In vivo irreversible and reversible transverse relaxation rates in human cerebral cortex via line scans at 7T with 250 micron resolution perpendicular to the cortical surface

Abstract

Understanding how and why MR signals and their associated relaxation rates vary with cortical depth could ultimately enable the noninvasive investigation of the laminar architecture of cerebral cortex in the living human brain. However, cortical gray matter is typically only a few millimeters thick, making it challenging to sample many cortical depths with the voxel sizes commonly used in MRI studies. Line-scan techniques provide a way to overcome this challenge and here we implemented a novel line-scan GESSE pulse sequence that allowed us to measure irreversible and reversible transverse relaxation rates—R₂ and R₂’, respectively—with extremely high resolution (250 μm) in the radial direction, perpendicular to the cortical surface. Eight healthy human subjects were scanned at 7T using this sequence, with primary visual cortex (V1) targeted in three subjects and primary motor (M1) and somatosensory cortex (S1) targeted in the other five. In all three cortical areas, a peak in R₂ values near the central depths was seen consistently across subjects—an observation that has not been made before, to our knowledge. On the other hand, no consistent pattern was apparent for R₂’ values as a function of cortical depth. The intracortical R₂ peak reported here is unlikely to be explained by myelin content or by deoxyhemoglobin in the microvasculature; however, this peak is in accord with the laminar distribution of non-heme iron in these cortical areas, known from prior histology studies. Obtaining information about tissue microstructure via measurements of transverse relaxation (and other quantitative MR contrast mechanisms) at the extremely high radial resolutions achievable through the use of line-scan techniques could therefore bring us closer to being able to perform “in vivo histology” of the cerebral cortex.

1. Introduction

Noninvasive studies of the laminar architecture of cerebral cortex would likely benefit from an improved understanding of how and why MR signals vary with cortical depth, an area of research often referred to as “laminar MRI” [1]. Acquiring these signals in vivo in humans is extremely challenging, however, since cortical gray matter is typically only a few millimeters thick, requiring voxels that are small enough to sample cortex at several depths while maintaining adequate signal-to-noise ratio (SNR) within the limits of realistic acquisition times.

One solution to this problem is to make the voxel size small along the radial direction (i.e., perpendicular to the cortical sheet)—a direction in which anatomical structure varies substantially—with much larger voxel size (providing higher SNR) in the tangential directions, along which structure varies little, as illustrated in Figure 1. Such anisotropic voxel geometries quite naturally result from the use of line-scan pulse sequences [2-4], making them potentially advantageous for anatomical laminar MRI studies as long as each line is prescribed orthogonally to the cortical surface, as we have recently argued [5].

FIG. 1.

Nissl stain of primary visual cortex, depicting the cytoarchitecture of cortical gray matter. The use of 1 mm isotropic voxels as shown in (A) means that very few cortical depths can be sampled at any given location on the cortical sheet, with substantial mixing of signals from different cortical layers as well as a high likelihood of partial-volume contamination from the adjacent white matter and cerebrospinal fluid (CSF). Substantially lowering the isotropic voxel size would address these concerns, but at the cost of a huge decrease in SNR, which scales linearly with voxel volume. On the other hand, the use of highly anisotropic voxels, as shown in (B) for the example of 0.25×3×3 mm voxels, means that a much larger number of cortical depths can be sampled, reducing the partial-volume effects described above without sacrificing SNR, as long as the voxels are suitably positioned and oriented with respect to the radial and tangential directions of the cortical surface, indicated in (A). Figure reproduced from Balasubramanian et al. [5].

Field strengths of 7T and above can be highly beneficial for laminar MRI studies, since the field-dependent gains in SNR can be translated into increases in spatial resolution. However, when measuring irreversible transverse relaxation rates R₂ = 1/T₂ with sequences in which each excitation RF pulse is followed by a train of refocusing RF pulses, the well-known increase in B₁⁺ spatial nonuniformity at higher field strengths [6, 7] can increase the likelihood of contamination from stimulated echoes, compromising the measurement of R₂ values [8]. Furthermore, a train of refocusing RF pulses at high field can easily lead to high specific absorption rate (SAR) levels exceeding patient safety limits. The use of sequences such as GESFIDE (“Gradient Echo Sampling of Free Induction Decay and Echo”) [9] or GESSE (“Gradient Echo Sampling of a Spin Echo”) [10] avoids the issues above, since these sequences involve sampling a single spin echo with multiple gradient echoes, and thus require only one refocusing pulse after each excitation pulse. As an added bonus, these methods enable the simultaneous measurement of both R₂ and reversible transverse relaxation rates R₂’ = 1/T₂’, and have been successfully used for R₂ and R₂’ mapping at 7T via 2D acquisitions [11, 12].

