Signatures of life course socioeconomic conditions in brain anatomy.

Socioeconomic status (SES) plays a significant role in health and disease. At the same time, early-life conditions affect neural function and structure, suggesting the brain may be a conduit for the biological embedding of SES. Here, we investigate the brain anatomy signatures of SES in a large-scale population cohort aged 45-85 years. We assess both gray matter morphometry and tissue properties indicative of myelin content. Higher life course SES is associated with increased volume in several brain regions, including postcentral and temporal gyri, cuneus, and cerebellum. We observe more widespread volume differences and higher myelin content in the sensorimotor network but lower myelin content in the temporal lobe associated with childhood SES. Crucially, childhood SES differences persisted in adult brains even after controlling for adult SES, highlighting the unique contribution of early-life conditions to brain anatomy, independent of later changes in SES. These findings inform on the biological underpinnings of social inequality, particularly as they pertain to early-life conditions.

INTRODUCTION

Low socioeconomic status (SES) contributes to negative health outcomes (Marmot & Bell, 2012), including cardiovascular disease (Kanjilal et al., 2006), diabetes (Stringhini et al., 2013), and decreased life expectancy (Stringhini et al., 2017). SES is further linked to differences in cognitive function (Aartsen et al., 2019). For instance, disadvantaged socioeconomic groups have an increased risk of dementia (Mayeda, Glymour, Quesenberry, & Whitmer, 2016), underscoring the putative link between brain health and SES (Resende, Guerra, & Miller, 2019). Evidence also points to a cumulative effect of socioeconomic disadvantage over time on health outcomes (Pollitt et al., 2008) highlighting the need to adopt a life course perspective when probing links between SES and physiological markers of health.

Links between SES and cognition support the notion that the brain is a plausible candidate for the biological embedding of SES, spurring several studies on the neural correlates of SES in the past decade. In the developing brain, childhood SES is tied to brain anatomy (Moriguchi & Shinohara, 2019) and function (Larsson, Solomon, & Kohn, 2015), such as reading abilities (Merz, Maskus, Melvin, He, & Noble, 2020). Specifically, hippocampal volumes correlate positively with SES (Hanson, Chandra, Wolfe, & Pollak, 2011); (Noble et al., 2015), as does cortical thickness (Lawson, Duda, Avants, Wu, & Farah, 2013). These observations suggest childhood SES may be associated with effects on language (Sarsour et al., 2011), reading abilities (Noble, Farah, & McCandliss, 2006), and mental health status (Reiss, 2013). Studies in adults are more limited, but generally support a link between SES and regional brain volumes (Elbejjani et al., 2017; Rzezak et al., 2015; Raizada & Kishiyama, 2010; Jednoróg et al., 2012), although a recent meta‐analysis served to highlight the diversity of specific SES neural correlates across studies (Yaple & Yu, 2020).

Studies on the neural imprints of SES nevertheless remain comparatively sparse (McDermott et al., 2019) and at times yield varied results (Farah, 2017; Yaple & Yu, 2020). Further, studies have more often relied on region‐of‐interest (ROI) analyses rather than a whole‐brain investigation (Farah, 2017; Yaple & Yu, 2020), leaving results open to bias (Poldrack, 2006). The tendency to limit analyses to specific regions may reflect the nature of the feature studied. SES presents a social construct, as opposed to a nosological entity, and therefore, corollary neural differences in the population should be subtle, requiring large‐scale studies to be identified at the whole‐brain level.

It further remains unclear whether SES‐related differences in late‐life reflect traces of childhood SES, as the latter may resolve with a higher SES in adulthood or maturation, or, conversely, persist into old age. A large body of literature has identified lingering effects of childhood deprivation on adult well‐being (Duncan, Ziol‐Guest, & Kalil, 2010; Magnuson & Votruba‐Drzal, 2008; Raphael, 2011), but, to date, few have queried the human brain to assess distal, neural traces of economic conditions in childhood (Tribble & Kim, 2019). While some studies uncover a positive association between childhood SES and increased hippocampal volumes in adulthood (R. T. Staff et al., 2012), others do not (Elbejjani et al., 2017; Lawson et al., 2017).

Neuroimaging studies on the effects of SES on the human brain have traditionally assessed morphometry characteristics using surface‐ and voxel‐based computational anatomy techniques (Noble et al., 2015; Lawson et al., 2013). Recent advances in quantitative magnetic resonance imaging (qMRI) methods allow for a more direct measurement of brain tissue properties that correlate with histological measures (Edwards, Kirilina, Mohammadi, & Weiskopf, 2018; Weiskopf, Mohammadi, Lutti, & Callaghan, 2015). Specifically, qMRI provides access to brain tissue properties—myelin, iron, and tissue water—that correlate with histological measures. Beyond this, relaxometry‐based qMRI minimizes spurious findings in voxel‐ and surface‐based morphometry related to spatially distributed intracortical myelin and iron (Lorio et al., 2016; Natu et al., 2019; Taubert et al., 2020). Thus, qMRI can be used to measure brain tissue properties with an enhanced precision relative to traditional anatomical MRI measures (Tabelow et al., 2019; Trofimova et al., 2021).

