Exploring the Interactions Between Neurophysiology and Cognitive and Behavioral Changes Induced by a Non-pharmacological Treatment: A Network Approach

ORIGINAL RESEARCH published: 29 July 2021 doi: 10.3389/fnagi.2021.696174 Edited by: Priyanka Shah-Basak, Medical College of Wisconsin, United States Reviewed by: Laura Lorenzo-López, University of A Coruña, Spain Dianne Patterson, University of Arizona, United States *Correspondence: Víctor Rodríguez-González victor[email protected] Received: 16 April 2021 Accepted: 13 July 2021 Published: 29 July 2021 Citation: Rodríguez-González V, Gómez C, Hoshi H, Shigihara Y, Hornero R and Poza J (2021) Exploring the Interactions Between Neurophysiology and Cognitive and Behavioral Changes Induced by a Non-pharmacological Treatment: A Network Approach. Front. Aging Neurosci. 13:696174. doi: 10.3389/fnagi.2021.696174 Exploring the Interactions Between Neurophysiology and Cognitive and Behavioral Changes Induced by a Non-pharmacological Treatment: A Network Approach Víctor Rodríguez-González1*, Carlos Gómez1,2,Hideyuki Hoshi3,Yoshihito Shigihara3, Roberto Hornero1,2,4 and Jesús Poza1,2,4 1Biomedical Engineering Group, Universidad de Valladolid, Valladolid, Spain, 2Centro de Investigación Biomédica en Red en Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN), Madrid, Spain, 3Precision Medicine Centre, Hokuto Hospital, Obihiro, Japan, 4IMUVA, Instituto de Investigación en Matemáticas, Universidad de Valladolid, Valladolid, Spain Dementia due to Alzheimer’s disease (AD) is a neurological syndrome which has an increasing impact on society, provoking behavioral, cognitive, and functional impairments. AD lacks an effective pharmacological intervention; thereby, non-pharmacological treatments (NPTs) play an important role, as they have been proven to ameliorate AD symptoms. Nevertheless, results associated with NPTs are patient-dependent, and new tools are needed to predict their outcome and to improve their effectiveness. In the present study, 19 patients with AD underwent an NPT for 83.1 ±38.9 days (mean ±standard deviation). The NPT was a personalized intervention with physical, cognitive, and memory stimulation. The magnetoencephalographic activity was recorded at the beginning and at the end of the NPT to evaluate the neurophysiological state of each patient. Additionally, the cognitive (assessed by means of the Mini-Mental State Examination, MMSE) and behavioral (assessed in terms of the Dementia Behavior Disturbance Scale, DBD-13) status were collected before and after the NPT. We analyzed the interactions between cognitive, behavioral, and neurophysiological data by generating diverse association networks, able to intuitively characterize the relationships between variables of a different nature. Our results suggest that the NPT remarkably changed the structure of the association network, reinforcing the interactions between the DBD-13 and the neurophysiological parameters. We also found that the changes in cognition and behavior are related to the changes in spectral-based neurophysiological parameters. Furthermore, our results support the idea that MEG-derived parameters can predict NPT outcome; specifically, a lesser degree of AD neurophysiological alterations (i.e., neural oscillatory slowing, decreased variety of spectral components, and increased neural signal regularity) predicts a better Frontiers in Aging Neuroscience | www.frontiersin.org 1July 2021 | Volume 13 | Article 696174 Rodríguez-González et al. Interactions Between MEG and NPT Outcome NPT prognosis. This study provides deeper insights into the relationships between neurophysiology and both, cognitive and behavioral status, proving the potential of network-based methodology as a tool to further understand the complex interactions elicited by NPTs. Keywords: non-pharmacological treatment (NPT), Mini-Mental State Examination (MMSE), Dementia Behavior Disturbance Scale (DBD-13), magnetoencephalography (MEG), networks, predict INTRODUCTION Dementia is a neurological syndrome that induces cognitive, behavioral, and functional alterations (Cummings, 2003). It is estimated that, in 2019, about 50 million people suffered from dementia worldwide, and this number is expected to increase to 132 million in 2050 (Alzheimer’s Disease International, 2019). Furthermore, its global economic impact is currently estimated at $1 trillion, and it is expected to be doubled by 2030 (Alzheimer’s Disease International, 2019). Alzheimer’s disease (AD) is the most common cause of dementia, with an exponentially growing incidence, especially in developed countries, due to the increase in life expectancy (Alzheimer’s Association, 2019). These figures show that AD is becoming a problem of utmost importance, highlighting the need to develop new treatments to help ameliorate the increasing impact of AD. Some pharmacological treatments for AD have been developed over the past few years (Alzheimer’s Disease International, 2019). Nonetheless, their effectiveness to mitigate dementia symptoms is very limited and patient-dependent and, in addition, they are often expensive (Qaseem et al., 2008; Alzheimer’s Association, 2019; Alzheimer’s Disease International, 2019). On the other hand, non-pharmacological treatments (NPTs) are showing promising results when dealing with AD-related cognitive alterations (Zucchella et al., 2018; Alzheimer’s Association, 2019). NPTs include a wide variety of strategies, ranging from physical training to cognitive stimulation, through psychological therapy (Dyer et al., 2018). As pharmacological therapies, they are not able to repair or stop the neuronal death caused by AD, but they are beneficial to patients with the disease (Dyer et al., 2018; Alzheimer’s Association, 2019). NPTs have been proven to effectively treat behavioral and psychological dementia symptoms, as well as to improve cognitive function and scores in depression tests (Oliveira et al., 2015; Dyer et al., 2018; Alzheimer’s Association, 2019). Therefore, NPTs are recommended as first-line managers to cope with behavioral and psychological symptoms of dementia, as they do not have adverse effects (Dyer et al., 2018). Nonetheless, their effectiveness has been shown to be patient-dependent (Kurz et al., 2011; Maki et al., 2018; Alzheimer’s Association, 2019). Many factors could influence the outcome of NPTs, such as previous cognitive level, symptom severity, or anti-psychotic use, but their impact is still unclear (Hsu et al., 2017). Therefore, being able to a-priori predict NPT outcome is a problem of paramount importance, since it would lead to personalized treatments and, consequently, to increased treatment efficiency. Neuroimaging techniques could be useful in this regard. They record neuronal activity on different levels, providing a quantitative framework to assess NPT influence on higher cognitive functions. Resting-state electroencephalography (EEG) and magnetoencephalography (MEG) have already been proven to be sensitive to changes induced by NPTs in brain activity (Amjad et al., 2019; Shigihara et al., 2020a,b), as well as to be potential predictors of NPT outcome (Amjad et al., 2019; Shigihara et al., 2020a,b). Both EEG and MEG are noninvasive neurophysiological techniques, though only MEG provides simultaneously high spatial and temporal resolution, as well as low distortion of scalp recordings due to the resistive properties of brain structures (Babiloni et al., 2009). MEG records brain activity in the range of milliseconds, which is of paramount importance to understand the function of a dynamic system like the brain (Babiloni et al., 2009). MEG recordings, and specifically resting-state signals, are often analyzed in patients with AD because they are able to detect the subtle changes that the disease provokes in neural activity (Engels et al., 2017; Mandal et al., 2018). Likewise, as previously mentioned, past studies found individual correlations between MEG-based parameters in specific brain regions and both, cognitive and behavioral variations due to NPTs (Amjad et al., 2019; Shigihara et al., 2020a,b). These results support the potential of MEG to quantify the effects of these therapeutic interventions. In the current research, we propose to further explore the complex interactions between the diverse variables under study by means of a network-related framework, which enables us to glimpse the footprint of the therapy in neural signals in a comprehensive and intuitive way. This approach is based on the generation of the so-called ‘‘association networks’’ that simplify the interpretation of the complex interactions between variables of diverse nature (Borsboom and Cramer, 2013; Fornito et al., 2016; Borsboom, 2017). Association networks are increasingly used as a tool for conceptualizing the interactions between symptoms in mental disorders, given their ability to capture all the intriguing complexity of these pathologies (Borsboom and Cramer, 2013; Borsboom, 2017). To the best of our knowledge, this is the first time that a network framework has been applied to assess the complex associations due to an NPT between neurophysiology, cognition, and behavior in AD. This framework provides a powerful tool to analyze the impact of NPT on neurophysiological signals and its potential predictors in a simple and integrated way. In this work, we hypothesize that NPT elicits several changes in different cognitive and behavioral dimensions, which in turn modify functional brain activity. The relationships between brain function and higher-order capacities are governed by a complex pattern of interactions between neurophysiological, cognitive, and behavioral variables. Consequently, new methodological Frontiers in Aging Neuroscience | www.frontiersin.org 2July 2021 | Volume 13 | Article 696174 Rodríguez-González et al. Interactions Between MEG and NPT Outcome approaches are needed to identify the changes in oscillatory brain activity that could be used to quantitatively assess NPT outcomes and, eventually, to design personalized therapeutic interventions. To address these issues, 19 patients with AD went through an NPT. Resting-state MEG activity, cognitive state, and behavioral status were evaluated at the beginning and end of the NPT. Different spectral and non-linear parameters of the MEG recordings were calculated to evaluate their interactions with the NPT outcome, which was measured by means of cognitive and behavioral tests. Specifically, we will address the following research questions: (i) are the association networks able to reflect the influence of the NPT on the relationships between neurophysiological and cognitive/behavioral variables?; (ii) what are the particular changes in the structure of the association networks due to the NPT?; and (iii) can the neurophysiological parameters predict the cognitive and behavioral changes associated with the NPT? MATERIALS AND METHODS Participants Nineteen patients with dementia from the geriatric health services facility ‘‘Kakehashi’’ (Obihiro, Japan) were recruited for this study. It is an official facility authorized by the Ministry of Health, Labor, and Welfare in Japan, recognized as a transient facility between hospitals and patients’ homes. The main role of this facility is to improve the physical and cognitive conditions of aged individuals to enable them to return to their homes. All the participants were diagnosed with AD, and two of them were also diagnosed with other pathologies: one with Parkinson’s disease, and the other one with vascular dementia. The diagnoses were carried out by clinicians before admission in the ‘‘Kakehashi’’ facility, and according to the National Institute on Aging- Alzheimer’s Association criteria (McKhann et al., 2011). If possible, patients’ medication remained unchanged during the NPT period. Patients underwent the NPT for 83.1 ±38.9 days (mean ±standard deviation, see Figure 1 for a graphical description of the NPT period), being treated every day by the NPT professionals. NPT is composed of five types of activities commonly used in geriatric health services facilities in Japan: 1. Physical training. It is aerobic exercise and resistance training, which are effective to improve cognitive function in aged individuals (Nagamatsu et al., 2012; Amjad et al., 2019). 2. Therapeutic role-playing. This therapeutic intervention is called ‘‘Otona-no-gakko’’ (‘‘School for adults’’) and it is used to both re-introduce patients to active life and enhance their daily motivation (Cotelli et al., 2012). 3. Nursing care. Nursing care provides proper eating, drinking, and a sanitary environment, which are essential to keep brain activity healthy. Furthermore, it has been previously suggested that diet has some relevant impact on AD (Rege et al., 2016; McGrattan et al., 2019). 4. Horticultural therapy. This therapy is based on gardening and planting activities to improve physical and cognitive FIGURE 1 | Schematic overview of the time course of the study. Two assessments took place during the study: at the beginning and the end of the therapy. Each assessment consisted of MEG recording and application of Mini-Mental State Examination (MMSE) and Dementia Behavior Disturbance (DBD)-13 tests, that were performed within 3 days. The period between assessments is defined as the time between the first and the second assessments. The non-pharmacological treatment (NPT) period is defined as the time between the admission in the facility and the second assessment. TABLE 1 | Sociodemographic and clinical information of the sample. Sociodemographic data Number of subjects 19 Age (years) 86.00 ±3.86 Gender (M:F) 7:12 NPT Period (days) 83.05 ±38.88 Pre Post MMSE 14.11 ±5.95 16.00 ±7.32 DBD-13 10.89 ±9.93 9.84 ±10.55 Data are shown as mean ±standard deviation. M, male; F, female; MMSE, Mini-Mental State Examination; DBD-13, Dementia Behavior Disturbance scale. conditions (Lu et al., 2020). Patients were familiar with these activities since our facility is located in an agricultural area. 5. Self-cognitive training. Self-cognitive training includes activities such as coloring books or crossword puzzles (Anderson and Grossberg, 2014). Each NPT session was designed each day by the experts to adapt it to the clinical features and mood of each patient (Maki et al., 2018), with a duration ranging from 20 to 40 min, according to the Japanese regulations. See Table 1 for a description of the sociodemographic and clinical information of the sample. All participants and their families or caregivers gave their informed consent to participate in the present study. The investigation was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki). The protocol was approved by the Ethics Committee of Hokuto Hospital. Cognitive and Behavioral Assessment Cognitive and behavioral performance was assessed twice for each patient, at the beginning and at the end of the NPT. Each assessment session consisted of two different tests conducted on the same day: an abbreviated version of the Dementia Frontiers in Aging Neuroscience | www.frontiersin.org 3July 2021 | Volume 13 | Article 696174 Rodríguez-González et al. Interactions Between MEG and NPT Outcome Behavior Disturbance Scale (DBD-13; Machida, 2012), and the Japanese Mini-Mental State Examination (MMSE; Folstein et al., 1975; Sugishita et al., 2010). The DBD-13 scale is a 52-point test consisting of 13 items of the original DBD-28 scale (Baumgarten et al., 1990; Machida, 2012). It measures the behavioral disturbance induced by dementia, assigning higher values to more behavior problems. The MMSE is a test with a maximum score of 30, which measures cognitive impairment by assessing different cognitive domains (Folstein et al., 1975; Sugishita et al., 2010). Lower values correspond to more