A different interpretation of the DIANA fMRI signal

Abstract

Direct detection of neural activity by functional magnetic resonance imaging (fMRI) has been a longstanding goal in neuroscience. A recent study argued that it is possible to detect neuroelectrical potentials using a specialized fMRI scanning approach the authors termed “direct imaging of neuronal activity” (DIANA). We implemented DIANA in anesthetized rats and measured responses to somatosensory stimulation, reproducing core findings of the original study. We show, however, that neural activity is neither sufficient nor necessary to produce such results. We use a combination of control conditions and simulations to demonstrate that DIANA signals can arise from nonideal aspects of the pulse sequence and specimen that help determine spatiotemporal characteristics of the data. Our analysis emphasizes a need for cautious interpretation and mechanistic evaluation of advanced fMRI techniques.

A reported ultrafast fMRI signal arises from artifacts in the imaging approach.

INTRODUCTION

Toi et al. (1) introduced the “direct imaging of neuronal activity” (DIANA) functional magnetic resonance imaging (fMRI) method to monitor fast correlates of brain function. DIANA is capable of producing two-dimensional scans with a temporal resolution of 5 ms under conditions where line-scan fast low-angle shot (FLASH) fMRI acquisition can be synchronized to repeated presentations of a stimulus (Fig. 1A) (2). Using the DIANA approach in anesthetized mice, Toi et al. revealed that transient fMRI responses of ~0.2% in amplitude could be observed within milliseconds of somatosensory, visual, and optogenetic stimuli. DIANA signals were seen in regions of the brain that show blood oxygenation level–dependent responses under analogous conditions in conventional mouse fMRI experiments (3, 4). Although the physical mechanism behind the DIANA effect was not determined, the fMRI signals appeared to be coincident with multiunit neuronal firing activity recorded from the same brain regions, raising the exciting possibility that the DIANA signals are causally related to electrophysiological signaling.

Fig. 1. Assessment of DIANA and DODI across conditions.

(A) Diagram of DIANA pulse sequence illustrating the stimulus trigger–associated delay in the context of rapid MRI line scan acquisition with 5 ms per scan. (B) Performance of DIANA in multiple samples under conditions indicated by the labels. Maps show t values for signal in the first six scans (yellow, times in milliseconds) with a threshold of t = 3 from individual subjects, all imaged with the same DIANA parameters (40 scans per repetition, 50 k-space lines, 50 repetitions), superimposed on grayscale MRI scans. Peaks in S1 and thalamus (Th) labeled in magenta. The bottom two results (dashed box) are inconsistent with a mechanism related to neural activity. (C) Comparison of average S1 signal time courses (n = 5) consistent with expected performance of DIANA. (D) Dependence of the DIANA t-value map from an individual animal on t threshold (yellow labels). (E) Comparison of mean (n = 5) S1 time courses unpredicted by an activity-dependent DIANA mechanism. (F) Diagram of the DODI pulse sequence illustrating omission of the stimulus trigger from the signal acquisition protocol and staggering of the stimulus (green arrowhead) from the trigger. (G) Performance of DODI in individual experiments analogous to (B). No voxels with t ≥ 3 are observed. (H) Mean S1 time courses (n = 5) obtained using DODI under conditions where DIANA produces expected results. (I) DODI t-value maps obtained at multiple thresholds, again illustrating lack of signals. All error bars represent SEM.

Since publication of the original DIANA study, other investigators have reported an inability to replicate its principal findings (5, 6). Because of this failed reproduction, some have speculated that the positive DIANA results might have arisen because of biased analysis or selective inclusion of data by Toi et al. This hypothesis is difficult to prove, however, and an alternative explanation is that the attempted reproductions differed subtly from the initial study in ways that prevented a genuine DIANA signal from being observed. Differentiating these two possibilities is important in part because of the need to ascertain whether the DIANA effect or phenomena like it might eventually be useful in future research.