Here we incorporate GESSE readouts into a (1D) line-scan sequence and present its first use for measuring in vivo laminar R₂ and R₂’ values—at 7T and with a radial resolution of 250 μm—in three well-studied cortical areas of the human brain: primary visual cortex (V1), primary motor cortex (M1) and primary somatosensory cortex (S1). We also attempt to interpret the resulting laminar profiles in terms of the intracortical distribution of myelin and iron in these three cortical areas, known from prior histology studies, as part of a broader effort to relate quantitative MR contrasts to the microstructure of cortical gray matter [13].

2. Methods

Eight healthy volunteers (2F/6M, ages: 24–37 years), having provided written informed consent in accordance with institutional guidelines, were scanned on a 7T whole-body scanner (Siemens Healthineers, Erlangen, Germany) equipped with custom-built radiofrequency (RF) coils for the head—a birdcage transmit coil and a 32-channel receive coil [14]. On each volunteer, a T₁-weighted dual-echo 3D magnetization-prepared rapid gradient echo (MPRAGE) scan [15] with adiabatic frequency-offset-corrected inversion (FOCI) pulses was acquired in ~8 minutes with the following parameters: TE₁/TE₂/TI/TR = 1.8/3.7/1100/2530 ms, readout bandwidth = 651 Hz/px, flip angle = 7°, generalized autocalibrating partial parallel acquisition (GRAPPA) phase-encoding acceleration factor = 2 and 0.75×0.75×0.75 mm³ voxel size [16]. The resulting images served as the anatomical localizer for the line-scan acquisitions that followed.

The general strategy for acquiring the line-scan GESSE data was largely the same as that used in our recent line-scan diffusion study [5] and is briefly summarized here. For subjects 1–3, the MPRAGE images were used to identify the calcarine sulcus and thus V1, which lies on either bank of this sulcus. We inspected both hemispheres in search of flat regions of cortex; for each of these subjects, we were able to find at least one location where the patches of V1 on opposing sides of the calcarine were flat on the scale of a few millimeters or more, as well as being parallel to one another. A single line was then prescribed, centered on the sulcus and oriented perpendicularly to these patches of V1. Similarly, for subjects 4–8, a single line was prescribed perpendicularly to flat patches of M1 and S1, which lie respectively on the anterior and posterior bank of the central sulcus (see Supplementary Figure 1). After determining the prescription of the line but before acquiring the line-scan data, transmit coil voltages were calibrated for each subject based on a B₁⁺ map [17] acquired at the beginning of the session and B₀ shimming was performed using a ~5-cm shim box centered on the region of interest, with further detail provided in Balasubramanian et al. [5].

The line-scan GESSE pulse sequence diagram is shown in Figure 2, where it can be seen that the slice-select gradients for the RF excitation and refocusing pulses are played orthogonally to one another. The intersection of the two slices defines a column of spins that generate signal, requiring spatial encoding in only one dimension, with very high spatial resolution achievable in this direction via frequency encoding [2-4]. The figure also shows the use of GESSE readouts [10] in this pulse sequence, with multiple gradient echoes sampling the signal on either side of the spin echo.

FIG. 2.

Line-scan GESSE pulse sequence diagram, with time on the horizontal axis and with the vertical axis for each row defined in the panel on the right (“RF-Signal Data”: transmit voltage; “Z Gradient”, “Y Gradient” and “X Gradient”: amplitude of the gradients generated by the three main gradient-coil systems; “ADC Signal Data”: binary variable indicating when the analog-to-digital converter is on versus off). The application of orthogonal slice-select gradients for the RF excitation and refocusing pulses results in signal from just a column of spins (see inset with yellow border). By frequency encoding along the column (i.e., the “line”), and with no need for phase encoding, a 1D Fourier Transform (FT) can be used during signal reconstruction. We have incorporated GESSE readouts into the basic line-scan sequence, sampling a single spin echo with multiple gradient echoes. In this study, we acquired 7 monopolar gradient echoes, with the 4^th gradient echo coinciding with the spin echo. (Monopolar readout gradients were chosen over bipolar gradients in order to avoid having to account for any spatial shifts between data from positive versus negative gradient lobes, resulting from off-resonance effects and finite readout bandwidth.)

