Layer-Specific Vulnerability is a Mechanism of Topographic Map Aging

Topographic maps form a critical feature of cortical organization, yet are poorly described with respect to their microstructure in the living aging brain. We acquired quantitative structural and functional 7T-MRI data from younger and older adults to characterize layer-wise topographic maps of the primary motor cortex (M1). Using parcellation-inspired techniques, we show that qT1 and QSM values of the hand, face, and foot areas differ significantly, revealing microstructurally-distinct cortical fields in M1. We show that these fields are distinct in older adults, and that myelin borders between them do not degenerate. We further show that the output layer 5 of M1 shows a particular vulnerability to age-related increased iron, while layer 5 and the superficial layer show increased diamagnetic substance, likely reflecting calcifications. Taken together, we provide a novel 3D model of M1 microstructure, where body parts form distinct structural units, but layers show specific vulnerability towards increased iron and calcium in older adults. Our findings have implications for understanding sensorimotor organization and aging, in addition to topographic disease spread.


Introduction
Topographic maps are a hallmark feature of the human brain, covering nearly half of the cortical surface (Sereno et al., 2022). Cortical microcircuits of topographic maps, such as columns and layers, therefore reveal critical insights into the fundamental mechanisms of aging and neurodegeneration. While macrostructural changes in volume and thickness are often described in cortical aging (Thambisetty et al., 2010;Storsve et al., 2014), such measures show inconsistent relationships with behavior and their mechanistic origins are often unclear (Martinez et al., 2015). Microstructural features of topographic aging, reliably measured using quantitative imaging, provide more specific information that can be better linked to ex-vivo human data and animal evidence. Nevertheless, there has rarely been any systematic description of the cortical layer microcircuits of topographic maps in living older adults.
Cortical layers serve as functionally-relevant units, where deep layers typically control output functions and superficial layers typically integrate information from other regions, with distinct cytoarchitectural profiles that are often uniquely affected by pathology. For example, disrupted cross-laminar processing in layer 5 has been shown to precede cell death in aging mice (Lison et al., 2014). While such animal studies have long recognised the importance of layer-specific mechanisms in aging, such success in the specificity of in-vivo human aging has been limited by resolution constraints. Topographic map architectures are also relevant for human aging, where local and global changes in the functional map architecture influence functional integration and separation, and are linked to behavior (Liu et al., 2021). The importance of topographic maps also extends more generally, where functional and structural map maintenance has been related to phantom limb pain in amputees (Makin et al., 2013).
Recent advances in ultra-high field MRI at 7 Tesla (7T-MRI) provide the sub-millimetre resolution required to investigate cortical layers in-vivo (Kuehn & Pleger, 2020). 7T-MRI has successfully been used to parcellate the human sensorimotor cortex (Dinse et al., 2015), detect input and output signal flows in M1 (Huber et al., 2017), and detect layer-specific low-myelin borders between adjacent topographic areas in the sensorimotor cortex (Kuehn et al., 2017). While 7T-MRI is a suitable tool to describe the microstructural architecture of topographic maps in-vivo, it has not yet been applied to a group of healthy older adults. The mechanisms that underlie topographic map aging in humans are therefore yet to be elucidated.
We take the primary motor cortex (M1) as a model system to characterize in-vivo microstructural aging of topographic maps. M1 is the thickest (3-4mm) cortical region (McColgan et al., 2020), therefore providing greater relative resolution per layer compared to any other cortical region. In addition, M1 has a clear large-scale topographic organization and a high signal-to-noise ratio in MR images. The structural architecture of M1 is complex and demands a comprehensive consideration of local variations in microstructure. Although M1 is classically depicted as one homogeneous cortical field (Brodmann, 1909), there are microstructural gradients across two dimensions of the cortical surface (Glasser et al., 2016). Recent research suggests that M1 may be subdivided into separate cortical fields corresponding to topographic areas (Sereno et al., 2022). The microstructure of M1 also varies across a third dimension, cortical depth, according to the distinct cytoarchitecture of cortical layers (Kuehn & Sereno, 2018). M1 is characterized by a particularly thick layer 5 that contains the heavily-myelinated Betz cells that connect directly to the spinal cord to control motor output. Layer 5, which can be subdivided into layer 5a and layer 5b, is often implicated in neurodegeneration (McColgan et al., 2020) and may therefore be relevant in microstructural topographic map aging.
There are a number of open questions associated with the microstructure of M1 in younger and older adults. First, it is unclear whether the different topographic areas representing major body parts would have similar or distinct microstructural profiles. This is important to clarify before investigating how aging affects M1 microstructure. Evidence from mice research has demonstrated varying effects of aging across topographic areas, where for example representations of the hindpaw are more vulnerable to age-related dedifferentiation than those of the forepaw (David-Jürgens et al., 2008). The inhomogeneity of M1 microstructure is further complicated by the presence of low-myelin borders that divide major topographic areas, such as the hand and face (Glasser et al., 2016;Kuehn et al., 2017). The role of these borders in cortical aging, particularly in the enlargement of body part representations, is currently unknown. Finally, it also needs to be clarified which cortical layers, and which topographic areas, are affected most in older adults to understand the precise architecture of topographic map aging.
To characterize microstructural aging with respect to layers and topographic areas, we applied parcellation-inspired techniques to sub-millimetre 7T-MRI data in healthy younger and older adults. We employed validated in-vivo proxies of cortical myelin (quantitative T1 (qT1), Stüber et al., 2014), iron (positive QSM (pQSM)), Langkammer et al., 2012) in addition to a proxy of diamagnetic substance (negative QSM (nQSM)), as measures of cortical microstructure.
Together with functional localisers, we extracted microstructural profiles of the major cortical fields of M1 (lower limb (LL), upper limb (UL), face (F)) (termed according to Glasser et al., 2017). Age-related differences in pQSM and nQSM values were of particular interest, since previous studies have shown increased iron Betts et al., 2016) and calcium (Jang et al., 2021;Kim et al., 2022) in aging. Increased iron has been related to poorer motor function in healthy older adults (Sullivan et al., 2009), whereas the biological source of the nQSM signal is currently under debate (see more details in the discussion). In order to investigate whether such age-effects are uniformly present across cortical layers or are layer-specific, we estimated anatomically-relevant cortical compartments in-vivo. Our approach was based on a comparison between in-vivo and ex-vivo M1 data (Huber et al., 2017), therefore providing a reasonable approximation of anatomically-relevant compartments and their computations (Persichetti et al., 2020). However, please note that there is a conceptual difference between our definition and the definition based on ex-vivo data, where cytoarchitecture is considered (Brodmann, 1909;Vogt & Vogt, 1919).
We applied a systematic approach to characterize microstructure within the different cortical fields of M1 representing the LL, UL, and F areas in a sample of healthy adults. (1) We first hypothesized that M1, in healthy younger adults, is comprised of microstructurally-distinct cortical fields corresponding to topographic areas (as suggested by (Flechsig, 1920), also see (Glasser et al., 2016;Kuehn et al., 2017;Sereno et al., 2022)). (2) We further hypothesized that the microstructure of these cortical fields is also distinct in older adults, but that (3) the low-myelin borders between them (previously shown in younger adults, see Kuehn et al., 2017) are degenerated in older adults. In addition, (4) we hypothesized that iron and diamagnetic substance in older adults would be elevated in a layer-and topographic area-specific way given the above cited evidence in rodents. (5) Finally, we hypothesized that those changes would show a relation to body part-specific motor function.
For all analyses, the focus is on the left (dominant) hemisphere, since all participants were right-handed. However, we also tested whether effects replicate for the right (non-dominant) hemisphere, which is particularly relevant for diseases in which pathology can onset in either hemisphere, or is related to handedness (Turner et al., 2011). To the best of our knowledge, this is the first study to apply recently-introduced techniques of in-vivo 3D parcellation (Alkemade et al., 2022;Kuehn & Sereno., 2018) to group brain data of different age groups. This will allow us to uncover the fundamental aspects of microstructural M1 architecture and aging in the human brain.

