Neural-Event-Triggered fMRI of large-scale neural networks

Available online at www.sciencedirect.com

ScienceDirect Neural-Event-Triggered fMRI of large-scale neural networks Nikos K Logothetis Brains are dynamic systems, consisting of huge number of massively interconnected elementary components. The activity of these components results in an initial conditionsensitive evolution of network states through highly nonlinear, probabilistic interactions. The dynamics of such systems cannot be described merely by studying the behavior of their components; instead their study benefits from employing multimodal methods. Neural-EventTriggered (NET) fMRI is a novel method allowing identification of events that can be used to examine multistructure activity in the brain. First results offered insights into the networks that might be involved in memory consolidation. On-going work examines the physiological underpinnings of the up and down modulation of metabolic activity, mapped with this methodology. Addresses Max Planck Institute for Biological Cybernetics, Imaging Science and Biomedical Engineering, University of Manchester, Manchester, United Kingdom Corresponding author: Logothetis, Nikos K ([email protected])

Current Opinion in Neurobiology 2015, 31:214–222 This review comes from a themed issue on Brain rhythms and dynamic coordination Edited by Gyo¨rgy Buzsa´ki and Walter Freeman

http://dx.doi.org/10.1016/j.conb.2014.11.009 0959-4388/# 2014 Published by Elsevier Ltd.

Introduction Neural activity patterns related to behavior occur at many different spatiotemporal scales and on different levels, ranging from molecules, genes and cells to neuronal microcircuits and large-scale neuronal networks that often involve distinct brain nuclei and regions. A computationally useful description of the hierarchical levels of organization considers the degree of covariance between neurons and the population size and extent of local neighborhoods [1,2]. The microscopic and mesoscopic levels predominantly refer to spiking and integrative dendritic activity, respectively. Macroscopic activity informs us about whole brain dynamics and interactions Current Opinion in Neurobiology 2015, 31:214–222

between large-scale neural systems, such as structures, nuclei and cortical regions. A predominant methodology for studies at the microscopic and mesoscopic-levels involves the use of single or multiple unit recordings. In these recordings, neuronal activity is often compared with sensory processing or behavior [3]. The development and optimization of electrode arrays and various intracranial imaging methods now permits recordings from an increasingly greater number of neurons, which provides information on larger neuronal ensembles [4]. In addition to microscopic and mesoscopic measurements, non-invasive brain imaging methods, such as electroencephalography (EEG), magnetoencephalography (MEG), positron emission tomography (PET) and functional magnetic resonance imaging (fMRI), are also used. These methods can be used in combination with biophysical modeling and multivariate pattern classification and decoding techniques to understand large-scale connectivity and the interactions of various neural structures and areas constituting the brain macroscopic system [5]. Improving our understanding of the function of neurons and microcircuits is important. Furthermore, understanding the anatomical and functional connectivity of large-scale networks is critical. The study of the brain’s complexity would also benefit from developing and employing methods combining simultaneous small and large-scale recordings of neural activity. Brains are complex dynamic systems par excellence. The brain system consists of many massively interconnected elementary components. The activity of these components results in an initial condition-sensitive evolution of network states through highly non-linear, probabilistic interactions. The dynamics of such systems cannot be described merely by studying the behavior of their elementary components. The system reflects a synergy-induced emergence of global patterns of neural activity. The study of these patterns requires the simultaneous acquisition of microscopic information at many different sites of brain networks. A spatial high resolution functional MRI scan with parallel imaging based on phased arrays of radiofrequency coils and reconstruction algorithms for acceleration volumes of 22 slices with 128  128 pixel in-plane resolution can be acquired with a temporal resolution of 0.5–1 s. It would undoubtedly be better to have 360 448 microelectrodes in a virtual three-dimensional grid within the brain rather www.sciencedirect.com

Neural-Event-Triggered fMRI of large-scale neural networks Logothetis 215

than tracking the metabolic changes in each voxel. However, this advance is unlikely to be achieved in the near future. The relationship between voxel-intensities and local field potentials is complicated and far from proportionate. Thus, the images require very careful interpretation [3]. The highly multidimensional information provided by fMRI may eventually provide useful constraints in the interpretation of brain states related to simultaneously acquired physiological data. These constraints can also be used in subsequent electrophysiological studies for an informed selection of brain regions in multisite recordings. This review briefly summarizes several characteristic indicators of brain-network selforganization and subsequently describes a first attempt to use so-called Neural-Event-Triggered fMRI (NETfMRI) for gaining insights into the brain states associated with various spontaneous neural events in different brain structures. Lastly, I briefly discuss problems with the interpretation of MR imaging data and the future perspectives of the NET-fMRI methodology.