Line-scan GESSE data were acquired in ~9 minutes with the following parameters: TR = 2 s, 256 repetitions, 7 monopolar gradient echoes with the 4^th gradient echo coinciding with the spin echo at TE = 50 ms (gradient-echo spacing ΔTE = 6.1 ms), readout bandwidth = 305 Hz/px, excitation/refocusing RF pulse duration = 2.0/2.5 ms with nominal flip angle = 90/180° and time-bandwidth product = 2.7/2.7. The voxel size along the line was 250 μm, with a 256-mm readout field of view and nominal line thickness of 3 mm, resulting in voxel dimensions of 0.25×3×3 mm³. The area of each crusher gradient bracketing the refocusing pulse was set to 62.6 ms·mT/m (see Balasubramanian et al. [5] for crushing considerations).

Line-scan signals were reconstructed and analyzed offline in MATLAB (MathWorks, Natick, MA, USA), given raw data from multiple coil channels as input. A 1D Fourier transform was applied to take the data from k-space into physical space (i.e., the “image” domain), using the fft function in MATLAB. For each voxel and for each gradient echo, complex-valued data were then averaged across the 256 repetitions prior to root-sum-of-squares coil combination.

Given a Lorentzian distribution of intra-voxel precession frequencies, GESSE time-domain signal magnitudes take the form

$$S(t)=k\text{exp}(-{\mathrm{R}}_2(t+\tau ))\text{exp}(-{\mathrm{R}}_2^{\prime }\mid t-\tau \mid )$$ [1]

where τ is the time between the excitation and refocusing pulses, t is the time after the refocusing pulse (with the spin echo therefore occurring at t = τ), k is a constant that includes the steady-state magnetization, and R₂ and R₂’ are the irreversible and reversible transverse relaxation rates, respectively [18]. Monoexponential fits to the signal on either side of the spin echo thus yield per-voxel R₂ and R₂’ values, as illustrated in Figure 3. Although this plot displays the signal logarithm ln(S), with monoexponentials appearing as straight lines, the actual fits were to the signal magnitude S, performed with the nonlinear least-squares Levenberg-Marquardt algorithm, as implemented in MATLAB's lsqnonlin function.

FIG. 3.

On a semi-logarithmic plot, representative line-scan GESSE data are shown with black dots, taken from a voxel within V1 from subject 1. Monoexponential fits to the signal on either side of the spin echo are shown with the purple straight lines, from which the R₂ and R₂’ values for this voxel can be extracted. The data points lie very close to the straight lines, indicating the suitability of the Lorentzian signal model (Eq. 1). Therefore, there was no need to consider alternative GESSE signal models such as the Gaussian [18] or the symmetric alpha-stable [52].

To account for the possibility of head motion, we ran a ~10-second three-plane localizer—the same scan that is routinely performed at the start of any session—immediately before and after the line-scan GESSE acquisition. If toggling between the “pre” and “post” three-plane localizer images revealed any appreciable movement, the subject was gently but firmly reminded of the importance of staying as still as possible, and the above scans were repeated. This was only necessary for two out of the eight volunteers in this study, likely reflecting the fact that these tended to be experienced subjects who had successfully participated in prior MRI studies; following the reminder to stay still, no evidence for head motion was found for the subsequent scans. To illustrate, axial images from the localizer scans are shown in Supplementary Figure 2 for one of the two subjects who was adjudged to have moved during the initial scans.

3. Results

The top panel of Figure 4 shows MPRAGE data from subject 1, resampled such that the prescribed line, overlaid in yellow, is oriented horizontally with respect to the image. The two patches of V1 on either side of the yellow circle within the calcarine sulcus CSF appear to be flat on a scale of at least 3 mm (the thickness of the line), as well as being parallel to one another, with the line perpendicular to these two opposing V1 patches. The line does also intersect cortex at other locations, but not orthogonally—substantial mixing of signals across cortical depths is therefore likely in these locations given our highly anisotropic 0.25×3×3 mm³ line-scan voxels, and so these “non-orthogonal cortices” will not be given any further consideration.

FIG. 4.