RESULTS AND DISCUSSION

To further investigate the reproducibility and origins of the DIANA signal, we applied the approach in a study of anesthetized rats undergoing unilateral electrical stimulation of the hindlimb in synchrony with the DIANA pulse sequence in a 9.4-T magnetic resonance imaging (MRI) scanner. In contrast to other reports (5, 6), we found that DIANA does in fact yield reproducible, transient responses in the brain. We observed signals in primary somatosensory cortex (S1) that follow stimuli with an average time to peak of 18.0 ± 1.3 ms and amplitude of 0.14 ± 0.02% (Fig. 1, B and C); the mean response was statistically significant with t test P = 0.002 (n = 5). At a t-value threshold of 1.0, the DIANA-associated signal changes appeared to stretch across most of the dorsal cortex, but at a more conservative t threshold of 3.0, the signal was restricted to a contralateral S1 region where peak activity is expected (7) (Fig. 1D). Smaller DIANA responses of 0.11 ± 0.03% were observed in parts of the thalamus, also contralateral to the stimulus (Fig. 1B and fig. S1); these signals peaked slightly before S1, 16.0 ± 4.2 ms after stimulation. No DIANA-associated responses (t test P ≥ 0.12, n = 5) were seen in control experiments performed without stimulus triggering, in animals imaged 1 hour after death, or in a water phantom (Fig. 1, B and C). These results are all consistent with results reported by Toi et al. and suggest that the DIANA method was functioning similarly in our experiments. As a further test, we verified that the positive DIANA signals of Fig. 1C display dependence on echo time (TE) but independence of flip angle, also consistent with the Toi et al. study (fig. S2).

Unexpectedly, however, significant responses of average magnitude 0.12 ± 0.03% (t test P = 0.007, n = 5) were also observed in animals that underwent DIANA imaging with stimulus triggering but with the electrical stimulus cable unplugged and stimulation therefore disabled. A mean response of 0.10 ± 0.02% (P = 0.003, n = 5) was seen furthermore in animals scanned 5 min after death, as defined by loss of periodic pulse oximetry signals. These observations (Fig. 1, B and E, and fig. S1B) imply that neuronal activity is not necessary for producing MRI signal changes in the DIANA experiments we implemented.

We noticed that the DIANA sequence used by Toi et al. includes a short time interval associated with triggering that is only present in line scans synchronized with stimulation, once every 200 ms. Although the trigger interval is only 12 μs [0.2% of the repetition time (TR) and 0.6% of the TE], the asymmetry imposed by the trigger could potentially create artifacts correlated with the stimulus presentation time. To test whether omission of the triggering interval changes the DIANA contrast, we designed an alternative pulse sequence which we call “DIANA omitting disruptive intervals” (DODI). DODI uses exactly the same line-scan FLASH acquisition as DIANA but sends a trigger only once every 10 200-ms stimulation cycles (Fig. 1F); intervening stimuli are then synchronized with the MRI acquisition using hardware operated independently of the scanner, and line scans within one entire cycle (200 ms) following each trigger are omitted from subsequent analysis. As a further precaution against trigger-related artifacts, we used external software to enforce a 100-ms delay between stimulus triggers controlled by DODI and consequent stimulus pulses generated by the constant current amplifier we used. We confirmed that DODI elicits the same stimulus pulses as DIANA by monitoring current induction in a wire loop placed in the field of view (FOV) during DIANA and DODI fMRI experiments performed on water phantoms (fig. S3).

