assess the relationship between brain and blood RNASE6 expression. RNASE6 expression in the brain cortex (N = 137) and hippocampus (N = 118) were evaluated against whole blood RNASE6 expression.

Post-hoc Biomarker Analyses
Given the relationship between APOE and AD biomarkers, we plan to examine the interaction between amyloid and tau positivity and any significant gene hits on cognition to better understand the biological mechanisms behind our cognitive results. For these analyses, we leveraged CSF measurements of amyloid and tau in VMAP and neuropathological measurements from ROS/MAP. Covariates for these analyses included age, sex, and PMI where relevant. Using similar regression models, we also investigated whether the gene x APOE -ε4 interaction had any effect on baseline CSF amyloid and tau levels, covarying for age and sex.

Results
The characteristics of individuals from VMAP (N = 324) and those from ROS/MAP who have DLPFC RNAseq (N = 535) are presented in Table 1 . Overall, a larger percentage of participants in VMAP are APOE -ε4 positive (34.8% vs. 23.1%, respectively) and have normal cognition (51.8% vs. 34.2%). VMAP participants are also younger than participants from ROS/MAP, on average. In contrast, a higher percentage of participants in ROS/MAP are tau and amyloid positive and they are more highly educated on average. It should be noted that cognition scores cannot be compared across studies because two different cognitive composites are used that are not scaled across studies. The total number of samples from each brain region in ROSMAP can be found in Supplementary Table 2.
The RNASE6 x APOE -ε4 interaction on baseline memory remained significant in sensitivity analyses (see Methods) when leveraging more stringent QC controlling for technical variation in RNA sequencing ( p = 2.23 × 10 −8 , Table 2 ). The interaction also remained significant when stratifying by diagnosis (i.e., normal cognition or MCI, Fig. 1 C, D) and covarying for education ( p -values < 0.00657, Table 2 ).

Replication in ROS/MAP
Leveraging data from an independent cohort, ROS/MAP, we examined the interaction between RNASE6 expression in brain tissue and APOE-ε4 genotype on global cognition at the final visit prior to death. Using bulk RNAseq data from 3 distinct brain regions: DLPFC, PCC, and head of the CN, we observed replication of the previous interaction on memory ( β = −0.35, p = 0.002) in the DLPFC ( Supplementary Fig. 1), though the observed effects were not present in the PCC or the CN ( p -values > 0.3; Table 3 ). Similar to VMAP, the RNASE6 x APOE -ε4 interaction was not significant longitudinally in ROS/MAP.

Correlation of Blood and Brain RNASE6 expression
To further investigate the relationship between blood and brain RNASE6 expression, we utilized data from the NIH GTEx project ( Lonsdale et al., 2013 ) in which RNASE6 expression was measured in both whole blood and brain tissue from the same individuals. Whole blood RNASE6 expression was significantly associated with RNASE6 expression in both the brain cortex (r = 0.30, β = 0.39, p = 0.0 0 04, Supplementary Fig. 3A) and hippocampus (r = 0.28, β = 0.40, p = 0.0024, Supplementary Fig. 3B).

Post-Hoc Analyses with Biomarkers
Like our initial findings within VMAP, we observed an interaction with amyloid positivity whereby higher levels of RNASE6 in blood were associated with worse baseline memory in VMAP ( β = −1.14, p = 0.001, Fig. 2 A). We also observed a similar interaction with tau positivity whereby higher levels of RNASE6 were associated with worse baseline memory ( β = −0.63, p = 0.04, Fig. 2 B).
We also wanted to examine whether RNASE6 expression influenced AD biomarker levels in APOE -ε4 carriers and non-carriers.
The main effect of RNASE6 was significantly associated with CSF A β 1-42 ( Fig. 3 A, β = 91.8, p = 0.02) such that higher RNASE6 levels in blood were correlated with reduced brain amyloid burden. However, when examining the RNASE6 x APOE -ε4 positivity interaction, this effect appeared to be in APOE -ε4 noncarriers only, though the interaction was non-significant ( p = 0.1, Fig. 3 A ). Though RNASE6 expression alone was not significantly associated with CSF tau or pTau, it significantly modified the relationship between APOE -ε4 and both CSF tau ( β = 230.1, p = 0.003) and CSF pTau ( β = 22.7, p = 0.01) levels such that APOE -ε4 carriers expressing higher levels of RNASE6 in blood have increased tau pathology at baseline ( Fig. 3 B, C ). None of these effects were observed using neuropathological measures of amyloid and tau in ROS/MAP.