Results in V1 from subjects 1–3. The top panel shows MPRAGE data for subject 1, with the prescribed line location and thickness for the subsequent GESSE line scan overlaid in yellow. The middle panel shows plots of R₂ and R₂’ versus position along the line for this subject and the bottom panel shows the corresponding plots for subjects 2 and 3. The two patches of V1 marked for each subject (light gray shaded regions) and the calcarine sulcus CSF in between (dark gray shaded region) were delineated based on that individual’s MPRAGE data. An intracortical R₂ peak can be appreciated within V1 in each case (purple arrows); however, no consistent pattern is apparent in the intracortical R₂’ profiles.

For the same subject, the middle panel of Figure 4 shows plots of R₂ and R₂’ along the line and the bottom panel shows the corresponding plots for subjects 2 and 3. The light gray shaded regions in these plots indicate the patches of V1 that were perpendicular to the line and were delineated based on each subject’s MPRAGE data. In each of these V1 patches, an R₂ peak can be seen, although the prominence of the peak does vary somewhat even within the same subject (e.g., compare the two R₂ peaks in V1 for subject 1). We remark that the observation of an intracortical R₂ peak has not been reported before, to our knowledge. For the R₂’ values, on the other hand, we see no consistent intracortical pattern across these six V1s; possible reasons for this lack of consistency are discussed in the next section.

The top panel of Figure 5 shows MPRAGE data from subject 4, with the prescribed line perpendicular to the flat patches of M1 and S1 targeted here. For this subject, the next panel shows plots of R₂ and R₂’ along the line, with the remaining panels showing the corresponding plots for subjects 5–8. In each subject, an intracortical R₂ peak is apparent in both M1 and S1, with the prominence of these peaks varying somewhat, as was the case in V1. Again, no obvious pattern is seen in the intracortical R₂’ profiles.

FIG. 5.

Results in M1 and S1 from subjects 4–8. The top panel shows MPRAGE data for subject 4, with the line prescription for the subsequent GESSE line scan shown in yellow overlay. The panel below shows plots of R₂ and R₂’ versus position along the line for this subject and the remaining panels show the corresponding plots for subjects 5–8. The patches of M1 and S1 marked for each subject (light gray shaded regions) and the central sulcus CSF in between (dark gray shaded region) were delineated based on that individual’s MPRAGE data. In each case, an intracortical R₂ peak can be seen within M1 and S1 (purple arrows), appearing somewhat broader in M1, with no consistent pattern apparent in the R₂’ profiles.

From both Figures 4 and 5, it can be seen that we have quite a large number of data points within cortex, relative to what is possible with more conventional acquisition schemes with voxel sizes that are typically ~1 mm for in vivo studies. This “dense sampling” of cortical depths, resulting from the use of appropriately positioned and oriented line-scan acquisitions with high radial resolution, is especially important for the investigation of thin cortical areas such as S1, which has an average thickness of only ~1.5 mm [19]. The 250-μm radial resolution used here would therefore be expected to give us ~6 data points within S1, and inspection of Figure 5 reveals that this does indeed appear to be the case for each subject.

Each patch of cortex highlighted in Figures 4 and 5 was partitioned into a “center” region (approximately the middle third of the voxels within that patch) and a “surround” region consisting of the remaining voxels within cortex, on either side of the central zone. (Note that this partitioning resulted simply from dividing the cortical depth into thirds and was not explicitly tied to the location of the R₂ peak.) A two-sample t-test was then performed to compare relaxation rates in the center versus the surround, and the resulting p-values for V1, M1 and S1 are shown in Tables 1, 2 and 3, respectively. From these tables, it can be seen that the difference in R₂ values between the center and the surround was statistically significant for each patch of cortex, whereas this was rarely the case for the corresponding R₂’ values, providing statistical support for the observations made above.

TABLE 1.