Self-organized brain activity Over the last decade, the focus of neuroimaging studies has shifted from functional localization to functional integration and brain connectivity [5]. The seminal studies of Raichle and colleagues on the default mode network [6] provided analyses of data from resting-state fMRI. These studies demonstrated the existence of various types of statistical dependencies between separated brain regions that reflect intrinsic functional connectivity. Structural and functional MRI data have been analyzed with increasingly rigorous and refined approaches of statistical mathematics and theoretical physics, including graph-theoretical methods and modeling in the framework of complex dynamic systems [2,7]. Systems neuroscience has a long history of electrophysiological investigations of spontaneous microscopic and mesoscopic neural activity in humans and animals. A characteristic emerging property in many brain networks is oscillatory activity [1,8]. This activity has been studied in great detail in various systems, including the thalamocortical system, basal ganglia, hippocampal formation and the brainstem. Rhythmic activity often reflects the interactions of populations of neurons [9,10]. However, in the case of the thalamocortical networks, it may also be generated by single neurons as a result of interplay between specific intrinsic currents [11]. The oscillations may occur in the same brain state and interact with each other either within the same or across different brain regions. These interactions are commonly hierarchical, and the phase of slower oscillations often modulates the power of the faster oscillations [12,13]. The spontaneous activity patterns largely emerge from the very same networks underlying various behaviors. Therefore, it is not surprising that on-going activity often constrains or determines responses to sensory stimulation [14]. Moreover, www.sciencedirect.com

frequency-band selective modulation [15] and large-scale synchronization [16] of rhythmic activity are believed to play a crucial role in cognitive processing. Brain oscillations are similar to most biological rhythms and are weakly chaotic. Their frequency range extends from infraslow (<0.01 Hz) to ultrafast rhythms over 300 Hz. EEG human studies and intracranial recordings in animals revealed at least ten interactive functionally relevant oscillation classes [8]. The total electrical output of these classes may occasionally develop into rotating [17] or traveling waves [18] depending on the nature and extent of network connectivity. One extensively studied oscillatory pattern is the slow oscillation (SO; in the low end of the traditional delta frequency-range). This pattern is recorded during quiescent sleep and was initially observed in the human EEG. The patterns were subsequently studied in detail with intracellular and extracellular cortical recordings in cats under different anesthesia regimes [19]. The slow oscillation (0.3–1.5 Hz) of neocortical neurons consists of fluctuations in the membrane potential between a depolarized (UP) and a hyperpolarized (DOWN) state. The depolarized state is usually marked by neuronal spiking [20]. This rhythm was initially thought to be entirely generated by the neocortex because it was readily found in the cortical slabs of anesthetized cats as well as following thalamic lesions, destruction of cortical afferents, and in studies of in vitro brain slices [21]. However, subsequent studies demonstrated that a dynamic interplay of the neocortical and thalamic oscillators is necessary for the complete expression of the actual physiological EEG rhythm [22]. A central feature of slow oscillations is the ability to temporally organize a number of other oscillatory events, including cortical K complexes (KC), spindles, and hippocampal sharp wave ripples (SPW-R) [10,21,23]. KC events primarily occur during slow wave sleep and are associated with a population burst-discharge of cortical neurons at all layers, including the corticothalamic pyramidal cells of the infragranular layers [21]. They may also be triggered by sensory stimuli using the same cortical machinery that produces spontaneous K-complexes during the slow oscillation [24]. Notably, KC rarely appear alone and often occur with other rhythms, such as those confined in the delta, alpha and spindle bands of EEG [21,24,25]. Because of their slow frequency range (0.5–0.9 Hz) and their characteristic shape, K complexes contribute to both the SO and delta rhythm and typically signal a shift from the down state to the up state. Spindles are the most frequent partners of the K complex. The synchronous occurrence of KC at the cortical level eventually favors the synchronization of other episodes triggered at distant sites such as the RTN to generate the alpha-like spindle oscillation [26]. Spindles were Current Opinion in Neurobiology 2015, 31:214–222

216 Brain rhythms and dynamic coordination

described in decorticated cats in reticular thalamus and basal ganglia [27]. In their study, Morison and Bassett convincingly showed that spindles can be recorded in the intralaminar thalamic region of a cat 3 days after the bilateral removal of the neocortex and optic nerves after the transection of the brain stem at the inter-collicular level. The best examples of irregularly occurring spontaneous episodes are the large-amplitude hippocampal LFP deviations (sharp-waves) typically associated with superimposed fast-field oscillations (from 100 to over 200 Hz depending on species) known as ‘ripples’ [28,29]. SPWRs are release phenomena. During active waking, the hippocampus (HP) is dominated by the theta rhythm. This rhythm is controlled by a network of cells extending from the brainstem to the medial septum and diagonal band of Broca (MSDB) to the hippocampus and the entorhinal cortex [9]. The MSDB structure modulates subsets of hippocampal interneurons and principal cells in the generation of the local theta rhythm [9]. The ripples mainly occur when the theta rhythm is suppressed [9]. Following theta-reduction, the CA3 network exhibits a highly synchronized population of spiking bursts that produce large LFP deflections in the dendrites of CA1 pyramidal cells of stratum radiatum. The massive depolarization of CA1 induces a short-lived dynamic interaction between these cell populations that yield the ripple-oscillations [28]. SPW-Rs are temporally linked to cortical spindles [30,31] and to SO. Therefore, SPW-Rs are considered to be part of a large-scale system of oscillatory networks. The coupling of these networks is thought to coordinate specific information transfer between neocortical and hippocampal cell assemblies [10,32,33]. An additional extensively studied class of spontaneous events associated with brainwide activity is the phasic electric potentials recorded from the pons, lateral geniculate nucleus (LGN), and occipital cortex. These electric potentials constitute a hallmark of rapid-eye-movement (REM) sleep [34]. Similar to sharp waves, so-called pontinegeniculate-occipital (PGO) waves are short (100 ms) but relatively large (>300 mV) LFP deflections. The first detailed description PGO came from electrophysiological recordings in pons, LGN and the cortex of cats [35,36]. PGO waves are related to several important brain functions including sensorimotor integration, dreaming, development and learning [37]. Their relation to large-scale networks has been recently demonstrated in human fMRI studies reporting REM-related activations in the pontine tegmentum, ventro-posterior thalamus and the primary visual cortex in the absence of any visual input [38].

