## Neural network decorrelation for healthy brain aging

James, L (2016) Society for Neuroscience## Nutrition and healthy brain functioning across the lifespan

James, L (2016) Society for Neuroscience## The number of cysteine residues per mole in apolipoprotein E affects systematically synchronous neural interactions in women's healthy brains

- Leuthold AC, Mahan MY, Stanwyck JJ, Georgopoulos A, Georgopoulos AP (2013)
*Experimental Brain Research*, 226:525–536. **Fig. 6.**The Z-statistic derived from a KS test of SNI distributions of all women with the SNI distribution of E2/2 is plotted against its rank, color-coded for apoE genotypes. Notice the broad distribution in the plot of individuals with other apoE genotypes. N = 130 women

Abstract

Apolipoprotein E (apoE) is involved in lipid metabolism in the brain, but its effects on brain function are not understood. Three apoE isoforms (E4, E3, and E2) are the result of cysteine-arginine interchanges at two sites: there are zero interchanges in E4, one interchange in E3, and two interchanges in E2. The resulting six apoE genotypes (E4/4, E4/3, E4/2, E3/3, E3/2, E2/2) yield five groups with respect to the number of cysteine residues per mole (CysR/mole), as follows. ApoE4/4 has zero cysteine residues per mole (0-CysR/mole), E4/3 has one (1-CysR/mole), E4/2 and E3/3 each has two (2-CysR/mole), E3/2 has three (3-CysR/mole), and E2/2 has four (4-CysR/mole). The use of the number of CysR/mole to characterize the apoE molecule converts the categorical apoE genotype scale, consisting of 6 distinct genotypes above, to a 5-point continuous scale (0-4 CysR/mole). This allows the use of statistical analyses suitable for continuous variables (e.g. regression) to quantify the relations between various variables and apoE.

Using such analyses, here, we show for the first time that apoE affects in a graded and orderly manner neural communication, as assessed by analyzing the relation between the number of CysR/mole and synchronous neural interactions (SNI) measured by magnetoencephalography (MEG) in 130 cognitively healthy women. At the one end of the CysR/mole range, the 4-CysR/mole (E2/2) SNI distribution had the highest mean, lowest variance, lowest range, and lowest coefficient of variation, whereas at the other end, 0-CysR/mole (E4/4) SNI distribution had the lowest mean, highest variance, highest range, and highest coefficient of variation. The special status of the 4-CysR/mole distribution was reinforced by the results of a hierarchical tree analysis where the 4-CysR/mole (E2/2) SNI distribution occupied a separate branch by itself and the remaining CysR/mole SNI distributions were placed at increasing distances from the 4-CysR/mole distribution, according to their number of CysR/mole, with the 0-CysR/mole (E4/4) being farthest away.

These findings suggest that the 4-CysR/mole (E2/2) SNI distribution could serve as a reference distribution. When the SNI distributions of individual women were expressed as distances from this reference distribution, there was a substantial overlap among women of various CysR/mole. This refocuses the placement of individual brains along a continuous distance from the 4-CysR/mole SNI distribution, in contrast to the common categorical assignment to a specific apoE genotype.

Finally, the orderly variation of SNI with the number of CysR/mole found here is in keeping with recent advances and ideas regarding the molecular mechanisms underlying the differential effects of apoE in the brain which emphasize the healthier stability conferred on the apoE molecule by the increasing number of cysteine-arginine interchanges, with 4-CysR/mole (E2/2) being the best case, as opposed to the instability and increased chance of toxic fragmentation of the apoE molecule with lower number of CysR/mole, with 0-CysR/mole (E4/4) as the worst case (Mahley and Huang in Neuron 76:871-885, 2012a). However, our results also document the appreciable variation of SNI properties within the various CysR/mole groups and individuals which points to the existence and important role of other factors involved in shaping brain function at the network level.