impaired cognition. MEG Recordings All MEG recordings were acquired at the Hokuto Hospital (Obihiro, Japan). As for the cognitive and behavioral assessments, brain signals were recorded twice: at the beginning and end of the treatment. MEG recordings, cognitive state, and behavioral status were evaluated within 3 days in order to: (i) accurately match each MEG recording with a cognitive and behavioral assessment; and (ii) make the intervals between MEG recordings and both cognitive and behavioral assessments as similar as possible. Thereby, 76.1 ±36.0 days (mean ±standard deviation) passed between the initial and final assessments. See Figure 1 for a graphical description of the period between assessments. For each subject, 5 min of resting-state brain activity was recorded using a 160-channel axial gradiometers MEG system (MEG Vision PQ1160C, Yokogawa Electric), with a sampling rate of 1,000 Hz and a low-pass filter at 200 Hz. Head position was registered with three fiducial markers placed on the patient’s head during the MEG scan: 5 mm above the nasion, and 10 mm in front of the tragus on each side of the head. Patients were asked to stay calm and awake with eyes closed, in a supine position during the recording. For security reasons, as well as to prevent somnolence, MEG recordings were monitored in real time. MEG Analysis Signals were preprocessed before the application of the source inversion algorithm. Next, different spectral and non-linear local activation parameters were calculated from the signals at the source level. Finally, these parameters were used to construct the networks based on the Spearman correlations between them. The next subsections describe the steps followed in the MEG analysis in detail. Preprocessing of MEG Signals To limit the presence of noise in the MEG recordings, signals were preprocessed using a 4-step pipeline (Rodríguez-González et al., 2020): (i) artifact removal using the SOUND algorithm (Mutanen et al., 2018; Rodríguez-González et al., 2019); (ii) finite impulse response (FIR) filtering: 1–70 Hz band-pass to limit noise bandwidth, and 49–51 Hz band-stop to remove line noise; (iii) artifact removal using independent component analysis; and (iv) visual selection of 5-s artifact-free epochs. Source Inversion Source-level signals were obtained using the Brainstorm toolbox, which is documented and freely available for download online under the GNU general public license1(Tadel et al., 2011). A forward model with 15,000 sources was created by means of boundary element model using the ICBM152 head template (Montreal Neurological Institute) and OpenMEEG software (Fonov et al., 2009; Gramfort et al., 2010; Douw et al., 2018). The head model was segmented into three tissues: brain, skull, and scalp, with conductivities of 1, 0.0125, and 1 Siemens per meter, respectively (Mahjoory et al., 2017). Sources were restricted to the cortex, and their direction was set normal to it (Mahjoory et al., 2017; Lai et al., 2018; Rodríguez-González et al., 2020). No noise recordings were available, so an identity matrix was used as noise covariance (Lai et al., 2018; Rodríguez-González et al., 2020). The 15,000 source-level time courses were projected into the 68 regions of interest (ROIs) provided by the Desikan-Killiany atlas, in order to have a manageable number of ROIs to work with Desikan et al. (2006),Lai et al. (2018), and Rodríguez-González et al. (2020). This source projection was done by averaging the reconstructed activation time courses of all the voxels in each ROI after flipping the sources of opposite direction (Lai et al., 2018; Rodríguez-González et al., 2020). As, we were working with resting-state signals, no a-priori assumptions about sources could be made. Thus, we used the weighted minimum-norm estimation (wMNE) algorithm, which restricts the solutions by minimizing the energy (L2norm) weighting deep sources to facilitate their identification (Lin et al., 2004). This method has been proven to be useful to reconstruct the underlying sources of resting-state MEG datasets (Lin et al., 2004). Feature Extraction Diverse signal processing methods have been widely used to describe the properties of brain activity. These methods characterize the electromagnetic fields generated by the synchronized neuronal pools responsible for the observed brain activity. In this study, we have used several local activation parameters, which measure the activation of single functional units (i.e., synchronized neuronal pools; Stam and van Straaten, 2012) They can be grouped in two main categories: (i) spectral parameters, which evaluate the time-frequency content of the recorded signal; and (ii) non-linear parameters, which measure relevant non-linear properties of the signal, such as variability, irregularity, or complexity. In this study, we have calculated a wide variety of parameters in both categories to fully characterize the properties of the resting-state MEG activity using its source reconstructed time courses on the 68 estimated ROIs. Spectral Parameters They are useful to characterize the spectral content of the signal. They were derived from the normalized power spectral density (PSDn), which was calculated using the Blackman-Tuckey method (Blackman and Tukey, 