Discussion
In this study, we observed the significant APOE4 -modifying effect of RNASE6 expression, in both blood and brain tissues, on cognition. Specifically, APOE -ε4 carriers expressing higher levels of RNASE6 in whole blood had worse baseline memory and executive function performance. We also replicated this novel discovery in an independent sample leveraging transcriptomic data from the dorsolateral prefrontal cortex. We also demonstrated that whole blood RNASE6 expression significantly correlates with brain RNASE6 expression. In addition, we found that higher RNASE6 levels are associated with poorer memory performance in individuals that are   amyloid-positive and/or tau-positive in comparison to individuals who are biomarker negative ( Fig. 2 A, B). RNASE6 also modifies the effect of APOE -ε4 on CSF tau and pTau levels, such that APOE -ε4 carriers expressing higher levels of RNASE6 have increased tau burden.
RNASE6 is a fascinating, novel inflammatory risk factor for AD. Excitingly, RNASE6 expression in whole blood significantly correlates with RNASE6 expression in the brain when using data measured within the same participants from GTEx ( Lonsdale et al., 2013 ), suggesting that it may have utility as a blood-based biomarker in lieu of brain samples.
RNASE6 protein exhibits antimicrobial activity ( Becknell et al., 2015 ). Overexpression of endogenous RNASE6 in mice is also associated with increased levels of reactive oxygen species as well as increased inflammatory factor secretion ( Fang et al., 2021 ). In addition, RNASE6 levels are increased in individuals with AD across several brain regions including the cerebellum, inferior frontal gyrus, and temporal cortex (AMP-AD Agora; https://agora.adknowledgeportal.org/genes/ (genes-router:genedetails/ENSG0 0 0 0 0169413). It has been established that immune function plays a role in cognitive decline ( King et al., 2018 ;Shen et al., 2019 ;Tao et al., 2018 ), and recent evidence has implicated the importance of neuroinflammation early in the AD cascade ( Pascoal et al., 2021 ) driving downstream neurodegeneration. Furthermore, different microglial pathways appear to be involved in the accumulation of amyloid and tau proteinopathies ( Patrick et al., 2021 ). RNASE6 was found to be upregulated in neurofibrillary tangle-bearing neurons suggesting that it may play a role in increasing tau pathology burden ( Miller et al., 2013 ). However, we also observe a trend in which higher RNASE6 expression in APOE-ε4 noncarriers correlates to reduced brain amyloid burden (i.e., higher CSF A β 1-42 , Fig. 3 A). Though we cannot conclude it without further study, we hypothesize that RNASE6 may play two distinct roles in both the neuroinflammatory-related clearance and deposition of AD neuropathology in APOE-ε4 carriers. In contrast, the observed effect of RNASE6 may simply be due to differences in the inflammatory milieu between APOE-ε4 carriers and non-carriers ( Friedberg et al., 2020 ).
Our significant results are restricted to cross-sectional cognitive performance. Although we looked at longitudinal outcomes, it should be noted that the RNASE6 x APOE -ε4 interaction on cognitive performance over time is nonsignificant (p.unadjusted > 0.05) in whole blood and brain tissue suggesting that RNASE6 may play a role in the overall biological state of aging and/or neurodegeneration instead of affecting the rate of cognitive decline during the neuropathological progression of AD. This is further supported by our discovery that the RNASE6 x APOE -ε4 interaction on cognition remains significant despite diagnosis (i.e., individuals with normal cognition or MCI).
Additionally, we note that carrying the APOE -ε4 allele does not modify the association between amyloid and cognition in our sample ( Supplementary Fig. 4). The amyloid x RNASE6 interaction results and the lack of an APOE -ε4 x amyloid interaction in our sample suggests the effects of APOE -ε4 observed here are likely reflecting the known effects of APOE in driving AD neuropathology, primarily through amyloidosis. Thus, one explanation for the lack of a longitudinal association between baseline RNASE6 and cognitive decline among APOE -ε4 carriers (despite a strong association with cross-sectional cognition) may be that the effect of RNASE6 among amyloid-positive individuals occurs very early in disease, but longitudinal analysis of RNASE6 , amyloid, and cognition would be needed to confirm this hypothesis. If RNASE6 expression is indeed a surrogate for an immune response, it may be that APOE -ε4 carriers have a higher susceptibility to a deleterious immune response to amyloid very early in disease.
However, there are other possibilities as to why we do not see longitudinal effects of the RNASE6 x APOE -ε4 interaction. First, we were unable to deconvolve the relationship between APOE -ε4 and amyloid in our analyses, although our results suggest the APOE -ε4 interaction effect is largely mediated by amyloid burden. Second, these analyses are limited by small sample sizes and the lack of longitudinal gene expression and biomarker data. Thus, studies utilizing additional RNAseq, cognition, and biomarker timepoints are needed for further examination of RNASE6 in this context. There is substantial evidence that neuroinflammatory pathways may be particularly relevant among APOE -ε4 carriers (reviewed in Kloske and Wilcock, 2020 ), and contribute to impaired amyloid clearance ( Qiao et al., 2001 ), enhanced gliosis ( Egensperger et al., 1998 ), and enhanced brain cytokine levels ( Lynch et al., 2003 ).
APOE -ε4 carriers have also displayed prolonged inflammatory responses in comparison to noncarriers ( Safieh et al., 2019 ). If RNASE6 expression is indeed a surrogate for an immune response, it can be hypothesized that APOE -ε4 carriers have a higher suscep- tibility to inflammation than noncarriers and also that our findings in baseline cognition may be due to prolonged inflammation before any changes to cognition. RNASE6 is also in a brain gene co-expression module with several AD genes including TREM2 and MS4A6A (AMP-AD Agora, see above), which are both highly expressed in microglia ( Deming et al., 2019 ;Hickman and El Khoury, 2014 ). In publicly available microglia data published previously, both RNASE6 and TREM2 are upregulated in an AD-associated microglial cluster ( Miller et al., 2013 ;Olah et al., 2020 ).
As aforementioned, the report discussing EFAM and its impact on APOE -ε4 ( Ma et al., 2022 ) provides an interesting convergence of observations from the peripheral and CNS resident immune systems; future work can explore whether these two factors (EFAM and RNASE6) are independently influencing APOE -ε4 or may synergize. In addition, future work on RNASE6 expression in microglia may help clarify the potential mechanistic pathway of the observed effect. It is also notable that another RNASE family gene, RNASE13 was previously associated with executive function resilience ( Mukherjee et al., 2014 ), suggesting that this family of proteins may be exciting targets for future investigation, particularly in response to pathology along an inflammatory pathway. This study has multiple strengths including our multimodal discovery analyses, independent replication, and the use of comprehensive longitudinal cognitive data from 2 deeply characterized aging studies. Furthermore, our findings that brain and blood RNASE6 expression are significantly correlated highlight the potential of blood transcriptomics to identify inflammatory and/or immune factors that have additional effects on the brain. Blood draws are also more accessible to individuals than in comparison to other procedures such as lumbar punctures, making similar analyses using blood transcriptomics increasingly viable in larger or more diverse populations. Also, recent studies not only allude that blood transcriptomics are useful in predicting AD  but also that brain markers of AD replicate in the blood, further supporting that RNASE6 may be an promising target for future study ( Iturria-Medina et al., 2020 ;Panitch et al., 2022 ).
However, there are limitations in our study that should be considered. The individuals in our sample are largely non-Hispanic white and highly educated, limiting the generalizability of our re-sults. We also lack functional data to support a specific mechanism of action; it is particularly challenging given that RNASE6 biology and the relationship between peripheral and brain RNASE6 expression is not well-characterized. Along with these considerations, the gene expression and pathology data used in our analyses are both cross-sectional; we cannot make a conclusion on how RNASE6 expression may affect AD neuropathology over time nor on how a diagnosis of AD may affect RNASE6 expression longitudinally. Future work with data captured both peripherally and centrally will be critical to extend these findings.