Spontaneous events and multi-structureactivity (MSA) profiles Decades of experimental work in animals and humans suggest that the aforementioned oscillatory patterns, inCurrent Opinion in Neurobiology 2015, 31:214–222

cluding single or multiple cycle short-lasting episodes, reflect state changes of self-organizing large-scale networks. These patterns regulate cognitive capacities, such as learning, memory encoding and consolidation and memory-guided decision making [39,40,41,42]. It is known that reactivation of neuronal ensembles that were active during awake experiences occur primarily during ripples [43–46]. Thus, the number of ripples increases after learning and the increase appears to predict memory recall both in rats [47,48] and in humans [49]. Conversely, the elimination of ripples by the electrical stimulation of HP during the post-learning SWS interferes with memory consolidation [50,51]. Similarly, disruption of PGO and REM-sleep appears to selectively interfere with the retention of procedural knowledge [52]. Evidently, the exact topologies of such networks and the emerging dynamic activity patterns resulting from their evolution over time are likely to be excellent indicators of specific brain functions and dysfunctions. However, electrophysiological recordings of neural events commonly involve one or more brain regions chosen by means of their anatomical connectivity patterns or by formerly established cooperative interactions of structures in the context of behavior. The global states associated with events remain elusive because of the dearth of methodologies permitting concurrent local recordings and wholebrain activity mapping. In an attempt to study the topology and dynamics of networks associated with events such as ripples, spindles or PGO waves, we have recently developed and applied so-called neural event triggered functional magnetic resonance imaging (NET-fMRI). This method permits concurrent multi-structure and multi-site intracranial recordings and the fMRI of whole-brain activity in 60– 70 regions of interest [ROI] corresponding to various cortical areas and subcortical structures or nuclei. These recordings can be performed in anesthetized and awake monkeys [53]. Figure 1 displays the ripple-triggered whole-brain activity map, the average time course of each ROI and the fraction of its voxels that were significantly up or down modulated within a peri-event time window. The fractional modulated volume and response-dynamics of each ROI, here termed as multi-structure-activity (MSA), offers some insights into the state of the brain networks associated with an event such as SPW-R. Sorting ROIs according to the developmental categorization of brain structures (e.g., metencephalon, mesencephalon, diencephalon, limbic and neo-cortex) revealed a modulation dichotomy between cortical and subcortical areas. SPW-Rs were consistently associated with robust activations of the neocortex and limbic cortex that took place concurrently with robust negative BOLD responses (NBR) in subcortical thalamic, associational (e.g., basal ganglia, cerebellum) and midbrain-brainstem neuromodulatory structures. Such downregulation of activity www.sciencedirect.com

Neural-Event-Triggered fMRI of large-scale neural networks Logothetis 217

Figure 1

(a)

(c) 3 3

120 Frequency in Hz

dlPFC medPFC orbPFC ParPrec Parlntra ParLat TmpPol TmpSTS TmpVis Premotor Motor S2 S1 TmpAu V5 V4 V2V3 pV1 pfV1 fV1 RetroSp PCC ACC dACC alns lns PirFo Ent HP basalAmy Amy NBM BNST MSDB HTh DS VS GP Pul LGN Tha VTA SN PAG InfCol SC MesE LC_CGn Raphe PontReg C-PBL alCb DCbN pflCb LatHemCb IntHemCb Vermis

4

160 SPW-R 140 100

(b)

(d)

2

80 60

1

40 20 –100

0

2

0 100 msec

1

0

–1

–2

–8

–6

–4

–2

t-Score 2 0

4

6

8

–3 –10

Activation Patterns on Standard-Brain

0

10

Time in seconds

–40

0

40

Modulated-ROI Fractions Current Opinion in Neurobiology

Ripple-Triggered MSA. (a) Time–frequency representation of the SPW-R events. (b) Average activation maps from all of the anesthetized-monkey sessions. PBR is observed in the neocortex and limbic cortex, while activity suppression is seen in the diencephalon, mesencephalon and metencephalon. (c) Time courses of each group ROI. Note the sign change in the transition from cortical to subcortical areas and the differences in response onset. (d) Fractions of activated voxels for each group ROI.

in many different structures at the time of SPW-R occurrence may be due to a state-transition favoring optimal hippocampal–cortical communication. The thalamic NBR might indicate an optimal brain-state during which the signal-to-noise ratio of hippocampal– cortical communication is maximized by minimizing neural activity related to sensory processing. The strong downregulation of metabolic activity was also observed in large portions of associational subcortical brain www.sciencedirect.com

structures, such as the basal ganglia (BG), which are related to the acquisition of skills and habits. BGsuppression likely involves a complex interplay of excitatory and inhibitory afferents, such as the medial temporal and pallido-thalamocortical projections to the prefrontal cortex or the subicular projections to the striatum [54]. In addition, activity suppression in BG may be due to diverse neuromodulatory regulation mechanisms, such as the differential dopaminergic (SN/VTA) or amygdala regulation of hippocampal and Current Opinion in Neurobiology 2015, 31:214–222