For each patch of V1, mean ± standard deviation of transverse relaxation rates (s⁻¹) in central and surrounding cortical depths, along with p-values resulting from t-test comparisons of center versus surround, with statistical significance indicated by an asterisk (α = 0.05). The labels “subject 1a” and “subject 1b” are used to indicate the intracortical data from subject 1 that appear to the left and to the right, respectively, of the CSF label in the middle panel of Figure 4 (and similarly for subjects 2 and 3). No correction for multiple comparisons was performed.

		subject 1a	subject 1b	subject 2a	subject 2b	subject 3a	subject 3b
R 2	center	27.3 ± 1.4	33.0 ± 2.8	32.2 ± 0.6	30.8 ± 2.2	26.7 ± 0.1	29.8 ± 0.4
	surround	25.7 ± 0.8	26.9 ± 2.0	28.0 ± 2.1	25.9 ± 2.5	23.6 ± 1.1	22.5 ± 2.4
	p-value	0.04*	0.01*	0.02*	0.02*	0.01*	0.01*
R2’	center	5.8 ± 0.3	7.7 ± 1.2	8.9 ± 0.2	7.2 ± 0.5	4.9 ± 0.6	6.6 ± 0.7
	surround	5.8 ± 1.6	5.7 ± 1.3	7.7 ± 1.3	7.2 ± 0.7	4.8 ± 2.7	2.1 ± 2.1
	p-value	0.49*	0.06*	0.13*	0.49*	0.48*	0.03*

TABLE 2.

For each patch of M1, mean ± standard deviation of transverse relaxation rates (s⁻¹) in central and surrounding cortical depths, along with p-values resulting from t-test comparisons of center versus surround, with statistical significance indicated by an asterisk (α = 0.05). No correction for multiple comparisons was performed.

		subject 4	subject 5	subject 6	subject 7	subject 8
R 2	center	27.8 ± 1.3	28.6 ± 0.8	29.6 ± 1.3	28.8 ± 1.2	30.7 ± 1.3
	surround	25.1 ± 2.1	26.5 ± 1.1	26.5 ± 1.3	25.6 ± 1.6	25.4 ± 1.4
	p-value	0.02*	0.01*	0.01*	<0.01*	<0.01*
R2’	center	7.2 ± 2.2	4.6 ± 1.0	6.1 ± 1.0	5.8 ± 0.9	6.8 ± 0.9
	surround	5.6 ± 1.8	3.9 ± 0.6	5.1 ± 0.7	4.6 ± 1.2	5.7 ± 2.0
	p-value	0.12*	0.10*	0.07*	0.06*	0.12*

TABLE 3.

For each patch of S1, mean ± standard deviation of transverse relaxation rates (s⁻¹) in central and surrounding cortical depths, along with p-values resulting from t-test comparisons of center versus surround, with statistical significance indicated by an asterisk (α = 0.05). No correction for multiple comparisons was performed.

		subject 4	subject 5	subject 6	subject 7	subject 8
R 2	center	26.6 ± 0.6	28.3 ± 0.3	24.3 ± 0.4	25.3 ± 0.4	27.9 ± 1.6
	surround	22.7 ± 2.0	23.8 ± 2.0	22.6 ± 1.3	23.1 ± 1.2	22.7 ± 2.9
	p-value	0.02*	<0.01*	0.04*	0.04*	0.03*
R2’	center	6.5 ± 2.5	3.8 ± 1.0	7.0 ± 0.4	5.4 ± 0.7	5.4 ± 0.7
	surround	4.0 ± 0.3	3.7 ± 0.4	5.8 ± 1.5	4.1 ± 0.7	4.5 ± 1.3
	p-value	0.08*	0.45*	0.13*	0.04*	0.20*

In the top row of Supplementary Figure 3, the intracortical R₂ profiles from Figures 4 and 5 are plotted on the same axis for each cortical area, after normalizing cortical depth to range between 0 and 1. Here it can be seen more clearly that while the profiles show some similarity overall, the exact location of the peak can vary appreciably from one patch of cortex in V1 to another, and similarly for M1 and S1. Therefore, averaging the profiles (a common practice in laminar MRI analyses) can smear out fine-scale features and broaden the R₂ peaks, as shown in the bottom row of the figure. The corresponding plots for R₂’ are presented in Supplementary Figure 4, once again failing to reveal any obvious and consistent pattern. Issues associated with combining data across different locations on the cortical sheet are discussed further in the following section.

4. Discussion and Conclusions

To our knowledge, in vivo measurements of irreversible and reversible transverse relaxation rates (R₂ and R₂’, respectively) as a function of cortical depth have not been previously reported in humans, especially at field strengths of 7T or above. The closest reports we were able to find in the literature were from Koopmans et al. [20] and Marques et al. [21], but even these two studies at 7T differ markedly from ours. For one thing, they examined the cortical depth dependence of R₂* (or its reciprocal, T₂*), entangling the effects of reversible and irreversible transverse relaxation. For another, these studies used isotropic voxel sizes of 0.65–0.75 mm, resulting in much greater mixing of signals across cortical depths or layers than that arising from our voxels with 250-μm radial resolution.