Conclusion
To conclude, we identified a gene, RNASE6, that modifies the association between APOE -ε4 and baseline cognition. In addition to supporting previous analyses implicating neuroinflammation in cognitive decline, our results suggest that data from blood transcriptomics can provide information about AD-relevant biological changes that may be occurring in the brain.

Conflicts of Interest
The authors report no conflicts of interest.

Funding Sources
This research was supported in part by grants K01AG049164, K24AG046373, R21AG059941, R01AG059716, R01AG034962, R01HL111516, R01NS100980, R01AG056534, RF1AG15819 from the Intramural Research Program, the National Institute on Aging, the National Institutes of Health, the Vanderbilt University Advanced Computing Center for Research and Education (ACCRE) instrumentation grant (S10OD023680), the Vanderbilt Institute for Clinical and Translational Research (VICTR) grant (UL1TR0 0 0445, UL1TR0 02243), and the Vanderbilt Memory & Alzheimer's Center.

Acknowledgements
The results published here are in whole or in part based on data obtained from Agora and the AD Knowledge Portal ( https: //adknowledgeportal.org ), which are supported by the NIA-funded AMP-AD consortium that shares evidence in support of AD target discovery.

Verification
The authors have not published or posted any related papers from the study described in this manuscript, and this manuscript is not under review at any other journal. We have presented preliminary work for this project at the Alzheimer's Association International Conference in July of 2021. All authors have substantially contributed to this project and have provided approval for submission.

Supplementary materials
Supplementary material associated with this article can be found, in the online version, at doi: 10.1016/j.neurobiolaging.2022. 06.011 .