Another feature of neuroimaging studies on SES is their focus on gray matter to detect the effects of exogenous variables on neural differences, but white matter may be more susceptible to plastic changes in adulthood (Fields & Bukalo, 2020) and therefore especially pertinent to neural correlates of social adversity (Chahal, Kirshenbaum, Ho, Mastrovito, & Gotlib, 2021). While some have sought SES‐related white matter differences in children (Ozernov‐Palchik et al., 2019) and adults (Johnson, Kim, & Gold, 2013), they have primarily employed tensor‐based models of diffusion‐weighted imaging. Tensor‐based measures of white matter microstructure lack a straightforward neuro‐biological interpretation (Wozniak & Lim, 2006) and are susceptible to inter‐site bias (Moyer, Steeg, Tax, & Thompson, 2020). Noninvasive in vivo white matter assessment remains a challenging endeavor (Heath, Hurley, Johansen‐Berg, & Sampaio‐Baptista, 2018) but magnetization‐transfer (MT) saturation offers a reliable marker of myelin content (Mancini et al., 2020; Melie‐Garcia et al., 2018; Natu et al., 2019). MT refers to the magnetization exchange between free protons and those bound to macromolecules such as myelin (Wolff & Balaban, 1994). While diffusion imaging indexes myelin via the movement of water in fiber tracts and is thus susceptible to several sources of measurement error (Tax et al., 2019), MT saturation maps quantify microstructural properties that are both sensitive and specific to the myelin fraction, rendering inter‐site variability low (Gracien et al., 2020). Imaging studies have found MT saturation correlates to ex‐vivo histological assessment of myelin in postmortem brains (Schmierer, Scaravilli, Altmann, Barker, & Miller, 2004; West et al., 2018), and, in addition, have the added benefit of being less susceptible to inter‐site variance (Lutti, Dick, Sereno, & Weiskopf, 2014). MT saturation's enhanced myelin sensitivity relative to diffusion imaging may therefore better serve in highlighting myelin variation in a population cohort, where differences are expected to be subtle.

In this study, we aimed to identify a neural embedding of childhood SES in an older population. To that end, we probed potential differences in gray matter and myelin content that correspond to SES variability in a population cohort of older adults, using quantitative MRI (Tabelow et al., 2019). We hypothesize that childhood SES will be reflected in late‐life neural markers even when adjusting for SES in adulthood. We investigate these differences at both the whole‐brain level, to query differences that may be evoked by qMRI's sensitivity; and also probe the hippocampus as an a priori ROI. We query this hypothesis by analyzing a large population cohort (n = 1,166) of older adults (mean age = 59.65 years) from one scanner site; employing quantitative neuroimaging using multiparametric maps; applying a data‐driven measure of SES; and exploiting a reliable marker of myelin. We further hypothesize that such differences will be observed in both gray and white matter, as quantified by MT saturation mapping. While previous studies have found mixed results regarding later‐life neural markers of childhood socioeconomic conditions, the characteristics of a large qMRI dataset may yet identify associated brain markers.

METHODS AND MATERIALS

Cohort

The study sample (BrainLaus) is a nested neuroimaging project within the CoLaus|PsyCoLaus general population cohort of the city of Lausanne, Switzerland (Firmann et al., 2008; Preisig et al., 2009). Specifically, the BrainLaus sample consists of CoLaus participants that were both eligible and willing to undergo MRI scanning. The BrainLaus study aims to scan participants at two time points, spaced 5 years apart. These two time points represent the third and fourth study time points of the greater CoLaus study. Analyses were performed on imaging data acquired between 2014 and 2018 and represented the first BrainLaus time point. A total of 1,274 participants were scanned at a single MRI site (Figure 1a).

FIGURE 1

Cohort characteristics. (a) The BrainLaus study comprises a subset of the PsyCoLaus cohort, which is itself a subset of the population cohort (Cohorte Lausanne, CoLaus). (b) The CoLaus Cohort includes a representative sample of the population, which is reflected in the BrainLaus subset, but for age. Here, age distributions for participants in the BrainLaus study are shown alongside age distributions for participants that did not undergo MR scanning

Cohort description

The CoLaus|PsyCoLaus study was designed to recruit a representative sample of the general population (Firmann et al., 2008). We sought to determine if the BrainLaus subset differed from the rest of the cohort on a number of key dimensions by examining differences between CoLaus|PsyCoLaus participants that were not included in the BrainLaus cohort (n = 5,401); and the BrainLaus cohort on all measures available for somatic variables (n = 1,274). (A full list of variables can be found in Appendix A). There was no significant difference in sex, education level, or last known occupation distributions between the two cohorts. We found a significant difference in age between the two cohorts, with an average age of 63 for CoLaus|PsyCoLaus participants without MRI scan and 59 for BrainLaus participants (Cohen's d = 0.4). This result underlines the necessity of including age as a nuisance variable in subsequent neuroimaging analyses that refer to epidemiological results drawn from the CoLaus|PsyCoLaus cohort (Figure 1b).

MRI data acquisition

The scanning protocol included a multiparameter mapping (MPM) relaxometry protocol (Taubert et al., 2020) and diffusion‐weighted acquisition that was not used in the current study. Approximate total scan duration lasted 4 ~ 5 min. Analyses were performed in SPM12 (http://www.fil.ion.ucl.ac.uk/spm/) using Matlab, 2017. Socioeconomic data were missing for 16 of the n = 1,182 participants whose neuroimaging data were retained. A total of n = 1,166 participants (mean age: 59.65 years; 622 females, 544 males) were included in our analyses.