The experiments of Fig. 1B were repeated using the DODI approach and resulted in no detectable fMRI signal changes (t test P ≥ 0.16, n = 5) under any conditions (Fig. 1, G to I, and fig. S1). Because these experiments use stimuli that are known to evoke reliable neural responses in somatosensory brain regions (8), the results thus indicate that neural activity is insufficient to produce DIANA-like fMRI responses under conditions where the short stimulus-triggering delay has been removed from the MRI pulse sequence. Notably, DIANA signals in a phantom and a human brain could also be observed using an implementation of the pulse sequence on a clinical 3-T scanner. For the human experiment, a more perturbative trigger interval of 60 μs was inserted into the pulse sequence, along with a 200-Hz radiofrequency (RF) offset, to ensure that larger DIANA artifacts could be observed under the comparatively low temporal signal-to-noise ratio (tSNR) conditions of this experiment. As with the high-field experiments, DIANA-like signals at 3 T were absent when the DODI pulse sequence, lacking the trigger interval, was used instead (fig. S4). In the human participant with trigger delay but without stimulation, DIANA peak shapes varied across the FOV, as with the 9.4-T rat data of Fig. 1 and fig. S1. These results indicate that the phenomenology of DIANA and DODI is not specific to small-bore fMRI at high magnetic field.

If the triggering delay that distinguishes DIANA and DODI experiments were the sole determinant of whether signals are observed, then DIANA responses would have been seen under all conditions and in all voxels probed in Fig. 1B. Because this was not the case either in our experiments or in the comparable results of Toi et al., we explored what other factors might play a role. First, we examined the relationship between the tSNR of voxels in the DIANA images and the strength of DIANA fMRI responses. The tSNR was asymmetric and biased toward the dorsal cortex, reflecting the imperfect sensitivity profile of our surface coil as well as intrinsic contrast in the rat brain (Fig. 2A). Local tSNR maxima are observed near both cortical and thalamic regions where DIANA signals are strongest in Fig. 1B. Overall, we found a positive correlation between tSNR and the t value associated with the DIANA signal from live rats (Fig. 2B), with a highly significant Pearson’s correlation coefficient of 0.49 (P = 0.0001, n = 2155 voxels in five animals). The tSNR from postmortem animals was much lower than from live animals, likely because of accumulation of deoxyhemoglobin (9) and explaining the lack of significant DIANA signals in these subjects. The relationship between DIANA and tSNR also provides a possible explanation for the tendency of DIANA responses to fall in brain regions where hemodynamic signal enhancements are typically observed. Although we saw DIANA signals under both stimulated and unplugged conditions, the mean signals observed during actual stimulation were modestly but systematically larger (Fig. 2, C and D), by about 30% (t test P = 0.046, n = 5). These results suggest that the triggering interval and tSNR together help determine the signal characteristics observed in DIANA experiments.

Fig. 2. Investigation of DIANA mechanism.

(A) Maps of tSNR for a representative live rat (top) and dead rat (bottom), 1 hour after euthanasia. Labels denote local tSNR maxima in S1 and thalamus (Th) contralateral to the stimulated limb. (B) Comparison of tSNR with t values for stimulus-associated DIANA signal change over voxels in five live rats (magenta) and five dead rats (gray), showing strong correspondence in live animals. (C) Mean time courses (n = 5) from a further set of animals imaged by DIANA under ketamine/xylazine in stimulated (magenta) or unplugged (black) conditions, showing a small change in amplitude following stimulation. (D) Pairwise comparison of amplitudes from animals in (B). (E) Average Bloch simulation results (50 runs per condition) comparing DIANA signals with and without RF spoiling (magneta and black, respectively). (F) Omission of spoiling from DIANA measurements produces an enhancement predicted by the simulation (black trace, n = 3), compared with normal DIANA (magenta, n = 5). (G) Dependence of mean simulated DIANA on the effective resonance linewidth (n = 50). (H) Dependence of average simulated DIANA responses on T₂ (n = 50). (I) Phase diagram comparing DODI (top) and DIANA (bottom). Black diagonals depict magnetization phase evolution during repeated pulses with flip angle α (gray verticals). Each pulse flips some of the magnetization around the horizontal phase = 0 axis, splitting it into inverted (dashed) and noninverted (solid) components. Incomplete spoiling can lead to propagation of the inverted components beyond subsequent excitations. In DODI, pulses are evenly spaced at the repetition time of the pulse sequence (TR), but in DIANA, the extra delay δ slightly extends TR (magenta label), leading to further phase trajectories (magneta lines) that are absent in DODI. Echos can form wherever these trajectories cross the horizontal axis, providing a mechanistic basis for the DIANA signal.