218 Brain rhythms and dynamic coordination

striatal activities [55,56]. Similarly, the deactivation of the pontine nuclei and cerebellar cortex could result from excitatory cortico-pontine or inhibitory subcortical projections onto specific subsets of GABAergic or glutamatergic pontine neurons, respectively [57]. An occasional concomitant activation (rather than deactivation) of the deep cerebellar nuclei could result from the deactivation of the inhibitory cortico-nuclear projections of the Purkinje neurons [57] and excitatory projections from various subcortical nuclei linked to the declarative memory system. All of these processes could optimize declarative memory consolidation by potentially reducing interfering activity related to other types of memory. Several studies have reported competition between learning and memory systems during the execution of various learning tasks [58,59]. There is evidence of post-learning antagonistic

interactions between memory systems in animal lesion and hippocampal inactivation studies [60]. The NETfMRI findings indicate that competition between memory systems may take place at the time of SPW-R occurrence during off-line states. The strong NBR of pons, LGN and foveal V1 activity during SPW-Rs occur despite the overall positive BOLD responses in the other primary sensory and associational cortices. These responses may be related to suppression of cholinergic neurons in the brainstem, including those inducing the PGO waves during the hippocampal–cortical dialog. REM and PGO are both related to procedural learning and synaptic consolidation. The network states related to synaptic and system consolidation may have synergistic or antagonistic interactions depending on learning/encoding phase, short-term retention of information, or long-term consolidation. The NET-fMRI

Figure 2

CA1 (Pl)

MUA Z-Score

(a) 25

(b) 2

20

1 0

15

–1 10 –2 5

–3

0

–4 –15 –10

–5

0

5

10

15

(c)

BOLD Z-Score

dLGN

3

(d) 2

2

1

1

0

0

–1

–1

–2

–15 –10

–5

0

5

10

15

–15 –10

–5

0

5

10

15

–15 –10

–5

0

5

10

15

Time in sec Current Opinion in Neurobiology

MUA and BOLD Responses in Hippocampus (left panels) and the dorsal Lateral Geniculate Nucleus (right panels). Blue (62%) and dark-red (38%) traces are the means of two subsets of ripple-triggered neural or fMRI activity obtained by clustering the dLGN responses. (a) Ripple-triggered MUA responses in the pyramidal cell layer of the hippocampus. (b) MUA in dLGN. (c) BOLD average BOLD response of the HP-ROI, and (d) BOLD response of the dLGN-ROI. Insets show the BOLD modulation and electrode position for each structure.

Current Opinion in Neurobiology 2015, 31:214–222

www.sciencedirect.com

Neural-Event-Triggered fMRI of large-scale neural networks Logothetis 219

results offer some first global-hints and any serious explanation of the underlying physiological process will require further research.

Interpretation of the up- and down-regulation of metabolic activity The NET-fMRI results raise the question of correct interpretation of fMRI signals, which correlate with changes in metabolism rather than neural activity. MR signal interpretation is more complicated during spontaneous activity. Oscillations and synchrony could, in principle, reduce the metabolic requirements by increasing the synaptic efficiency. However, the oscillations and the gamma rhythm are associated with high-energy demands that require high oxidative energy metabolism, strong mitochondrial performance, and sufficient supply with oxygen and nutrients [61,62]. In general, the variance of positive BOLD responses (PBR) in the cortex is best explained by the dynamics of different band-limited-power signals derived from the LFP in anesthetized or drug-free animals. The interpretation of the LFP itself may also require caution because it reflects both sensory input and neuromodulation-induced changes in the local excitation-inhibition balance [63–65]. However, the relationship of neuronal inhibition and BOLD responses is not straightforward. The inhibition may increase or decrease energy consumption depending on the extent of local interneuron involvement [3]. Certain interneuronal classes directly control blood flow regulation [66], and the hemodynamic mechanisms of NBR are different from those of PBR [67]. Nonetheless, NBR itself may be seen as a reasonable marker of reduction of population-activity because it is often correlated with decreases in multiunit activity (MUA) [3,68]. On-going experiments in our laboratory with combined hippocampal, thalamic, cortical recordings and MR imaging show a robust correlation between up/down modulations and changes in MUA activity (Figure 2). Figure 2a,b shows ripple-triggered multiunit activity in the hippocampus and lateral geniculate nucleus, respectively. The responses in both structures were subdivided into two groups by clustering. Blue and dark-red traces show responses occurring in 62% and 38% of ripple-episodes, respectively. Figure 2c,d shows the corresponding BOLD responses. In both subtypes, there is a robust reduction in thalamic MUA, which fully supports the structure’s NBR. At first glance, this finding appears to be in disagreement with a large number of observations, suggesting certain degrees of synchrony between hippocampal ripples and thalamocortical spindles [10]. However, the slow MUA reduction often occurs with a brief, ripple-synchronized increase in MUA (see dark-red trace in Figure 1b). Therefore, we hypothesize that a plasticity-related activity enhancement of case selective neuronal subwww.sciencedirect.com

populations may occur concurrently with an overall decrease in population spiking. This strategy would increase the SNR of targeted processes. The decreases in MUA were also observed in some other brain regions with characteristic NBR, including the primary visual cortex, associational thalamus and pontine region. Cumulatively, these observations suggest that in most NBR areas revealed with NET-fMRI a reduction of MUA is highly probable and is potentially associated with diverse short-lasting activity changes at the time of SPW-R.