Quantitative MT maps were calculated using a multi‐echo 3D FLASH (fast low angle shot) protocol at a 1 mm, isotropic resolution (Weiskopf et al., 2013). The MRI data was acquired with T1‐, PD‐, and MT‐weighted contrast (respective repetition time/flip angle [FA] of 23.7 ms/21°C, 23.7 ms/6°C, and 23.7 ms/6°C [MT]). For the MT‐weighted contrast, an off‐resonance Gaussian MT saturation RF pulse (4 ms, FA = 220°C, 2 KHz frequency offset) was applied before nonselective excitation. Multiple echo images were acquired with echo times ranging from 2.2 to 19.7 ms (except for the MT‐weighted scans where the maximum echo time was 17.2 ms, due to the application of the MT saturation pulse). We used GRAPPA parallel imaging (acceleration factor of 2) in anterior–posterior phase encoding direction and 6/8 partial Fourier acquisitions in the partition direction (left–right). The protocol also included the acquisition of MRI data for the mapping of the radio‐frequency excitation field B1 (Lutti et al., 2014). This data was acquired using the technique described in (Lutti et al., 2012). Acquisition settings were identical to those described in (Taubert et al., 2020).

MRI data preprocessing

Acquired MRI data underwent automated preprocessing in the multichannel unified segmentation Bayesian framework of SPM12 yielding GM and WM probability maps derived from MT and PD* maps. To achieve higher anatomical precision, we used additionally the diffeomorphic spatial registration DARTEL based on all individual GM and WM tissue maps (Ashburner & Friston, 2005) to then apply the derived spatial registration parameters onto gray matter volume and MT saturation maps and warp these to standard MNI space. Aiming to preserve the initial total MT saturation signal, we followed the default settings for implementation of an established weighted‐averaging procedure using in‐house software tools (Draganski et al., 2011 ).

Image quality assessment

Data quality assessment in neuroimaging is a crucial antecedent to data analysis (Alfaro‐Almagro et al., 2018; Esteban et al., 2017). Given the size and average age of the cohort, as well as the plurality of MRI data acquired, a multistep image quality procedure was applied to our initial sample of n = 1,274. In a first instance, we computed regional averages for MT, R2*, and gray matter volumes for each participant by applying individual inverse deformation fields to anatomical derived from the MICCAI 2012 Grand Challenge and Workshop on Multi‐Atlas Labeling (https://my.vanderbilt.edu/masi/about-us/resources-data), yielding 129 values for each participant, for each dataset. Individual average values falling outside a range of ±4 SDs from the group mean of a specific brain region were flagged (n = 55). In a second instance, we examined differences between individual GM, WM, and CSF segmentations and corresponding canonical tissue probability maps. We first binarized individual tissue segmentations before conducting this procedure. Resulting images were then vectorized, assigned a value of 1 for all voxels >0 and summed. Participants for whom this total difference exceeded the group's average difference by 2 SDs were subsequently flagged (n = 31). Finally, we performed a visual inspection of datasets that showed high SDs of the R2* parameter in white matter. This index has been shown to exhibit a high correlation with motion history during data acquisition (Castella et al., 2018). The criterion for a high SD was set to a conservative cut‐off, which flagged approximately 500 potentially problematic datasets. As our cohort tended toward an older population, we expect more movement than average. Therefore, we visually examined these 500 datasets to identify gross movement, physiological anomalies, or other artifacts. This visual rating identified n = 25 problematic datasets. Neuroimaging datasets that failed one or more of the above quality check were excluded from analysis (n = 80). An additional six participants did not have complete neuroimaging datasets, and a further six were found to have been scanned with a different head coil and were also excluded from the data analysis pool. Finally, SES data were missing for n = 16 of the retained neuroimaging datasets, leaving a total of n = 1,166 participants included in the final analysis (Appendix D).

Neuroimaging data analysis

Analyses were performed in SPM12, using Matlab, 2017. We designed three multiple regression analyses in SPM for each neuroimaging dataset to examine brain differences linked to SES in the cohort. In the first two models (Model 1 and Model 2), we included either adult SES (aSES) or childhood SES (cSES) as a covariate of interest. For the third (Model 3), we designed a full model that included both aSES and cSES. By including the two SES variables, we can assess the unique contribution of one or the other to neural outcome variables. Importantly, we did not orthogonalize these two measures as no firm principle can attribute primacy to one or the other. Finally, age, sex, and total intracranial volume (TIV)—a proxy for head size—were included in the design as nuisance variables (Peelle, Cusack, & Henson, 2012).

Our approach to statistical analysis was informed by a desire to balance both Type I and Type II errors (Eklund, Nichols, & Knutsson, 2016; Kang, Blume, Ombao, & Badre, 2015; Noble, Scheinost, & Constable, 2020). Therefore, we performed both whole‐brain and ROI analyses, detailed below. We tested for the overall contribution of SES to differences in neural data by estimating coefficients using threshold‐free cluster enhancement (TFCE) and applying nonparametric tests (5,000 permutations) to probe for significance (Smith & Nichols, 2009). Analyses were performed with a significance threshold of p = 0.05, FWE corrected for multiple comparisons across the whole search volume, comprising either the brain's entire gray matter or white matter (Ashburner & Friston, 2005). For the whole brain analysis, we report TFCE and t‐statistic results that survive FWE correction with a threshold of p = 0.05.