To gain further insight into these effects, we simulated the DIANA experiment using the Bloch equations (10), which describe the evolution of magnetization in MRI scans. The simulation approach is schematized in fig. S5, with further details described in Materials and Methods. The calculations confirmed that small signal peaks could be observed within milliseconds of the DIANA stimulation trigger but that analogous signals are absent from the simulated DODI signal (Fig. 2E). The magnitude of the simulated DIANA signals is strongly affected by RF spoiling, as implemented using a standard phase cycling regime (11)—a predicted result that we could experimentally verify using fMRI in rats (Fig. 2F). The strengths of simulated DIANA signals also depend on the dispersion of magnetic resonance frequencies in the simulated specimen (Fig. 2G), a factor that could explain why the comparatively homogeneous phantom probed in Fig. 1B did not produce an experimentally observed DIANA signal. To further examine this idea, we performed DIANA on a phantom containing an air bubble and saw that signals could indeed be observed in regions of presumed magnetic heterogeneity near the bubble (fig. S6). The amplitude of the simulated DIANA signal also depends on the transverse relaxation time (T₂) (Fig. 2H), a factor that could also contribute to activity-dependent enhancement of DIANA signal (Fig. 2C) in the presence of hemodynamic effects on T₂ (12, 13). Parameters that affect simulated DIANA peak height also alter the time to peak (fig. S7), indicating that multiple factors likely codetermine the variable timing of DIANA signals experimentally observed among brain regions. More generally, qualitative characteristics of the DIANA signal in simulations could be could be reconciled with a phase diagram that predicts the formation of coherent echos by residual unspoiled transverse magnetization propagated between excitations in the DIANA but not DODI acquisition schemes (Fig. 2I) (14).

Toi et al. proposed that T₂ changes in activated neurons might contribute to rapid signal changes observed by DIANA. These changes could in principle form the basis of DIANA-related fMRI approaches, even if the pulse sequence artifact of the original DIANA is corrected. To investigate this issue further, we therefore replicated a test of cell stimulation followed by T₂-weighted MRI reported in the original paper. Immortalized T cells (Jurkat cells) were incubated with varying concentrations of supplemental potassium chloride, expected to produce rapid membrane depolarization, for 20 or 90 min before pelleting and imaging by T₂-weighted MRI at 9.4 T. A positive relationship between potassium concentration and T₂ was observed after 90 min (correlation coefficient = 0.97, P = 0.001, n = 6 conditions) but not 20 min (correlation coefficient = −0.16, P = 0.76) of KCl incubation (Fig. 3, A and B). Morphological analysis of the cells before versus after potassium treatment (Fig. 3, C and D) shows that cells were significantly larger after incubation with potassium for 90 min than before (t test P < 0.0001, n = 392 cells before and 331 cells after incubation). These results suggest that cell swelling but not membrane depolarization is linked to the T₂ effects observed in these experiments and that the changes occur on a timescale that is unlikely to cause fast signals observable in DIANA-based fMRI.

Fig. 3. Investigation of stimulus-dependent T₂ changes in cultured cells.

(A) T₂ maps of Jurkat T cells imaged 20 min (left) and 90 min (right) following addition of labeled concentrations of KCl in millimolar. (B) Average T₂ values (n = 3) measured at the two time points. (C) Sample micrographs of Jurkat cells before (top) or after (bottom) 90-min incubation with added 141 mM KCl. Scale bar, 50 μm. (D) Quantification of cell diameters before (n = 392) and after (n = 331) incubation with 141 mM KCl. All error bars, SEM.

