Perception affects the brain’s metabolic response to sensory stimulation

Processing of incoming sensory stimulation triggers an increase of cerebral perfusion and blood oxygenation (neurovascular response) as well as an alteration of the metabolic neurochemical profile (neurometabolic response). Here we show that perceived and unperceived isoluminant chromatic flickering stimuli designed to have similar neurovascular responses as measured by blood oxygenation level dependent functional MRI (BOLD-fMRI) in primary visual cortex (V1) have markedly different neurometabolic responses as measured by functional MRS. In particular, a significant regional buildup of lactate, an index of aerobic glycolysis, and glutamate, an index of malate-aspartate shuttle, occurred in V1 only when the flickering is perceived, without any relation with behavioral or physiological variables. Wheras the BOLD-fMRI signal in V1, a proxy for input to V1, was insensitive to flickering perception by design, the BOLD-fMRI signal in secondary visual areas was larger during perceived than unperceived flickering indicating increased output from V1. These results indicate that the upregulation of energy metabolism induced by visual stimulation depends on the type of information processing taking place in V1, and that 1H-fMRS provides unique information about local input/output balance that is not measured by BOLD-fMRI. Significance statement Visual perception has a measurable metabolic effect in the primary visual cortex (V1). Here we show that the upregulation of energy metabolism induced by isoluminant chromatic flickering depends on subjective visual perception. Within V1, perceived and unperceived stimuli that are contrast-matched to elicit similar blood-oxygenation level-dependent fMRI response are associated with clearly distinct neurochemical profiles. Specifically, regional accumulations of lactate (an index of aerobic glycolysis) and glutamate (an index of malate-aspartate shuttle) only develop during perceived stimuli, together with a larger activation of secondary visual areas. Our results imply a dissociation between metabolic and functional response, and indicate that that the upregulation of energy metabolism induced by visual stimulation depends on the type of information processing taking place in V1.


Introduction
In the brain, sensory stimulation is associated with a substantial increase of regional functional 46 hyperaemia (i.e. cerebral blood flow, CBF) as well as energy metabolism of glucose, the main 47 cerebral energy substrate (1). The metabolic response to stimulation includes an oxidative 48 component, as measured by the cerebral metabolic rate of oxygen (CMRO2), and a non-oxidative 49 component, as reflected by lactate accumulation (2). Cortical lactate levels have been shown to 50 increase during visual stimulation, simultaneously to the acceleration of the malate-aspartate 51 shuttle, a process termed aerobic glycolysis (i.e. lactate production independent of oxygen 52 availability) (3)(4)(5)(6)(7)(8)(9)(10)(11). Despite intense research, the neurophysiological mechanisms underlying the 53 up-regulation of glycolytic metabolism of glucose are still largely unknown (12). Most 54 importantly, the impact of information processing to the metabolic response of the cerebral cortex 55 to sensory stimulation has not been fully investigated thus far. In particular, nothing is known about 56 the modulatory effect exerted by the perception of different stimuli on regional brain energy 57 metabolism. 58 Sensory perception is thought to rely on the complex interplay of neural circuits that process 59 information in a cortical layer-and area-mediated manner involving thalamo-cortical, intracortical, 60 cortico-cortical and cortico-thalamic feedforward/feedback loops (13). Sensory stimuli transduced 61 by sensory organs reach specific thalamic nuclei that relay information to primary sensory cortices, 62 which in turn filter and eventually transmit information to secondary sensory areas (14). These 63 transactions are dependent on the particular features of different incoming stimuli, thus it is 64 possible that the relevant neurovascular and neurometabolic counterparts are correspondingly 65 distinct (15). 