ДОЛГОВРЕМЕННЫЕ КОРРЕЛЯЦИИ АМПЛИТУДЫ АЛЬФА КОЛЕБАНИЙ ПРЕДСКАЗЫВАЮТ ЭФФЕКТИВНОСТЬ РАБОЧЕЙ ПАМЯТИ

Научная статья
DOI:
https://doi.org/10.23670/IRJ.2019.90.12.033
Выпуск: № 12 (90), 2019
Опубликована:
2019/12/18
PDF

ДОЛГОВРЕМЕННЫЕ КОРРЕЛЯЦИИ АМПЛИТУДЫ АЛЬФА КОЛЕБАНИЙ ПРЕДСКАЗЫВАЮТ ЭФФЕКТИВНОСТЬ РАБОЧЕЙ ПАМЯТИ

Научная статья

Беляева В.1, Ермолова М.2, Новиков Н.3, Гуткин Б.4, Феурра М.5, Феделе Т.6, *

6 ORCID: 0000-0001-7574-8062;

1 Лаборатория неврологических решений, Департамент медицинских наук и технологий, Швейцарский федеральный технологический институт, Цюрих, Швейцария;

2, 3, 4, 5, 6 Институт Когнитивных Нейронаук, Центр Нейроэкономики и Когнитивных Исследований,

Научно-Исследовательский Университет Высшая Школа Экономики, Москва, Россия;

4 Исследовательский центр PSL, Школа высшего образования, Париж, Франция

* Корреспондирующий автор (fedele.tm[at]gmail.com)

Аннотация

Способность успешно выполнять задачу поддерживается путем подавления лишней информации. Предыдущие исследования показали, что альфа-колебания (8-15 Гц) играют ключевую роль в защите важной информации от отвлекающих факторов во время выполнения задачи на рабочую память (РП). В данном исследовании мы изучили, как временная динамика альфа-ритма в состоянии покоя предсказывает успешность РП в модифицированной парадигме Штернберга с задержкой сравнения.

Мы предполагаем, что поведенческие характеристики могут быть предсказаны динамикой колебательной активности в покое. Для этого мы использовали долговременную корреляцию (LRTC), новый и надежный подход к количественной оценке баланса между возбуждением и торможением в нейронных сетях.

Активность коры мозга испытуемых в покое была записана перед выполнением задания на РП с помощью электроэнцефалографии (ЭЭГ). LRTC измерялись в покое в альфа-диапазоне, чтобы предсказать индивидуальные различия в успешности РП.

Мы обнаружили, что LRTC лобной доли в состоянии покоя предсказывают индивидуальный уровень точности в задаче на РП. Важно отметить, что в спектральной мощности ЭЭГ в состоянии покоя никакого эффекта не наблюдалось.

В то время как лобная кора, как известно, участвует в подавлении отвлекающих факторов, мы показываем в данной статье, что способность переключаться между вниманием и сенсорным подавлением может быть измерена уже во временной нейронной динамике как характеристика отдельного человека.

Ключевые слова: рабочая память, ЭЭГ, альфа колебания, долговременные корреляции.

LONG-RANGE TEMPORAL CORRELATIONS IN THE AMPLITUDE OF ALPHA OSCILLATIONS PREDICT WORKING MEMORY PERFORMANCE

Research article

Beliaeva V.1, Ermolova M.2, Novikov N.3, Gutkin B.4, Feurra M.5, Fedele T.6, *

6 ORCID: 0000-0001-7574-8062;

1 Decision Neuroscience Laboratory, Department of Health Sciences and Technology, Eidgenössische Technische Hochschule (ETH), Zurich, Switzerland;

2, 3, 4, 5, 6 Institute of Cognitive Neuroscience, Centre for Cognition and Decision Making, National Research University Higher School of Economics, Moscow, Russia;

4 Research University Ecole Normale Superieure, Paris, France

* Corresponding author (fedele.tm[at]gmail.com)

Abstract

The ability to successfully perform a task is supported by inhibition of irrelevant information. Evidence showed that alpha oscillations (8-15 Hz) play a key role in protecting relevant information from the appearance of distractors during a working memory (WM) task. Here, we investigated how the temporal dynamics of alpha rhythm at rest predict the WM performance in a modified delay-match Sternberg paradigm.

We hypothesize that the behavioral performance can be predicted by the dynamics of oscillatory activity at rest. To this end we used Long-Range Temporal Correlation (LRTC), a novel and reliable approach to quantify the balance between excitation and inhibition in neuronal networks.

Subjects were recorded with scalp EEG during rest before undergoing the WM task. LRTC were measured at rest in the alpha band in order to predict the inter-individual differences in the WM performance.