1958; Ruiz-Gómez et al., 2018; Rodríguez-González et al., 2020). The parameters computed in this category are listed below: •Relative power (RP). It summarizes the neural activation in a certain frequency range, relative to the full spectral content of 1http://neuroimage.usc.edu/brainstorm Frontiers in Aging Neuroscience | www.frontiersin.org 4July 2021 | Volume 13 | Article 696174 Rodríguez-González et al. Interactions Between MEG and NPT Outcome the signal. RP was calculated in the well-known conventional frequency bands: delta (δ, 1–4 Hz), theta (θ, 4–8 Hz), alpha (α, 8–13 Hz), beta 1 (β1, 13–19 Hz), beta 2 (β2, 19–30 Hz), and gamma (γ, 30–70 Hz). •Median frequency (MF). MF is the frequency that divides the PSDn into two halves of equal power. It is commonly used to measure the global signal slowing provoked by AD disruptions (Poza et al., 2007; Dauwels et al., 2011). •Individual alpha frequency (IAF). It measures the frequency where the alpha peak can be found. It is calculated as the frequency that divides the extended alpha band (4–15 Hz) into two halves of equal power (Klimesch, 1999; Poza et al., 2007). Alpha peak is related to higher cognitive functions, so this parameter is widely used to assess cognitive disfunction (Klimesch, 1999; Poza et al., 2007). •Spectral entropy (SE). This parameter measures the flatness or uniformity of the PSDn using Shannon entropy. It has been proven that patients with AD show a less distributed spectral content of the PSDn than controls, which suggests less variety of neural oscillatory components (Poza et al., 2008b; Gómez and Hornero, 2010). •Spectral edge frequency (SEF). It is quantified as the upper limit of the PSDn. This parameter is calculated as the frequency that comprises 95% of the power of the PSDn, and is identified as the bandwidth of the signal (Poza et al., 2007). Due to the slowing and the reduction in the variety of neural oscillatory activity associated with AD, this parameter has been used to characterize brain signals in patients with dementia (Poza et al., 2007). Non-linear Parameters Non-linearity is a fundamental property of complex systems, such as the brain (Stam, 2005). Non-linear analyses of brain signals are then commonly used to describe the alterations produced by a neuropathology like AD. The non-linear parameters assessed in this study are: •Lempel-Ziv complexity (LZC). It is a coarse-grain complexity measure. LZC estimates the complexity by counting the number of subsequences that the binarized version of the analyzed signal contains (Lempel and Ziv, 1976). It assigns higher values to more complex time series (Fernández et al., 2010, 2011). A decrease in complexity has been associated with AD progression (Gómez et al., 2006; Fernández et al., 2010). •Sample entropy (SampEn). SampEn is an irregularity measure that assigns higher values to more irregular time sequences. It has two tuning parameters: the sequence length and the tolerance, which were respectively set to 1 and 0.25·std (std: standard deviation of the signal) based on previous studies (Gómez et al., 2009; Hornero et al., 2009; Rodríguez-González et al., 2020). A decrease in irregularity has been observed in the neural activity of patients with AD (Escudero et al., 2009; Gómez et al., 2009; Hornero et al., 2009). •Central tendency measure (CTM). This parameter is useful to quantify the variability of a signal. It is based on calculating the second-order differences diagram of the time series and then counting the points within a radius. In the present study, the radius has been set to 0.025, based on previous analyses (Ruiz-Gómez et al., 2018; Rodríguez-González et al., 2020). CTM assigns higher values to less variable signals. Previous studies have reported that AD is associated with lower CTM values (Ruiz-Gómez et al., 2018). In addition to the spectral and non-linear parameters, a new measure is presented in the current study: the spatial Shannon entropy (SSE). Specifically, the SSE computes the entropy of the spatial distribution of values for a given local activation parameter. The spatial distribution of the considered parameter is estimated as the normalized histogram of its values considering the 68 ROIs. The calculation of the SSE of a local activation parameter enables us to quantify the changes induced by the NPT in the spatial patterns of brain oscillatory activity. A parameter with similar values across the brain (i.e., showing a delta-like distribution of values) would yield a low SSE value, whereas a high SSE value would be obtained by a parameter with a wide range of variation (i.e., displaying a uniform distribution). It is noteworthy that in the previous examples the parameters could have similar mean values, but their SSE values would be different. Hence, the SSE was computed for each spectral and non-linear parameter; the SSE of a given parameter will be referred to as S with the parameter name in brackets, e.g., the SSE of the IAF will be denoted as S(IAF). Construction of Association Networks In this study, we have generated different networks to account for the potential relationships between the neurophysiological, cognitive, and behavioral parameters. Thereby, the network nodes were individual variables (all the neurophysiological