In a second instance, we performed a small volume correction (SVC) analysis focusing on the hippocampus, a region that consistently emerges in studies probing the neural correlates of SES (Hanson et al., 2011; Jednoróg et al., 2012; Ursache & Noble, 2016; Elbejjani et al., 2017; Piras, Cherubini, Caltagirone, & Spalletta, 2011; Yaple & Yu, 2020; R. T. Staff et al., 2012; Lawson, Mathys, & Rees, 2017) in Model 3. Coefficients were estimated using the TFCE methods applied to a hippocampal mask comprised of left and right hippocampi. We constrained our a priori region set to the hippocampus to minimize the risk of errors of reverse inference (Poldrack, 2011). Regional masks applied in the SVC analysis were derived from the MICCAI 2012 Grand Challenge and Workshop on Multi‐Atlas Labeling (https://my.vanderbilt.edu/masi/about-us/resources-data).

SES variables

The CoLaus|PsyCoLaus longitudinal cohort study collected a wide range of sociodemographic variables from which we derived SES measures. SES can be indexed in several ways; however, consensus holds that three observable variables can serve as valid measures of the underlying construct, namely, education, income, and occupation levels (Winkleby, Jatulis, Frank, & Fortmann, 1992; Oakes & Rossi, 2003). Because SES is a multifactorial construct, it is commonly indexed by composite measures (Shavers, 2007; Mueller & Parcel, 1981; Stumm et al., 2020) that can be weighted empirically before being aggregated into one score (Cowan et al., 2012; United Nations Economic Commission for Europe, 2018). In this study, we focused on the above three facets of SES to construct a composite measure. CoLaus|PsyCoLaus demographic data included information on mean income (in six intervals), education (three levels), and self and partner's last known occupation. Education levels were ranked according to highest level completed in the following manner: mandatory school or apprenticeship (low); high school diploma or upper secondary education (middle); and university degree and above (high). Occupations were ranked according to the European Socioeconomic Classification (ESEC) scale (https://www.iser.essex.ac.uk/archives/esec/user-guide) (nine levels) and assigned a corresponding numerical value; own income was taken to be highest household income between spouses, where applicable. Measures of childhood SES included father's occupation (ranked according to the ESEC scale); highest parental education (three levels); and a measure of childhood household financial status as proxy of childhood income (Appendix B). This last measure included a sum of nine positive and negative answers for family lifestyle and conditions, such as ownership of a car and having insufficient heating. The following variables were scaled into tertiles and assigned values ranging from 0 to 2. Adult occupation, taken as highest household occupation, mean income, paternal occupation, and childhood finances. To further obtain a precise, data‐driven measure of adult and childhood SES constructs specific to our cohort, we extracted variance contributions from each of the SES components listed by performing two PCAs for the trio of adult and childhood SES variables. We found that, in adulthood, education explained most of the variance (63.7%), followed by income (21.84%) and occupation (14.5%). In childhood, household income explained most of the variance (70.39%), followed by education (16%) and occupation (13.6%). We then created a composite measure of adult SES and one of childhood SES by weighting tertile measures of income, education, and occupation with their respective variance contributions before summing them. This procedure allowed for the range of possible SES variables to increase from 3 to 48, with a concomitant increase in information, as formalized by entropy, from 1.58 to 4.68 and 1.55 to 3.67 bits, for adult and childhood SES, respectively. By weighting the SES composite measure components by their respective variance weights, we produced a single, precise, sample‐specific measure of SES to include as an independent variable in our analyses (see Appendix E for analyses on unweighted composite SES).

RESULTS

There was no significant difference in sex, education level, or last known occupation distributions between the CoLaus|PsyCoLaus and the BrainLaus subsample. We found a significant difference in age between the two subsamples, with an average age of 63 for CoLaus|PsyCoLaus participants without MRI scan and 59 for BrainLaus participants (Cohen's d = 0.4). This result underlines the necessity of including age as a nuisance variable in subsequent neuroimaging analyses that refer to epidemiological results drawn from the CoLaus|PsyCoLaus cohort (Figure 1b).

Model 1—Brain differences associated with adult SES

MT differences associated with adult SES

Adult SES was tied to decreases in MT density in the right entorhinal cortex (Table 1; Figure 2).

TABLE 1

Whole brain voxel level analysis of MT load and gray matter in relation to adult SES (aSES)

Region	Cluster size k	p(FWE‐corr)	t	Coordinates
				x	y	z mm
MT aSES negative correlation
R Entorhinal cortex	73	0.005	4.84	20	0	−45

Note: Results shown above are significant at a threshold of p = 0.05, FWE corrected for multiple comparison using the TFCE method.

Abbreviations: GM, gray matter; MT, magnetization transfer.

FIGURE 2

Results of GLM 1 (adult SES). (a) Results of a negatively signed one‐sample t test (t‐statistic) of adult SES on MT. (b) Results of a positively signed one‐sample t test of adult SES on gray matter volume (TFCE). Colorbars indicate t values. All maps shown are thresholded at p = 0.05, FWE‐corrected for multiple comparisons

Gray matter volume differences associated with adult SES

Adult SES correlated positively with gray matter volume in several regions, including right postcentral gyrus, left precuneus, left thalamus, and right cerebellum (exterior) (Table 1; Figure 2). SVC analyses on the hippocampus reveal greater bilateral gray matter in association with SES (Table 4).

We also probed the possible interaction of SES with age to query a differential aging effect modulated by SES (Steptoe & Zaninotto, 2020), but found no results in either MT or gray matter, when correcting for multiple comparisons.

Model 2—Brain differences associated with childhood SES

MT changes associated with childhood SES

Childhood SES (cSES) correlated significantly positively with MT in right superior parietal lobule. In white matter, cSES correlated positively with MT near the pallidum/ventral tegmentum and bilateral precentral gyrus (Table 2). The pattern found in MT notably delineates the sensorimotor network (van den Heuvel & Hulshoff Pol, 2010) (Table 2; Figure 3).