**The effect of apolipoprotein E (apoE) genotype on synchronous neural interactions (SNI) in healthy brains**- Mahan MY, Leuthold AC, Stanwyck JJ, Georgopoulos A and Georgopoulos AP (August 2013)
*Frontiers in Neuroinformatics*, 7. DOI:10.3389/conf.fninf.2013.09.00064. International Neuroinformatics Coordinating Facility (INCF) Neuroinformatics Conference, Stockholm, Sweden. **Fig. 1.**Dendrogram derived from hierarchical tree clustering of the number of cysteine residues per mole (CysR/mole) SNI distribution distances. Notice the distinct division of E2/2 with 4-CysR/mole, the clustering of all other apoE genotypes, and the orderly placement in the tree of apoE genotypes according to the number of CysR/mole, from zero to four.**Assessing dynamic functional connectivity using time-varying graphs in magnetoencephalography data with an application to healthy aging**.- Mahan MY, Georgopoulos AP (August 2015) 3rd Annual BICB Industry Symposium. Minneapolis, MN. Second Place Prize.
**Effects of age on lagged cross-correlations among neural activities measured by magnetoencephalography (MEG) in the resting state**.- Mahan MY, Loe ME, Leuthold AC, Georgopoulos AP (August 2014) 2nd Annual Biomedical Informatics and Computational Biology Industry Symposium. First Place Prize.
**Apolipoprotein E genotypes and dynamic neural communication in healthy brains**- Leuthold AC, Mahan MY, Georgopoulos AP (November 2013) . Annual meeting for the Society for Neuroscience, San Diego, CA.
**The effect of apolipoprotein E (apoE) genotype on synchronous neural interactions (SNI) in healthy women brains**.- Mahan MY, Leuthold AC, Stanwyck JJ, Georgopoulos A, Georgopoulos AP (September 2013) 10th Annual Women's Health Research Conference, Minneapolis, MN.
**Apolipoprotein E genotypes differentially affect synchronous neural interactions in healthy brains**.- Mahan MY, Leuthold AC, Stanwyck JJ, Georgopoulos A, Georgopoulos AP (January 2013) 5th Annual Biomedical Informatics and Computational Biology Research Symposium, Rochester, MN.
**Brain health index: an integrative assessment of brain status derived from multimodal measurements of brain function, structure, and chemistry**.- Mahan MY, Georgopoulos AP (November 2012) National Academies of Sciences Keck Future Initiatives (NAKFI) Conference, Irvine, CA. Invited Participant.

Abstract

In this study, we analyzed the effect of apolipoprotein E (apoE) genotype on SNI distributions in cognitively healthy subjects of various ages to determine the relations between apoE genotype and neural communication. ApoE is involved in lipid metabolism in the brain but its effects on brain function are not understood. Three apoE isoforms (E4, E3, and E2) are the result of cysteine-arginine interchanges at two sites: there are zero interchanges in E4, one interchange in E3, and two interchanges in E2. The resulting six apoE genotypes yield five groups with respect to the number of cysteine residues per mole (0-4 CysR/mole). The use of the number of CysR/mole to characterize the apoE molecule converts the categorical apoE genotype scale, consisting of 6 distinct genotypes above, to a 5-point continuous scale, allowing the use of statistical analyses suitable for continuous variables.

Using such analyses, here we show for the first time that apoE affects in a graded and orderly manner neural communication, as assessed by analyzing the relation between the number of CysR/mole and synchronous neural interactions (SNI) measured by magnetoencephalography (MEG) in 130 cognitively healthy subjects. By investigating the statistical properties along the range of CysR/mole SNI distributions, the 4-CysR/mole (E2/2) SNI distribution was found to have unique properties. The special status of the 4-CysR/mole distribution was reinforced by the results of a hierarchical tree analysis (see figure 1) where the 4-CysR/mole (E2/2) SNI distribution occupied a separate division by itself and the remaining CysR/mole SNI distributions were placed at increasing distances from the 4-CysR/mole distribution, according to their number of CysR/mole, with the 0-CysR/mole (E4/4) being farthest away.