parameters, the score in the cognitive examination—i.e., MMSE–, and the score in the behavioral test—i.e., DBD-13), and the network edges (or weights) were the associations between them. These associations were estimated as the Spearman rank correlations between pairs of variables to detect both linear and non-linear monotonic interactions; age and gender were introduced in the correlation analysis as covariates to control for their effect. Non-significant correlations (i.e., network edges with p-values >0.05) were removed from the network (Zhang, 2011; Barberán et al., 2012). For the sake of simplicity, negative correlations were converted to positive, as we were interested in the association, and not in the nature of that association. Afterward, networks were constructed using Gephi software2. The width of the edges was linked to the magnitude of the relationship, with a wider edge meaning stronger association. The Force Atlas 2 algorithm was employed to group nodes with higher correlations while taking nodes with lower correlations away (Jacomy et al., 2014). No regularization algorithm was applied, as we were interested in exploring all the associations, especially those of cognitive and behavioral parameters, even if they are weaker than others. 2https://gephi.org/ Frontiers in Aging Neuroscience | www.frontiersin.org 5July 2021 | Volume 13 | Article 696174 Rodríguez-González et al. Interactions Between MEG and NPT Outcome Four different associations networks were generated: •Pre-NPT network. The nodes were the neurophysiological, cognitive, and behavioral parameters obtained before the application of the NPT. •Post-NPT network. The nodes were the neurophysiological, cognitive, and behavioral parameters obtained after the application of the NPT. •Changes network. The nodes were the variation of neurophysiological, cognitive, and behavioral parameters during the NPT, i.e., the difference between their values before and after the NPT. •Prediction network. The nodes were the neurophysiological parameters before the application of the NPT, and the variation of the cognitive and behavioral parameters during the NPT. Furthermore, networks stability was analyzed. To do that, the probability of obtaining similar results was assessed by using a bootstrapping methodology with 2,000 iterations to generate new random network instances (Efron, 1992; Epskamp et al., 2018). In each iteration, a new network is created by randomly selecting 19 subjects of the original dataset, being possible to select the same subject more than once (Efron, 1992). From the four bootstrapped instances (one for each network), the 95% confidence interval of the edges’ weights was reported (Jimeno et al., 2020). Another network was generated using the bootstrapped samples from Pre-NPT and Post-NPT networks: the Variability network. This network shows the differences in the association pattern between Pre-NPT and Post-NPT. Each network edge from the Variability network was computed as the test statistic obtained when comparing the bootstrapped samples from Pre-NPT and Post-NPT networks for that particular edge. Thus, the higher the edge weight in the Variability network, the higher the differences between Pre-NPT and Post-NPT networks. Although all the edge weights were statistically significant (p-values <0.05, Wilcoxon signed rank test), only the 5% strongest connections (i.e., showing the biggest differences between Pre-NPT and Post-NPT networks) were displayed. RESULTS Changes in Cognition and Behavior Induced by the NPT The average MMSE value before conducting the NPT was 14.11 ±5.95 (mean ±SD), while the average for DBD-13 was 10.89 ±9.93. Then, after applying the NPT, the average value for the MMSE was 16.00 ±7.32, and the average for DBD-13 9.84 ±10.55. The effectiveness of the NPT was assessed by comparing the MMSE and DBD-13 before and after conducting the NPT. Both, MMSE and DBD-13 show a statistically significant improvement after the NPT, with MMSE significantly increasing (p-value = 0.0323, one-tailed Wilcoxon signed rank test) and DBD-13 significantly decreasing (p-value = 0.007, one-tailed Wilcoxon signed rank test). FIGURE 2 | Parameter network of the patients with Alzheimer’s disease (AD) before the NPT (Pre-NPT network). Wider edges correspond with stronger associations. Nodes corresponding to cognitive and behavioral parameters are shown in blue, while nodes of neurophysiological parameters are shown in orange. FIGURE 3 | Parameter network of the patients with AD after the NPT (Post-NPT network). Wider edges correspond with stronger associations. Nodes corresponding to cognitive and behavioral parameters are shown in blue, while nodes of neurophysiological parameters are shown in orange. Changes in Network Structure Induced by the NPT In order to evaluate the changes induced by the NPT in the patients’ network structure, three different networks were evaluated: (i) Pre-NPT network (Figure 2); (ii) Post-NPT network (Figure 3); and (iii) Variability network (Figure 4). The figures depicting the stability of the networks shown in Figures 2, 3can be found in the Supplementary Figures 1, 2. Frontiers in Aging Neuroscience | www.frontiersin.org 6July 2021 | Volume 13 | Article 696174 Rodríguez-González