66 The thalamic lateral geniculate nucleus (LGN) mediates visual stimuli with temporal frequencies 67 at least up to 90 Hz to the layer IV of V1 (16)(17)(18)(19)(20)(21), which in turn relays to output layers II/III and 68 V where temporal filtering occurs (22), consistent with the notion that visual perception requires 69 the activation of visual areas downstream V1 (i.e. secondary visual cortices). In agreement with 70 these arguments, it has been repeatedly reported that invisible visual flickering is still able to 71 activate V1 even without any perceptual effects (23), as revealed by in vivo electrophysiology in 72 non-human primates (16) as well as behavioral evidence (24) and BOLD fMRI (25) in humans. 73 High (30 Hz) frequency visual stimulation has been found to selectively suppress multi-unit 74 activity (MUA) in cat V1 as compared to low frequency (4 Hz) visual stimulation (26). 75 Importantly, local field potentials (LFPs) and tissue oxygen response, which directly contribute to 76 the generation of the BOLD signals (27), were preserved at both frequencies . 77 In the present study, we combined blood-oxygenation level dependent (BOLD) functional 78 magnetic resonance imaging (fMRI) and proton functional magnetic resonance spectroscopy (1H-79 fMRS) in humans and exploited the well known effect of temporal frequency on visual perception. 80 Specifically, we examined the functional and metabolic responses of the primary visual cortex 81 (V1) to perceived or unperceived isoluminant chromatic flickering stimulations obtained by using 82 temporal frequency either below (7.5 Hz; PF, perceived flickering) or above (30 Hz;UF,83 unperceived flickering) the critical flicker fusion (CFF) threshold of ~15 Hz for rod-mediated 84 vision (28). Based on experimental evidence and metabolic modeling, we have previously 85 proposed that the local input-output balance between neuronal synaptic/spiking (or 86 subthreshold/suprathreshold) activity is a primary determinant in the up-regulation of aerobic 87 glycolysis (29)(30)(31). We thus hypothesized that the loss of visual perception is accompanied by 88 fundamental changes in the metabolic responses of human V1. 89 90

Subjects perception of the visual stimuli 92
To achieve perceptual isoluminance between green and red color (necessary for loss of perception 93 at 30 Hz), we adjusted the brightness of the green color for each individual subject, which was 94 remarkably similar across subjects (green/red brightness ratio 71.9±1.2%, range 70.1% to 73.5%; 95 see Table 1). After this procedure, 100% of the subjects confirmed that their perception of the 30 96 Hz frequency stimulus steadiness was equivalent to the resting condition. Overall, the subject's 97 perception was a gray/colored checkerboard that in the colored squares showed either a fast green 98 and red alternation during PF epochs, or a static yellow during UF epochs (Movie S1). As a further 99 confirmation, while in the scanner the subjects were unable to distinguish the 30 Hz red-green 100 flickering checkerboard (used in the actual experiments) from a color-matched static yellow 101 checkerboard (used for testing only). Specifically, the perception of the steady yellow color versus 102 the 30 Hz red-green flickering was indistinguishable, as assessed by asking the subjects to guess 103 the origin of the stimulus for 10 consecutive trials (average of correct responses 52±16%, not 104 different from chance level, p=0.62). All subjects reported to distinctly perceive the green and red 105 color when the checkerboard was flickering at 7.5 Hz. None of the subject perceived the intrinsic 106 flickering of the screen due to the refresh rate (60 Hz). 107

Subjects attention 112
To examine the possible modulation of the BOLD signal by attention (32), we measured 113 pupillometry and task-performance data during all experiments. All subjects maintained gaze on 114 the fixation cross during the whole epochs ( Figure 1A

124
Mean pupil diameter was not statistically different across conditions (One-Way ANOVA, p=0.98). Error bars 125 correspond to SD. (F) Task performance in terms of response delay was not statistically different across conditions 126 (One-Way ANOVA, p=0.36). Error bars correspond to SD. (G) There was no correlation between task performance and mean gaze displacement (r 2 <0.06, p>0.47). (H) There was no correlation between task performance and mean 128 pupil diameter (r 2 <0.08, p>0.49).