We found that LRTC at rest over frontal cortex were predictive of the level of accuracy in the WM task across individuals. Importantly, no effect was visible when considering spectral EEG power at rest.

While frontal cortex is known to be involved in suppressing distractor interference, we show here that the ability to switch between attention and sensory suppression can be measured already in the neural temporal dynamic as a feature of the single individual.

Keywords: working memory, EEG, alpha oscillations, long range temporal correlations.

Introduction

The ability to efficiently filter relevant from irrelevant information supports the individual working memory (WM) performance. Indeed, in our everyday life it is crucial to have mechanisms that can inhibit distracting information. Inhibiting information that is irrelevant is a core executive function of the working memory system [12]. Behavioral and electrophysiological studies showed that suppressing distractors during a WM task [1], [20] and the dropping unnecessary information from WM [14], [20] tend to improve the retrieval of information that is relevant to the task. This suggests that mechanisms involving cognitive control have an important role in protecting relevant versus irrelevant information. Evidence showed that suppression of irrelevant information has a neural correlate which is represented by alpha oscillatory activity in parieto-frontal network. The increase of alpha activity protects against anticipated distracters [1]. Indeed, oscillations at different frequencies reflect different behaviors and cognitive processes that underly different states of neural network activity as typically measured with electroencephalography (EEG) and magnetoencephalography (MEG) [4], [11]. We should further note that spontaneous neural oscillatory activity in brain is highly variable in its duration, as well as amplitude, frequency and recurrence. It is believed that such oscillations reflect functional states, and thus we can assume that at rest, these spontaneous patterns which seem to be unpredictable, show statistical similarities (specifically power law scaling behavior of a particular observable) defined as “Long Range Temporal Correlations” (LRTC) [16], [18]. In this paper we hypothesize that resting-state alpha-rhythm temporal dynamics maybe predictive of behavioral performance. We used a modified version of WM-task implemented in [1], a delay-match Sternberg paradigm which elicited enhancement in parieto-occipital alpha as an indicator of attention suppression during the maintenance period of visual working memory. Here we test whether the neural dynamic that describes how excitation and inhibition are balanced in the single individual at rest can  indicate the ability to efficiently employ cognitive control.

Methods

All experiments in this study conformed to the relevant guidelines and regulations. Informed consent was obtained from all participants. The study was approved by the local Ethics Committee of the Higher School of Economics, Moscow.

The study involved 15 subjects aged from 18 to 26 years (9 women, mean age = 22.5, SD = 2.3). Subjects were right-handed, with no history of neurologic or psychiatric disorders, and with good or corrected eye-sight. 2 participants’ resting state data had technical problems, so 13 participants’ data was used for the analysis (8 women, mean age = 22.6, SD = 2.4).

The paradigm was a modified version of the Sternberg task [24]. In each trial a set of 5 symbols was shown sequentially to the subject. Each presentation lasted 33 ms with an inter stimulus interval of 1.1 s. At the end, one probe symbol was presented. The task was to identify whether the probe was part of the proposed set. We implemented three different conditions: 1) Load 5 condition, where all 5 symbols had to be memorized; 2) strong distractor condition, where the 5th item was a distractor which had to be ignored (the distractor was a symbol similar to the presented set) 3) weak distractor condition, where the 5th item was a distractor which again had to be ignored (the distractor was a symbol easily distinguishable from to the presented set). At 1.1 s. from the onset of the 5th symbol, the probe was staying on the screen until the answer was given by button press. Accuracy and reaction time were recorded as measures of performance. For the analysis of reaction time, only correct trials were used. One session of the task consisted of 180 trials, with 60 trials for each condition.

The experiment was conducted in 2 days. The first day was a training day, when participants trained on 3 sessions of a behavioral task. The second day consisted of 2 parts: recording of resting state EEG and behavioral task. The EEG/EMG data were recorded with BrainAmp amplifiers and BrainVision Recorder software (Brain Products GmbH, Munich, Germany). For the recording of EEG, a standard montage with 64 electrodes was employed with mastoid electrodes used to record oculography. Throughout the experiment skin resistance was kept under 10 kOhm. Resting state recording included 10 minutes of eyes-open resting with eyes fixated on a black screen.

EEG data was preprocessed in EEGLAB toolbox within MATLAB software. The EEG signal was filtered between 2 and 40 Hz and re-referenced to an average reference. The data from oculography electrodes was not included in the average reference calculation. Signal was visually inspected and periods contaminated by motor artifacts were removed from the data. Heavily contaminated channels were excluded from the analysis. Eye movements were corrected by independent component analysis.