et al. Interactions Between MEG and NPT Outcome FIGURE 4 | Variability network, showing the 5% strongest connections corresponding to the network edges that obtained the most significant differences between Pre-NPT and Post-NPT networks. Wider edges correspond with stronger associations. Nodes corresponding to cognitive and behavioral parameters are shown in blue, while nodes of neurophysiological parameters are shown in orange. Figure 2 displays the relationships between the parameters under study without the influence of the NPT, as they were calculated with the samples obtained at the beginning of the therapeutic intervention. It can be observed that while DBD-13 is disconnected (i.e., it has no relationship with any other parameter), MMSE is related with another 11 parameters: RP(Gamma), MF, IAF, SE, SEF, LZC, SampEn, CTM, S(RP(Gamma)), S(RP(Beta 1)), and S(CTM). Interestingly, no associations were observed for any RP parameter apart from RP(Gamma). On the other hand, Figure 3 shows the association network, but including the influence of the NPT, as it has been calculated with the parameters obtained after conducting the NPT. The Post-NPT network has a higher number of connections on cognitive and behavioral parameters in comparison with the Pre-NPT network: MMSE now has 13 connections, while DBD-13 has 11. Of note, eight out of the 13 associations of MMSE were maintained from the Pre-NPT network (MF, IAF, SE, LZC, SampEn, CTM, S(RP(Beta 1)), and S(CTM)), while the other five were new associations (RP(Delta), RP(Beta 1), RP(Beta 2), S(RP(Beta 2)), and S(SampEn)). In contrast to the Pre-NPT network, three parameters based on RP are now associated with the MMSE, but RP(Gamma) is no longer significant. Furthermore, the significant relationships for DBD-13 that can FIGURE 5 | Network of changes. As indicated by the “Var” suffix, this network displays the relationships between the variation of the parameters under study during the NPT (i.e., ParameterPost −ParameterPre). Wider edges correspond with stronger associations. Nodes corresponding to cognitive and behavioral parameters are shown in blue, while nodes of neurophysiological parameters are shown in orange. be appreciated in the Post-NPT network are with: RP(Gamma), SE, SEF, LZC, SampEn, CTM, S(RP(Gamma)), S(SEF), S(LZC), S(SampEn), and S(CTM). To get deeper insights on the changes induced by the NPT in the parameter network, the Variability network was constructed, depicting the 5% strongest differences between Pre-NPT and Post-NPT networks (Figure 4). As expected, the parameter whose relationships have changed most between both networks is DBD-13, with 11 connections, while MMSE only showed one connection. Interestingly, six out of those 11 connections (S(RP(Delta), S(RP(Beta 2)), S(RP(Gamma)), S(LZC), S(SampEn), and S(CTM)) are spatial entropies. Relationship Between Neurophysiological and Cognitive and Behavioral Changes The network of changes can be observed in Figure 5. This network describes the associations between the variation of the parameters under study (neurophysiological, cognitive, and behavioral) by computing: ParameterPost − ParameterPre. A positive value indicates an increase in the parameter provoked by the NPT, while a negative value is associated with a decrease. The figure showing the stability of the network depicted in Figure 5 can be found in the Supplementary Figure 3. It could be observed that apart from the DBD-13 - MMSE relationship, DBD-13 displays four connections RP(Beta 1), S(IAF), S(SE), and S(RP(Beta 1)), while MMSE only one S(RP(Beta 2)). Interestingly, four out of five significant associations involve spatial entropies: S(RP(Beta Frontiers in Aging Neuroscience | www.frontiersin.org 7July 2021 | Volume 13 | Article 696174 Rodríguez-González et al. Interactions Between MEG and NPT Outcome FIGURE 6 | Prediction network, showing the ability of the assessed neurophysiological parameters to predict the outcome of the NPT. The “Var” suffix, which appears associated with DBD and MMSE, means the variation of the parameters under study during the NPT (i.e., ParameterPost − ParameterPre). Wider edges correspond with stronger associations. Nodes corresponding to cognitive and behavioral parameters are shown in blue, while nodes of neurophysiological parameters are shown in orange. 1)), S(IAF), S(SE) with DBD, and S(RP(Beta 2)) with MMSE. Besides, three associations involve beta band: RP(Beta 1), S(RP(Beta 1)), and S(RP(Beta 2)). No association involves any non-linear parameter. Predictability of the NPT Outcome by Means of the Neurophysiological Parameters Figure 6 contains the Prediction network. It depicts the ability of the neurophysiological parameters under study to predict the outcome of the NPT, measured by the variation of the cognitive and behavioral parameters (MMSE and DBD- 13). It could be appreciated that, aside from the relationship that links cognitive and behavioral parameters, MMSE has nine significant relationships (RP(Delta), RP(Beta 1), MF, SE, SampEn, S[RP(Delta)], S[RP(Alpha)], S(RP(Beta 1)), and S(CTM)), while DBD-13 has only one (RP(Beta 1)). Of note, only two out of these 10 parameters involve non-linear parameters (MMSE-SampEn and MMSE-S(CTM)), and four of them involve