The DIANA approach of Toi et al. (1) is novel in applying line-scan MRI to probe brain function on a millisecond timescale. Here, we have shown how sensitive this method is to subtle nonidealities in the pulse sequence and to spatially dependent properties of the specimen under investigation. Notably, we demonstrate that DIANA-like signals can be observed reproducibly in the brain but that neuronal activity is neither necessary nor sufficient for producing them. The tendency of DIANA responses to overlap with specific brain areas arises principally from the nonuniform sensitivity profile of surface coils used in these experiments, combined with a further bias toward DIANA detection in the presence of hemodynamic T₂ effects. Although we cannot explain the precise temporal relationships among DIANA signals reported by Toi et al., we did observe that DIANA peak times vary somewhat among brain regions in both animals and humans and that multiple physical factors can cause similar variation in simulations. We speculate that correspondence of electrophysiological and DIANA signals reported in the original study was largely a coincidence, unrelated to a close physical coupling mechanism but rooted in a shared relationship of these two readouts to spatially colocalized hemodynamic responses (15). Although the goal of probing neural activity with millisecond temporal resolution using fMRI remains unmet, we hope that the contribution of Toi et al. will inspire further efforts to achieve this ambitious and worthy objective.

MATERIALS AND METHODS

Animal preparation and stimulation

All animal procedures were conducted in accordance with National Institutes of Health guidelines and with the approval of the MIT Committee on Animal Care (protocol number 0721–059-24). All experiments were performed with female Sprague-Dawley rats, age 8 to 12 weeks, supplied by Charles River Laboratories (Wilmington, MA). Twenty rats were used for in vivo imaging experiments described here.

Rats were anesthetized and intubated for mechanical ventilation. For both anesthesia and paralysis, a mixture of dexmedetomidine (0.05 mg/ml) and for paralysis pancuronium bromide (1 mg/ml) was injected into each rat before scanning [intraperitoneal bolus (1 ml/kg) for induction, intraperitoneal infusion (2 ml/kg per hour) afterwards]. Data from these rats are shown in Fig. 1 and fig. S1. Another group of rats was anesthetized with a cocktail of ketamine/xylazine (80 mg/kg, 8 mg/kg) intraperitoneally instead of dexmedetomidine to more closely mimic anesthesia used by Toi et al. (1). These results are shown in Fig. 2 (B and C). During all scans, body temperature was kept stabile using a heating pad with water temperature set at 39°. Animals were monitored by pulse oximetry to measure blood oxygenation and heart rate, using a foot clip on the contralateral side to the simulation side. Postmortem experiments were performed after 5 min and 1 hour after loss of signal of the vital sign monitoring. Before these experiments, animals were euthanized by removing the oxygen and turning on 5% isoflurane.

For all experiments involving hindlimb stimulation, stimulation was delivered using two electrodes (anode and cathode) inserted subcutaneously into the right hind paw with a current of 0.5 mA, pulse duration of 0.5 ms, and interpulse interval of 200 ms, matching the fMRI TR. Constant current pulses were generated using an A-M Systems Model 2100 stimulus isolator and pulse timing was governed using a Cambridge Electronic Design 1401 data handling unit controlled by the accompanying Spike2 version 7.19 software.

