Twenty-three patients with idiopathic PD (12 women) were recruited at the Outpatient Clinic, Department of Neurology, University of Lübeck. For details of recruitment, see Figure 1. One patient was excluded because the Parkinson Neuropsychometric Dementia Assessment (PANDA, Kalbe et al., 2008) indicated cognitive impairment. EEG recordings from seven patients had to be excluded because of excessive muscle activity during preparation for speaking. Another patient was excluded because the audio recording was missing due to a technical failure. The data sets of the remaining 14 patients (8 women) were analyzed for this study. Mean age ± standard deviation was 69.5 ± 8.0 years (age range: 52–81 years).

All patients were diagnosed by an experienced neurologist; the initial diagnosis was made between 3 years and 16 years before this study was performed. All patients received antiparkinsonian medication at the time of the study (mean equivalent L-dopa dose: 677 mg/day). To assess the severity of motor symptoms, part III of the Unified Parkinson’s Disease Rating Scale (UPDRS) was performed in all patients (Fahn and Elton, 1987). The mean UPDRS III score ± SD was 17 ± 11 (minimum: 3, maximum: 44).

Twenty-seven healthy individuals (13 women) were recruited for an age-matched control group. Twenty-five participants were recruited through the subject database of the Department of Neurology, two participants were healthy spouses of the patients recruited for this study. EEG recordings from 10 healthy participants had to be excluded because of excessive muscle activity during preparation for speaking. One healthy participant was excluded because of an exceptionally delayed speech onset. In this participant, the time between the onset of the visual cue and the onset of speaking was on average 1965 ms (for the rest of the control group, speech latency was 659 ± 153 ms). Another healthy participant was excluded because of an exceptionally high number of incorrect speech responses (72%). The data sets of the remaining 15 healthy participants (7 women) were analyzed for this study. Mean age ± SD was 69.7 ± 7.0 years (age range: 55–80 years).

All included patients and control participants were right-handed, had normal or corrected-to-normal vision and no history of neurological or psychiatric disorder (except PD in the patient group). The five cognitive tasks of the PANDA (Kalbe et al., 2008) showed no cognitive impairment. The Beck Depression Inventory (Beck et al., 1961) indicated no depressive symptoms. To assess verbal fluency at the behavioral level, all participants were asked to produce as many words beginning with “m” as possible within 2 min as part of the Regensburger Wortflüssigkeits-Test (Aschenbrenner et al., 2000).

This study was carried out in accordance with the recommendations of the Research Ethics Board of the University of Lübeck with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the REB. Participants received a remuneration of 40 Euro.

A visually cued overt speech production task was performed during the EEG recording. Participants were seated in a comfortable chair with their eyes about 80 cm in front of a 20 inch LCD computer screen. To cue speech production, the syllables papapa or pataka were shown in the middle of the screen for 2000 ms. Patients were instructed to speak aloud the required utterance as soon as the cue appears. The onset-to-onset time interval between cues was 5000 ms. To reduce task-switching effects and to facilitate the task, a blocked presentation was chosen. Each block was preceded by the instruction “Jetzt kommt papapa (Now we’ll present papapa)” or “Jetzt kommt pataka (Now we’ll present pataka)”. Each block consisted of five papapa and five pataka trials. Five blocks were presented in pseudo-randomized order for each condition. In total, 25 papapa and 25 pataka trials were recorded. All stimuli were presented using Presentation software (Neurobehavioral Systems, Berkeley, CA, USA)¹.

All utterances were recorded using a microphone and stored for off-line analysis with the open source audio software Audacity². All utterances were examined by the same listener (CW) and double-checked by another researcher (PS). Trials without a correct, intelligible utterance were excluded from further data analysis. In the patient group, on average 0.9 trials per participant were excluded because a correct verbal response was missing (minimum: 0 trials, maximum: 5 trials). In the control group, on average 1.7 trials per participant were excluded due to a missing correct response (minimum: 0 trials, maximum: 9 trials). Speech latency (the time difference between onset of the instruction and onset of overt speech) was determined for all correct responses and all participants. On an individual basis, speech latencies were z-transformed and trials with exceptionally short or long latencies (z < −3 or z > 3) excluded. On average, 0.9 trials per participant in the patient group (minimum: 0, maximum: 2) and 0.7 trials per participant in the control group (minimum: 0, maximum: 2) were excluded after z-transformation of latencies. The average individual speech latency was finally calculated for every participant.

EEG data were recorded in a sound-dampened and electromagnetically-shielded room using an elastic 32 electrode cap (Electro-Cap International, Eaton, OH, USA)³ and a Synamps amplifier (Compumedics Neuroscan, Singen, Germany)⁴. To record brain electric activity, 29 tin electrodes were mounted on the scalp according to the 10/20 system with modified combinatorial nomenclature (American Electroencephalographic Society, 1994). Scalp electrodes were referenced online against the left mastoid. To monitor vertical and horizontal eye movements, electrodes were placed above and below the left eye and on the outer canthus of each eye. Data was sampled at 250 Hz, with a bandpass filter of 0.01–50 Hz. Electrode impedances were kept below 5 kΩ.