TABLE 2

Whole brain voxel level analysis of MT load and gray matter volumes in relation to childhood SES (cSES)

Region	Cluster size k	p(FWE‐corr)	t	Coordinates
				x	y	z mm
MT CSES positive correlation
R superior parietal lobule	40	0.006	4.77	21	−45	68

Note: Results shown above are significant at a threshold of p = 0.05, FWE corrected for multiple comparison using the TFCE method.

Abbreviations: GM, gray matter; MT, magnetization transfer.

FIGURE 3

Results of GLM 2 (Childhood SES) in MT maps and gray matter volumes. (a) Results of both positively (TFCE) and negatively (t‐statistic) signed one‐sample t tests of childhood SES on MT. (b) Results of both positive (TFCE) one‐sample t tests on childhood SES on in gray matter. All maps shown are thresholded at p = 0.05, FWE‐corrected for multiple comparisons

Gray matter volume differences associated with childhood SES

Childhood SES correlated positively with gray matter volume in several regions, including right cerebellum, left postcentral gyrus, right lingual gyrus, brainstem, left precentral gyrus, left inferior temporal gyrus, and left occipital fusiform gyrus (Table 2; Figure 3). SVC analyses reveal significant associations with cSES in the left hippocampus (Table 4).

As with adult SES, no results emerged for an age by childhood SES interaction in either MT or gray matter, when correcting for multiple comparisons.

Model 3—Full model

To better inform our hypothesis, we further analyzed both child‐ and adulthood‐SES in the same model. These two variables are significantly correlated (r = 0.536, p < 0.001), and thus we computed their variance inflation factor (VIF) to determine if the presence of multicollinearity could be tolerated, finding a value of 1.40, which falls below a conservative cutoff of 5 (Mumford, Poline, & Poldrack, 2015). At the whole brain level, significant results were found for positive associations between childhood SES and MT in the left pallidum and precentral gyrus, and gray matter in bilateral cerebellum, left cuneus, left postcentral gyrus and right thalamus, and middle temporal gyrus (Table 3; Figure 4).

TABLE 3

Whole brain voxel level analysis of MT and gray matter in relation to SES in a full model including both childhood and adult SES

Region	Cluster size k	p(FWE‐corr)	TFCE	Coordinates
				x	y	z mm
MT cSES positive t test
L pallidum	1,311	0.033	1,173.28	−12	−2	−6
		0.033	1,167.98	−6	−6	−9
		0.038	1,137.08	−14	−9	−12
L precentral gyrus	992	0.041	1,086.14	−30	−24	58
		0.041	1,073.47	−15	−16	68
		0.043	1,056.36	−18	−32	66
L precentral gyrus	104	0.048	1,018.33	−44	−15	44
		0.049	1,016.70	−38	−12	48
GM cSES positive t test
l cuneus	24,255	0.001	2,604.20	−4	−93	12
		0.001	2,419.18	3	−60	4
		0.001	2,415.93	−9	−93	26
L cerebellum	5,861	0.002	2026.42	−21	−62	−58
		0.002	1938.24	−33	−75	−52
		0.030	1,066.86	−46	−64	−34
R cerebellum	3,596	0.002	1929.93	20	−60	−62
		0.007	1,469.33	36	−74	−52
		0.007	1,450.83	27	−72	−52
L postcentral gyrus	663	0.018	1,252.65	−8	−36	78
		0.022	1,145.61	−21	−32	75
		0.048	951.81	−30	−33	58
R middle temporal gyrus	707	0.031	1,055.50	57	−4	−16
		0.031	1,052.70	51	−6	−10
		0.033	1,039.95	54	0	−26
Brainstem	548	0.033	1,034.89	6	−30	−9
		0.039	994.53	−4	−30	−8
R postcentral gyrus	84	0.044	965.48	9	−40	78
R thalamus	224	0.047	953.73	24	−28	9
		0.048	951.87	16	−28	12
R inferior temporal gyrus	77	0.048	948.40	54	−54	−16
L thalamus	56	0.049	938.82	−10	−30	14
R superior parietal lobule	30	0.050	927.97	22	−42	72
R transverse temporal gyrus	16	0.050	926.23	46	−10	6

Note: Results shown above are significant at a threshold of p = 0.05, FWE corrected for multiple comparison using the TFCE method.

Abbreviations: GM, gray matter; MT, magnetization transfer.

FIGURE 4

Results of GLM 3 for childhood SES including adult and childhood SES as covariates in MT maps and gray matter volumes. (a) Results of positive childhood SES correlates in MT. (b) Results of t tests on childhood SES in gray matter. Colorbars indicate TFCE‐values. All maps shown are thresholded at p = 0.05, FWE‐corrected for multiple comparisons

Small volume correction analysis, model 3—Adult SES

SES correlated positively with the left hippocampus gray matter volume (Table 4).

TABLE 4

Small volume correction analysis of MT load and gray matter volumes in relation to childhood and adult SES in Model 3

Region	Cluster size k	p(FWE‐corr)	TFCE	Coordinates
				x	y	z mm
MT cSES
Positive T test
L hippocampus	1	0.021	153.18	−16	−16	−16
L hippocampus	1	0.022	151.26	−15	−14	−16
L hippocampus	2	0.024	144.55	−18	−18	−15
L hippocampus	1	0.026	138.78	−18	−14	−14
GM aSES
Positive T test
L hippocampus	256	0.011	203.06	−16	−9	−22
L hippocampus	3	0.045	125.61	−20	−27	−10
GM cSES
Negative T test
R hippocampus	301	0.016	205.34	36	−28	−9

Note: TFCE analysis results constrained to an anatomical mask for the hippocampus (left and right). Results shown above are significant using a threshold of p = 0.05, FWE corrected for multiple comparisons constrained to the search volume.