Small-bore fMRI scanner operation

High-field MRI acquisition was performed with a 20-cm-bore 9.4-T Bruker small animal scanner. A custom-made 30-mm surface coil was used as a receiver, while a Bruker volume coil was used for transmission. Field inhomogeneity was minimized via the MAPSHIM protocol in Paravision 6.0.1 software. Anatomical scans were acquired using a T₂-weighted rapid acquisition with refocused echos pulse sequence with 18 slices of 1-mm thickness, 25 mm by 25 mm FOV, TE = 33 ms, TR = 2 s, and 8 averages. Except where noted in the text, functional imaging (DIANA or DODI) was performed with the following parameters: FOV 25 mm by 25 mm, image size 50 × 50, slice thickness = 1 mm, flip angle (α) of 4°, TE = 2 ms, TR = 5 ms. Thirty repetitions were acquired for in vivo experiments and 30 to 50 repetitions were acquired for phantom experiments. The excitation pulse length was 1.4 ms, with a time-bandwidth factor of 4200 Hz-ms and 2-ms slice selection gradient duration. For all experiments, RF spoiling was applied using the RF spoiling module in the Bruker Paravision 6.0 scanner control software, except for the trials designed to measure signal without RF spoiling. The Bruker spoiling method implements a quadratic phase cycling program to increment the phase of each excitation pulse by n × 117°, for the nth excitation, with the receiver phase tracking the excitation; the phase program was extended to cover the entire acquisition series without resetting. Default gradient spoiling was implemented using the MRT_UpdateSpoiler function referenced by the Bruker FLASH pulse sequence. To probe T₁-related contrast, DIANA was performed using α = 2°, 4°, and 8°, as in Toi et al. (1). To probe T₂-related contrast, the acquisition was implemented with TE = 2, 4, or 6 ms, all with a TR of 10 ms.

For all in vivo experiments, the FOV used for fMRI was placed over an appropriate region of S1 cortex to detect hindlimb stimulus-induced activation using an atlas-aligned T₂-weighted anatomical scan as a reference. Imaging experiments were conducted under three conditions: (i) with stimulation; (ii) under nonstimulated conditions where the trigger was omitted from the pulse sequence; (iii) and under conditions where the stimulation trigger was retained but the stimulation cable to the animal was unplugged from the stimulation amplifier. One group of five animals was used for all DIANA experiments in Fig. 1, a second cohort of five rats was used for all DODI experiments of Fig. 1, a third group of five rats was used for the experiments of Fig. 2B, and three further animals were used for the RF spoiling test of Fig. 2E.

Phantom experiments were performed under conditions matching (iii), but with a water-filled 2-ml Eppendorf tube scanned in place of the rat and a 12 mm by 12 mm FOV with matrix size 40 × 40 was used. To test for synchronization between the MRI acquisition and the stimulation, we performed experiments on a water-filled 50-ml Falcon tube. A solenoid coil with a diameter of 1 cm (10 turns) was placed inside the Falcon tube and connected to the same stimulator used for functional imaging experiments in rats. Imaging was then performed in the presence of stimulation with parameters identical to those used in DIANA and DODI experiments. Magnetic induction caused by the coil was detectable as signal changes using both acquisition approaches. Synchrony between these distortions and the scanner operation was confirmed by examining time courses of the resulting signal change.

Clinical fMRI scanner operation

DIANA and DODI pulse sequences were implemented on a Siemens Prisma 3T. For phantom imaging, a saline-filled 50-ml Falcon tube was imaged in a saddle coil transceiver, using the following parameters: 60 mm by 60 mm FOV, matrix size 64 × 64, slice thickness 2 mm, TR = 7.5 ms, TE = 3.0 ms, α = 4°, and phase cycling increment 50°. A delay of 10 μs was inserted at every 100th TR for DIANA-style acquisition, and 400 repetitions of the entire set of time points and k-space lines were acquired. For human head imaging, a 32-channel receive-only head coil and body coil transmitter were used for acquisition with following parameters: 19.2 cm by 19.2 cm FOV, matrix size 64 × 64, slice thickness 3 mm, TR = 7.5 ms, TE = 3.0 ms, α = 4°, and phase cycling increment 50°. Gradient spoiling was implemented using the default algorithm of the FLASH pulse sequence on the Siemens Prisma platform. A delay of 60 μs was inserted at every 64th TR, and 200 overall repetitions were acquired. The longer DIANA delay of 60 μs was used to boost detectability of the DIANA artifact in the presence of tSNR limitations arising from the comparatively short acquisition duration and propensity for motion, with respect to the 3-T phantom experiments. For analogous reasons, head imaging data were acquired with a small frequency offset (200 Hz) that was not used in other experiments. This also increased the artifact size and was necessary for robust detection of DIANA signals in the human individual.