Participants
40 healthy volunteers including 20 younger adults (< 35 years of age; 8 female) and 20 older adults (> 70 years of age; 11 female) were enrolled in the present study. After data quality check (see section 2.5.1 for details), a total of 35 participants, including 17 younger adults (8 female) with a mean age of 25 years (SD = 2.7 years) and 18 older adults (11 female) with a mean age of 71 years (SD = 4.0 years) remained for analysis. Additional demographic information and group differences are shown in Table 1 During the functional MRI session (session 2), functional and additional SWI data were collected. The same sequence and parameters as above were used to collect SWI slab images (covering M1 and S1) with a 0.5 mm isotropic resolution. In addition, we obtained whole-brain functional images with a 1.5 mm isotropic resolution (81 slices, field-of-view read = 212 mm, echo time = 25 ms, repetition time = 2000 ms, GRAPPA 2, interleaved acquisition) using an EPI gradient-echo sequence.

Experimental Design
The functional imaging (session 2) involved a blocked-design paradigm, where the participants were instructed to move their left or right foot, left or right hand, or tongue. Participants were trained on the movements outside the scanner and wore fingerless braces, covering the hands and forearms, to reduce large movement of the hands during the precise movement of the fingers and other body parts. Instructions were shown on a screen inside the scanner (gray background, black color). They were instructed to prepare for movement (e.g. 'prepare right hand') before carrying out the movement (e.g. 'move right hand') for 12 seconds, followed by 15 seconds of rest. Each movement was repeated four times, resulting in a total of 20 trials. The total scan time for the MRI session (2) was approximately 90 minutes.

