Clinicopathologic Heterogeneity and Glial Activation Patterns in Alzheimer Disease

This cross-sectional study evaluates the association between corticolimbic vulnerability to tau pathology as a continuous trait with clinicalpathologic heterogeneity and glial activation patterns in neuropathologically diagnosed Alzheimer disease.

This supplementary material has been provided by the authors to give readers additional information about their work.

eMethods eMethods 1. CorticoLimbic index (CLix) development
We have previously defined a mathematical algorithm to classify Alzheimer's disease (AD) subtypes in our discovery cohort of 889 AD cases based on their corticolimbic patterns of tangle accumulation. 3However, a conceptual method was lacking that would enable us to visualize how extreme a hippocampal sparing AD or limbic predominant AD case was relative to a more representative typical AD case.By removing the extreme AD phenotypes, we would theoretically be better positioned to reduce attributable variance and improve our ability to interpret relevant clinicopathologic findings in the more common form of typical AD.To account for the individuality of corticolimbic patterns in AD, we introduce a novel re-expression of the AD subtype algorithm as a continuous index termed CLix.
As previously described in the discovery cohort, 3 tangle counts were used for subtyping AD cases based on corticolimbic patterns of tangle accumulation.The posterior hippocampus at the level of the lateral geniculate nucleus was reviewed to obtain tangle counts in the CA1 and subiculum.Cortical tangle counts were obtained from association cortices, including superior temporal, inferior parietal, and middle frontal.The original algorithm assessed the regional severity of tangle accumulation in an individual compared to that in the overall AD cohort, as well as the ratio of hippocampal subsector accumulation to cortical regions within each case.In the refined algorithm, we used data from AD cases in the larger Florida Autopsied Multi-Ethnic AD (FLAME-AD) series (n=1361) as the new reference population and re-expressed the algorithm in terms of a continuous numeric score that we refer to as CLix.
The original discovery cohort overlaps with the FLAME-AD series by 54% (n=739/1361).The algorithm refers to percentiles of tangle count distributions and derived quantities, noting that the 50th percentile is the median (eTable 4 ).Tangle counts in the hippocampal subsectors (H) and cortical regions (C) were used, both being integral to the implementation of Braak tangle staging. 5The inputs of the algorithm are the hippocampal and cortical counts and their means, as well as the ratio (R) of the hippocampal and cortical means, all relative to their respective medians in the reference population.CLix re-expresses the algorithm as a continuous numeric value between 0 and 40 such that those with values of less than 10 are classified as hippocampal sparing AD, while those with values of 30 or higher are classified as limbic predominant AD (Figure 1A).The continuous measure is useful in that it places AD cases on a spectrum, with some for which the subtype is clear (<10: hippocampal sparing, ≥10 to <30: typical, ≥30: limbic predominant) while those with scores close to the cut-off of 10 or 30 are more borderline.Thus, even cases classified as typical AD can be assessed on a continuum.
The CLix creation algorithm and imputation for missing tangle counts values were described in detail within attached files on GitHub (Translational Neuropathology Lab.GitHub).To describe the CLix algorithm briefly.For a new case, we have   * and To assign a CLix score, this  * is then measured in terms of the distribution of  from the reference dataset resulting in . is the proportion of the reference dataset that is less than  * .We handle  one of the following three ways: -If  is equal to 50, we multiply by 0.4 to rescale and account for  being a proportion and the CLix score is calculated as 0.4 * .-If  is less than 50, the CLix score will be 0.4 times the maximum of ,  , and  . is the largest of the proportion of  in the reference dataset that is less than  * divided by 2, with  1,2,3. is the second largest of the proportion of  in the reference dataset that is greater than or equal to  * divided by 2, with  1,2,3,4.-If  is greater than 50, the CLix score will be 0.4 times the minimum of ,  , and  . is the second smallest of the proportion of  in the reference dataset that is greater than or equal to  * divided by 2 and subtracted from 100, with  1,2,3. is the second smallest of the proportion of  in the reference dataset that is less than or equal to  * divided by 2 and subtracted from 100, with  1,2,3,4.