Conclusions and perspectives NET-fMRI is a new method that allows events to be detected, identified and used to examine MSA in the entire brain. Its application may prove highly informative even if the initial results are suggestive rather than decisive regarding the comprehension of network topology and dynamics. It should be noted that neither the activity maps nor the sequences of up and down-regulation indicate a causal relationship between the trigger event and the network activity changes. The state of widespread networks likely depends on a large number of variables. A subset of these variables may be gradually characterized following intensive future experimentation. Interestingly, large-scale observations may improve our understanding of the detailed nature of single neural episodes. SPW-R events are associated with the strong activation of the entire cortex. Is each event correlated with such extensive upregulation? In principle, global activation may simply be the result of averaging hundreds of events that individually have specific cortical targets. Alternatively, events may have different subtypes that each reflect different hippocampal networks and are associated with different MSA and different cognitive functions. SPW-R events have mainly been associated with memory consolidation. However, recent experimental evidence suggests that they are involved in other cognitive capacities, such as memory-driven decision making and working memory [41,69,70]. Such functions may be mediated by different networks and neuronal types. The connectivity patterns and dynamics of these networks could potentially influence the time-frequency characteristics of SPW-R. As mentioned above, SPWRs emerge from the interactions of CA3 and CA1 neurons. The synchronized bursting of CA3 neurons (inducers) yields a large LFP depolarization in the CA1 dendritic fields. The resulting interaction of excitatory–inhibitory cells in CA1 (generators) produces fast oscillation. The effects of inducers and generators are evident in the time– amplitude–frequency characteristics of sharp waves and ripples, respectively. Quantitative analysis of the NET-fMRI data revealed four subtypes of SPW-Rs differing in the relationship of Current Opinion in Neurobiology 2015, 31:214–222

220 Brain rhythms and dynamic coordination

sharp-wave-phase and peak oscillatory-activity [71]. The ripples were synchronized to different phases of the sharp-wave or even appeared without any inducer signs. The variations in the inducer-generator activity were further confirmed in detailed spike-field coherence analyses. Moreover, the idea that such differences may be an expression of different network topologies and dynamics was supported by the examination of the MSA typically measured with NET-fMRI. The MSA showed subtype-specific modulations of responses in the neocortex and in the noradrenergic and serotonergic neuromodulatory centers (locus coeruleus, raphe) that are known to significantly modulate the hippocampal circuits [72,73,74].

Conflict of interest statement

The ‘classical’ ripples aligned to the peak of their SPWs and were associated with enhanced neocortical activation and stronger down regulation of LC and raphe. This result suggests that these episodes occur more often during sleep states [75,76]. In contrast, ripples occurring without noticeable inducer signs show weaker downregulation of neuromodulatory centers. These data indicate their potential involvement in active memoryrelated behaviors [40,41,69] or in decision-making under unexpected uncertainty-conditions [77]. Such functional differences may be due to different network organizations with different inducers or generators of SPW-R [78].

1.

The optimization of NET-fMRI will also requires rigorous classification and pattern recognition methods for an unconditional detection and identification of events. The classic approach relies on scrutiny of predefined frequency bands. However, this approach may be unable to capture relevant phenomena in the absence of prior studies. In an effort to address this issue, we recently introduced (Besserve et al., submitted for publication) a principled LFP data-analysis relying on the extraction of dynamic components with characteristic power spectra using the Itakura-Saito non-negative matrix factorization technique. We then detected transient events with dictionary learning algorithms. The methodology permits the precise detection and identification of known thalamocortical and hippocampal events. Importantly, these findings also reveal previously unknown types of coupling and new MSA patterns. These new patterns include the pre-activation of neocortical areas before delta-events and the lack of pre-activation during spindles. Finally, statistical learning and complex dynamic systems methodologies may eventually enable the characterization of local episodes based on the concurrently recorded MSA profiles at any given time. The multidimensionality of the fMRI data compensates for the poor (1– 0.5 s) sampling rate and offers robust probability density functions reporting the occurrence or sparsity of individual events. Current Opinion in Neurobiology 2015, 31:214–222

Nothing declared.

Acknowledgements I thank Michel Besserve, Juan Felipe Ramirez-Villegas, and Oxana Eschenko for reading the early draft of the manuscript and providing useful suggestions. This research was supported by the Max Planck Society. I apologize to those whose work we have not been able to cite for reasons of space.

References and recommended reading Papers of particular interest, published within the period of review, have been highlighted as:  of special interest  of outstanding interest Freeman WJ: Making sense of brain waves: the most baffling frontier in neuroscience. In Biocomputing. Edited by Parelus P, Principe J, Rajasekaran S. Kluwer; 2008:33-55.

2. 

Deco G, Jirsa VK, Robinson PA, Breakspear M, Friston KJ: The dynamic brain: from spiking neurons to neural masses and cortical fields. PLoS Comput Biol 2008, 4. A good review discussing the computational models at different spacetime scales. The authors describe the benefits of modeling at microscopic, mesoscopic and macroscopic levels and discuss the value of the mean-field and related formulations of dynamics with essential and complementary roles as forward models that can be inverted given empirical data.

