Resting-state Functional Connectivity and Spontaneous Brain Co-activation Xiao Liu, Ph.D.

Transcription

1 Resting-state Functional Connectivity and Spontaneous Brain Co-activation Xiao Liu, Ph.D. Assistant Professor Department of Biomedical Engineering Institute for CyberScience

2 Functional Magnetic Resonance Imaging (fmri) Signal Rest T 2* /T 2 Decay Activated Neural Activity TE CBF CBV CMRO 2 T 2* /T 2 Weighted MR Images Activated Rest Introduction Background Methods Results Discussions Difference Activation Map Ogawa S et al. PNAS, 1990&1992 Bandettini PA, et al. MRM 1992 Kwong KK et al. PNAS 1992

3 Resting-state fmri Connectivity: The First Study Session #1 Finger Tapping Reference Time Course Paradigm HRF (External Modulation) Functional Map (Task-evoked Response) Session #2 Resting State Reference Time Course Spontaneous fmri Signal (Endogenous Fluctuation) Correlation Map (Spontaneous Correlation) Introduction Background Methods Results Discussions Biswal et al. MRM (1995)

4 Other Networks Auditory Cordes D et al. AJNR (2000) Language Visual More RsfMRI Correlation Correlational Patterns Mantini et al. PNAS (2007) Hampson et al. HBM (2002) Functional Connectivity Resting-state Networks Default Mode Attention Fox MD et al. PNAS (2006) Memory Fransson P et al. HBM (2005) Vincent JL et al. J Neurophysiol (2006) Introduction Background Methods Results Discussions

5 Significance 1. Basic Neuroscience 2. Clinical Application Raichle Science (2006) Whitefield-Gabrieli et al. PNAS (2009) Brain s Dark Energy Functional Connectivity Network Organization Characterize Brain States Schizophrenia Parkinson s Disease Alzheimer s Disease Autism Introduction Background Methods Results Discussions

6 Trend What causes rsfmri signal correlations & their network pattern? Introduction Background Methods Results Discussions

7 Agenda Part I Q: What causes network-specific correlations and their temporal dynamics? resting-state networks and co-activation patterns a temporal decomposition method Part II Q: What causes global rsfmri signal, non-specific correlation, and their behavioral relevance? global rsfmri signal, non-specific coactivation, and their relation to vigilance underlying neuronal events Introduction Background Methods Results Discussions

8 Non-stationary Brain Connectivity Functional connectivity = Temporal fmri signal correlation (Averaged relationship) B A C Average B A C t 1 B A C t 2 B A C t 3 B A C t m B A C t n Co-activation Pattern Background Methods Results Discussions

9 RsfMRI Signal Correlations Vary over Time Extra dimension of information? Chang et al. NeuroImage (2010) Stationary correlations give similar information as structural connectivity Structural Connectivity Functional Connectivity from HCP Co-activation Pattern Background Methods Results Discussions

10 Limitations Shorter time window Larger temporal variations solely by signal-to-noise ratio reduction? non-neuronal events OR brain connectivity? Pairwise correlation high-order correlation: co-activations of multiple regions Neuron 1 Neuron 2 Neuron 3 Case #1 Case #2 Alternative way to understand nonstationary functional connectivity? particularly important for neuroimage data the number of voxels (N) >> the number of time points (T) N. (N-2)/2 >> N. T (pairwise correlations) (actual measurements) Co-activation Pattern Background Methods Results Discussions

11 Replicate RSN Patterns with A Few Time Points BOLD [S.D.] fmri signal from the posterior cingulate cortex (PCC) seed PCC Time [sec] Threshold 0.5 r -0.5 temporal mean CorrMap from 123 volumes Average of 18 volumes Single volume Group Level r = CorrMap from 100% data Average of 15% data Co-activation Pattern Background Methods Results Discussions Liu X. and Duyn J.H., PNAS (2013)

12 Distinct Patterns at Different Time 2 S.D. BOLD PCC Frame 2 Frame 1 * mpfc * mpfc 50 sec Time Frame 2 Frame 1 PCC Co-activation Pattern Background Methods Results Discussions Liu X. and Duyn J.H., PNAS (2013)

13 Classifying fmri Volumes According to Their Spatial Patterns CAP i +1 CAP i CAP = Co-Activation Pattern Co-activation Pattern Background Methods Results Discussions Liu X. and Duyn J.H., PNAS (2013)

14 Temporal Decomposition of Default Mode Network Overall Average of 15% data PCC-CAP 3 PCC-CAP 4 PCC-CAP 1 PCC-CAP 2 PCC-CAP 7 PCC-CAP 8 PCC-CAP 5 PCC-CAP BOLD [S.D.] -0.8 caudate nucleus = + Hippocampus parahippocampal gyrus Co-activation Pattern Background Methods Results Discussions Liu X. and Duyn J.H., PNAS (2013)

15 Temporal Decomposition of Default Mode Network Z > 6 MFG PCC-CAP 1 PCC-CAP 2 PCC-CAP 1 PCC-CAP 3 SFG PCC-CAP 1 PCC-CAP 4 Co-activation Pattern Background Methods Results Discussions Liu X. and Duyn J.H., PNAS (2013)

