It has been recently shown that the apparent diffusion coefficient (ADC) of intracellular metabolites in the primate brain¹ and human² grey and white matter is quite stable as the diffusion time t_d is increased from a few tens of ms to ~1 sec. This led us to conclude that the vastest fraction of each metabolite pool is not restricted in small subcellular domains (cell bodies, organelles…) but is instead “freely” diffusing along cell fibers. Here we investigate if this (admittedly) simplistic view is consistent with metabolite data collected up to unprecedented high b in the mouse brain, outside the low b monoexponential regime.Data analyzed here are those reported in a separate abstract³. Briefly, acquisitions were performed in a large voxel of the mouse brain using our new STE-LASER sequence³, based on a diffusion-weighted stimulated echo block followed by a LASER localization block, yielding no cross terms. 10 mice were scanned at 11.7 T Bruker scanner (G_max=752 mT/m) using a cryoprobe, at TE=33.4 ms and t_d=63.2 ms, up to b_max=60 ms/µm² (q_max=1 μm^-1) (Fig.1). Individual scan phasing was performed and experimental macromolecule spectrum was included in LCModel’s basis-set. Unlike biological water, metabolites are mainly intracellular, and membranes are nearly impermeable to metabolites. Cellular processes can be described in first approximation as a collection of long cylinders (Fig.2A) with radius a, and intracellular diffusivity D_intra^cyl. We assume cylinders randomly oriented to calculate signal attenuation in the narrow pulse approximation^5,6. This theoretical attenuation is used to fit experimental data as a function of q to estimate D_intra^cyl and a.Signal attenuation as a function of q is reported in Fig.2B for each metabolite, together with best fits. Extracted parameters are reported in Table 1. Results show that randomly oriented cylinders account well for measured attenuation, giving fiber radii consistent with axons, dendrites and astrocytic processes, and D_intra^cyl in the expected range (0.30-0.45 µm²/ms ^1,2,4,5). The only exception is NAA, for which the model extracts a=0. Although a=0 has been used as an assumptionto fit NAA diffusion in the past⁴, we think this is not satisfactory, considering that this does not hold for other metabolites (including glutamate which is supposed to be, like NAA, predominantly neuronal). Explanations based on finer structural features (e.g. spines⁶) affecting diffusion of a single cytosolic pool also seem unlikely, because they should apply to other metabolites diffusing in the same cells. NAA is synthetized in neuronal mitochondria, so it may be significantly present in mitochondria (which represent 5-10% of the cytoplasmic volume), resulting in a second NAA pool, with limited exchange with cytosolic NAA. As mitochondrial matrix is viscous⁷ and densely filled with membranes, it is expected that T₂ is very short. Hence, mitochondrial NAA should become invisible at long TE. Consistently, we have measured that increased TE (73.4 ms) led to small but significantly stronger signal attenuation at high b for NAA (Fig.3B), but not for other metabolites³. When analyzing NAA signal acquired at longer TE with cylinders (Fig.3B-C), the model returns a=0.62 µm (and D_intra^cyl=0.335 µm²/ms), which is now realistic, strongly suggesting that a highly restricted NAA pool has become invisible. Going one step further, these a and D_intra^cyl values can be injected in a modified model, where a log-normal distribution of spherical compartments⁸ (accounting for organelles/mitochondria) is added to cylinders whose properties were determined at long TE, to fit data at short TE (Fig. 3A-B). New free parameters are the diffusivity inside spheres D_intra^sph, the volume fraction of these spheres ν_sph, and the mean radius μ and s.d. σ of spheres. Best fit of NAA attenuation at short TE yields (Fig.3C): ν_sph=2.0%, D_intra^sph=0.168 µm²/ms, μ=0.019 µm, σ=0.028 µm. Values of ν_sph and D_intra^sph are consistent with the known mitochondrial volume fraction and the higher viscosity in mitochondria⁷. Extracted radii are much smaller than typical mitochondria size, but very consistent with the typical distance between cristae, which must cause most of the restriction inside mitochondria.The non-monoexponential signal attenuation of intracellular metabolites in the mouse brain can be essentially described by diffusion in long and thin cylinders, yielding realistic D_intra and fiber diameters, thus consolidating what we had previously proposed based on ADC measurements performed at very long t_d (and low b). However, this simple model seems to require some small “correction” for NAA at very high b, under the form of a small fraction of the NAA signal originating from a highly restricted (with short T₂) compartment, e.g. a mitochondrial pool.