Breadth and age-dependency of relations between cortical thickness and cognition

Abstract

Recent advances in neuroimaging have identified a large number of neural measures that could be involved in age-related declines in cognitive functioning. A popular method of investigating neural-cognition relations has been to determine the brain regions in which a particular neural measure is associated with the level of specific cognitive measures. Although this procedure has been informative, it ignores the strong interrelations that typically exist among the measures in each modality. An alternative approach involves investigating the number and identity of distinct dimensions within the set of neural measures and within the set of cognitive measures before examining relations between the 2 types of measures. The procedure is illustrated with data from 297 adults between 20 and 79 years of age with cortical thickness in different brain regions as the neural measures and performance on 12 cognitive tests as the cognitive measures. The results revealed that most of the relations between cortical thickness and cognition occurred at a general level corresponding to variance shared among different brain regions and among different cognitive measures. In addition, the strength of the thickness-cognition relation was substantially reduced after controlling the variation in age, which suggests that at least some of the thickness-cognition relations in age-heterogeneous samples may be attributable to the influence of age on each type of measure.

Introduction

A large number of measures of brain structure and brain function have been found to be negatively related to age, and many of these measures have also been found to be related to measures of cognitive functioning. Consider measures of cortical thickness, as assessed by the distance between the gray matter–cerebral spinal fluid (CSF) boundary and the gray matter–white matter boundary. Because it is postulated to reflect the density of neurons, dendrites, spines, synapses, and glial cells, cortical thickness is a potentially important neural substrate of cognition.

Negative relations between adult age and measures of cortical thickness have been reported in numerous studies (e.g., Ecker et al., 2009, Fjell et al., 2006, Fjell et al., 2009, Fjell et al., 2014, Hogstrom et al., 2013, Hutton et al., 2009, McGinnis et al., 2011, Salat et al., 2004, Salat et al., 2009, Tustison et al., 2014, Westlye et al., 2011), and many studies have also reported positive relations between measures of cortical thickness and cognitive functioning (e.g., Choi et al., 2008, Desrivieres et al., 2015, Ehrlich et al., 2012, Engvig et al., 2010, Fjell et al., 2006, Haier et al., 2009, Karama et al., 2009, Karama et al., 2011, Narr et al., 2007, Schilling et al., 2013, Walhovd et al., 2006, Westlye et al., 2009, Westlye et al., 2011; but see Colom et al., 2013). Based on these 2 sets of findings, it is tempting to postulate that age-related reductions in cortical thickness in specific neuroanatomical regions are involved in age-related reductions in particular types of cognitive functioning. However, we suggest that it is important to consider 2 issues when making these types of inferences; level of analysis, and the degree to which the relation between the 2 types of measures might be dependent on the relation of each measure with age.

Although sometimes considered individually, most neuroanatomical measures derived from different brain regions are highly related with one another, and most cognitive measures are highly related with one another. This lack of independence implies that some of the relations observed with an individual measure could be shared with influences that affect many measures and are not unique to the target measure. However, shared and unique influences cannot be distinguished unless multiple measures are examined in some type of organizational structure.

Consider Fig. 1 which portrays 3 possible organizations with sets of neural measures and cognitive measures. Panel A illustrates a situation with no structure in either the neural or cognitive measures. Neural-cognition relations could be investigated within a framework such as this by examining all possible combinations of neural measures and cognitive measures. However, this is almost never done because of the extremely large number of possible neural measures that could be obtained across different regions of the brain. Instead, analyses are often conducted to determine which clusters of neural measures are significantly related to particular cognitive measures. Any structure that emerges with this approach is therefore based on relations the neural measures have with that set of cognitive measures and does not necessarily reflect the intrinsic dimensionality of the neural measures, independent of their relations with other types of measures.

An alternative approach to investigate neural-cognition relations is portrayed in panel B in which the 2 types of measures are first grouped into factors representing shared individual difference variance, and then neural-cognition relations are examined at the level of factors instead of individual measures. Unlike the situation portrayed in panel A, interrelations among each type of measure are examined to determine a set of dimensions that parsimoniously represents the structure of individual differences within each type of measure before examining relations between the 2 types of measures.

A third possible structure is illustrated in panel C in which the measures are organized into both specific factors and a common factor. In this latter structure, the common factor represents influences shared among all measures, and the specific factors represent influences shared among subsets of measures that are independent of what is common across all measures. In the psychometric literature, this type of model is known as a bifactor (or nested-factors or orthogonal common factor) model, and the common factor is often designated g.

Bifactor and related hierarchical models of cognitive functioning have been investigated in a large number of studies, including several examining influences associated with adult age (e.g., Hildebrandt et al., 2011, Salthouse and Davis, 2006, Salthouse and Ferrer-Caja, 2003, Salthouse, 2009, Schmiedek and Li, 2004). A consistent finding in these studies has been a moderately large influence of age on the common factor, with additional influences on one or more cognitive ability factors. Several studies have also examined relations of neural measures with structural models of cognitive ability. For example, Karama et al. (2011) and Menary et al. (2013) found that most of the relations between cortical thickness and cognition in samples of adolescents were evident at the level of the common cognitive factor, and Booth et al. (2013) and Penke et al. (2010) found that most of the relations between measures of white matter integrity and cognitive measures in older adults were at the level of the common cognitive factor.

Although cortical thickness is often measured across very small cortical regions, broad areas of significant thickness-cognition relations have been reported in many studies (e.g., Choi et al., 2008, Desrivieres et al., 2015, Karama et al., 2011, Menary et al., 2013, Narr et al., 2007, Walhovd et al., 2006), and some researchers have used a measure of the average thickness across multiple brain regions when examining relations with cognition (e.g., Hedden et al., 2014, Hutton et al., 2009, Schnack et al., 2015). We are aware of only 1 study in which interrelations of cortical thickness measures were examined. In that study, Ecker et al. (2009) found that a hierarchical model with first-order factors corresponding to different lobes provided a good fit to cortical thickness data based on 28 gyrus-defined brain regions. Significant interrelations of measures of regional volume (e.g., Kennedy et al., 2009, Raz et al., 2005) and white matter integrity (e.g., Li et al., 2012, Lovden et al., 2013, Penke et al., 2010, Penke et al., 2012, Wahl et al., 2010) have also been found in several recent studies, which implies that the measures could be organized into a relatively small number of factors.

Because very few measures within a given modality are unrelated to other measures in that modality, it is important to consider relations among the measures within a given modality when investigating relations between neural measures and cognitive measures. The method advocated here is to first determine the organizational structure with each type of measure, and then examine relations at the broadest or most general levels in the structures before considering any relations that might exist at more specific levels.

As noted earlier, reductions in cortical thickness with increased age could be postulated to contribute to age-related differences in cognitive performance. However, because many neural and cognitive measures are related to age, at least some of the relations between neural and cognitive measures could be attributable to the relations each type of measure has with age. This possibility can be investigated by comparing the cortical thickness–cognition relations before and after statistical control of the variability in age. The reasoning is that, if the neural-cognition relations are substantially reduced when there is little variation in age, at least some of the relations could be inferred to be attributable to the associations of both the neural and cognitive measures with age.

To summarize, the primary goal of the present study was to demonstrate the usefulness of a proposed analytical procedure for investigating neural substrates of age-cognition relations involving examination of the structure of cortical thickness measures and of cognitive measures, and investigating the levels in the respective structures at which the 2 types of measures are related to another. In addition, the thickness-cognition relations were examined before and after control of the variation in age to investigate the role of age on those relations.