130
The hit/miss ratio for the cross rotation task during the stimulation protocol was essentially 1.0, as 131 expected due to the simplicity of the task, for the entire duration of the experiment. In particular, 132 the delay of the response was not statistically different for rest, PF and UF epochs (ranging, on 133 average, between 360 ms and 460 ms), both during fMRI (One-Way ANOVA, p=0.09 and p=0.77 134 for run 1 and 2, respectively) and 1H-fMRS (One-Way ANOVA, p=0.  To achieve the same BOLD response in V1 during PF and UF, we reduced the stimulation contrast 145 for the 7.5 Hz condition to 75% relative to the 30 Hz condition ( Figure 2A). As expected, we found 146 that the average BOLD timecourse ( Figure 2B To better characterize the effect of the two different stimulations, we estimated the main effect of 167 the flickering frequency. The main effect of PF appeared in the lateral occipital cortices, but not in 168 V1 ( Figure 3B and Table 2). Although the electrophysiological activity in V1 could not be directly 169 assessed in our experiment, based on the literature (27)

Dissociation between metabolic and BOLD responses in V1 during loss of visual perception 195
To assess whether metabolism was sensitive to stimulus processing within V1 we performed 196 single-voxel 1H-fMRS. The reproducible BOLD response in the occipital lobe elicited by the PF 197 and UF stimulations allowed a very accurate VOI positioning for 1H-fMRS. High quality and 198 artifact-free spectra (water linewidth 7.2±0.6 Hz, Figure 4A) were reliably obtained from almost 199 all subjects (Table 1 and  μmol/g, respectively, during the PF stimulation corresponding to an increase of about 28% and 202 3%, respectively, over the baseline (paired two-sample t-test, qFDR=0.001), whereas they both 203 remained at their basal levels (-0.04±0.13 μmol/g, qFDR=0.42 for lactate, and 0.03±0.20 μmol/g, 204 qFDR=0.63 for glutamate) during the UF stimulation. The lactate and glutamate responses were 205 significantly different (paired two-sample t-test, qFDR=0.01 for lactate and qFDR=0.003 for 206 glutamate) among the two stimulation conditions ( Figure 4B). No other metabolites among those 207 quantified showed a reliable stimulation-dependent change (Table 3). We were unable to detect a 208 reliable change for aspartate (paired two-sample t-test, qFDR=0.98). 209 To confirm our fMRI result of similar BOLD effect during PF and UF, we examined linewidth 210 narrowing of total creatine (tCr) and total N-acetylaspartate (tNAA) signals. We found a 211 stimulation-induced decrease of tCr and tNAA linewidth during both PF and UF ( Figure  To substantiate our 1H-fMRS result, we determined the difference spectra between conditions, 216 which mainly consisted of uncorrelated noise and only a few correlated residuals ( Figure 4E). 217 Specifically, the difference between PF and rest spectra showed a signal in the region 218 corresponding to lactate, and in spectral regions corresponding to glutamate; both signals were 219 absent in the difference between UF and rest spectra. The difference spectra also featured some 220 narrow peaks corresponding to the main singlets of the spectrum, particularly tCr and tNAA, as a 221 result of BOLD-induced line narrowing (36). Similar residuals on tCr and tNAA were recognizable 222 in the difference spectrum between UF and rest, but not in the difference spectrum between the 223 two active conditions, again consistent with the evidence of a similar BOLD effect on spectral 224 linewidth elicited by either of the stimulations. Overall, the only correlated signals that survived 225 in the difference spectrum between PF and UF were lactate and glutamate, which strongly supports 226 the significance of the concentration changes based on LCModel quantifications. 227 228

253