16 Not Limited By Seeding CAP i +1 CAP i Co-activation Pattern Background Methods Results Discussions

17 Not Limited By Seeding CAP i +1 CAP i 30 CAPs Co-activation Pattern Background Methods Results Discussions Liu X. et al. Front Syst Neurosci (2013)

18 Two Anti-Correlated Networks? Or Multiple versus One? CAP 2 CAP 3 FEF IPS CAP 4 CAP 9 SMA CAP 6 Fox et al. PNAS (2005) Z 6 20 PCG Co-activation Pattern Background Methods Results Discussions Liu X. et al. Front Syst Neurosci (2013)

19 Thalamocortical Co-Activations Z Visual CAP 2 CAP 3 Pulvinar Pulvinar [18, -26, 10] [18, -26, 10] CAP 26 LGN [24, -22, -4] Sensorimotor CAP 19 CAP 23 VPL [-14, -22, 4] VPM [-14, -16, 4] Co-activation Pattern Background Methods Results Discussions Liu X. et al. Front Syst Neurosci (2013)

20 Thalamocortical Co-Activations (Negative) CAP 19 IC 6 CorrMap Thalamic Reticular Nucleus? Z r Z Co-activation Pattern Background Methods Results Discussions Liu X. et al. Front Syst Neurosci (2013)

21 Functional Relevance of the CAPs: An Example CAP 16 Motor SMA Medial IPS Medial IPS plays a critical role in visuomotor coordinate transformation Grefkes et al. J Anat. (2005) Co-activation Pattern Background Methods Results Discussions

22 Occurrence Rate versus Correlation: A Simulation Case # Correlation Occurrence % +67% Co-activation Pattern Background Methods Results Discussions

23 Occurrence Rates of CAPs: Males versus Females Occurrence Rate ** p < 0.01, Bonferroni corrected CAP # CAP 23 Co-activation Pattern Background Methods Results Discussions Liu X. et al. Front Syst Neurosci (2013)

24 Further Exploration in This Area We are trying to develop computational methods/models to quantify and study temporal dynamics of brain networks. Hidden Markov Chain CAP 1 CAP 2 CAP 3 CAP 4 CAP 5 Temporal Graph Theory Spontaneous brain activity measured by fmri Co-activation Pattern Background Methods Results Discussions

25 Summary Resting-state network patterns result from co-activation patterns (CAPs) at discrete time points. CAPs explain non-stationary functional connectivity whether (or how many) critical points with clear patterns are included (change of signal-to-noise ratio), and what types of patterns (dynamics of neuronal activity) are include in the time window A new data-driven approach few assumptions and data transformations more specific information regarding brain co-activations robust against motion artifacts Occurrence rate of CAPs differentiating conditions or populations Co-activation Pattern Background Methods Results Discussions

26 Agenda Part I Q: What causes network-specific correlations and their temporal dynamics? resting-state networks and co-activation patterns a temporal decomposition method Part II Q: What causes global rsfmri signal, non-specific correlation, and their behavioral relevance? global rsfmri signal, non-specific coactivation, and their relation to vigilance underlying neuronal events Introduction Background Methods Results Discussions

27 Global RsfMRI Signal & Non-specific Correlations Global Signal Background Methods Results Discussions Fox, M et al., J Neurophysiol, 2009

28 Anti-correlation and Global Signal Regression PCC-CAP 2 PCC-CAP 8 PCC-CAP 5 CorrMpa Artifact With GSR Fox et al. PNAS (2005) Map Statistics Distribution Murphy et al. NeuroImage (2009) Without GSR 0. 4 CAP Decomp Correlation CAP Decomp The whole-brain co-activation occurs preferentially with sensory systems! BOLD [S.D.] Global Signal Background Methods Results Discussions Liu et al. ISMRM 2013 #2251

29 Neural Component in Global rsfmri signal Global rsfmri signal is inversely correlated with vigilance light sleep >> awake (e.g., Horovitz SG et al., HBM, 2008) eyes-closed > eyes-open caffeine hypnotic drugs (e.g., Jao T et al., NeuroImage, 2013) (Wong CW et al., NeuroImage, 2013) (e.g., Litaca CS et al., NeuroImage, 2013) Gamma-band LFP power recorded locally is correlated with global fmri Scho lvinck ML et al., PNAS, 2010 Global Signal Background Methods Results Discussions

30 Why Important? Non-specific correlations confounded with network-specific correlations Local connectivity change due to network dysfunction? 1. What is the neural correlate of the global fmri signal? An example: functional connectivity in schizophrenia OR Global changes due to vigilance change? Whitefield-Gabrieli et al. PNAS (2009) 2. Why is it sensitive to vigilance change? Yang et al. PNAS (2014) Global Signal Background Methods Results Discussions