Data analysis

Analysis of behavioral data

We used ez package [9] in RStudio to run repeated measures Analysis of variance (ANOVA) with factor Condition (Load5, Strong, Weak) and two dependent variables: accuracy and reaction time (RT). Difference between conditions was compared with paired-sample t-test. Greenhouse-Geiser correction was applied for data, which violated the assumption of sphericity.

Behavioral measurements were recorded during EEG session for 15 subjects. Accuracy was calculated as percentage of correct answers in all trials. RT was analyzed only for correct responses. Trials, where 5th stimulus was the same as the probe, were removed from analysis.

To find main effect of LRTC on general working memory performance we averaged accuracy and RT by subject in three conditions of our task. We took behavioral data from 13 subjects, to correlate with LRTC scores, while 2 subjects did not have resting state recordings (for reasons described in Subjects section).

Long Range Temporal Correlations (LRTC)

In order to investigate, how temporal dynamics of alpha oscillation impact working memory performance, we calculated Long Range Temporal Correlations (LRTC) in the alpha spectral range during a 10 minutes resting state session preceding the working memory task.

First, we defined for each participant the individual alpha peak. For the single subject, we computed the power spectral density of each EEG channel and averaged across channels. We used Welch estimation with Hanning window of 2 s. We visually identified the alpha peak for each subject. Individual alpha peak varied from 8.5 Hz to 13 Hz (mean = 10.7, SD = 1.03, median = 11). To calculate LRTC we used Neurophysiological Biomarker Toolbox (NBT) toolbox v0.5.5, which functionality is described elsewhere [8].

First, we filtered the signal for each electrode within following range: individual alpha frequency ± 2 Hz. Then we extracted amplitude envelope with Hilbert transform a(t) and calculated cumulative sum of the signal Y(t) as

29-12-2019 21-48-53  (1)

The whole rest recordings were divided into windows τ ranging from 3 to 57 seconds with 50% overlap. These windows were equally spaced on the logarithmic scale. The signal Y(t) was detrended and the fluctuation function F(τ) in each time window τ computed according to

29-12-2019 21-49-00  (2)

This procedure was repeated for each time window τ. The window length and the fluctuation were represented on a double logarithmic scale, and a least-squares line was fitted to these values. The slope of this line is the scaling exponent ν, which ranges from 0.5 to 1 and indicated the presence of LRTC.

The LRTC scaling exponent was calculated for each channel within each subject during rest.

Correlation between LRTC and behavioral data

The matrix of LRTC scaling exponents (with dimension subject x channels) was then used for correlation with accuracy and RT (vectors with dimension 1 x subjects). This step was implemented with Fieldtrip toolbox [17] in Matlab. For each EEG electrodes we obtained a Pearson’s correlation across subjects between subject’s LRTC scaling exponents in each electrode and behavioral outcomes. In order to account for multiple comparisons, we used cluster statistics [13]. Clusters were defined as two or more neighboring electrodes that demonstrated a significant correlation with p < 0.05. A null distribution was generated using the Monte Carlo method from 1000 permutations of the original data. As a result, we obtained clusters of electrodes with LRTC scaling exponents correlated with behavioral data.

For visualization purposes, LRTC scaling exponents of one representative electrode in significant cluster were correlated again with behavioral measurement.

Results

Behavioral data from EEG session

We found no significant impact of Condition on accuracy (F(2,26) = 1.19, p = .31). Interestingly, Condition significantly influenced RT (F(1.3,16.77) = 4.5, p = .03) (Fig. 1). Subjects were faster giving answers in Weak condition, than in Strong (p = .008) or Load5 trials (p = .01). The Strong and Load5 conditions did not differ significantly (p = 0.34). This is in line with the expected task complexity.

 

29-12-2019 21-51-11

Fig. 1 – Difference in RT among three experimental conditions

Notes: Asterisk represents significant difference  

Correlation of LRTC and behavioral data

The major finding of the paper is the correlation of resting LRTC with the mean accuracy achieved during the behavioral WM task. We found significant frontal cluster (p = 0.044) with 4 electrodes (F3, F7, AF3, F5), when correlated LRTC scaling factor (Fig. 2A). LRTC scores in F3 electrode, which had the highest correlation coefficient, correlated with accuracy. Significant negative correlation (R = -.68, p = .009) was observed: subjects with higher LRTC scores in frontal region performed worse in working memory task (Fig. 2B).

No significant clusters were found for correlation of LRTC scores and RT.