Data Quality Inspection
The MP2RAGE data of n = 5 adults (n = 3 younger, n = 2 older) were excluded due to low SNR or severe truncation artifacts affecting M1, leaving a total of n = 35 participants (n = 17 younger, n = 18 older) for the myelin analyses. In addition, the SWI data of n = 5 adults (n = 3 younger, n = 2 older) were excluded due to severe motion and truncation artifacts affecting M1, leaving a total of 31 participants (n = 14 younger, n = 16 older) for the SWI analyses.

Structural Preprocessing
Structural images were processed using CBS Tools (version 3.0.8)  implemented in MIPAV (version 7.3.0) (McAuliffe et al., 2001). We first registered the slab MP2RAGE image to the whole-brain MP2RAGE image using the 'Optimized Automated Registration' module in MIPAV and ANTs (version 1.9.x) (Avants et al., 2011). Registration quality in the sensorimotor areas were checked by two raters before proceeding. The slab and whole-brain images were then merged, resulting in whole-brain MP2RAGE images with improved resolution in the slab region. Using CBS Tools, the MP2RAGE skull stripping module was employed to remove extra-cranial tissue, and the MP2RAGE dura estimation module was used to estimate the dura mater, which was manually refined in ITK-SNAP. The topology-preserving anatomical segmentation (TOADS) algorithm (Bazin & Pham, 2008) was used to segment the UNI images into different tissue types based on a voxel-by-voxel probability approach. The cortical reconstruction, using implicit surface evolution (CRUISE) module, was then used to estimate the boundaries between the WM and GM and between the GM and CSF (Han et al., 2004), resulting in levelset images. Using these images, we created the subject-specific cortical surfaces used for mapping the microstructural measures.

SWI Processing
Quantitative Susceptibility Maps (QSM) were reconstructed using QSMbox (version 2.0) images was checked by two raters before proceeding. We separated the QSM data into positive QSM (pQSM) and negative QSM (nQSM) values following a previously described approach .

M1 Masks
ROI masks of M1 in both hemispheres were manually delineated, based on the MP2RAGE image, using the manual segmentation tool in ITK-SNAP. Based on anatomical landmarks, the M1 region was defined to include all relevant topographic areas (foot, hand, face/bulbar). The omega-shaped knob, known as the 'hand knob', serves as a reliable and robust landmark of the hand area in M1 with 97-100% accuracy (Yousry et al., 1997). We first identified the hand knob (located at the anterior wall of the central sulcus) in axial slices and masked all slices in which it was visible. Then, to include the face (bulbar) area, we masked slices inferior to the hand knob until the precentral gyrus was no longer visible (Donatelli et al., 2019). Finally, to include the foot area, we masked slices superior to the hand knob until the precentral gyrus was no longer visible. Since the foot area extends superiorly to the longitudinal fissure (Penfield & Rasmussen, 1950), we masked the paracentral lobule (PCL) while avoiding the supplementary motor area and the primary somatosensory cortex (Spasojević et al., 2013). Finally, the masks were refined in the coronal and sagittal views.

Surface Mapping
We divided the cortex into 21 layers according to the equivolume approach implemented in the volumetric layering module (Waehnert et al., 2014; of CBS Tools, using the volume-preserving layering method and an outward layering direction. We then sampled qT1, acquired using the MP2RAGE sequence, and signed QSM values at each layer, before mapping the values onto inflated cortical surfaces (see Fig. 1) using the surface mesh mapping module with the closest-point method.

Defining Cortical Layers
We calculated mean curvature using the simple curvature function in ParaView (version 5.8.0) and regressed this out of qT1 values for each layer, on an individual basis, following a previously published approach (Sereno et al., 2013). Using group-averaged decurved qT1 profiles, we applied a data-driven approach to define anatomically-relevant cortical compartments (Huber et al., 2017) (see Fig. 2). Layer 5 (L5) was identified based on a plateau in decurved qT1 after a steep increase in decurved qT1, reflecting the sharp increase in myelin content from the superficial layers to the heavily-myelinated L5 (depths 4-13). In addition, we distinguished between layer 5a (L5a) (4-7) and layer 5b (L5b) (8-13) based on the presence of two small qT1 dips that are considered to represent the two layer compartments (Huber et al., 2017). All depths above L5a were then labeled as the superficial layer (Ls, 1-3), which we suggest includes anatomical layers 3 and 4 but not anatomical layers 1 and 2, since the latter are particularly sparse and inaccessible with MRI (Huber et al., 2017). Finally, we defined layer 6 (L6, 14-17) based on a sharp decrease in decurved qT1, after which values plateaued again indicating the presence of WM (18-21). Taken together, we here refer to Ls, L5a, L5b and L6 when referring to cortical compartments that presumably refer to anatomical layers 3 and 4 (Ls), anatomical layer 5a (L5a), anatomical layer 5b (L5b) and anatomical layer 6 (L6), where, however, an exact delineation of the layers would need ex-vivo validation.  Dinse et al., 2015) and cell histological staining (right) (Vogt & Vogt, 1919). Note that these cortical layers are defined based on in-vivo MRI data and may not correspond exactly to the anatomical layers as defined by ex-vivo myelo-and cytoarchitecture. Also note that here, low qT1 values represent high myelin whereas in the other graphs in the article, values are plotted reversed such that high values represent high myelin.