Small volume correction analysis, model 3—Childhood SES

Childhood SES correlated positively with the left hippocampus in white matter. Childhood SES further correlated negatively with right hippocampal volume (Table 4).

DISCUSSION

In this study, we examined the relationship between life‐course SES and structural brain properties using MRI‐derived estimates indicative of myelin content and gray matter volume in mid‐ and old‐age individuals from the general population. In contrast to previous studies, we adopted a life‐course perspective and hypothesized that neural traces of childhood SES remain when controlling for adult SES. We found that both childhood and adult SES separately correlated with gray matter volume and myelin differences. The effect of childhood SES on gray matter volume and myelin content was independent of adult SES circumstances. Childhood SES was associated with robust neural differences even when controlling for adult SES. Our results support the hypothesis that childhood SES leaves a neural imprint even in adulthood and more generally, corroborate the latent effect model for the impact of childhood SES on adult outcomes (Nelson & Gabard‐Durnam, 2020).

Studies on neural imprints of SES have yielded variable results, as highlighted in a recent meta‐analysis (Yaple & Yu, 2020), which may be due in part to limited sample sizes and acquisitions across different sites. Two key studies have attempted to overcome this problem (Noble et al., 2015; McDermott et al., 2019), finding, in the first, a positive correlation between parental education and hippocampal volume, without adjusting for family income, and an association between cortical surface area and both parental income and education. In the second study, widespread cortical surface area and hippocampus similarly correlate with higher SES, as assessed by the Hollingshead score. While these two studies yield concordant results, they are both found in pediatric cohorts and do not investigate white matter measures. Our results support hippocampal involvement in SES differences, with higher bilateral hippocampal gray matter correlating with higher adult SES. However, only left hippocampal gray matter and myelin correlated with childhood SES; right hippocampal gray matter displayed a negative association with childhood SES, and at the same time, a positive association between MT and the same variable. This discrepancy in the relationship between hippocampus and SES may highlight a specificity in effects on brain tissue properties, as hippocampal volume may cede to increases in myelination (Natu et al., 2019).

An innovative feature of our study is the use of MT saturation maps to extract estimates of gray matter volumes (Helms, Draganski, Frackowiak, Ashburner, & Weiskopf, 2009) and myelin content (Melie‐Garcia et al., 2018), which in part serves to dampen scanner site variability (Focke et al., 2011; Gracien et al., 2020). Our results therefore offer an added reliability over findings obtained with traditional MRI methods, particularly with regards to myelin quantification methods. Further, myelin may be a more pertinent metric for function across the lifespan (Ziegler et al., 2019; Chen, Chen, Hsu, & Tseng, 2020).

SES differences in brain's myelin

Most studies on in vivo structural brain properties linked to SES focus on gray matter volume or cortical thickness measures. However, myelin plays a crucial role in brain function and dysfunction (Fields & Bukalo, 2020) and therefore should not be neglected. As shown in our study, MT values covaried with SES in regions distinct from gray matter volume changes, highlighting myelin's independent status in the brain. Our results support a recent study showing a relationship between neighborhood deprivation, and rate of myelination, as assessed by MT, across childhood and adolescence (Ziegler et al., 2019). Further, we find a positive association of myelin in regions comprising the sensorimotor network with SES. This network has previously been associated to cognitive impairment (Agosta et al., 2010) and MT in old age correlated with motor performance (Seidler et al., 2015). Aging induces cognitive decline (Park, O'Connell, & Thomson, 2003), even in early adulthood (Salthouse, 2009) as well as decreases in motor performance (Thompson, Blair, & Henrey, 2014). Converging evidence highlights the increasing association between cognitive and sensorimotor functions with aging (Li & Lindenberger, 2002). One possibility is that childhood SES may provide a buffer to functional decline in old age via increased myelination of the sensorimotor network.

Regional specificity of SES neural differences

Brain regions found to covary with SES play important roles in cognition, memory, and motor function. The pallidum plays a role in reward and motivation (Smith, Tindell, Aldridge, & Berridge, 2009), as well as motor function (Gillies et al., 2017). The hippocampus plays a significant role in memory (Knierim, 2015) as do regions of the temporal lobe (Wong & Gallate, 2012), also implicated in language functions (Davey et al., 2016). The hippocampus in particular has previously been found implicated in psychosocial adversity (Tottenham & Sheridan, 2009) and is also known to be especially susceptible to plasticity (Leuner & Gould, 2010). Differences in any of these structures may therefore have considerable functional implications. The nature of the study prevents claims of causality between SES and implicated regions. Further, neuroimaging studies are prone to errors of reverse inference (Poldrack, 2011). The exact biological pathway between childhood economic status and health in adulthood remains ill‐defined (Matthews & Gallo, 2011; Foulkes & Blakemore, 2018). Nonetheless, we speculate that the hippocampus in particular may be related to SES by way of enriched environments in childhood (Cassarino & Setti, 2015). SES effects on hippocampal volumes in children are mediated by caregiver quality (Luby et al., 2013) and higher SES may stem stress‐related effects on the hippocampus (McEwen, 2012). SES‐related access to extracurricular activities in childhood and green space may also enhance cognition by way of the temporal lobe (Hillman, Erickson, & Kramer, 2008), a link that may be mediated by motor regions (Voss, Nagamatsu, Liu‐Ambrose, & Kramer, 2011). SES is also related to aberrant reward responses (Hanson et al., 2016; Oshri et al., 2019), which may reflect our findings in the pallidum.