 

29-12-2019 21-51-37

Fig. 2 – Correlation between LRTC and mean accuracy

Notes: A) Significant frontal cluster for correlation between LRTC scaling factor and mean accuracy. Topographies show the strength of Spearman’s correlation for all electrodes. White dots indicate electrodes of significant cluster (F3, F7, AF3, F5). B) The slope of the least-squares fitted lines (black) corresponds to correlation between LRTC scores in F3 electrode found in significant cluster and mean accuracy

Conclusion

LRTC have been described as an indicator of the balance between excitation and inhibition in neural networks [10], [19], this blnce is a prerequisite for the efficient interaction among neural substrates [2], [21], [22]. Optimal excitation/inhibition balance allows for stable maintenance of cognitive states and modulation of neural communication (Eriksson et al., 2015). To date, LRTC have been linked to cognitive abilities to perform perceptual [18], motor [23] and decision-making [3] tasks.

In order to study the role of LRTC in working memory and attentional switch, we adapted a paradigm previously used to demonstrate the role of parieto-occipital alpha oscillations in attentional suppression [1]. Our experiments showed that while the neural oscillatory dynamics at rest in these perceptual areas did not correlate with behavioral performance, we found that we were able to predict individual performance accuracy using LRTC in left frontal cortical areas.  In particular, we find that the behavioral accuracy correlates with the LRTC scaling factor at rest. The negative sign of this correlation is contrasts  recent findings [12], where LRTC were positively correlated with performance in a 2-back WM task. However, our protocol explicitly targets perceptual working memory, with higher working load and entails the switch of attentional state with the presentation of a distractor during the maintenance period. Moreover, while previous studies highlight the involvement of a large network, here we observe a specific role in frontal sites. Interestingly, this is in line with models of top-down modulation of prefrontal circuits in perceptual working memory [6], [7]. Direct recordings from neurons in monkeys outlined the critical role of dorsolateral prefrontal cortex (dlPFC) in distractor suppression [25]. Moreover, in the same study, reversible inactivation of the dlPFC resulted in larger performance deterioration than inactivation of areas which contribute to the perceptual selection of salient information.

Taken together, the critical state of frontal areas might be an indicator of the individual ability to exert cognitive control, which can be measured in the temporal dynamics of neural oscillation at rest. While further work is required to better characterize the relation of LRTC in the prediction of cognitive control, we demonstrated that it is possible to establish a link between neural dynamic at rest and attentional switch via top-down modulation in the framework of working memory.

Финансирование Исследование было выполнено в рамках проекта, поддержанного грантом РНФ (контракт 17-11-01273). Funding This work was supported by Russian Science Foundation grant (No: 17-11-01273).
Конфликт интересов Не указан. Conflict of Interest None declared.