Raw data were re-referenced to the average activity of both mastoid electrodes. An independent component analysis (ICA) was performed using an extended information maximization algorithm (Bell and Sejnowski, 1995) as implemented in EEGLAB (Delorme and Makeig, 2004)⁵ to identify and remove components related to eye blinks and ocular movements (Hoffmann and Falkenstein, 2008).

To calculate time-frequency spectra, single-trial data were convolved with a complex Morlet wavelet using the open source MATLAB toolbox FieldTrip (Oostenveld et al., 2011)⁶. The width of the wavelet was set at seven cycles. To avoid contamination of EEG data by speech-related muscle artifacts, individual EEG epochs were truncated at speech onset. Thus, the number of analyzed trials and hence the signal-to-noise ratio decreased after approximately 400 ms. Figure 2 illustrates the number of trials (mean ± SD) available at a given time point for the papapa and the pataka conditions, separately.

After wavelet transform, oscillatory power in the studied frequencies (1–50 Hz, linear increase) was computed for each trial in the time window −1000 ms to speech onset and averaged for each subject time-locked to the onset of the visual cue before calculating a grand average. The interval −100 ms to 0 ms (onset of visual cue) served as baseline for all computations. A baseline close to the onset of the visual cue was chosen to minimize contamination by muscle artifacts related to the preceding trial.

After inspection of the time course of β-activity (Figure 7), we decided to test changes in β-power in the time window of 200–400 ms after onset of the visual cue.

An ANOVA was calculated for the averaged power in the β-band (16–31 Hz) and the 200–400 ms time window using the between-subjects factor group (patients vs. controls) and the within-subjects factors condition (papapa vs. pataka), anteriority of EEG electrodes (with 3 levels: F7/F3/Fz/F4/F8; T7/C3/Cz/C4/T8; P7/P3/Pz/P4/P8) and laterality of EEG electrodes (with 5 levels: F7/T7/P7; F3/C3/P3; Fz/Cz/Pz; F4/C4/P4; F8/T8/P8) using IBM SPSS Statistics (version 23)⁷. Greenhouse-Geisser correction was performed when necessary. To estimate the effect size of ANOVA results, partial η² was calculated.

No significant difference between patients and controls regarding the number of words produced in the verbal fluency test (Regensburger Wortflüssigkeits-Test, Aschenbrenner et al., 2000) was found (16 ± 6 vs. 17 ± 4, p = 0.69). During the EEG experiment, speech latency of included trials was not significantly different between patients and controls (651 ± 145 ms vs. 659 ± 153 ms; p = 0.89).

Raw EEG data of two PD patients are shown in Figure 3. Figure 3A displays the first four papapa trials of a patient whose data were included in the final analysis. By contrast, Figure 3B displays the first four papapa trials of another patient whose data were excluded due to massive muscle artifacts.

Figure 4 illustrates the time-frequency spectra for the papapa condition in the control (A) and the patient group (B). Figure 5 illustrates the time-frequency spectra for the pataka condition in the control (A) and the patient group (B). The time-frequency spectra in Figure 6 show the differences of spectral power between patients and controls for the papapa condition (A) and the pataka condition (B). Figure 7A depicts the averaged time courses of spectral power in the β-band for both experimental conditions (papapa vs. pataka) and groups (patients vs. controls). Figure 7B demonstrates individual changes in β-band power (averaged across the time window of 200–400 ms) for patients vs. controls and for both speech production conditions. The topological distribution of changes in β-power in the time window of 200–400 ms relative to the baseline (−100 to 0 ms) for controls and patients is illustrated in Figure 7C.

In the ANOVA, β-power was significantly higher in PD patients than in controls in the time window of 200–400 ms after the visual cue (significant effect of group; F_(1,27) = 10.90, p = 0.003, partial η² = 0.288). The main effect of condition and the interactions condition × group, condition × anteriority, condition × group × anteriority, condition × laterality, condition × group × laterality, condition × anteriority × laterality and condition × group × anteriority × laterality were not significant (p > 0.05).

Overt speech production is a highly complex task, accomplished by a distributed bilateral cortical and subcortical neural network (for a review, see Kemmerer, 2015). The brain areas that are crucial for the different stages of speech production are well known. Using functional magnetic resonance imaging (fMRI), activation was detected in the primary motor cortex, supplementary motor area, cingulate motor area, thalamus, basal ganglia, insula, temporal lobe and cerebellum during non-lexical speech production (Sörös et al., 2006). The information flow between these areas of the speech production network, in contrast, is only partially understood. Models of speech production propose detailed motor plans that activate dozens of muscles in a well-defined order on a millisecond time-scale and that are constantly updated by auditory and somatosensory feedback (Hickok, 2012).