spatial entropies (S[RP(Delta)], S[RP(Alpha)], S(RP(Beta 1)), and S(CTM)). Supplementary Figure 4 depicts the stability of the network depicted in Figure 6. These relationships are of great importance because, as mentioned in the Introduction section, predicting the NPT is crucial. Thus, to obtain deep insights on them, and disentangle the nature of these associations, we plotted scatterplots for every significant relationship (involving cognitive or behavioral parameters) obtained in the previous section, reporting the specific correlation values (ρ). These scatterplots are shown in Figure 7. Remarkably high relationships between parameters can be observed, with a mean value of 0.56. The strongest relationships can be observed for associations involving MMSE and delta and beta bands: MMSE-RP(Delta) (ρ=−0.69, pvalue = 0.002, Spearman rank correlation), MMSE-RP(Beta 1) (ρ= 0.57, p-value = 0.017, Spearman rank correlation), MMSE- S(RP(Delta)) (ρ=−0.56, p-value = 0.019, Spearman rank correlation), and MMSE-S(RP(Beta 1)) (ρ= 0.70, p-value = 0.002, Spearman rank correlation). DISCUSSION In the present study, we have assessed the effects of an NPT in the neurophysiology of patients with AD, as well as whether its outcome is predictable by means of MEG-based parameters. Our results hold three main findings related to the three research questions posed in the introduction: (i) the NPT alters the structure of the association networks, unveiling relationships between DBD-13 and neurophysiological parameters: RP(Gamma), SE, SEF, LZC, SampEn, CTM, S(RP(Gamma)), S(SEF), S(LZC), S(SampEn), and S(CTM); (ii) the changes induced by the NPT are related to the changes in the DBD-13, suggesting an impact of the NPT in the behavioral symptoms of AD; and (iii) the value of nine neurophysiological parameters (RP(Delta), RP(Beta 1), MF, SE, SampEn, S[RP(Delta)], S[RP(Alpha)], S(RP(Beta 1)), and S(CTM)) before going through the NPT are related to the NPT outcome, suggesting a potential predictive power of the aforementioned parameters to foresee the response of the patients to the NPT. NPT Induces Several Changes in the Structure of the Association Networks The Pre-NPT network displayed in Figure 2 shows that, before conducting the NPT, the neurophysiological parameters are associated with the MMSE, but not with the DBD-13. This could be explained as both tests are measuring the alterations provoked by dementia in different cognitive domains. On the one hand, DBD-13 measures strictly behavioral disturbances defined as ‘‘the outward manifestation of some underlying cognitive, psychological, or physiological deficit—regardless of etiology—likely to cause stress to those caring for the patient’’ (Baumgarten et al., 1990). On the other hand, MMSE quantifies cognitive impairment in a more global sense, by means of different cognitive dimensions, such as attention, orientation, language, perception, calculus, or the ability to follow simple instructions (Folstein et al., 1975). Therefore, these results suggest that cognitive disturbances measured in a broader sense are directly related to the neurophysiological state. Nevertheless, this is not the case for behavioral disturbances, where this relationship could be mediated or obscured by external factors, such as the environment, relationship with caregivers, or daily life habits. It is worth mentioning that all the spectral and non-linear parameters, apart from those derived from the RP (except RP(Gamma)), show statistically significant associations with Frontiers in Aging Neuroscience | www.frontiersin.org 8July 2021 | Volume 13 | Article 696174 Rodríguez-González et al. Interactions Between MEG and NPT Outcome FIGURE 7 | Scatterplots representing the relationship between the neurophysiological parameters computed before applying the NPT (x-axis) and the changes in cognitive and behavioral variables after the NPT (y-axis). In the top part of each panel, the value of the Spearman rank correlation for each specific pair of parameters is plotted. Dashed lines represent the linear regression of the data. the MMSE. RP values do not contain information about the complete oscillatory activity, but only of a certain frequency band, typically associated with a limited number of cognitive functions (Uhlhaas et al., 2008). As it is known that AD induces alterations in several cognitive domains (Alzheimer’s Association, 2019; Alzheimer’s Disease International, 2019), it could be hypothesized that the absence of associations between MMSE and RP could be provoked, at least partially, because they are only reflecting particular cognitive dimensions of AD disruptions. Noteworthy, associations were found for RP(Gamma), as well as for its spatial entropy. The gamma band has been proven to play an important role in several higher cognitive functions (Bartos et al., 2007; Martorell et al., 2019). Besides, this frequency band is also affected by AD neuropathology. Previous studies reported that AD patients’ brain activity is associated with an enhanced gamma power (van Deursen et al., 2008; Wang et al., 2017), an increase in long distance gamma connectivity (Ba˛sar et al., 2017), and an increase Frontiers in Aging Neuroscience | www.frontiersin.org 9July 2021 | Volume 13 | Article 696174