eMethods 2. Retrospective clinical abstraction
We retrospectively collected data on demographics and clinical progression from existing clinical records provided to the brain bank by participants and/or family members.Years of education were collected.The age at onset of first cognitive symptoms was calculated by subtracting the date of birth from the approximate date at onset of cognitive symptoms.The date at onset was then subtracted from the eMethods 3. Neuropathologic procedures and digital pathology Tissue samples were obtained during standardized neuropathologic evaluation of the brains by a single neuropathologist, Dennis W. Dickson, the director of the Mayo Clinic brain bank.Formalin-fixed paraffin-embedded tissue sections from the posterior hippocampus, superior temporal cortex, inferior parietal cortex, and middle frontal cortex were cut into 5 m thin slices and mounted on positively charged glass slides (Figure 1).As previously described, 1 standard neuropathologic examinations and procedures were performed to determine the Braak tangle stage 5 and Thal amyloid phase. 7Thioflavin-S, a fluorescent dye that binds to β-pleated sheets of which both tangles and amyloid-β plaques are composed, was used to assess tangle counts with a 40X objective (0.125 mm 2 microscopic field) and amyloid plaque density (to a maximum of 50 per microscopic field) with a 10X objective (3 mm 2 microscopic field) using an Olympus BH2 fluorescent microscope (Evident Corporation, Tokyo, Japan).Available data for TDP-43 positivity are reported for FLAME-AD from screening efforts performed in the amygdala.For the digital pathology subgroup, limbic predominant age-related TDP-43 encephalopathy neuropathologic changes (LATE-NC) staging was assigned to each case using TDP-43 immunohistochemistry on the amygdala, hippocampus, and middle frontal cortex. 8Cerebrovascular disease was measured using the modified Kalaria scale, 9 as previously described. 10We excluded cases in the digital pathology subgroup (n=60) with high cerebrovascular burden (ie, microinfarcts and/or infarcts) as indicated by a cortical score of 5 to 6 and/or basal ganglia score of 3 to 4.
Tissue slides from the posterior hippocampus, superior temporal, inferior parietal, and middle frontal cortices were scanned on an Aperio AT2 scanner (Leica Biosystems, Buffalo Grove, IL, USA) and ImageScope software (Leica Biosystems, version 12.4.2.7000) was used for drawing annotations to digitally quantify neuropathology.The hippocampal CA1 was traced using the lacunosum layer as the superior border, the alveus served as the inferior border, and the midway of the dentate gyrus separated CA1 from the subiculum.All stains of the same tissue sample were traced simultaneously with neuroanatomic guidance from diagnostic hematoxylin and eosin (H&E)-stained sections to ensure analysis of matching brain regions and pathologies.Sections of association cortices (superior temporal, inferior parietal, and middle frontal) were annotated in the gray matter using the strait of the gyrus, where the superficial surface and gray/white junction lay parallel, as a landmark for consistency across samples.All tracing of annotation layers was performed blinded to AD subtype.Two custom designed color-deconvolution macros (AT8 and CD68) and three positive-pixel count macros (GT-38, 6F/3D, GFAP) (Figure 1, eTable 5 ) were applied. 13The macros identify positive staining for tau, amyloid-β, GFAP, and CD68 burden on tissue sections, which is then converted into a percent burden of the total area annotated that represents positive staining per area annotated.Representative images are provided in (Figure 1) displaying characteristic staining of the neuropathologic lesions and their corresponding macro markup images.

eMethods 4. TREM2 R47H variant
Total DNA was extracted from frozen cerebellum using standard protocols with AutoGen FlexSTAR (AutoGen, Holliston, MA, USA).Cases were previously genotyped using NeuroChip (Infinium NeuroConsortium Array, Illumina, Inc., San Diego, CA, USA) to determine TREM2 R47H variant status.Genotyping of the NeuroChip signal-intensity data was performed using GenomeStudio version 2.0.5 with default settings.Genotype calls were exported from GenomeStudio to PLINK 14 format and annotated using ANNOVAR. 14,15enotypes at the TREM2 R47H position were then extracted.