SES differences and the hippocampus

Differences in hippocampal gray matter volume associated with SES have previously been reported in a number of studies (Jednoróg et al., 2012; R. T. Staff et al., 2012). Our results partially support this relationship, but we detect an inverse relationship between right hippocampal gray matter and childhood SES. We otherwise find positive relationships between bilateral hippocampal gray matter and adult SES. Childhood SES in our cohort was nonetheless also associated negatively with MT in bilateral temporal pole, which, like the hippocampus, forms part of the temporal lobe system, and is uniquely sensitive to age‐related decline (Pelletier et al., 2017). Childhood SES further correlates positively with greater MT in left hippocampus (and right hippocampal gray matter), raising the possibility of interplay between myelin and gray matter. For instance, cortical gray matter reduction occurs in healthy adolescence along with an increase of myelination (Giorgio et al., 2010). In the same model, adult SES correlated with increased left hippocampal volume. Our results thus suggest a more complex interaction in temporal lobe regions in relation to SES, with differential effects of myelin, gray matter and early or late‐life SES implicated in disparate neural profiles.

Limitations of the study

Here, childhood SES was assessed using adult recall that is susceptible to faulty memories (Havari & Mazzonna, 2015). Household income in childhood and adulthood are further indexed by different measures and it can be argued that the one for childhood skews towards assessing disadvantage, although this bias may be redressed as retrospective assessments tend to favor a more optimistic view of how things were (Mitchell, Thompson, Peterson, & Cronk, 1997). We also define SES with a composite measure, which does not identify unique risk factors (Hackman, Farah, & Meaney, 2010). Further, we cannot associate function to SES‐related neural differences, as the current dataset does not include cognitive or behavioral variables, nor does it include mother's occupation as a potential indicator. Finally, our study design precludes the possibility to control for context beyond SES in early life that can impact neural structure. In spite of these limitations, our results are based on a precious dataset, as not all large‐scale neuroimaging data have childhood, or conversely, adult data, and few include quantitative MRI maps.

CONCLUSIONS

Our study adds to the growing literature on brain correlates of SES. Known associations between childhood adversity and late‐life outcomes strongly suggest a causal process set into the arrow of time. By highlighting a neurophysiological embedding of childhood SES in old age, our results add credence to the lasting physical incorporation of exogenous social factors into the brain.

CONFLICT OF INTEREST

The authors report no biomedical financial interests or potential conflicts of interest.

TABLE A1

Demographic and somatic measures at baseline

Variable	Timepoint
Sex	Baseline
Age	Baseline
How many years lived in Switzerland	Baseline
Marital status (married, single, divorced, widowed)	Baseline
Education level (low, mid, high)	Baseline
Education level (years)	Baseline
Occupation level (low, mid, high)	Baseline
Number of children	Baseline
Swiss born (yes or no)	Baseline
Date of arrival in Switzerland	Baseline
Mini‐mental state exam score (>60 years only)	Baseline
Minutes walked to work per day	Baseline
Physical activity (weekly frequency of activity >20 min, five categories)	Baseline
Height	Baseline
Weight	Baseline
BMI	Baseline
Adiponection levels	Baseline
Leptin levels	Baseline
Ferritin levels	Baseline
Transferrin levels	Baseline
c‐reactive protein levels	Baseline
Interleukin 1 levels	Baseline
Interleukin 6 levels	Baseline
Tumor necrosis factor alpha levels	Baseline

TABLE A2

Demographic and somatic measures at timepoint 1

Variable	Timepoint
Age	Timepoint 1
Occupation level (9 categories, ESEC)	Timepoint 1
Occupation position (low, mid, high, not working)	Timepoint 1
Occupation position (low, mid, high, not working, housewife)	Timepoint 1
Last known occupational position (low, mid, high, not working)	Timepoint 1
Social benefits (yes or no)	Timepoint 1
Cardio‐myopathy (yes or no)	Timepoint 1
Valvular heart disease	Timepoint 1
Heart failure	Timepoint 1
Arrythmia	Timepoint 1
Coronary artery disease	Timepoint 1
Angina (yes or no)	Timepoint 1
Myocardial infarction	Timepoint 1
Mini‐mental state exam score (>60 years only)	Timepoint 1
Height	Timepoint 1
Weight	Timepoint 1
BMI	Timepoint 1
Insulin levels	Timepoint 1
Adiponection levels	Timepoint 1
Leptin levels	Timepoint 1