These results support the idea that the number of CysR/mole is an important quantitative factor underlying the effect of apoE on SNI. In addition, these findings suggest that the 4-CysR/mole (E2/2) SNI distribution could serve as a reference distribution. When the SNI distributions of individual subjects were expressed as distances from this reference distribution, there was a substantial overlap among subjects of various CysR/mole. This orderly variation of SNI with the number of CysR/mole is in keeping with recent advances and ideas regarding the molecular mechanisms underlying the differential effects of apoE in the brain which emphasize (a) the healthier stability conferred on the apoE molecule by the increasing number of cysteine-arginine interchanges, with 4-CysR/mole (E2/2) being the best case, as opposed to (b) the instability and increased chance of toxic fragmentation of the apoE molecule with lower number of CysR/mole, with 0-CysR/mole (E4/4) as the worst case.

Overall, we show for the first time that the apoE genotype affected the SNI distribution in a systematic and graded fashion, according to the number of CysR/mole in the apoE molecule.

Abstract

To examine the brain as the dynamic network it is, functional connectivity analyses need to include the temporal dimension as part of graph construction. Time-varying graphs are a valuable tool for such purposes. Here, we present two methods, lag-based and window-based, to construct time-varying graphs from magnetoencephalography (MEG) data and apply these methods to assessing dynamic functional connectivity across the lifespan. MEG recordings were collected from 140 healthy women (32-97 years old) for two sessions. These MEG time series were processed using an ARIMA(30,1,3) model to render the series stationary and nonautocorrelated. Nodes for both methods were defined to be the individual sensors (n = 248). For the lag-based method, each subject's crosscorrelations (CCs) were computed for all sensor pairs (n = 30,628) for ±20 lags and significant CCs were retained. For the window-based method, significant zero-lag crosscorrelations were computed and retained for 20 non-overlapping time windows. Then, we performed a multivariate regression to assess the effect of age, where the graph metrics (i.e., strength, diversity, degree, etc.) were the dependent variables and age was the independent variable. Through an iterative, stepwise procedure those metrics that varied systematically with age were detected and retained. This analysis was performed for both methods and the results were compared. In addition, to determine reliability, an intraclass correlation between sessions was calculated for each metric. Discussion is focused on evaluating the two methods for constructing time-varying graphs; exploring the reliability of metrics across these methods; and identifying patterns of dynamic functional connectivity with age.

Abstract

A central effort of our research is focused on investigating brain function across the lifespan. For that purpose, we use magnetoencephalography to record high-fidelity resting-state brain activity at high temporal resolution. This yields 248 sensors x 60,000 ms matrix of neural activity recorded simultaneously from the cerebral cortex. To estimate the strength and direction of neural interactions, we calculate pair-wise cross-correlation functions (CCF) (N = 30,628) between the prewhitened 248 sensor time series for ±50 ms lags. We have found in previous studies that the zero-lag cross-correlations carry sufficient information to discriminate among brain diseases, and change systematically across the lifespan. In this study, we investigated the age-dependent changes in cross-correlations calculated for lags ±50 ms. This yielded 101 cross-correlations for each one of the 30,628 sensor pairs. The strength, sign, and lag of each cross-correlation was noted, and more general patterns of the cross-correlogram were determined and quantified (e.g. contiguous cross-correlations, multiple significant peaks in the cross-correlogram, systematic driving of a sensor on others, etc.). We then regressed each one of these measures and quantitative features of the cross-correlogram against the age of 133 brain-healthy women subjects (28-94 years old). We discovered highly significant associations between CCF attributes and subject age. These served as the basis to construct a model of how brain communication patterns change with age, in such a way that brain function remains healthy, namely, a model of healthy brain aging. This model was further corroborated using longitudinal measurements taken from study subjects every year (http://healthybrain.umn.edu).

Updated
April 18, 2017