Functional Data Processing
The functional data were retrospectively motion-corrected using the Siemens 'MoCo' correction.
The data were preprocessed using SPM12, including smoothing with a Gaussian kernel of 2 mm, slice-timing to correct for differences in image acquisition time between slices, and realignment to reduce motion-related artifacts. The functional volumes were averaged and manually registered to the qT1 images using ITK-SNAP, based on anatomical landmarks.
First-level analysis was used to generate t-statistic maps (t-maps) based on contrast estimates for each body part (e.g. left hand = [1 0 0 0 0]). The peak cluster of each t-map was saved as a binary mask. In the few cases where the peak cluster only included the face area of one hemisphere (tongue movement typically elicits a single peak cluster including the face area of both hemispheres), we included the largest cluster in the face area of the other hemisphere in the resulting mask. The t-maps and peak cluster masks were then registered to the qT1 images in ANTs, using the registration matrices previously generated in ITK-SNAP, before being mapped onto the same inflated cortical surfaces as used for the structural data (see Fig. 1). We applied the binarised peak cluster masks to the t-maps to create functional localisers which contained t-values only in the peak cluster area. We refined the localisers by removing overlapping voxels between different localisers, where the body part with the highest t-value retained the overlapping voxel (i.e., 'winner-takes-it-all' approach). These refined functional localisers (shown in Fig. 3A) were then binarised and multiplied with the layer-wise qT1 and QSM values, resulting in values for each cortical field at four different cortical layers.

Myelin Border Analysis
Based on previous evidence of low-myelin borders between the hand and face areas of M1 (Glasser et al., 2016;Kuehn et al., 2017), we defined the myelin border based on the highest qT1 value (lowest myelin) between the locations of the peak t-values of the UL and F localisers.
We extracted the qT1 values at the border and calculated the average body part qT1 values (UL+F/2).

Behavioral Tests of Motor Function
A subset (n = 27) of participants (12 younger, 15 older) underwent body part-specific behavioral tests of motor function (see Fig. 3B). The 6-Minute-Walking-Test (6MWT) was used to measure walking distance, as an estimate of gross motor function of the legs (Enright, 2003). Participants were instructed to walk as far as possible around a 30-meter measuring tape in six minutes, without speed-walking or running. The total distance walked was multiplied by weight in kilograms as previously described (Bernstein et al., 1994). We used a hand-held dynamometer to measure hand strength twice in each hand, before averaging to result in one measurement per hand (Peolsson et al., 2001). We used pegboards to measure hand and finger dexterity,

Statistical Analysis
Statistical analyses were performed using IBM SPSS Statistics (version 26, IBM, USA).
Mixed-effects ANOVAs were used to assess between-group (age group) and within-group (layer, cortical field) differences for each measure of cortical microstructure (qT1, pQSM, nQSM). The significance level for all statistical tests was set to the 5% threshold (p < .05).
Significant main effects and interactions of the ANOVAs were investigated using post-hoc tests and the Holm-Bonferroni method (Holm, 1979) was used to correct for multiple comparisons. In the cases where the assumption of sphericity was violated (Sig. < 0.05) according to Mauchly's test, we applied a sphericity correction. The correction used depended on the epsilon value calculated by the Greenhouse-Geisser correction (Field, 2013). If ε < 0.75, then the Greenhouse-Geisser correction was used, while the Huyn-Feld correction was used if ε > 0.75.
As a result of this correction, the reported degrees of freedom and F-values are sometimes not reported as whole numbers. Two-tailed independent-samples t-tests were used to measure group differences in the size of the myelin border and in behavioral measures of motor function, while one-tailed independent-samples t-tests were used to measure group differences in the mean dice coefficient of the overlap between functional representations. Linear regression models were used to investigate the influence of demographic variables (gender, education, cognitive function) on the cortical microstructure of M1.

Body Part-Specific Behavioral Motor Impairments in Older Adults
We quantified body part-specific motor function in younger and older adults (see section 2.6 for details and Fig. 3B for visualization, see Supplementary Table 1 for statistics).