With an isotropic voxel size of ~0.75 mm, one cannot expect to sample more than two cortical depths at a typical location in S1, which has an average thickness of ~1.5 mm [19]. Thus, in order to study the cortical depth dependence of MR signals with such voxels, the strategy commonly used in laminar MRI studies exploits the fact that the cortical depth of the centroid of these relatively large isotropic voxels will naturally vary from one location on the cortical sheet to another [22], depending on the interplay between the prescription of the voxels and the cortical folding pattern in each individual; by combining data across different locations on the cortical sheet (but usually staying within a cortical area such as V1), one can hope to glean some information on how these MR signals vary with cortical depth, very much like looking at a moving average.

How best to combine data across different cortical locations is unclear, however, and we note that the Koopmans and Marques studies referenced above used different methods in this regard. This likely explains why they obtained different results in V1: whereas Koopmans et al. observed a local maximum in R₂* values near the middle cortical depths [20_{], Marques et al. report a nearly monotonic decrease of R2}* values as one goes from the gray-white interface towards the pial surface [21]. It is not clear which of the two methods is better, but if the goal in combining data across different cortical locations is to minimize the mixing of signals across different cortical layers, then one must also account for the fact that layer depths are known to vary substantially across cortex in a manner dependent on the folding pattern, with the granular layer shifted towards the pia mater in gyral crowns and towards the white matter in sulcal fundi [23]. Even after accounting for this shift [24], Hinds et al. [25] observed a systematic, residual variation in the depth of the stria of Gennari in human V1, complicating the prediction of layer positions from the cortical folding pattern alone. These issues could lead to the inadvertent blurring out of intracortical features, as illustrated in Supplementary Figure 3.

Note that we can avoid combining data across different cortical locations—sidestepping the problems described above—through the use of line-scan acquisitions where each line is oriented perpendicularly to the cortical surface, with high resolution in the radial direction providing a dense sampling of cortical depths and relatively large voxel size in the tangential directions providing sufficient SNR to enable the investigation of individual cortical locations within a given subject. The major limitation of such a strategy is the lack of spatial coverage across the cortical sheet: since any given line is unlikely to intersect cortex orthogonally at several locations, each line will likely provide depth-related information on at most an opposing pair of nearby cortical locations, e.g., the two banks of V1 in subjects 1–3 or the patches of M1 and S1 in subjects 4–8. Since there is considerable dead time in our line-scan GESSE implementation, one could improve its spatial coverage by acquiring multiple non-parallel lines within each TR, with the flexibility of having each line prescribed independently and targeting a different pair of cortical locations. This would of course require a much more elaborate line-prescription strategy than that needed for a single-line acquisition, mostly likely involving offline planning based on a subject’s prior anatomical scan. Such a “multi-line” acquisition is still unlikely to compete with the spatial coverage provided by a more conventional 2D or 3D isotropic acquisition scheme, or even that provided by a suitably-oriented single-slice acquisition with high in-plane isotropic resolution, and we therefore anticipate that line-scan and conventional acquisition strategies will ultimately complement each other, with the former sampling many cortical depths at a few locations and the latter sampling a few depths at many cortical locations.

The intracortical R₂ peak in V1, M1 and S1—which we believe has not been described before—was consistently visible in each subject in our study and we now attempt to interpret this R₂ peak in terms of quantities familiar from histology: namely, myelin and iron content. We make a further distinction between non-heme storage iron (e.g., ferritin) versus heme iron in the form of deoxyhemoglobin, the presence of which within the microvasculature can induce irreversible transverse relaxation in the surrounding tissue [26], with regions with higher microvascular density thus potentially exhibiting higher R₂ values.

