Neuroepigenetic signatures of age and sex in the living human brain

Age- and sex-related alterations in gene transcription have been demonstrated, however the underlying mechanisms are unresolved. Neuroepigenetic pathways regulate gene transcription in the brain. Here, we measure in vivo expression of the epigenetic enzymes, histone deacetylases (HDACs), across healthy human aging and between sexes using [11C]Martinostat positron emission tomography (PET) neuroimaging (n = 41). Relative HDAC expression increases with age in cerebral white matter, and correlates with age-associated disruptions in white matter microstructure. A post mortem study confirmed that HDAC1 and HDAC2 paralogs are elevated in white matter tissue from elderly donors. There are also sex-specific in vivo HDAC expression differences in brain regions associated with emotion and memory, including the amygdala and hippocampus. Hippocampus and white matter HDAC expression negatively correlates with emotion regulation skills (n = 23). Age and sex are associated with HDAC expression in vivo, which could drive age- and sex-related transcriptional changes and impact human behavior.

]Martinostat uptake increases with age in the white matter after partial volume correction, supporting the non-partial volume corrected findings. a Voxel-wise correlations of SUVR with age, controlled for sex (n=41) were restricted to white matter and region-based voxel-wise correction was applied to correct for partial volume effects. Statistical maps were overlaid onto the MNI 1mm template in radiological orientation (family wise error rate (FWE) corrected for multiple comparisons, PFWE< 0.05; non-parametric permutation testing n=10,000 permutations). Red-yellow represents regions significantly increased with age. b SUVR was extracted from cerebral white matter based on automated FreeSurfer segmentation in native space, using the geometric transfer matrix method to correct for partial volume effects. Spearman correlation analysis between age and SUVR (n=41). analysis between SUV extracted from the whole brain and SUV extracted from the pons, based on automated FreeSurfer segmentation in native space (n=41). c Distribution volume (VT) time activity curves (TACs) were derived from a two-tissue compartmental model (2TCM) using metabolite-corrected arterial plasma as an input function. Individual subject SUV maps were normalized to mean SUV from the pons as SUVRPons. Pearson correlation analysis was performed between regional VT values and SUVRPons values for each subject (n=7). Each circle represents a separate brain region (n=14 brain regions). Supplementary Note 1. Brain volumes. It is well-established that regional brain volumes are altered in normal aging 1 . As expected, total gray matter (GM) volume normalized to intracranial volume (ICV) decreased with age (Spearman's r= -0.66, P< 0.0001) and cerebrospinal fluid (CSF) volume normalized to ICV increased with age (Spearman's r= 0.48, P= 0.0014, Supplementary Figure 2) in subjects imaged in this study (n= 41). No statistically significant difference was observed in white matter (WM) volume normalized to ICV with age (Spearman's r= 0.0085, P= 0.96, Supplementary Figure 2). CSF volume normalized to ICV was higher in males than females (P= 0.045, Supplementary Figure 12b). As SUVR values could be affected by changes in brain volume, total GM normalized to ICV, WM normalized to ICV, and CSF normalized to ICV were added as regressors of non-interest to all SUVR whole brain voxel-wise analyses.

Supplementary Note 2.
Body mass index. To support that the age-related [ 11 C]Martinostat uptake differences observed in our study were not driven by body mass index (BMI), we assessed

Supplementary Note 3. Pons normalization. To further verify that [ 11 C]Martinostat uptake differences
were not driven by age-related brain atrophy, the pons was chosen as a secondary normalization region.
Pons volume did not change with age in subjects imaged in this study (Spearman's r= 0.0036, P= 0.98, Supplementary Figure 6a). SUV in the pons was strongly associated with SUV in the whole brain (Spearman's r= 0.95, P< 0.0001, Supplementary Figure 6b). We determined that standard uptake value (SUV) normalized to pons mean (SUVRPons) from 60-90 min post radiotracer injection is an appropriate surrogate measure for the distribution volume (VT) in a subset of healthy young and old adults. Regional SUVRPons were strongly correlated with VT values derived from a two-tissue compartmental model (2TCM), using metabolite-corrected arterial plasma as an input function (all Pearson's r≥ 0.90, all P< 0.0001, Supplementary Figure 6c). Voxel-wise analysis correlating SUVRPons with age was performed while controlling for sex. SUVRPons was increased with age throughout the cerebral white matter, recapitulating our primary finding (PFWE< 0.05, Supplementary Figure 7a). In addition to regional increases, voxel-wise analysis also showed decreased SUVRPons with age limited to the CSF, and small portions of opercular and insular cortices adjacent to the CSF, which were likely driven by age-associated partial volume effects. (PFWE< 0.05, Supplementary Figure 7b).
Pons normalization was also used to verify that [ 11 C]Martinostat uptake differences were not driven by sex-related differences in brain volume. Voxel-wise analysis of [ 11 C]Martinostat SUVRPons was performed between groups controlling for age (Z score >2.3, Pcluster< 0.05, Supplementary Figure 13a) and recapitulated our primary findings. SUVRPons was higher in females compared to males in the frontal medial cortex, amygdala, hippocampus, parahippocampal gyrus, and thalamus (Supplementary Figure   13a). SUVRPons was lower in females compared to males in cerebellar white matter (Supplementary   Figure 13a). This finding was recapitulated for the majority of the brain regions using non-parametric analysis and a statistical threshold of PFWE< 0.05 (Supplementary Figure 13b).

Supplementary Methods
Radiotracer synthesis according to cGMP guidelines, as detailed in a previous publication 2 . MR data analysis. In order to determine regions of interest in native space, the MEMPRAGE images were reconstructed with FreeSurfer's automated segmentation and parcellation, version 6.0 (http://surfer.nmr.mgh.harvard.edu/) 10 . Brain volumes for gray matter, white matter and cerebrospinal fluid as well as estimated total intracranial volume were obtained using Freesurfer. Diffusion MR (dMRI) images were preprocessed utilizing in house scripts, which include the following: axis alignment, centering, eddy current correction and, motion correction (https://github.com/pnlbwh/pnlutil). dMRI images were visually quality checked for noise, severe motion, and signal drops. All images were masked to exclude non-brain areas with 3D Slicer 11 . To assess white matter microstructure, generalized fractional anisotropy (gFA) maps were calculated from the orientation distribution function (ODF) that were estimated at each voxel utilizing regularized spherical harmonics 12  algorithms. Whole brain voxel-wise analysis was performed with unsmoothed gFA images in MNI space using FSL's randomise 16 .