Identifying Anatomically-Relevant Cortical Compartments in M1
We described the topographic microstructure of M1 using qT1 as a proxy for cortical myelin content (Stüber et al., 2014), pQSM as a marker for cortical iron content (Langkammer et al., 2012), and nQSM as a measure of cortical diamagnetic substance (Deh et al., 2018). A sub-millimeter isotropic resolution of 0.5 mm allowed us to characterize intracortical contrast with high precision (see Fig. 1 for an overview of extracted data). Given known correlations between curvature and layer-wise qT1 (Sereno et al., 2013) (see Fig. 2A & 2B for our data), we followed a previous approach to regress out effects of curvature from 'raw qT1' values (see . We then applied a data-driven approach (Huber et al., 2017) to divide M1 into four anatomically-relevant compartments, based on intracortical myelin content, that we refer to as 'layers' (Ls = superficial layer; L5a = layer 5a; L5b = layer 5b; L6 = layer 6; see Fig. 2E).

Demographic Variables Do Not Predict Variance in M1 Microstructure
To

M1 is Comprised of Distinct Cortical Fields
In order to target our first hypothesis (In healthy younger adults, M1 is comprised of microstructurally-distinct cortical fields corresponding to topographic areas), we first tested whether M1 is best described as a single cortical field (as suggested by Brodmann, (Brodmann, 1909)), or whether it is comprised of several fields (as suggested by Flechsig, 1920; see also Sereno et al., 2022). It has previously been reported that the UL and F areas of M1 are separated by low-myelin borders (Glasser et al., 2016;Kuehn et al., 2017). However, it has not yet been investigated whether the microstructure of these areas significantly differs, which would be necessary to define them as separate 'cortical fields'. To investigate this, we computed ANOVAs with the within-subjects factors cortical field (LL, UL, F) and layer (Ls, L5a, L5b, L6) on qT1, pQSM and nQSM values as in-vivo proxies of cortical microstructure in younger adults.
With respect to qT1 values, there are significant main effects of layer (F (1.17, 18.63) = 576.42, P < 10 -16 , η p 2 = .97) and cortical field (F (1.30, 20.83) = 5.34, P = .024, η p 2 = .25), as well as a significant interaction between layer and cortical field (F (2.12, 33.98) = 12.78, P < 10 -5 , η p 2 = .44; note that we report values of the dominant (left) hemisphere in the main text, but see Supplementary Tables   2 & 3 for the corresponding analyses on the non-dominant (right) hemisphere). The main effect of layer was expected and is driven by a significant decrease in qT1 (i.e., an increase in cortical myelin) with cortical depth that is due to the high myelination in the deep cortex near the white matter (Dinse et al., 2015). More specifically, Ls shows higher qT1 (i.e., less myelin) than L5a 10 -13 , η p 2 = .95) and cortical field (F (2, 26) = 5.05, P = .014, η p 2 = .28), but there is no significant interaction between layer and cortical field (F (2.5, 32.50) = 1.50, P = .235, η p 2 = .10, see Supplementary Tables 6 & 7 for the right hemisphere). Note that lower (i.e. more negative) nQSM reflects more diamagnetic tissue contrast (Deh et al., 2018). Taken together, we show that the F, UL and LL areas of M1 show significant differences in all three quantitative tissue contrasts in younger adults, which we summarize in a novel 3D model of the microstructural architecture of healthy M1 (see Fig. 4A). We show that the F area, overall, is characterized by high iron and diamagnetic substances. We also show layer-specific differences between cortical fields, for example, the F area shows low myelin in L6. In line with our hypothesis (1), our results indicate systematic differences in the microstructural profiles of the cortical fields of M1 in younger adults; we therefore refer to these areas as 'cortical fields' in the following text when investigating age-related changes in the M1 architecture (i.e., LL field, UL field, F field), in line with a recently-introduced parcellation atlas based on functional data (Sereno et al., 2022).
Overall, we show that the microstructural differences between the cortical fields (LL, UL, F) of M1, as reported above, can also be detected in older adults. In addition, older adults show more diamagnetic substance (nQSM) than younger adults. This effect is atopographic (i.e., even across cortical fields) but layer-specific (i.e., specific to Ls and L5a, see  across the cortical surface and with depth (Ls = superficial layer; L5a = layer 5a; L5b = layer 5b; L6 = layer 6). Main effects: * and ** indicate significance at the 5% and 1% significance levels, respectively. Interactions: * indicates statistical significance at 5% level, corrected for multiple comparisons using the Holm-Bonferroni method.