3.

Logothetis NK: What we can do and what we cannot do with fMRI. Nature 2008, 453:869-878.

4.

Buzsaki G: Large-scale recording of neuronal ensembles. Nat Neurosci 2004, 7:446-451.

5.

Friston KJ: Modalities, modes, and models in functional neuroimaging. Science 2009, 326:399-403.

6.

Raichle ME, MacLeod AM, Snyder AZ, Powers WJ, Gusnard DA, Shulman GL: A default mode of brain function. Proc Natl Acad Sci U S A 2001, 98:676-682.

7.

Bullmore E, Sporns O: Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci 2009, 10:186-198.

8.

Buzsaki G, Logothetis N, Singer W: Scaling brain size, keeping timing: evolutionary preservation of brain rhythms. Neuron 2013, 80:751-764.

9.

Buzsaki G: Theta oscillations in the hippocampus. Neuron 2002, 33:325-340.

10. Sirota A, Buzsa´ki G: Interaction between neocortical and hippocampal networks via slow oscillations. Thalamus Relat Syst 2005, 3:245-259. 11. Steriade M, Llinas RR: The functional states of the thalamus and the associated neuronal interplay. Physiol Rev 1988, 68: 649-742. 12. Bragin A, Jando G, Nadasdy Z, Hetke J, Wise K, Buzsaki G: Gamma (40–100 Hz) oscillation in the hippocampus of the behaving rat. J Neurosci 1995, 15:47-60. 13. Leopold DA, Murayama Y, Logothetis NK: Very slow activity fluctuations in monkey visual cortex: implications for functional brain imaging. Cereb Cortex 2003, 13:422-433. 14. Arieli A, Sterkin A, Grinvald A, Aertsen A: Dynamics of ongoing activity: explanation of the large variability in evoked cortical responses. Science 1996, 273:1868-1871. 15. Singer W, Gray CM: Visual feature integration and the temporal correlation hypothesis. Annu Rev Neurosci 1995, 18:555-586. 16. Fries P: A mechanism for cognitive dynamics: neuronal communication through neuronal coherence. Trends Cogn Sci 2005, 9:474-480. www.sciencedirect.com

Neural-Event-Triggered fMRI of large-scale neural networks Logothetis 221

17. Ermentrout GB, Kleinfeld D: Traveling electrical waves in cortex: insights from phase dynamics and speculation on a computational role. Neuron 2001, 29:33-44. 18. Massimini M, Huber R, Ferrarelli F, Tononi G: Sleep slow oscillations as traveling waves: origins and pathways of propagation in humans. Sleep 2004, 27:32. 19. Steriade M, Nunez A, Amzica F: A novel slow (<1 Hz) oscillation of neocortical neurons in vivo: depolarizing and hyperpolarizing components. J Neurosci 1993, 13:3252-3265. 20. McCormick DA, Bal T: Sleep and arousal — thalamocortical mechanisms. Annu Rev Neurosci 1997, 20:185-215. 21. Steriade M, Amzica F: Slow sleep oscillation, rhythmic Kcomplexes, and their paroxysmal developments. J Sleep Res 1998, 7(Suppl 1):30-35. 22. Crunelli V, Hughes SW: The slow (<1 Hz) rhythm of non-REM sleep: a dialogue between three cardinal oscillators. Nat Neurosci 2010, 13:10-18. 23. Molle M, Yeshenko O, Marshall L, Sara SJ, Born J: Hippocampal sharp wave-ripples linked to slow oscillations in rat slow-wave sleep. J Neurophysiol 2006, 96:62-70. 24. Amzica F, Steriade M: Cellular substrates and laminar profile of sleep K-complex. Neuroscience 1998, 82:671-686. 25. Halasz P: K-complex, a reactive EEG graphoelement of NREM sleep: an old chap in a new garment. Sleep Med Rev 2005, 9:391-412.

40. Singer AC, Carr MF, Karlsson MP, Frank LM: Hippocampal SWR  activity predicts correct decisions during the initial learning of an alternation task. Neuron 2013, 77:1163-1173. The authors studied the behavior of rats involved in a spatial alternation task. They demonstrated that the coordination of neural activity during SPW-Rs, is grater in SPW-Rs preceding correct, as compared to incorrect trials. Such findings suggest that reactivation during awake SWRs contributes to the evaluation of possible choices during memory-guided decision making. 41. Karlsson MP, Frank LM: Awake replay of remote experiences in the hippocampus. Nat Neurosci 2009, 12:913-918. 42. Bourdiec AS-L, Muto V, Mascetti L, Foret A, Matarazzo L, Kusse´ C, Maquet P: Contribution of sleep to memory consolidation. Future Neurol 2010, 5:325-338. 43. Wilson MA, McNaughton BL: Reactivation of hippocampal ensemble memories during sleep. Science 1994, 265:676-679. 44. Kudrimoti HS, Barnes CA, McNaughton BL: Reactivation of hippocampal cell assemblies: effects of behavioral state, experience, and EEG dynamics. J Neurosci 1999, 19:4090-4101. 45. Nadasdy Z, Hirase H, Czurko A, Csicsvari J, Buzsaki G: Replay and time compression of recurring spike sequences in the hippocampus. J Neurosci 1999, 19:9497-9507. 46. Skaggs WE, McNaughton BL: Replay of neuronal firing sequences in rat hippocampus during sleep following spatial experience. Science 1996, 271:1870-1873.