TABLE A3

Demographic and somatic measures at timepoint 2

Variable	Timepoint
Age	Timepoint 2 (brain imaging timepoint)
Date of exam	Timepoint 2 (brain imaging timepoint)
Social benefits disability	Timepoint 2 (brain imaging timepoint)
Social benefits retirement	Timepoint 2 (brain imaging timepoint)
Employment status (currently working, yes/no)	Timepoint 2 (brain imaging timepoint)
Occupation position (ESEC category)	Timepoint 2 (brain imaging timepoint)
Hypertension (yes or no)	Timepoint 2 (brain imaging timepoint)
Diabetes (yes or no)	Timepoint 2 (brain imaging timepoint)
Cardio‐myopathy (yes or no)	Timepoint 2 (brain imaging timepoint)
Valvular heart disease	Timepoint 2 (brain imaging timepoint)
Heart failure	Timepoint 2 (brain imaging timepoint)
Arrythmia	Timepoint 2 (brain imaging timepoint)
Coronary artery disease	Timepoint 2 (brain imaging timepoint)
Angina (yes or no)	Timepoint 2 (brain imaging timepoint)
Myocardial infarction	Timepoint 2 (brain imaging timepoint)
Alcohol consumption (yes or no)	Timepoint 2 (brain imaging timepoint)
Alcohol consumption, weekly rate	Timepoint 2 (brain imaging timepoint)
Smoking status (smoker, non‐smoker, former smoker)	Timepoint 2 (brain imaging timepoint)
Weight	Timepoint 2 (brain imaging timepoint)
Height	Timepoint 2 (brain imaging timepoint)
BMI	Timepoint 2 (brain imaging timepoint)
BMI category (underweight, normal, overweight)	Timepoint 2 (brain imaging timepoint)
Bioimpedance measure	Timepoint 2 (brain imaging timepoint)
High density lipoprotein cholesterol	Timepoint 2 (brain imaging timepoint)
Low density lipoprotein cholesterol	Timepoint 2 (brain imaging timepoint)
Triglycerides level	Timepoint 2 (brain imaging timepoint)
Glucose level	Timepoint 2 (brain imaging timepoint)
c‐reactive protein levels	Timepoint 2 (brain imaging timepoint)
Interleukin 1 levels	Timepoint 2 (brain imaging timepoint)
Interleukin 6 levels	Timepoint 2 (brain imaging timepoint)
Tumor necrosis factor alpha levels	Timepoint 2 (brain imaging timepoint)

TABLE C1

Financial status in childhood, questionnaire

Question	Numerical coding
Family owning a car during participant's childhood	Yes = 1
Family owning a TV during participant's childhood	Yes = 1
Family employing a maid during participant's childhood	Yes = 1
Family owning a dish‐washer during participant's childhood	Yes = 1
Family owning a telephone during participant's childhood	Yes = 1
Family having enough heat when cold during participant's childhood	Yes = 1
Family member participating to social/cultural association during participant's	Yes = 1
Family going on holidays (outside of home) during participant's childhood	Yes = 1
Family owning their home during participant's childhood	Yes = 1

TABLE E1

MT statistics for weighted vs un‐weighted childhood SES composite score

MT values in gray matter masks
Weighted childhood SES						Unweighted childhood SES
Cluster p(FWE)	Cluster k	Peak Z	Coordinates			Cluster p(FWE)	Cluster k	Peak Z	Coordinates
			x	y	z mm				x	y	z mm
0.371	366	4.62	21	−45	68	0.751	147	4.1	21	−45	68
0.846	100	3.87	34	36	14	0.961	31	3.52	34	36	14
0.711	166	3.86	68	−26	−26	0.975	19	3.46	−34	32	14
		3.25	60	−30	−15	0.88	82	3.44	15	−98	4
0.17	592	3.8	−2	−4	−8	0.978	16	3.43	−45	−58	38
		3.51	−12	−2	0	0.969	24	3.36	60	−46	−21
0.748	148	3.79	2	−88	6	0.936	49	3.34	−6	−3	−4
		3.1	2	−93	−4	0.984	10	3.21	26	−34	68
0.507	274	3.74	22	−92	3	0.976	18	3.2	6	−3	−4
		3.59	32	−94	3	0.99	4	3.15	−12	−4	−4
		3.57	16	−98	6	0.992	2	3.1	68	−26	−26
0.903	69	3.61	14	−4	42	0.993	1	3.09	−14	−86	20
0.878	83	3.59	54	28	−4
0.665	189	3.54	6	−24	69
		3.44	12	−34	72
0.97	23	3.44	−56	−45	27
0.931	52	3.44	−64	−15	−27
0.94	46	3.44	38	−54	34
0.981	13	3.43	−24	−16	57
0.98	14	3.39	−12	−4	39
0.975	19	3.36	32	26	33
0.942	45	3.33	26 ‐	33	68
0.973	21	3.33	60	−46	−21
0.949	40	3.33	21	−22	56
0.979	15	3.3	−57	−50	−4
0.985	9	3.28	51	−76	−2
0.883	80	3.26	−28	−30	60
0.984	10	3.23	30	−15	63
0.979	15	3.22	−12	−36	72
0.991	3	3.21	58	48	0
0.984	10	3.18	−2	−90	26
0.99	4	3.16	−34	33	15
0.982	12	3.13	−3	−24	74
0.993	1	3.11	39	3	−24
0.993	1	3.09	0	−87	33
0.993	1	3.09	−14	−86	20

TABLE E2

Gray matter volume statistics for weighted versus un‐weighted childhood SES composite score

Gray matter volumes
Weighted childhood SES						Unweighted childhood SES
Cluster p(FWE)	Cluster k	Peak Z	Coordinates			Cluster p(FWE)	Cluster k	Peak Z	Coordinates
			x	y	z mm				x	y	z mm
0.002	150	5.12	−6	−54	72	0.001	235	5.45	−6	−54	72
0.01	54	4.79	−12	−21	−6	0.019	25	5.14	−18	−22	−18
0.035	6	4.66	−18	−24	−18	0.012	45	4.71	15	−50	75
0.045	1	4.52	9	−52	74	0.025	16	4.67	32	−78	−54