Diffusion-weighted spectroscopy (DWS) of the brain is a highly motion-sensitive MR method as a consequence of the large voxel size and low metabolite diffusion coefficients, which require large b-values. While translational motions cause coherent phase shifts, which can be readily corrected retrospectively, rotations induce phase gradients over the measurement volume, leading to a rapid signal attenuation and consequently to ADC overestimation [3]. In this work, we analyse the influence of synthetic rotational motion on DWS measurements in vitro, characterise motion-induced signal loss in vivo, and explore prospective as well as rejection-based retrospective correction schemes.Experiments were performed on a 3T Prisma system (Siemens Healthcare, Germany) with a maximum gradient strength of 80mT/m. DWS measurements were performed with a customised diffusion-sensitised STEAM sequence augmented with prospective motion correction (PMC) based on Moiré Phase Tracking (MPT) [4]. Postprocessing of acquired raw data consisted of the following steps: coil combination, phasing of individual acquisitions, eddy current correction, HLSVD for removal of residual water peaks and quantification by LCModel [5]. Phantom measurements were conducted at room temperature on a phantom with 1-octanol, which has similar diffusion properties as in vivo brain metabolites [6]. A rotation of the diffusion rephasing gradient (with respect to the diffusion dephasing gradient) by a small angle Θ<1° around the z-axis served as synthetic rotational motion [7]. Sequence parameters were TE=35ms, TM=100ms, TR=1.5s, δ=14.7ms, Δ=117.5ms, VOI=1cm³, diffusion gradient directions [1,0,0], [1,1,-5] (in the scanner coordinate system), 8 b-values between 43 and 7933s/mm². An in vivo experiment was performed with a similar protocol (VOI=20x25x25mm³ (Fig. 3), one diffusion direction [0,1,0] with b=(170,679,1523,2992)s/mm² in a white-matter brain region of a healthy subject (Fig. 2). WET water suppression with suboptimal parameters was used to enable zero-order phase correction of individual averages based on the residual water peak. Motion information was recorded by the MPT system with a frame rate of 80 frames/s and a PMC-based position update of the MRS voxel was performed once per TR prior to the excitation pulse. For further improvement, single DWS averages were retrospectively rejected based on two different criteria:range of measured rotational motion between excitation and acquisition and amplitude reduction of the residual water peak (RW).Phantom experiment: According to [3,7] a small rotation $$$\theta<<1$$$ gives rise to a linear phase gradient $$$\Delta k_i=\gamma \sum_{i,j} \epsilon_{ijk}\int G_j\theta_kdt$$$ leading to intra-voxel dephasing and consequently to signal attenuation $$$\alpha=\prod_i\left|\text{sinc}\left(\Delta k_i L_i/2\right)\right|$$$ (for constant spin density). The synthetic rotation experiment shown in Fig. 1 confirms the theoretical prediction with reasonable accuracy. With b>6000s/mm2 , even a rotation of 0.1° causes an attenuation of about 50%. In vivo experiment: In experiments acquired without deliberate motion ('motionless') the RW time series exhibits amplitude attenuation increasing with the b-value (Fig. 2). The time series of Θ contains two periodic components: low-frequency respiratory motion with a periodicity of roughly 2-3 TR and an amplitude of 0.05°-0.1°, and high-frequency motion (presumably due to cardiac pulsation) with an amplitude of roughly 0.02°. In all 'swallow motion' experiments, larger rotations of Θ=0.1°-0.7° within one TR were observed. This motion strongly correlates with RW attenuation (r=0.8), leading to vanishing RW signal with b=2992s/mm². The resulting motion-corrupted spectra exhibit an attenuation of the macro-molecular peak at 0.9ppm, which should not be affected by diffusion [7] (Fig. 3). After application of both rejection criteria the filtered motion-corrupted spectra appear visually similar to the 'motionless' spectrum. Motion gave rise to an ADC overestimation of 40% for tNAA compared to 'motionless' measurement. With both rejection criteria, ADCs similar to the ‘motionless’ case could be obtained . However, the motion-based rejection criterion lead to a slight ADC increase as compared to the 'motionless' measurement (Fig. 4).Phantom and in vivo experiments demonstrate increasing susceptibility to rotation-induced signal attenuation for increasing b-values, in accordance with theoretical prediction. The strong correlation between swallow motion and signal attenuation of RW suggest that both rejection criteria are about equally efficient in correcting for subject motion during the scan. Potentially, an unsuppressed water peak (e.g. utilizing metabolite cycling) could be an even more robust marker for retrospective rejection of corrupted spectral averages [9]. PMC as performed in this work (update once per TR) can effectively prevent motion-induced voxel displacements during typically long DWS measurements. Such displacements are particularly severe for DWS experiments in small confined structures such as the corpus callosum. Feasibility of additional real-time correction between dephasing and rephasing gradients, which might mitigate motion-induced signal attenuation, will be investigated in future work, but remains a challenging task due to limited precision and non-negligible latency of optical motion tracking.