The cortical gray matter of the brain features one of the highest metabolic rates of all organ tissues 254 of the human body. Although energy is recognized as a limiting factor for the human cerebral 255 cortex (37,38), the increase in lactate concentration occurring upon sensory stimulation isn't the 256 result of limited oxygen availability (39,40), as it is for skeletal muscle. Why the cerebral cortex 257 upregulates glycolytic metabolism for sensory information processing is unknown, but it is well-258 established that glycolysis serves specific neurophysiological and neurobiological purposes, such 259 as axonal vesicle transport, vesicle recycling, action potential waveform modulation, reuptake of 260 neuroactive compounds, and dendritic spine remodeling (reviewed in (31)). Furthermore, lactate 261 is known to be implicated in cognitive processes occurring during waking activity, like learning 262 and memory (41)(42)(43)(44), although the exact underlying mechanisms are still debated (45,46). 263 In the present study, we asked whether alterations in visual perception are also reflected in 264 metabolic changes within the primary visual cortex in humans. We report that stimulus perception 265 affects the lactate and glutamate response in V1. The PF and UF stimulations elicited, by 266 experimental design, the same average BOLD signals increase inside the spectroscopic voxel, 267 indicating an equivalent degree of neurovascular coupling and possibly of local synaptic activity 268 in the two experimental conditions (47). Yet, we observed a significant increase in the regional 269 lactate and glutamate levels only during the PF stimulus, with no appreciable change during the 270 UF stimulus compared with resting conditions. A temporal dissociation between BOLD and lactate 271 changes has been previously reported during repeated photic stimulations (48). In that case, BOLD 272 response was preserved, but not the lactate and glutamate accumulation, possibly due to 273 habituation of neuronal firing. Metabolic adaptation, in terms of glutamate levels, in the presence 274 of constant BOLD and electrophysiological activity was also reported in epilepsy (49). These 275 results suggest that the physiological mechanisms underlying BOLD signals and energy 276 metabolism do not necessarily overlap under all experimental conditions. 277 Our results suggest that lactate and glutamate may be dissociated from BOLD changes when 278 cortical input and output are differentially modulated by e.g., intracortical inhibition (26). In fact, 279 an overall switch toward inhibition is expected to reduce the energy request of the brain, thus 280 impacting on metabolic rates independently of BOLD response (2). Accordingly, changes in 281 excitation/inhibition equilibrium have been proposed as a source of variability of the CBF/CMRO2 282 coupling ratio (50). In more general terms, the metabolic response seems capable of differentiating 283 neural states that are intrinsically distinct, although they induce the same BOLD response. Similar 284 BOLD signal changes in correspondence to substantially different alterations in tissue lactate and 285 glutamate level during visual flickering could not be predicted a priori unless assuming 286 fundamental differences in information processing during stimulation. In particular, the firing rate 287 of layer IV neurons that receive input from LGN is higher during UF than PF, as evidenced by the 288 synchronization of these neurons to the stimulus frequency (16)(17)(18)(19)(20)(21), yet we found that lactate and 289 glutamate increases during PF, not UF. These findings indicate that the stimulation-induced effect 290 on metabolite concentrations is happening downstream the input stage in layer IV, and specifically 291 during the intracortical processing involving output layers II/III. This conclusion is supported by 292 the fact that high-frequency flickering (30-60 Hz) abolished MUA, a measure of neuronal spiking 293 (putatively from layer II/III) (26), but not single neuron recordings from layer IV (16). 294 Using compartmentalized metabolic modeling, we have recently obtained evidence that the above-295 mentioned outcome can be explained by distinct signaling mechanisms underlying spiking and 296 synaptic activity (e.g., pre-and postsynaptic ionic currents) that are indeed frequency-dependent 297 (30,51). We previously reported that chromatic and achromatic flickering at the same temporal 298 frequency elicited the same neurochemical response in V1 despite differences in the responding 299 neuronal populations (i.e., blob versus interblob) (10). Together with the results that we report 300 here, these observations suggest that lactate and glutamate, and hence aerobic glycolysis, are both 301 sensitive to cortical processing (e.g., input/output or excitation/inhibition balance) rather than the 302 targeted neuronal population. Our results can be interpreted in keeping with the concept that 303 increased glutamate levels reflect an upregulation of the malate-aspartate shuttle (MAS) (52), 304 although we were unable to report