Hand-Face Myelin Borders Is Not Degenerated in Older Adults
To test our third hypothesis (The low-myelin border between the hand and the face is degenerated in older adults), we compared averaged qT1 sampled in the UL and F fields with the averaged qT1 sampled at the UL-F border, at each layer, between younger and older adults.
There are no significant age differences in any layer, but there is a trend in Ls towards a greater border in older adults (Ls: older adults = -778.14 ± 212.61; younger adults = -643.95 ± 216.08, To additionally investigate whether greater functional overlap between body part representations occurred in older adults, despite intact (or even slightly elevated) structural borders, we calculated the mean dice coefficient of the overlap between the functional localisers of the UL field and the F field (Dice, 1945). Taken together, we do not confirm our hypothesis that a degenerated hand-face structural border in older adults would relate to larger cortical representations / to more overlap between the UL and F areas. We instead show a trend towards a more pronounced hand-face boundary in older compared to younger adults in the superficial layer of M1, whereas the other layers do not show a trend towards a difference. Please note that this effect in the superficial layer does not survive a correction for multiple comparisons.

Layer-Specific Vulnerability in Older Adults
To test our fourth hypothesis (Iron in older adults is elevated in a layer-and topographic area-specific way), we investigated whether increased iron in older adults' M1 is uniformly present across cortical layers, or whether it occurs in particular cortical layers only. In addition, we investigated whether the age-related increase in iron is even across cortical fields, or whether some cortical fields show a selective vulnerability. We performed an ANOVA with the factors cortical field (LL, UL, F), layer (Ls, L5a, L5b, L6) and age (younger, older) on pQSM values, which served as a validated in-vivo proxy for cortical iron content (Langkammer et al., 2012).

Age-Related Iron Differences are Independent of Cortical Atrophy
To investigate whether cortical atrophy can explain the age-related differences in pQSM values (iron content), we calculated the ANOVA on pQSM values, as described above, while controlling for mask size (i.e. cortical field mask sizes used to extract microstructural profiles). We show that the main effect of age remains significant (F (1,28) = 20.71, P < 10-5, ηp2 = .43). The interaction between age and cortical layer (F (1.31,28) = 3.66, P = .016, ηp2 = .12) also remains significant, and is still driven by older adults showing the strongest effect size (largest Hedge's g), that is, most iron accumulation in L5a (older = .0014 ± .0029; younger = -.0024 ± .0024; df = 28, t = 4.60, P < 10-5, g = 1.64, 95% CI [.81 2.44]). Our findings are supported by previous evidence that age-related QSM effects are largely independent of brain atrophy measured with 7T-MRI .
Taken together, we show that increased iron in older adults' M1, a hallmark of cortical aging (Hallgren & Sourander, 1958), is particularly pronounced in L5a, suggesting output layer vulnerability. We further show that this effect is independent of atrophy in the topographic areas of M1. We also show that other than the increased diamagnetic substance, age-related iron increase is not atopographic but topographic in the left hemisphere, because there was most age-related iron accumulation in the LL cortical field. This effect was not present in the right (non-dominant) hemisphere.

The Functional Relevance of M1 Microstructure
To target our fifth and final hypothesis (Microstructural M1 changes show a relation to body part-specific motor function), we performed correlational analyses for each cortical field, for younger and older adults separately, between microstructural measures and motor function (see Fig. 8). Given the subset of participants with behavioral data was small (n = 12 for younger adults, n = 15 for older adults) and the number of tests was large (n = 27), we report correlation coefficients without p-values to provide an overview of potential relationships. shows strong (r > .5) negative correlations with pQSM values in Ls and L5a layers. More iron in the Ls and L5a in older adults was therefore associated with more dropped pins (i.e., worse accuracy). In the F field in older adults, tongue movement errors (as measured by TT) show strong (r > .5) negative correlations with pQSM values in L5b and L6. More iron in L5b and L6 in older adults was therefore associated with more movement errors (i.e., worse accuracy).
Conversely, the 6MWT scores did not show any strong (r > .5) correlations with the LL field in older adults. Our data therefore indicate that increased iron in older adults may have a link to poorer accuracy in motor tasks, while the layer-specific differences between cortical fields need further investigation.