26. Contreras D, Destexhe A, Sejnowski TJ, Steriade M: Spatiotemporal patterns of spindle oscillations in cortex and thalamus. J Neurosci 1997, 17:1179-1196.

47. Eschenko O, Ramadan W, Molle M, Born J, Sara SJ: Sustained increase in hippocampal sharp-wave ripple activity during slow-wave sleep after learning. Learn Memory 2008, 15: 222-228.

27. Morison RS, Bassett DL: Electrical activity of the thalamus and basal ganglia in decorticate cats. J Neurophysiol 1945, 8:309314.

48. O‘Neill J, Senior TJ, Allen K, Huxter JR, Csicsvari J: Reactivation of experience-dependent cell assembly patterns in the hippocampus. Nat Neurosci 2008, 11:209-215.

28. Buzsaki G, Horvath Z, Urioste R, Hetke J, Wise K: High-frequency network oscillation in the hippocampus. Science 1992, 256:1025-1027.

49. Axmacher N, Elger CE, Fell J: Ripples in the medial temporal  lobe are relevant for human memory consolidation. Brain 2008, 131:1806-1817. The authors report the first recordings of ripples in humans during a cognitive task. They recorded intracranial LFP with macroelectrodes from the hippocampus and rhinal cortex epilepsy patients during an afternoon nap. The data show that ripples can be detected both in the hippocampus and in the rhinal cortex. The authors demonstrate that ripples are indeed related to behavioral performance in humans.

29. O‘Keefe J, Nadel L: The Hippocampus as a Cognitive Map. Oxford University Press: Oxford; 1978, . 30. Siapas AG, Wilson MA: Coordinated interactions between hippocampal ripples and cortical spindles during slow-wave sleep. Neuron 1998, 21:1123-1128. 31. Axmacher N, Mormann F, Fernandez G, Elger CE, Fell J: Memory formation by neuronal synchronization. Brain Res Rev 2006, 52:170-182.

50. Girardeau G, Benchenane K, Wiener SI, Buzsaki G, Zugaro MB: Selective suppression of hippocampal ripples impairs spatial memory. Nat Neurosci 2009, 12:1222-1223.

32. Wierzynski CM, Lubenov EV, Gu M, Siapas AG: State-dependent spike-timing relationships between hippocampal and prefrontal circuits during sleep. Neuron 2009, 61:587-596.

51. Ego-Stengel V, Wilson MA: Disruption of ripple-associated hippocampal activity during rest impairs spatial learning in the rat. Hippocampus 2010, 20:1-10.

33. Isomura Y, Sirota A, Ozen S, Montgomery S, Mizuseki K, Henze DA, Buzsaki G: Integration and segregation of activity in entorhinal–hippocampal subregions by neocortical slow oscillations. Neuron 2006, 52:871-882.

52. Karni A, Meyer G, Rey-Hipolito C, Jezzard P, Adams MM, Turner R, Ungerleider LG: The acquisition of skilled motor performance: fast and slow experience-driven changes in primary motor cortex. Proc Natl Acad Sci U S A 1998, 95:861-868.

34. Datta S: Cellular and chemical neuroscience of mammalian sleep. Sleep Med 2010, 11:431-440.

53. Logothetis NK, Eschenko O, Murayama Y, Augath M, Steudel T, Evrard HC, Besserve M, Oeltermann A: Hippocampal–cortical interaction during periods of subcortical silence. Nature 2012, 491:547-553.

35. Brooks DC, Bizzi E: Brainstem electrical activity during sleep. Anat Record 1963, 145:211. 36. Bizzi E, Brooks DC: Pontine reticular formation — relation to lateral geniculate nucleus during deep sleep. Science 1963, 141:270. 37. Datta S: Avoidance task training potentiates phasic pontinewave density in the rat: a mechanism for sleep-dependent plasticity. J Neurosci 2000, 20:8607-8613. 38. Miyauchi S, Misaki M, Kan S, Fukunaga T, Koike T: Human brain activity time-locked to rapid eye movements during REM sleep. Exp Brain Res 2009, 192:657-667. 39. Kirov R, Weiss C, Siebner HR, Born J, Marshall L: Slow oscillation electrical brain stimulation during waking promotes EEG theta activity and memory encoding. Proc Natl Acad Sci U S A 2009, 106:15460-15465. www.sciencedirect.com

54. Graybiel AM: Habits, rituals, and the evaluative brain. Annu Rev Neurosci 2008, 31:359-387. 55. McGaugh JL: The amygdala modulates the consolidation of memories of emotionally arousing experiences. Annu Rev Neurosci 2004, 27:1-28. 56. Packard MG, Teather LA: Amygdala modulation of multiple memory systems: hippocampus and caudate-putamen. Neurobiol Learn Memory 1998, 69:163-203. 57. Brodal P, Bjaalie JG: Salient anatomic features of the corticoponto-cerebellar pathway. Prog Brain Res 1997, 114:227-249. 58. Poldrack RA, Rodriguez P: How do memory systems interact? Evidence from human classification learning. Neurobiol Learn Memory 2004, 82:324-332. Current Opinion in Neurobiology 2015, 31:214–222