significant differences in aspartate levels likely due to 305 insufficient sensitivity of MRS at 3T. Notably, MAS does not necessarily correlate with 306 mitochondrial respiration and cerebral blood flow, as evidenced by the findings that (i) oxidative 307 metabolism and CMRO2 are enhanced at both low and high flickering frequencies (26), and (ii) 308 glutamate and aspartate, but not lactate, correlate with BOLD signals during low-frequency (~8 309 Hz) visual stimulation (10,53). To what extent aerobic glycolysis and MAS can dissociate from 310 oxidative phosphorylation in the brain remains to be established, but previously published data 311 strongly indicates that the increase in lactate and glutamate levels is the consequence of the surge 312 in glycolysis and the ensuing accumulation of NADH taking place during neuronal activation (54). 313 While the similar BOLD response in V1 is a direct consequence of our experimental design, we 314 additionally found that secondary visual cortices were recruited to a larger extent during PF 315 stimulus than UF stimulus, supporting the argument that their activation is necessary for perception 316 (55)(56)(57). The preferential response of secondary visual areas to perceived stimuli that we observe 317 here broadly confirms previous results of a strong BOLD activity in ventral ("visual-for-318 perception" processing) and dorsal ("visual-for-action" processing) streams during visible stimuli 319 and a reduction of their activity in conditions of invisible stimulation (58). 320 The role of primary visual cortex in stimulus perception is an intense area of research (59)(60)(61)(62)(63). 321 Previous fMRI studies investigating the dependence of V1 BOLD activity on visual perception 322 produced controversial results, which might be related to the difficulty of disentangling perception 323 from spatial attention. Indeed, it has been shown that attention, but not perception, modulates the 324 BOLD signal in human V1 (32). In fact, perception was found to enhance the BOLD response 325 within V1 for specific experimental designs (e.g., flash suppression of perception) (64). Our study 326 employed a substantially different stimulation paradigm that specifically allowed us, by adjusting 327 the stimulation contrast, to induce the same BOLD response, and inductively an equivalent mean 328 degree of synaptic activity, within V1 under the two experimental conditions (47). Most 329 importantly, we did not use any additional stimulus (e.g., visual masking) to suppress or modulate 330 visual perception. On the contrary, we used a simple way to modulate perception for extended 331 periods of time (required by 1H-fMRS) while maintaining attention at a nearly constant level, as 332 evidenced by pupillometry and task performance. In particular, we employed the well-known 333 dependence of BOLD response in V1 to flickering, which at full-contrast peaks at 4-8 Hz and 334 settles around 70% of its maximal value even at frequencies above 30 Hz (65) respectively) (66)(67)(68)(69)(70). Interestingly, the preferred frequency of secondary visual areas is just below 339 the CFF for chromatic flickering (~25 Hz) and luminance flickering (~50 Hz) (71). In our 340 experiments, the isoluminant chromatic flickering stimulation at 30 Hz is above the CFF and 341 accordingly we found that the BOLD response in secondary visual areas drops substantially 342 compared with the concurrent BOLD response in V1. Overall, by using visual stimulations below 343 and above the CFF and adjusting the contrast of the low-frequency stimulus we were able to 344 modulate perception alongside BOLD activity in secondary visual areas with an unchanged BOLD 345 activity in V1. 346 Our study has some limitations. For instance, the fMRI measurements have been obtained using 347 30 s epochs, while the fMRS measurements have been obtained using 4 min epochs. Long 348 acquisition times are required for achieving a good sensitivity of the 1H-fMRS measurements and 349 a reliable metabolite quantification, even at magnetic fields of 3T. Nonetheless, results in both 350 humans (48) and rats (72) have showed that prolonged (i.e., minutes) stimulations elicit a persistent 351 BOLD response with minimal attenuation. Although we cannot exclude a certain degree of 352 neuronal adaptation, we confirmed that the BOLD effect during the fMRS acquisition was indeed 353 showing non uniform responses. Indeed, although the main differential