31 Global Signal in ECoG Data Eyes-closed Rest Ketamine/Medetomidine Propofol Sleep Electrocorticography (ECoG) Liu, X. et al. Cere. Cor Global Signal Background Methods Results Discussions

32 Global Signal in ECoG Data Eyes-closed Rest Ketamine/Medetomidine Propofol Sleep Electrocorticography (ECoG) global averaged spectrogram removed Z sec 1 sec g b a q d g-blp b-blp a-blp q-blp d-blp Broadband Global Signal Background Methods Results Discussions or 53 Cross-electrode Correlation Event-like process? 2. Spectral characteristics? Clustering 3. Relation to vigilance? Average Average 0.5 r -0.5 Liu, X. et al. Cere. Cor. 2014

33 A Sequential Spectral Transition (SST) Pattern Frequency [Hz] Δ Power [db] mv 10 sec 2 mv 1 sec Channel-Averaged Spectrogram 10 sec High (42 87 Hz) 1 sec Global Signal Background Methods Results Discussions Middle (9 21 Hz) Low (<4 Hz) Liu, X. et al. NeuroImage, 2015

34 Averaged Pattern of Sequential Spectral Transition (SST) Correlation Frequency [Hz] Δ Power [db] Δ Power [db] Δ Power [db] Δ Power [db] Correlation Sleep (Monkey C) sec -1.4 sec Time [s] Eyes-closed Middle vs. Low Middle vs. High Global Signal Background Methods Results Discussions Sleep (Monkey G) sec -1.2 sec Eyes-open Middle vs. Low Sleep EC EO Low-Frequency SSI Middle vs. High Sleep Time [s] EC EO Liu, X. et al. NeuroImage, 2015

35 SST-like Structure at the Induction of Propofol Anesthesia Experiment 1 Experiment 2 Global Signal Background Methods Results Discussions Liu, X. et al. NeuroImage, 2015

36 Sequential Changes Wake-promoting System Sleep-promoting System Saper CB. et al. Nature Mice Takahashi K. et al. Neuroscience Global Signal Background Methods Results Discussions

37 Spatial Pattern and Feedback Hypothesis Δ Power [db] Correlation Sleep Eyes Closed High Frequency A release from such a control 1 2 [-5, 0] sec Middle Frequency A loss of feedback inhibitory control Middle 1 leading in time Inhibitory feedback control of the mid-frequency activity??? Klimesch W et al. Brain Res Rev 2007 Jensen O and Mazaheri A. Front Hum Neurosci 2010 Time [s] Middle 1 vs High 1 Middle 1 vs High 2 Middle 2 vs High 1 Middle 2 vs High 2 Liu, X. et al. NeuroImage, 2015 Global Signal Background Methods Results Discussions

38 Spatial Pattern of Global Co-activation: ECoG versus fmri Sleep Eyes Closed Global Signal Background Methods Results Discussions

39 Concurrent fmri-electrophysiology data Frequency [Hz] Epochs Local Field Potential (LFP) Time [s] Time [s] MION-CBV (sign flipped) 2 5 Normalized CBV [S.D.] Data from Scho lvinck ML et al., PNAS, 2010 Global Signal Background Methods Results Discussions

40 Concurrent fmri-electrophysiology data Time-frequency LFP pattern Global fmri average Global Signal Background Methods Results Discussions

41 Spatial Pattern of Global Co-activation: ECoG versus fmri Sleep Eyes Closed Human Connectome Project Data: 2mm isotropic, TR = 0.72 sec Global Signal averaging large peaks Anterior commissure Sensorimotor Nucleus basalis of basal forebrain Optic tract Auditory Anterior commissure Visual Anterior commissure Optic tract Global Signal Background Methods Results Discussions Optic tract Haines DE. Neuroanatomy 5 th Edition

42 Spatial Pattern of Global Co-activation: ECoG versus fmri a Z Dorsal Midline Thalamus b Substantia Nigra (SN) Global Signal Background Methods Results Discussions

43 Summary Global signal (spatially non-specific correlation) global average of ECoG power is characterized by a recurrent sequential spectral transition (SST) pattern. SST is strong during the light sleep state, weak but still present during eyes-closed condition, but largely absent during eyes-open condition, similar to the state-dependency of global fmri signal SST induces large, nearly whole-brain changes in fmri signals sensory regions show largest fmri changes during global coactivations, consistent with the spatial pattern of high-frequency gamma power changes at SSTs. nucleus basalis of the basal forebrain, which is the wake-promoting center of the brain, show de-activation at the whole-brain coactivation. the global signal may directly or indirectly affect resting-state connectivity quantification, and needs to be taken care properly. Global Signal Background Methods Results Discussions

44 Acknowledgement NIH/NINDS/LFMI Jeff Duyn Alan Koretsky Afonso Silva Catie Chang Peter van Gelderen Jacco de Zwart Duan Qi Dante Picchioni Hendrik Mandelkow Natalia Gudino Erika Raven Roger Jiang Collaborators David Leopold (NIMH) Toru Yanagawa (RIKEN) Naotaka Fujii (RIKEN) Acknowledgement Background Methods Results Discussions