MRI pulse sequences

DIANA fMRI experiments were implemented using the exact pulse sequence (ppg file) used by Toi et al. or closely modified from the standard Bruker FLASH ppg file to confirm the deterministic importance of the stimulus trigger delay in giving rise to the DIANA signal. The pulse sequence implements loops in the following order: (i) over time points between stimuli (40 time points with 5-ms TR in our implementation), with trigger output generated once per loop, accompanied by a 12-μs delay; (ii) over lines of k-space (50 in our implementation); and (iii) over repetitions (30 to 50 in our experiments). The DODI pulse sequence is similar to the DIANA pulse sequence but implements slightly different loops: (i) over blocks consisting of one stimulus trigger and 10 repetitions of the all time points between pairs of stimuli, with stimulus timing synchronized by the external pulse generator; (ii) over lines of k-space; and (iii) over further repetitions. Data from the first iteration of the DODI loop 1 acquisition are discarded before further processing, so that effects of the trigger used in this loop are avoided. The total number of repetitions in DODI is therefore nine from loop 1 times the value used for loop 3 (5 to 20 as for DIANA experiments). Bruker-format DIANA and DODI pulse sequences produced by the authors for this study are available upon request.

DIANA and DODI acquisitions on the clinical scanner were implemented on the basis of the Siemens FLASH pulse sequence, by interchanging the averaging and phase-encoding loop. To simulate the stimulus trigger in DIANA, a delay was introduced into periodic repetitions in the innermost acquisition loop to mimic the analogous delay in the Bruker-format DIANA sequence. We note that the Siemens Prisma trigger module would not ordinarily introduce a delay like this, so it was necessary to explicitly modify the pulse sequence to emulate the Bruker-based DIANA implementation.

MRI time series analysis

Data from DODI and DIANA experiments were analyzed identically. Images were reconstructed by a custom-made script in Python by using the fid, acqp, visu_pars, and reco files from Bruker. Binary data in the fid file were unpacked and reshuffled into the correct order. An inverse fast Fourier transform was applied with phase correction supplied by the reco file to reconstruct the images. Image time series were then preprocessed using the following steps: (i) the mean of all repetitions for each dataset was computed; (ii) a temporal smoothing was applied using a Gaussian kernel (10-ms width) to each average image time series; and (iii) rat brain data were aligned to the Waxholm atlas (16). Rat fMRI time courses were obtained as the signal averaged over voxels in the contralateral S1 hindlimb region and a thalamic region, both as defined in the Waxholm space. The thalamic region combined posterior, ventral posteriomedial, and posterolateral nuclei, as specified by the atlas. Percent signal change values and time courses were defined with respect to the mean of a baseline computed from time frames 1 to 20, before each stimulus pulse. Peak response amplitudes reported in the text were computed as the difference between this baseline and time points of the peak signal with errors computed as the SEM over animals in each case.

Response maps for individual datasets, presented in Fig. 1 (B and G), were computed by performing Welch’s t test between each voxel amplitude at each time point and the baseline as defined above, over repetitions of the acquisition. For rat experiments with 30 repetitions, results with t > 2.0 are significant with P ≤ 0.05. Before calculation of the t map, a spatial smoothing with a Gaussian kernel of one voxel was applied. Maps of t values are displayed as overlays over corresponding original DIANA or DODI images. Values of tSNR were computed by dividing the average image signal by the SD after quadratic detrending of the time series. Statistical comparisons were conducted by using two-tailed t tests between different groups (paired or unpaired). Descriptive statistics are reported in the text as means ± SEM over animals or data points, as noted.

