Cerebrum

Manual labeling at different scales was performed by a trained expert using ITK-SNAP software. Details of intracranial cavity mask and macrostructures (cerebrum, cerebellum, and brainstem) can be found in the corresponding original papers (Manjón et al., 2014 (link); Romero et al., 2015 (link)). Lateral ventricles and subcortical structures were manually segmented from scratch using ITK-SNAP software using the 3 orthogonal views to avoid any inconsistency in 3D. Lateral ventricles label were thresholded using a threshold of 100 over the intensity-normalized images to get a consistent label definition (note that choroid plexus was not included in our lateral ventricles definition). All subcortical structures were segmented according to the current common definition criteria with the exception of hippocampus that was segmented using EADC protocol (Frisoni and Jack, 2011 (link)).
We further increased the number of available priors in the library by flipping them along the mid-sagittal plane using the symmetric properties of the human brain. Therefore, a total number of 100 labeled training templates (original and flipped) have been created as done in BEaST paper (Eskildsen et al., 2012 (link)).

We tested two types of statistical models to address our two questions, for each brain metric map (i.e. 7 maps in total): (1) Are there structural brain metrics that are associated with the ability to adopt asymmetry during the split-belt walking; and (2) Are there different brain-behavior relationships for younger and older adults? All models consisted of linear models with covariates constructed in SPM12 (estimated in CAT12 for surface metrics). Therefore we had a total of 14 separate models (7 brain metric maps x 2 questions) in which we tested for a correlation between each brain metric map and the gait parameters. The distinction between models for questions 1 and 2 was in the organization of the covariates of interest (i.e. the symmetry scores). For question 1, we tested for clusters in the brain maps that shared similar correlations across both age groups, and for question 2, we were interested in the interaction of age group and the gait parameter being tested. The symmetry scores included three gait parameters that inform us about ML balance control including (1) $\mathrm{\Delta }$ CoM, (2) $\mathrm{\Delta }$ Step-CoM, and (3) $\mathrm{\Delta }\int$ CoM-CoP. All symmetry scores were included in each of the models where the brain metric was the dependent variable. Age group and sex were used as covariates. Total intracranial volume was also used as a covariate for models that involved cerebrum gray matter and cerebellum volume³⁴ . We re-estimated all models with the Threshold-Free Cluster Enhancement (TFCE; http://dbm.neuro.uni-650jena.de/tfce) toolbox using the default 5,000 permutations^{52 (link)}. For each model, statistical significance was established at p<.05, after family-wise error (FWE) correction. Since we included all the gait metrics in each brain metric model, multiple comparisons across the gait metrics are accounted for, and the family-wise error correction corrects for multiple comparisons across brain voxels. Significant brain clusters were anatomically identified based on the appropriate atlas for that brain metric: the Automatic Anatomical Labeling (AAL3) atlas^{53 (link)} for cerebrum gray matter, SUIT^{40 (link),41 (link)} for cerebellum volume, DK40^{54 (link)} for surface measures (cortical thickness, sulcal depth, fractal dimension, and gyrification), and JHU-ICBM tract atlas^{55 (link),56 (link)} for fractional anisotropy.

Experiments with pig brains were permitted by the Health and Veterinary Office Münster (Reg.-No. 05 515 1052 21). Pig brain was obtained from a local butcher and separated into the following anatomical parcels: cerebrum, cerebellum, hypothalamus, and brain stem/spinal cord. Each tissue section was washed with distilled water, roughly cut into 10 × 10 × 10 mm pieces and homogenized using a blender (VDI 12, VWR International, Hannover, Germany). Homogenates were stored at − 20 °C. For adjustment of pH to 5–9, 0.5 M tris(hydroxymethyl)aminomethane (Tris-base, Serva, Heidelberg, Germany) buffer was prepared with hydrochloric acid (HCl, Honeywell Riedel–de Haen, Seelze, Germany). Reference tissue homogenates (RTHs) with controlled pH (pH-RTHs) were composed of RTHs and buffer (w/v), as displayed in supplementary Table S1 before spiking of PPIX. For the preparation of RTHs without pH control (pH ~ 7), 200 to 600 mg of the homogenates were directly spiked with PPIX (Enzo Life Sciences GmbH, Lörrach, Germany) stock solution (300 pmol/µl in dimethyl sulfoxide, Merck KGaA, Darmstadt, Germany) to the desired concentrations (0.0, 0.5, 0.75, 1.0, 2.0, 3.0 and 4.0 pmol/mg) and homogenized using a vortex mixer. RTHs and pH-RTHs were transferred to a Petri dish forming tissue samples of about 4 × 4 × 2 mm. Hyperspectral measurements were performed immediately using the same parameters as for tissue biopsies. The software calculated the PPIX contribution in µg/ml based on the calibration with liquid phantoms^{18 (link)–20 (link),25 (link)–28 (link)}. A unit related to the sample weight is superior and more common in solid tissue samples like homogenates or brain biopsies, because these samples are routinely weighed for analysis in the laboratory. Thus, we refer to pmol/mg for the spiked samples of homogenates experiments and evaluate them in relation to the calculated PPIX contribution from HI in µg/ml. All generated raw data analyzed during this study are included in the supplementary data file.

Cerebrum and cerebellum ACh concentrations were measured by the colorimetric method using sandwich ELISA Kits (catalog #E4452, BioVision Inc.^®) (catalog # ab287811) (Milpitas, CA, United States), and expressed as µmol/mg protein according to the manufacturer’s instructions.