Список литературы / References

  1. Bonnefond, M. Alpha oscillations serve to protect working memory maintenance against anticipated distracters. / M. Bonnefond, O. Jensen // Curr.Biol. - 2012. - Vol. 22. - P. 1969-1974.
  2. Beggs, J. M. Neuronal avalanches are diverse and precise activity patterns that are stable for many hours in cortical slice cultures. / Beggs, J. M., Plenz, D. // Journal of Neuroscience. - 2004. - Vol. 24. - No. 22. - P. 5216 –5229.
  3. Colosio, M. Neural mechanisms of cognitive dissonance (revised): an EEG study. / Colosio, M., Shestakova, A., Nikulin, V.V., Blagovechtchenski, E., Klucharev, V // Journal of Neuroscience. - 2017. - Vol. 37. - No. 20. - P. 5074–5083.
  4. Engel, A. K. Dynamic predictions: oscillations and synchrony in top-down processing. / Engel, A. K., Fries, P., Singer, W. // Nat.Rev.Neurosci. - 2001. - Vol. 2. - P. 704-716.
  5. Eriksson, J. Neurocognitive architecture of working memory. / Eriksson, J., Vogel, E.K., Lansner, A., Bergstrom, F., Nyberg, L. // Neuron. - 2015. -Vol. 88. - P. 33–46.
  6. Gazzaley, A. Top-down modulation: bridging selective attention and working memory. / Gazzaley, A., Nobre, A. C. // Trends in cognitive sciences. – 2012. – Vol. 16. – No. 2. – P. 129-135.
  7. Geng, J. J. Attentional mechanisms of distractor suppression // Current Directions in Psychological Science. – 2014. – Vol. 23. – No. 2. – P. 147-153.
  8. Hardstone, R. Detrended fluctuation analysis: a scale-free view on neuronal oscillations. / Hardstone, R., Poil, S. S., Schiavone, G., Jansen, R., Nikulin, V. V., Mansvelder, H. D., & Linkenkaer-Hansen, K. // Frontiers in physiology. - 2012. - Vol.3. - P. 450.
  9. Lawrence, M. A. ez: Easy Analysis and Visualization of Factorial Experiments. / Lawrence, M. A. // CRAN R Project. -2016. Retrieved from https://cran.r-project.org/web/packages/ez/
  10. Linkenkaer-Hansen, K. Long-range temporal correlations and scaling behavior in human brain oscillations. / Linkenkaer-Hansen, K., Nikouline, V. V., Palva, J. M., & Ilmoniemi, R. J. // Journal of Neuroscience. – 2001. – Vol. 21. – No. 4. – P. 1370-1377.
  11. Llinas, R. R. Bursting of thalamic neurons and states of vigilance. / Llinas, R. R., Steriade, M. // Journal of Neurophysiology. -2006. - Vol.95. - P. 3297-3308.
  12. Mahjoory, K. Power and temporal dynamics of alpha oscillations at rest differentiate cognitive performance involving sustained and phasic cognitive control. / Mahjoory, K, Cesnaite, E, Hohlefeld, F.U., Villringer, A., Nikulin, V.V. // Neuroimage. -2019. - Vol. 88. - P. 35-144.
  13. Maris, E. Nonparametric statistical testing of EEG-and MEG-data. / Maris, E., & Oostenveld, R. // Journal of Neuroscience Methods. -2007. - Vol.164. - No. 1. - P. 177-190.
  14. Matsukura, M. Attention effects during visual short-term memory maintenance: protection or prioritization? / Matsukura, M., Luck, S. J., Vecera, S. P. // Perceptual Psychophysics. - 2007. - Vol. 69. - P. 1422-1434
  15. Miyake, A. The unity and diversity of executive functions and their contributions to complex "Frontal Lobe" tasks: a latent variable analysis. / Miyake, A., Friedman, N. P., Emerson, M. J., Witzki, A. H., Howerter, A., & Wager, T. D. // Cogn Psychol. - 2000. - Vol. 41. - P. 49-100.
  16. Nikulin, V. V. Long-range temporal correlations in electroencephalographic oscillations: Relation to topography, frequency band, age and gender. / Nikulin, V. V. & Brismar, T. // Neuroscience. - 2005. - Vol. 130. -P. 549-558.
  17. Oostenveld, R. FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. / Oostenveld, R., Fries, P., Maris, E., Schoffelen, J. M. // Computational intelligence and neuroscience. - 2011. - Vol.1.
  18. Palva, J. M. Neuronal long-range temporal correlations and avalanche dynamics are correlated with behavioral scaling laws. / Palva, J. M., Zhigalov, A., Hirvonen, J., Korhonen, O., Linkenkaer-Hansen, K., Palva, S. // Proc.Natl.Acad.Sci.U.S.A. - 2013. - Vol.110, - P. 3585-3590.
  19. Poil, S.S. Critical-state dynamics of avalanches and oscillations jointly emerge from balanced excitation/inhibition in neuronal networks. / Poil, S.S., Hardstone, R., Mansvelder, H.D., Linkenkaer-Hansen, K. // Journal of Neuroscience. - 2012. - Vol. 32. - No 29. - P. 9817–9823.
  20. Schneider, D. On the neural mechanisms underlying the protective function of retroactive cuing against perceptual interference: Evidence by event-related potentials of the EEG. / Schneider, D., Barth, A., Getzmann, S., Wascher, E. // Biological Psychology, - 2017 -V.124. -P. 47-56.
  21. Shew, W. L. Information capacity and transmission are maximized in balanced cortical networks with neuronal avalanches. / Shew, W. L., Yang, H., Yu, S., Roy, R., & Plenz, D. // Journal of neuroscience. – 2011. – Vol. 31. – No. 1. – P. 55-63.
  22. Shew, W.L. The functional benefits of criticality in the cortex. / Shew, W.L., Plenz, D. // Neuroscientist. - 2013. - Vol.19. - No 1. - P. 88–100.
  23. Smit, D. J. A. Long-range temporal correlations in resting-state alpha oscillations predict human timing-error dynamics. / Smit D. J. A., Linkenkaer-Hansen K., de Geus E. J. C. // Journal of Neuroscience. – 2013. – Vol. 33. – No. 27. – P. 11212-11220.
  24. Sternberg, S. High-speed scanning in human memory // Science. – 1966. – Т. 153. – №. 3736. – С. 652-654.
  25. Suzuki, M. Distinct neural mechanisms of distractor suppression in the frontal and parietal lobe. / Suzuki, M, Gottlieb, J. // Nat. Neurosci. -2013. - Vol.16. - No 1. - P. 98-104.