Figure 6 shows the myelin and non-heme iron content in and around V1, along with the microvascular density, taken from prior histology studies [27, 28]. Note that all of these quantities tend to peak near the central layers of V1, making it difficult to uniquely associate any one of them with the intracortical R₂ peak we observe—a point that is equally applicable to the detection of a central peak in V1 via any other MRI method or contrast mechanism. Note, however, the dramatic increase in myelin content as one crosses from gray matter into white matter (see left panel of Figure 6). We, on the other hand, see very little difference in R₂ values across the gray-white interface, as evident from inspection of the outer boundaries of the light gray shaded regions in Figure 4. From this, we conclude that myelin content does not appear to have much of an influence on the R₂ values we observe, and is therefore unlikely to be a primary factor in forming the intracortical R₂ peak seen here. (The situation might have been different had our acquisition been sensitive to myelin water [29], but very little signal contribution from this short-T₂ component is likely given the relatively long echo times used here.)

FIG. 6.

Histology images showing myelin and non-heme iron content in V1, taken from Fukunaga et al. [27], along with intravascular india ink data from Duvernoy et al. [28], from which microvascular density can be inferred. Note that these quantities all exhibit a peak near the central layers of V1, which contain the stria of Gennari. Note also the dramatically higher myelin content in white matter versus gray matter—any MR measure sensitive to the amount of myelin would therefore be expected to exhibit strong gray-white contrast.

Figure 7 shows histology data for M1 and S1 [28, 30], where we once again see a huge difference in myelin content between gray and white matter. This is in contrast to the negligible difference in R₂ values across the gray-white boundary seen in Figure 5, in further support of the decision to discount myelin concentration as a possible explanation of our observed intracortical R₂ peak. In S1, both non-heme iron content and microvascular density peak near the central layers, much like in V1. An important difference is seen in M1, however: here, microvascular density is quite homogeneous across cortical depth, as remarked upon by Duvernoy et al. [28], and thus would be expected to impact R₂ equally at all depths. It is only non-heme iron content that exhibits a peak near the central layers of M1, S1 and V1, with little difference across the gray-white border, making it the “histological quantity” most likely to account for the intracortical R₂ peak we have reported in this study.

FIG. 7.

Myelin and non-heme iron content in M1 and S1, adapted from Stüber et al. [30]. The intravascular india ink image from Duvernoy et al. [28] shows that while microvascular density peaks near the central layers of S1, it is quite homogeneous across cortical depths in M1. The non-heme iron stain image (top center) was converted to grayscale in order to produce the intensity profiles through M1 and S1 shown in the bottom plots, which qualitatively compare quite favorably with the R₂ profiles shown in Fig. 5. For these intensity profiles, note that a normalized cortical depth of 0 corresponds to the boundary between white matter (WM) and gray matter (GM), whereas a depth of 1 corresponds to the pial surface, the boundary between GM and CSF in the in vivo setting. Note also that for this iron stain data, the tissue unfortunately appears to have been sectioned rather obliquely relative to the posterior bank of the central sulcus (purple arrow), obscuring the pattern in that part of S1.

Several comments and caveats are in order with regard to our interpretation of non-heme iron being the likely source of the intracortical R₂ peak we observe. (i) This interpretation is consistent with our prior work relating the expected age-dependent non-heme iron content in the basal ganglia [31] to the in vivo R₂ values measured in these structures at 7T using a conventional 2D GESSE sequence [12]. (ii) We note that Fukunaga et al. [27] and Marques et al. [32] also implicated non-heme iron, rather than myelin or microvascular deoxyhemoglobin content, as the likely source of intracortical contrast in their studies; they, however, were looking at phase images derived from gradient-echo scans, not R₂ values from spin-echo acquisitions. (iii) The histology data in Figures 6 and 7 do show a small difference in iron content across the gray-white boundary for the three cortical areas considered here (but one that is much smaller than the difference in myelin content or microvascular density). We, however, fail to see any appreciable difference in R₂ values across the boundary, something we cannot explain at present (but see the next two points). (iv) We have treated non-heme iron as a single entity here, but different formulations of non-heme iron, perhaps dependent on the nature of its cellular distribution [33], may have distinguishable MR properties [13], and it may be beneficial to account for such differences once they are better understood. (v) We also have to keep in mind the sensitivity of transverse relaxation processes to various other quantities such as water and protein content, and so the full explanation of the R₂ patterns seen here may ultimately require us to take these and other factors into consideration.