effect of frequency was 357 located far from the calcarine sulcus, it also involved some mid hemispheric areas presumably 358 included in the spectroscopic voxel. As an illustration, heterogeneity in the response to stimulation 359 within V1 might be due to, e.g., eccentricity (67), which we however minimized by using a 3° 360 foveal stimulation. There is also an hemispherical lateralization for the spatial frequencies (73), 361 but this is not likely to apply to our study, as we used the same radial checkerboard (i.e., including 362 many spatial frequencies) for both PF and UF. The finding that the changes in BOLD signal during 363 PF and UF were nearly double in the anatomical (i.e., Brodmann atlas-based) V1 compared with 364 the spectroscopic VOI indicates a substantial partial-volume effect. The associated loss of 365 sensitivity entails that the lactate concentration change that we report here might in fact be under-366 estimated. Finally, we cannot exclude that feedback input to V1 from secondary visual areas might 367 be taking place during stimulus perception (74). However, under our experimental design we were 368 able to fully control BOLD response in V1 by only acting on stimulus contrast, without altering 369 the subjective perception of the stimuli, which indicates that the BOLD activity in V1 largely 370 reflected stimulus features. 371 In conclusion, in this study, we report for the first time that visual perception has a measurable 372 metabolic effect on V1. Our findings imply that BOLD fMRI and 1H-fMRS are complementary 373 techniques, capable of highlight different aspects of neural activation and stimulus processing, e.g., 374 under conditions in which synaptic and spiking activity are partially disentangled due to an altered 375 balance between excitation and inhibition. 376 Although we focused on the visual pathways, it is conceivable that our results can be translated to 377 other sensory modalities. For example, tactile (75) or auditory (76,77) stimulations during 378 vegetative state can still activate primary somatosensory cortex (S1) or primary auditory cortex 379 (A1), respectively, in the absence of perception and without the involvement of higher-order 380 associative areas. 381 We suggest that the cortical metabolic profile might be an indicator of sensory perception, in 382 keeping with the dynamics of lactate and glutamate across the sleep-wake cycle (31,40,78,79). 383 In particular, elevated brain lactate and glutamate levels are associated with wakefulness and 384 memory formation, which naturally require the processing of incoming sensory stimuli, like the 385 control exerted by the central visual pathways for either gating or filtering out behaviorally relevant 386 or irrelevant visual information. In particular, aerobic glycolysis and lactate might reflect cortical 387 information processing and, in turn, intracortical communication, in agreement with the relation 388 between regional metabolic rates of glucose utilization and resting-state network dynamics in the 389 cerebral cortex (80)(81)(82)(83)(84). order shim terms were adjusted using the automatized Siemens routine based on field map 399 acquisition and fitting. For MRS, shimming was optimized using FASTMAP with EPI readout 400 (85) and manually refined when necessary to reach a water linewidth of less than 9 Hz. 401

Subjects 402
A total of 32 healthy volunteers were initially enrolled for this study after they gave informed 403 consent, according to the Helsinki declaration and to European Union regulations, and following 404 the approval by the Ethics Committee of Fondazione Santa Lucia (Rome). Exclusion criteria 405 included any kind of previous neurological or psychiatric disease and impaired visual acuity. 406 Thirteen subjects were discarded either upon online fMRI processing (see below) or in early 407 quality control on data, because of motion (3) or suboptimal anatomical features, with the most 408 activated area in an unfavourable position respect to the coil sensitive volume (10). Nineteen 409 healthy volunteers (10 females, 9 males; age 25±4 years, mean±SD; age range 20 to 35 years) were 410 thus considered for this study. Sample size calculations performed before the study assumed a two-411 tail paired t-test design, a power of 0.95 and were based on an effect magnitude for lactate change 412 (photic stimulation vs rest) of 0.20 ± 0.15 μmol/g as estimated in our previous works at 7T (10).