222 Brain rhythms and dynamic coordination

59. Albouy G, King BR, Maquet P, Doyon J: Hippocampus and striatum: dynamics and interaction during acquisition and sleep-related motor sequence memory consolidation. Hippocampus 2013, 23:985-1004. 60. Schroeder JP, Wingard JC, Packard MG: Post-training reversible inactivation of hippocampus reveals interference between memory systems. Hippocampus 2002, 12:280-284. 61. Kann O: The energy demand of fast neuronal network oscillations: insights from brain slice preparations. Front Pharmacol 2011, 2. 62. Lord LD, Expert P, Huckins JF, Turkheimer FE: Cerebral energy metabolism and the brain’s functional network architecture: an integrative review. J Cereb Blood Flow Metab 2013, 33: 1347-1354. 63. Lippert MT, Steudel T, Ohl F, Logothetis NK, Kayser C: Coupling of neural activity and fMRI-BOLD in the motion area MT. Magn Reson Imaging 2010, 28:1087-1094. 64. Goense JBM, Logothetis NK: Neurophysiology of the BOLD fMRI signal in awake monkeys. Curr Biol 2008, 18:631-640. 65. Logothetis NK, Pauls JM, Augath MA, Trinath T, Oeltermann A: Neurophysiological investigation of the basis of the fMRI signal. Nature 2001, 412:150-157. 66. Buzsaki G, Kaila K, Raichle M: Inhibition and brain work. Neuron 2007, 56:771-783. 67. Goense J, Merkle H, Logothetis NK: High-resolution fMRI reveals laminar differences in neurovascular coupling between positive and negative BOLD responses. Neuron 2012, 76:629-639. 68. Shmuel A, Augath MA, Oeltermann A, Logothetis NK: Negative functional MRI response correlates with decreases in neuronal activity in monkey visual area V1. Nat Neurosci 2006, 9:569-577. 69. Dupret D, O‘Neill J, Pleydell-Bouverie B, Csicsvari J: The reorganization and reactivation of hippocampal maps predict spatial memory performance. Nat Neurosci 2010, 13:995-1002. 70. Yu JY, Frank LM: Hippocampal–cortical interaction in decision making. Neurobiol Learn Memory 2014. 71. Ramirez-Villegas JF, Logothetis N, Besserve M: Diversity of sharp-wave ripple complexes reveals activation of differentiated brain-wide functional networks. Cybernetics MB; 2014; submitted.

Current Opinion in Neurobiology 2015, 31:214–222

72. Freund TF, Gulyas AI, Acsady L, Gorcs T, Toth K: Serotonergic control of the hippocampus via local inhibitory interneurons. Proc Natl Acad Sci U S A 1990, 87:8501-8505. 73. Ul Haq R, Liotta A, Kovacs R, Rosler A, Jarosch MJ, Heinemann U,  Behrens CJ: Adrenergic modulation of sharp wave-ripple activity in rat hippocampal slices. Hippocampus 2012, 22: 516-533. Norepinephrine (NE) is released in the hippocampus from the afferents of the locus coeruleus, in particular during novelty detection. The authors activated CA1/CA3 with orthodromic and antidromic stimulation of the stratum radiatum and Schaffer collaterals, respectively, and examined the modulatory effects (through adrenoreceptor-agonists) of NE on stimulus-induced SPW-Rs in the CA3 and CA1 of rat hippocampal slices. 74. Varga V, Losonczy A, Zemelman BV, Borhegyi Z, Nyiri G, Domonkos A, Hangya B, Holderith N, Magee JC, Freund TF: Fast synaptic subcortical control of hippocampal circuits. Science 2009, 326:449-453. 75. Mcginty DJ, Harper RM: Dorsal raphe neurons — depression of firing during sleep in cats. Brain Res 1976, 101:569-575. 76. Takahashi K, Kayama Y, Lin JS, Sakai K: Locus coeruleus neuronal activity during the sleep-waking cycle in mice. Neuroscience 2010, 169:1115-1126. 77. Payzan-LeNestour E, Dunne S, Bossaerts P, O‘Doherty JP: The  neural representation of unexpected uncertainty during valuebased decision making. Neuron 2013, 79:191-201. A human fMRI study examining brain activity related to unexpected uncertainty in a state reflecting conditions of the violation of top-down predictions that are expected to be correct. The authors reported the activation of multiple cortical areas but also of the noradrenergic brainstem nucleus locus coeruleus. Importantly, the prefrontal cortical regions known to project directly to the locus coeruleus were not activated during the state of unexpected uncertainty. This result suggests that the modulatory effects during this state might be attributed to LC. 78. Schomburg EW, Ferna´ndez-Ruiz A, Mizuseki K, Bere´nyi A,  Anastassiou CA, Koch C, Buzsa´ki G: Theta phase segregation of input-specific gamma patterns in entorhinal–hippocampal networks. Neuron 2014, 84:470-485. This study is an investigation of interregional coordination of gamma oscillations using high-density electrophysiological recordings in the rat hippocampus and entorhinal cortex. The authors report that 30–80 Hz gamma dominated CA1 local field potentials (LFPs) on the descending phase of CA1 theta waves during navigation with 60–120 Hz gamma at the theta peak. The authors suggest that Gamma synchronization is a mechanism for interregional communication and rapidly loses efficacy at higher frequencies.

www.sciencedirect.com