Unlike the case with R₂, our intracortical R₂’ profiles fail to exhibit any consistent pattern across subjects, as we noted in the previous section. One might have instead expected to see an intracortical peak in the R₂’ profiles (as well as in the R₂ profiles), based on prior studies reporting a robust relationship between R₂’ and non-heme iron content. However, we note that these studies typically focused on the iron-rich structures of the basal ganglia, either in vivo (e.g., Gelman et al. [34]) or ex vivo (e.g., Brammerloh et al. [35]). In the cortical areas we investigated (M1, S1, and V1), not only is the non-heme iron content far lower [31], but the spatial distribution of the iron deposits may well be quite different in these layered structures than in the basal ganglia; in cortex, the spatial distribution of non-heme iron may not induce substantial magnetic field variations or inhomogeneities over distances greater than the diffusion length (~10 μm in brain parenchyma for the echo times used here) resulting in only small increases (if any) in the reversible relaxation rate [36]. Furthermore, small changes in R₂’ could easily be obscured given the additional sensitivity of reversible transverse relaxation rates to macroscopic field variations (with a spatial scale above ~1 mm). Several factors could contribute to high variability across subjects in these macroscopic inhomogeneities and hence in the R₂’ measurements. These include variability in the quality of B₀ shimming, as well as in the proximity to large veins with high deoxyhemoglobin content. Macroscopic field gradients are also highly sensitive to the orientation of tissue with respect to the main magnetic field (e.g., see Lee et al. [37])—a sensitivity inherited by R₂’ and thus by R₂* (usually defined as R₂ + R₂’)—with the orientation dependency of cortical T₂* (= 1/R₂*) values having been well documented previously (e.g., see Cohen-Adad et al. [38]). Variability in the angle between the main magnetic field and the surface normal of each patch of cortex targeted in this study could therefore have led to further variability in the R₂’ values we measured. As an aside, we note that the usual definition of R₂* as simply the sum of R₂ and R₂’ has been called into question in the context of biological tissues [39], due to differences in the mesoscopic signal contributions to gradient-echo acquisitions (typically used to measure R₂*) versus spin-echo acquisitions (typically used to measure R₂).

Although our focus here has been on structure rather than function, we note that line-scan techniques have also been used to investigate the variation of fMRI signals with cortical depth. Initial work in this area was in rats [40, 41] with reports of line-scan fMRI studies in humans only appearing very recently [42, 43]. Gradient-echo sequences were typically employed in these studies, with line scanning enabled through the use of slab-selective RF saturation pulses to suppress much of the signal outside the line. High temporal resolution was achieved by setting the TR to 50–200 ms, motivated by the desire to investigate cortical depth-dependent latencies in the hemodynamic response to sensory stimulation. In principle, the line-scan GESSE sequence presented here could also be used to study hemodynamic latencies, but in practice, lowering the TR to 50–200 ms would be technically challenging: at these short TRs, the use of a 90° excitation flip angle would be costly in terms of SNR (due to inadequate recovery of the longitudinal magnetization). One could instead seek a flip angle that is optimal with regard to SNR; however, unlike gradient-echo acquisitions, optimal excitation flip angles for spin-echo acquisitions will far exceed 90° for short TRs [44, 45], given gray matter T₁ values of 1–2 seconds [46]. This is problematic for two reasons: (i) high flip angles combined with low TRs are likely to lead to SAR limit violations, and (ii) simple excitation RF pulses (such as the commonly-used sinc pulse) provide poor slice profiles (or, in our case, line profiles) for flip angles well above 90° [47]. These obstacles could possibly be overcome in the future through the use of parallel transmit techniques [48] and tailored RF pulses [49] to mitigate SAR issues and improve slice/line profiles.

In summary, we have reported here in vivo measurements of R₂ and R₂’ within gray matter in humans, at 7T and with an unprecedented radial resolution of 250 μm, made possible through the use of a novel line-scan GESSE pulse sequence with the lines prescribed perpendicularly to the cortical patches of interest. The observed R₂ peak within cortical areas V1, M1 and S1, seen consistently across subjects, lends itself well to an interpretation in terms of intracortical non-heme iron content. This nicely complements our recent high-radial-resolution line-scan diffusion study [5] in which we were able to relate diffusion tensor properties as a function of cortical depth to the known fiber architecture within M1 and S1. Even if our “non-heme iron interpretation” ultimately proves to be incorrect or incomplete, the R₂ patterns shown in this report are arguably interesting in their own right and, in combination with other MR features, may prove valuable for the in vivo delineation of cortical layers and areas [50] or for the interpretation of non-quantitative T₂-weighted signals acquired in vivo at high resolution [51].