Discussion
Topographic maps are a hallmark feature of the human brain, covering almost half of the cortical surface (Sereno et al., 2022). Cortical microcircuits of topographic maps, such as columns and layers, reveal critical insights into the mechanisms of aging and neurodegeneration, yet are poorly characterized in living older adults. We used M1 as a model system to characterize microstructural topographic map aging by applying parcellation-inspired techniques to sub-millimetre structural MRI data of healthy younger and older adults. We demonstrate systematic differences in the microstructural tissue profiles between the lower limb (LL), upper limb (UL) and face (F) areas in younger adults, suggesting that M1 is comprised of microstructurally distinct cortical fields. We show that these cortical fields are also distinct in older adults, and that the low-myelin borders between them do not degenerate in healthy aging.
We also show that increased iron in older adults, a hallmark of cortical aging (Hallgren & Sourander, 1958), is particularly prominent in layer 5a of both hemispheres and in the LL field of left M1. Overall, we provide a novel 3D model of the microstructural architecture human M1, where layer-specific vulnerability is a central mechanism of cortical aging.
In the present study, we used a data-driven approach to estimate anatomically-relevant cortical compartments using structural data with a sub-millimeter isotropic resolution of 0.5 mm (explained in more detail in section 2.5.6). It must be noted that there is a conceptual difference between our definition of layers and the definition based on ex-vivo data, where cytoarchitecture is considered (Brodmann, 1909;Vogt & Vogt, 1919). Nevertheless, our approach is based on a comparison between in-vivo and ex-vivo M1 data (Huber et al., 2017) and we were able to characterize intracortical contrast with high precision (see Fig. 2). We therefore provide a reasonable approximation of layers and their computations (Persichetti et al., 2020). While previous studies often distinguish between input and output structures, we extended this by distinguishing between layer 5a and layer 5b. We are cautious in our interpretation of the different effects within layer 5, however, and therefore draw conclusions more generally about the output layer 5. Future studies should apply this layering approach to ex-vivo data, where also cytoarchitectural data exist, to test the reliability of this measure.
We here provide a novel model of the human 3D microstructure in M1 (see Fig. 4A). More precisely, we show that quantitative markers of cortical microstructure significantly differ between body parts, which we interpret to show that M1 is comprised of microstructurally distinct cortical fields that represent major body parts. These differences sometimes occur across cortical layers (e.g. the F field is characterized by high iron content and low nQSM) while they are sometimes layer-specific (e.g. the F field shows low myelin in L6 compared to the UL and LL fields). Our findings challenge classical depictions of human M1 as a single cortical field (Brodmann, 1909), suggesting instead that M1 is comprised of distinct cortical fields, each with distinct microstructural and functional features (Flechsig, 1920;Sereno et al., 2022). These microstructural differences may relate to differences in the cytoarchitecture and associated myeloarchitecture between cortical fields, for example where the size, shape and clustering of the heavily-myelinated Betz cells differ along the M1 strip (Rivara et al., 2003), and where differences in the myeloarchitectonic patterns have been identified (Flechsig, 1920).
Critically, we do not find an interaction between age and cortical field in myelin (qT1) or diamagnetic substance (nQSM) content, suggesting that the differences between cortical fields are also present in older adults. In addition, we do not show evidence of degenerated low-myelin borders between the hand and the face area in M1 in older adults. Our hypothesis that a degenerated myelin border relates to more functional overlap is therefore not confirmed by our data. The finding that cortical fields remain microstructurally distinct in healthy aging highlights limits to age-related plasticity within M1 (Sereno, 2005), providing a mechanistic explanation as to why maps are often maintained after deprivation (Makin & Bensmaia, 2017;Striem-Amit et al., 2018), and why body representations are often preserved in older age (Riemer et al., 2019). Moreover, since the low-myelin borders differ between individuals, they may explain the high inter-individual variability in neurodegenerative diseases involving topographic disease spread (Schreiber et al., 2021).
While previous studies have highlighted the vulnerability of M1 to age-related iron accumulation (Hallgren & Sourander, 1958;Acosta-Cabronero et al., 2016;Betts et al., 2016), we extend these findings to show that this occurs unevenly across cortical depth in M1. More specifically, we demonstrate that age-related increases in iron occur most strongly in what we defined as layer 5a of the left hemisphere. Layer 5, which is here divided into layer 5a and layer 5b, is responsible for the output of motor function via the heavily-myelinated Betz cells (McColgan et al., 2020). Interestingly, iron accumulation in the deeper L5b is often implicated in neurodegenerative diseases affecting motor control (McColgan et al., 2020). Our findings suggest that age-related differences in iron accumulation within the output layer 5 may distinguish healthy aging from neurodegeneration in M1. Further research, perhaps using invasive techniques in animal models, should investigate the effects of aging on layer 5 in more detail.
In addition to age-related increases in iron, we also demonstrate that the superficial layer and layer 5a of M1 show lower nQSM (that is, more substance) in older adults compared to younger adults. The biological source of nQSM is currently under debate but has been shown to reflect myelin (Deh et al., 2018) and calcium (Wang et al., 2017;Jang et al., 2021;Kim et al., 2022).
Our data support the latter interpretation, by showing age-related increase in nQSM signal (i.e. more negative) without age-related differences in a validated proxy of myelin content (Stüber et al., 2014). Calcium deposits are well-evidenced in healthy aging and are associated with negative consequences on cognitive function (Thibault et al., 2007;Toescu & Verkhratsky, 2007). Interestingly, calcium deposits in older adults have been shown to occur largely near vascular structures (Jang et al., 2021), which may explain why the age-related increase in nQSM signal was strongest in the superficial layer of M1, where large vessels are located.