eMethods 5. ADRC and MCSA study cohort descriptions
Clinical and neuroimaging procedures The neuroimaging group (n=93) was derived from Mayo Clinic Alzheimer's disease research center (ADRC) and Mayo Clinic Study of Aging (MCSA).The ADRC study participants are recruited through dementia referral services.Whereas the MCSA is a populationbased, epidemiological cohort that was designed to investigate the prevalence, incidence, and risk factors for mild cognitive aging and dementia among the residents of Olmsted County, Minnesota 16 .Using the Rochester Epidemiology Project (REP) medical records linkage system infrastructure 17,18 to enumerate Olmsted County population, study participants were randomly invited to participate in the MCSA using an age-and sex-stratified sampling frame.Demographics and clinical progression measures were databased prospectively.To compare cognitive impairment findings to FLAME-AD, we examined longitudinal MMSE decline as points lost per year in individuals with three or more MMSE test scores available.The clinical diagnosis of the participants was ascertained at the time of magnetic resonance imaging (MRI) using previously published criteria. 19An atypical clinical syndrome was classified for individuals with an antemortem clinical diagnosis of primary progressive aphasia, frontotemporal dementia, posterior cortical atrophy, and corticobasal syndrome.A recently described dysexecutive syndrome in AD 20 was retrospectively evaluated in the clinical records of the neuroimaging group.
All MRIs were acquired on 3T Siemens and GE scanners.The acquisition protocols and analysis for MRI (n=93) and tau positron emission tomography (PET, n=19/93) were previously published. 21,22The voxel level analyses were conducted using unified segmentation in SPM12 23 with MCALT templates and settings for structural MRI and tau PET scans (https://www.nitrc.org/projects/mcalt/).For neuroimaging analyses, a false discovery rate was corrected at p<0.05 for MRI and for tau-PET.The significant voxels with the following thresholds were displayed using MRIcroGL software.

Neuropathologic procedures
Neuropathologic sampling followed Consortium to Establish a Registry for Alzheimer's Disease (CERAD) recommendations and National Institute on Aging-Alzheimer's Association criteria for AD neuropathologic change assessment. 24,25Formalin-fixed, paraffinembedded 5-µm-thick tissue sections were stained with hematoxylin and eosin, as well as Bielschowsky silver stain.As described in the eMethods, thioflavin-S microscopy was used to quantify the number of tangles in posterior hippocampal subsectors (CA1, subiculum), superior temporal, inferior parietal, and middle frontal.Tangles counts were inputted into the CLix package to assign a CLix score and AD subtype.The discovery series was used to originally define the AD subtype algorithm (n=889). 3In the current study, the Florida Autopsied Multi-Ethnic Alzheimer's disease (FLAME-AD) series was used as the updated reference population (n=1361) for the development of the corticolimbic index (CLix).H = Hippocampus, C = Cortex, R = Ratio between the hippocampus and cortex.

eTable 3 .
Clinicopathologic Characteristics of the Digital Pathology Subgroup

eTable 4 . 5 . 6 . 1 . 2 . 3 .
Reference Percentiles and Thioflavin-S Positive Tangle Counts for AD Subtype Classifier Algorithm eTable Characteristics of Primary Antibodies Used in this Study eTable Slides Stained and Evaluated for the Digital Pathology Subgroup eTable 7. CLix Distribution Among AD Subtypes Stratified by Ethnoracial Status eTable 8. Regional Neuropathologic Measures in the AD Digital Pathology Subgroup eFigure Overview of 3 Study Groups Evaluated With CLix eFigure Structural MRI and Tau-PET Scans Across CLix Scores eFigure Regional Quantitative Measures and Distributions in the Digital Pathology Subgroup Among AD Neuropathologic Subtypes eReferences.