Experimental protocol 428
Visual stimulations were presented in eight (fMRI) or four (fMRS) epochs, with stimulation 429 epochs (either PF or UF) interleaved by rest (isoluminant, uniform grey images) epochs. Since the 430 CFF for loss of perception is higher for luminance than for chromatic flickering (25), before each 431 session the brightness of green squares during the UF condition (i.e. isoluminant condition) was 432 adjusted interactively by the subject, who piloted increasing and decreasing brightness ramps and 433 was instructed to identify the brighness levels corresponding to loss of luminance flickering 434 perception of the resulting yellow. The green level was then set midway between the two 435 perceptual vanishing levels. The stimulus contrast was adjusted in preliminary acquisitions on five 436 subjects, in order to induce a comparable BOLD response to PF and UF in V1. During the initial 437 fMRI sessions epochs lasted 30 seconds each (total 4 minutes), while for fMRS acquisitions epochs 438 were 4 minutes long (total 16 minutes). Each subject underwent 1 fMRI and 2 fMRS runs (10 439 minutes apart, without moving the subject from inside the scanner); the order of PF and UF 440 conditions was counterbalanced within each subject, and the initial stimulation type was 441 randomized between subjects. 442 Task 443 In order to maintain visual fixation and keep the attentional state constant, the subjects were asked 444 to focus on a central target (a cross) and to press a button whenever the target rotated. Subjects 445 were specifically instructed to maintain their attention on the fixation cross rather than focusing 446 on reaction times (i.e., speed to push the button). To confirm the absence of any detectable brain pathology in our subjects, T1 and T2 weighted 461 anatomical scans were acquired with a standard volume birdcage coil after the end of the functional 462 scans. Anatomical scans included an MPRAGE acquisition (resolution 1.0×1.0×1.0 mm 3 , para-463 axial slices, in-plane FOV 256×160 mm 2 , TE=2.48 ms, TR=2150 ms, TI=1000 ms, FA=8º), that 464 was later used during the post-processing for normalization purposes. 465 fMRS data acquisition 466 The spectroscopic voxel (size 25×20×20 mm 3 ) was localized in the most activated area within V1, 467 based on both anatomical scan and results of the online fMRI processing. The voxel was located 468 either left or right of the interemispheric fissure to minimize the cerebrospnial fluid fraction in the 469 VOI. Two MRS sessions were acquired with an optimized, in-house written STEAM sequence 470 (TE=7 ms, TM=50 ms, TR=3000 ms, FA=70º) which included outer volume saturation and 471 VAPOR water suppression (86,87). An eight-step phase cycle was used; transients were averaged 472 within each phase cycle, and each phase cycle was saved separately for further processing. Water 473 unsuppressed data were acquired from the same voxel for eddy currents compensation (88). In 474 order to minimize T1 weighting, the flip angle was kept below the calculated Ernst angle in both 475 fMRI and fMRS acquisitions. 476 Pupillometry 477 In order to monitor attentional state with a physiological parameter, we acquired pupillometry data 478 using an eye-tracking system (Applied Science Laboratories, model 504) equipped with remote 479 pan/tilt optic infrared module and a video camera that was custom-adapted for use in the scanner. 480 Subject gaze position and pupil size data were processed as previously described (33 (Asp), creatine (Cr), γ-Aminobutyric acid (GABA), glucose, glutamine (Gln), glutamate (Glu), 506 glycine, glycerylphosphorylcholine, glutathione (GSH), lactate (Lac), myo-inositol (Ins), N-507 acetylaspartate (NAA), N-acetylaspartylglutamate (NAAG), phosphocholine, phosphocreatine, 508 phosphorylethanolamine, scyllo-inositol, and taurine (Tau). Glucose, an important marker of 509 energy metabolism, whose changes have also been reported in previous 7T studies (3,9), was not 510 included in the basis set due to highly unreliable quantification observed in preliminary tests. 