There is also a complex relationship between magnetic susceptibility and amyloid beta accumulation. While amyloid accumulation has been shown to be diamagnetic using ex-vivo 7T-MRI (Gong et al., 2019), it has also been shown to give rise to paramagnetic signal change based on a more specific approach comparing ex-vivo MRI at 9.4T and 14.1T (Tuzzi et al., 2020), where the latter may relate to iron accumulation near amyloid deposits. It is unclear whether our data support the interpretation of age-related nQSM increases as amyloid accumulation, given that healthy older adults would be expected to show minor or no amyloid accumulation in M1. Note that age-related and layer-specific differences cannot be due to group differences in layer width, since the size of the layer compartments was equal across age groups, even though they were defined separately within each group. These results are also unlikely to be explained by age-related differences in cortex morphology, since other microstructural tissue profiles that were extracted using the same masks, such as qT1, are stable across age groups. In addition, a control analysis showed that even when taking individual mask size into account, the age-related iron difference is still significant. Taken together, with strongest iron accumulation in layer 5, we here indicate that layer-specific vulnerability is a central mechanism of topographic map aging.
In terms of age-related differences in iron across cortical fields of M1, we show a small effect, where the LL field shows the highest effect and the UL field shows the smallest. This effect is only significant in the left (dominant) hemisphere. However, this finding is supported by animal research, where the hindpaw, but not the forepaw, representation of rats shows early signs of aging (David-Jürgens et al., 2008). This may explain why older adults show particularly deteriorated walking behavior, which is associated with reduced independence in late life (de Bruin & Schmidt, 2010). Since the UL field shows the smallest age effect in the dominant hemisphere, we suggest that use-dependent plasticity (i.e. of the dominant hand) may preserve topographic maps in the face of cortical aging. This is supported by the absence of differences in age-related iron accumulation between cortical fields of the right hemisphere (i.e. non-dominant body representations). This finding has some implications for basic research on the motor system, for example, it challenges the use of the hand area as a model system to study motor learning (Dumel et al., 2018) given the foot area of the dominant hemisphere may degenerate earlier. Our results are also relevant for neurodegenerative diseases where iron accumulation in specific cortical fields is considered as a disease-specific marker (Ravits & La Spada, 2009), and has been linked to symptom severity (Costagli et al., 2016;Kwan et al., 2012) and topographical disease spread (Schreiber et al., 2021). The present study provides a comprehensive model of the 3D microstructure of M1 in younger and older adults, to be used as a reference point for defining disease-specific features of pathology.
Finally, we performed correlational analyses to investigate how (age-related differences in) the cortical microstructure of M1 relates to motor function. It is worth noting that these analyses were done on reduced sample sizes for which behavioral data were available (n = 12 for younger adults, n = 15 for older adults). These preliminary analyses indicate that more iron in older adults relates to worse behavioral accuracy, and that the relationships between iron and motor behavior may be layer-dependent in older adults. More specifically, we show that increased iron in the superficial layer and layer 5a of the UL field is strongly associated with decreased accuracy in hand movements (r > .5), whereas increased iron in the layer 5b and layer 6 is associated with decreased accuracy in the tongue movement task (r > .5). This is in line with previous evidence of reduced hand dexterity with iron accumulation in aging (Li et al., 2015). Why in different cortical fields, more iron in specific layers relates to decreased accuracy needs to be clarified by future research. In both age groups, we found no strong correlations (r > .5) between microstructural measures and performance on the 6MWT, suggesting that this measure is not specific enough to relate to microstructural measures of the LL field. This may be due to the fact that we extracted microstructural profiles from each hemisphere separately, but walking is a bilateral motor task involving different body parts such as the foot, the limb and the knee, and where also the structure of the corpus callosum may be influential. Further research should use larger sample sizes to further describe the relationships between layer-specific iron content and motor function.

Conclusions
Taken together, our study highlights the importance of a 3D approach to topographic map aging that recognises layers and cortical fields as functionally-relevant units (Kuehn & Sereno, 2018), which show distinct age effects. Using recent advances in in-vivo microstructural imaging, we present a novel model of human M1 microstructure that includes distinct cortical fields and low-myelin borders, both in younger and older adults. We highlight that iron accumulation in layer 5, and diamagnetic substance (calcium) accumulation in layer 5 and the superficial layer, demonstrate layer-specific vulnerability in cortical aging. We suggest that further work should aim to disentangle the cortical field-specific iron accumulation in aging from that of neurodegeneration, where this is currently considered a disease-specific marker. We argue that layer-specific vulnerability is a central mechanism of topographic map aging, which inspires novel therapeutic interventions in M1 and beyond.

Data Availability
Anonymised data can be made available upon request.
completed the data processing. A.N. carried out the data investigation and statistical analysis.
A.N. wrote the manuscript. J.D., E.K., S.S and M.W. reviewed and edited the manuscript. E.K., S.S. and S.V. supervised the study. All authors contributed to the article and approved the submitted version.