T₂ measurements from KCl-treated cells

Jurkat immortalized T cells were used to measure the T₂ relaxation time before and after incubation with different concentrations of potassium chloride, as in the study by Toi et al. (1). Cells were cultured in RPMI 1640 cell media (Thermo Fisher Scientific) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin (Thermo Fisher Scientific). Incubation with varying concentrations potassium ions was performed by combining the cells with media supplemented by 0, 1, 4.2, 49.4, 95.8, or 141 mM KCl. After 20 or 90 min of incubation, 10⁷ cells per condition were pelleted into individual wells of a 96-well assay plate and scanned by MRI. A Bruker 7T scanner was used with a volume coil to obtain multiecho spin echo images of a single 0.5-mm slice positioned at the half-height of the cell pellets. Further acquisition parameters were as follows: FOV 58 mm by 58 mm, image size 256 × 256, α = 90°, TE = 8–240 ms, TR = 5 s. Data were acquired 20 and 90 min following the end of the KCl incubation period. T₂ values were computed by curve fitting exponential decays to the data as a function of TE. Microscopic analysis was performed from cells before (pre) or following (post) incubation with added 141 mM KCl for 90 min. Cell diameters were obtained by manual measurement of dimensions from the micrographs. Results from these experiments combine data from three individual repetitions, with averages and SEMs over repetitions reported where relevant.

Bloch simulation

MRI signal from DIANA and DODI acquisition schemes was simulated by integrating the Bloch equations (10) in a rotating frame demodulated by the resonance frequency, using custom scripts coded in MATLAB and schematized in fig. S5. Magnetization dynamics were described by

$$\frac{d}{\mathit{dt}}\overrightarrow{M}=\left(\begin{array}{ccc}\frac{-1}{T_2}& \mathrm{\gamma }∆B_z& 0\\ -\mathrm{\gamma }∆B_z& \frac{-1}{T_2}& 0\\ 0& 0& \frac{-1}{T_1}\end{array}\right)\overrightarrow{M}+\left(\begin{array}{c}0\\ 0\\ \frac{M_{\infty }}{T_1}\end{array}\right)$$

where T₁ and T₂ are the relaxation times, $\overrightarrow{M}$ is the instantaneous magnetization, M_∞ is the magnetization at thermal equilibrium, ∆B_z is an offset of the main magnetic field strength (B₀), where appropriate, to account for gradients and off-resonance effects. RF pulses were modeled as instantaneous rotations by flip angles α about axes of varying phase around z. RF spoiling was simulated by incrementing the excitation pulse phase by i × 117° per excitation, where i is the excitation number (17). To parallel the experiments, the flip angle was 4°, TR was 5 ms, TE was 2 ms, and T₁ and T₂ relaxation times were 1250 and 50 ms, respectively, unless noted. The DIANA sequence was simulated by adding an extra delay of 10 μs for every 40th TR, while the DODI sequence was simulated without this delay. For each repetition of the simulated pulse sequence, the magnetization at TE was rotated by the same phase used for RF spoiling, to simulate receiver phase cycling, and then saved.

Each simulation computed dynamics for a distribution of 50 spins, each modeled at 200 spatially resolved positions, for a total of 10,000 spin trajectories per run. Each spin was assigned a random frequency chosen from a Gaussian probability distribution with a mean of −100 Hz and an SD of 300 Hz, unless otherwise noted. Gradient spoiling was modeled by applying a position-dependent phase offset to each spin at the end of each TR, such that the full range of phase changes was 2π. Two-dimensional Gaussian noise was added to the transverse magnetization recorded at each TE, using an SD of 0.1% of the magnetization along x and y axes.

The full pulse sequence was repeated 2000 times per run, equivalent to 50 cycles of stimulation every 40 scans. Results from each run were defined as the mean transverse magnetization computed over all spins and positions at each TE, for the final 40 scans. Mean and SEM values were computed by averaging simulated DIANA or DODI signal over 50 independent simulations. Results were temporally aligned to mock stimulation corresponding to the lengthened TR values included in the DIANA simulation runs. Percent signal changes are defined with respect to the baseline before these alignment points, as for the DIANA and DODI experiments.

Supplementary Materials

This PDF file includes:

Figs. S1 to S7

Legend for data S1

Other Supplementary Material for this manuscript includes the following:

Data S1