511 Metabolite spectra were simulated using GAVA (94), including information on the sequence pulse 512 program. The basis set included also a subject-specific macromolecular (MM) signal, that was 513 acquired on each subject in the occipital region, using a double inversion recovery approach 514 (STEAM, TI1=1700 ms, TI2=520 ms, TE=7 ms, TM=50 ms, TR=2000 ms, FA=90º) (95), that 515 resulted in almost complete metabolite nulling, averaged between subjects, and then modeled with 516 Hankel-Lanczos singular value decomposition. LCmodel quantifications with Cramér-Rao lower 517 bounds (CRLB) above 30% were discarded, except for Lac for which the threshold was set at 35%. 518 Since this study is focused on epoch to epoch metabolic changes, absolute quantification with 519 water referencing was not performed to avoid the associated uncertainty. Metabolites were rather 520 referred to the internal creatine signal, assumed to be 7.5 μmol/g in the VOI. Finally, concentrations 521 measured in homologous epochs were averaged, obtaining for each subject four concentrations, 522 corresponding to two stimulation conditions and to the relevant rest reference (the resting epoch 523 immediately successive to a condition). Eleven metabolites were quantified in at least 15 subjects 524 (80% of participants). These included aspartate, total creatine, GABA, glutamate, glutamine, 525 glutathione, lactate, myo-inositol, N-acetylaspartate, total choline, taurine. In order to exclude 526 BOLD adaptation during the 4-min duration of the fMRS epochs, we determined the BOLD effect 527 during fMRS scanning as the kernel size (in Hz) that minimized the amplitude of the difference spectra between stimulated epochs (either PF and UF) and the preceding resting epoch. All spectra 529 were then averaged according to three categories: rest, PF and UF conditions, and spectral 530 differences were calculated between conditions. 531 Statistics 532 For pupillometry and task performance results, statistical comparisons were made using Student's 533 t-test and One-Way ANOVA on the rest, UF, and PF conditions. No post-hoc test was necessary. 534 For fMRI results, correction for multiple comparisons in functional voxel-based analysis was 535 performed using False Discovery Rate (FDR) correction. Resulting clusters were also checked 536 through Monte Carlo Simulation using the AFNI tool Alphasim (89) after estimation of residuals 537 smoothness. 538 For fMRS results, statistical analysis was restricted to those reliably quantified metabolites 539 associated with energy metabolism that showed consistent funcional changes in previous fMRS 540 studies (4,9,10), namely Lac, Glu, and Asp. Metabolite concentration changes referred to the 541 corresponding resting epoch and between different active conditions were tested using paired two-542 sample t-tests, with FDR correction for 9 multiple comparisons. 543 Data were presented as the mean ± standard deviation (SD). A p-value, or a qFDR-value where 544 relevant, of less than 0.05 was considered as statistically significant.

Ethics statement 567
All experiments with human subjects performed by the authors complied with all applicable ethical 568 standards, including the Helsinki declaration and its amendments, institutional/national research 569 committee standards, and international/national/institutional guidelines.   Left: task performance in terms of response delay was not statistically different across conditions (One-Way ANOVA, p>0.09). Error bars correspond to SD. Center: there was no correlation between task performance and mean gaze displacement (r 2 <0.06, p>0.47). Right: there was no correlation between task performance and mean pupil diameter (r 2 <0.08, p>0.49). Sessions are shown in different rows (Top: fMRI Run 1; Middle: fMRI Run 2; Bottom: 1H-fMRS Run 2).

Figure 4 -figure supplement 1
Averaged 1H-fMRS spectra across individual epochs (Left: REST; Center: PF; Right: UF) of the subset of subjects (N=16) who had a reliable quantification for both Lac and Glu. <insert Table S1 here followed by a page break > Movie S1 (separate file).
The movie shows the pupil of a representative subject during a stimulation cycle including UF and PF epochs, as well as the relevant physiological and behavioral responses.