eTables eTable 1 .
Clinicopathologic characteristics of the FLAME-AD series used to develop the CLix methodology forCLix ≥10 to <30) Data are medians (25 th percentile to 75 th percentile) unless noted.TDP-43 screening was performed in the amygdala corresponding to LATE-NC stage 1.APOE=gene encoding apolipoprotein E. g=grams.LATE-NC=limbic-predominant age-related TDP-43 encephalopathy neuropathologic change.MMSE=Mini-Mental State Examination.TDP-43=transactive response DNA binding protein of 43 kDa.yr=years.CLix ≥10 to <30) are medians (25 th percentile to 75 th percentile) unless noted.TDP-43 screening was performed in the amygdala corresponding to LATE-NC stage 1.The MMSE longitudinal decline was only available for a subset of the digital pathology subgroup (3 hippocampal sparing, 7 typical, 3 limbic predominant).Lewy body disease was not included as part of the digital pathology subgroup exclusion criteria.APOE=gene encoding apolipoprotein E. g=grams.LATE-NC=limbic-predominant age-related TDP-43 encephalopathy neuropathologic change.MMSE=Mini-Mental State Examination.TDP-43=transactive response DNA binding protein of 43 kDa.yr=years.

eTable 4 .
Reference percentiles and thioflavin-S positive tangle counts for AD subtype classifier algorithm

eFigure 2 .
Structural MRI and tau-PET scans across CLix scoresCLix score is a spatial score representative of corticolimbic tangle involvement that extends well to structural MRI (left) and tau-PET (right).Example cases from the neuroimaging group were all neuropathologically diagnosed with high likelihood AD. (Top) Low CLix score corresponded to relatively preserved limbic structures with extensive cortical tau-PET uptake as shown by representative images in a 61-year-old man imaged 2.0 years prior to death.(Middle) A moderate CLix score corresponded to affected limbic structures with medial temporal lobe and cortical tau-PET uptake as shown by representative images in an 86-year-old man imaged 0.8 years prior to death.(Bottom) High CLix score corresponded to greatly affected limbic structures with minimal cortical tau-PET uptake as shown by representative images in an 89-year-old man imaged 1.6 years prior to death.Abbreviations: AD, Alzheimer's disease; CLix, CorticoLimbic index; MRI, magnetic resonance imaging; PET, positron emission tomography.Note: The color bar values indicate the value of the T-statistic with higher tracer uptake shown in warmer colors.Arrow on tau PET map points to posterior cingulate and cuneus region.

CLix distribution among AD subtypes stratified by ethnoracial status Alzheimer's disease neuropathologic subtypes in FLAME-AD (n=1361)
Medians (1st quartile to 3rd quartile) and counts (%) are reported, and P values result from Kruskal-Wallis rank sum test for continuous variables.Two Native American decedents (CLix=18, 22) and one Asian decedent (CLix=22) from FLAME-AD were not included in the analysis due to the small sample size.AD=Alzheimer's disease.CLix=CorticoLimbic index.FLAME-AD=Florida Autopsied Multi-thnic Alzheimer's disease.

Overview of three study groups evaluated with CLix Antemortem
evaluation of clinical heterogeneity was evaluated in the FLAME-AD series (middle) and neuroimaging group (left), whereas postmortem evaluation of neuropathologic heterogeneity was evaluated in the FLAME-AD group (middle) and digital pathology subgroup (right).Pertinent information regarding sample size, goal, methodologies applied, tables, figures, and appendix reference are ;included for each of the three groups analyzed.Abbreviations: AD, Alzheimer's disease; ADRC, Alzheimer's Disease Research Center; CLix, CorticoLimbic index; FLAME-AD, Florida Multiethnic Alzheimer's disease; MCSA, Mayo Clinic Study of Aging; MRI, magnetic resonance imaging; PET, positron emission tomography.