Multiple lines of evidence, using electroencephalography, magnetoencephalography and electrocorticography, have demonstrated that electrical brain oscillations are crucial for information flow within distributed brain networks (Pfurtscheller and Lopes da Silva, 1999; Fries, 2005; Schnitzler and Gross, 2005; van Wijk et al., 2012; Cheyne, 2013). One of the most widely replicated finding is that preparation for and execution of limb movements are correlated with a gradual decrease of oscillatory power in the β-band, starting about 1 s before movement onset (Jasper and Penfield, 1949; Pfurtscheller, 1981; Leocani et al., 1997; for a review, see Hari and Salmelin, 1997; Cheyne, 2013; Kilavik et al., 2013).

Similar to voluntary limb movements, the preparation of speech production is also associated with a decrease of β-power (Gehrig et al., 2012; Jenson et al., 2014; Mersov et al., 2016). Mersov et al. (2016) studied brain rhythms of fluent speakers and adults who stutter using magnetoencephalography before and during cued overt reading of words in a carrier phrase. During speech preparation, a decrease of β-power (in this study: 15–25 Hz) was found in the bilateral cuneus of fluent individuals. In addition, a decrease of β-power was found in the mouth motor cortex (Mersov et al., 2016). Similarly, Gehrig et al. (2012) used magnetoencephalography to investigate the preparatory phase of overt and covert reading in healthy individuals. The authors found a decrease of β-power in the bilateral parietal lobe and the left articulatory motor region during preparation, beginning about 350 ms after the onset of a visual preparation cue (Gehrig et al., 2012).

Although the dynamics of β-power have been investigated for decades, the functional significance of event-related fluctuations of oscillatory brain rhythms has not been fully elucidated (Cheyne, 2013). The pre-movement decrease of β-power is thought to reflect the preparation for the motor response (Pfurtscheller and Lopes da Silva, 1999; Cheyne, 2013; Kilavik et al., 2013). In contrast, an increase of β-power in the motor cortex preserves the current motor state and inhibits the initiation of new motor plans (Neuper and Pfurtscheller, 2001; Engel and Fries, 2010). Several studies suggest a causal relationship between β-oscillations and motor functions (van Wijk et al., 2012). Stimulation of the subthalamic nucleus at 20 Hz slowed the development of maximal grip force in PD patients relative to healthy controls (Chen et al., 2011). Even in healthy adults, non-invasive transcranial alternating-current stimulation at 20 Hz over the contralateral motor cortex slows voluntary hand movements (Pogosyan et al., 2009).

With the advent of deep brain stimulation for the treatment of PD and the possibility to record local field potentials through the stimulation leads, an increase of β-band oscillations in cortico-basal ganglia circuits at rest (Brittain and Brown, 2014), even at an early stage of the disease (Pollok et al., 2012), has been described. Excessive β-band oscillations are believed to be directly linked to chronic dopamine depletion (Mallet et al., 2008) and are now regarded as neurophysiological signature of PD (Oswal et al., 2013). In PD patients who underwent bilateral implantation of deep brain stimulation electrodes in the subthalamic nucleus and who were off dopaminergic medication, a significant correlation was found between the power of local field potentials in a 8–35 Hz frequency band and clinical severity of PD as assessed by the total UPDRS III score (Neumann et al., 2016). Treatment with dopamine or deep brain stimulation of the subthalamic nucleus has been shown to decrease or even normalize excessive β-band oscillations (Giannicola et al., 2010; Jenkinson and Brown, 2011; Quinn et al., 2015). In PD patients off medication, the onset of pre-movement β-power decrease (Pollok et al., 2012; Meziane et al., 2015) was delayed during the preparation for limb movements. Heinrichs-Graham et al. (2014) used magnetoencephalography to study brain oscillations during planning, execution and termination of a tap of the right index finger in unmedicated PD patients and healthy age-matched controls. PD patients had a significantly smaller decrease of β-power prior and during movement than controls (Heinrichs-Graham et al., 2014).

While cortical and subcortical activity associated with limb movement has been widely investigated in PD, only a small number of studies have examined the neural correlates of speech production in PD. Functional neuroimaging research using positron emission tomography (PET; Pinto et al., 2004) and fMRI (Rektorova et al., 2007; Arnold et al., 2014) found increased activation, mainly in motor areas, in PD patients compared to controls. Overt speech production was associated with increased activation in the supplementary motor area (Pinto et al., 2004), primary orofacial sensorimotor cortex (Rektorova et al., 2007) as well as left dorsal premotor cortex and inferior frontal gyrus (Arnold et al., 2014) in patients vs. controls. For the study of brain oscillations, Hebb et al. (2012) recorded local field potentials in the subthalamic nucleus of PD patients during a continuous speech-language task, including naming the months of the year and counting upward from one. The authors observed a bilateral decrease of β-power preceding and during overt speech production in the subthalamic nucleus (Hebb et al., 2012). A recent study by Wojtecki et al. (2017) recorded subthalamic nucleus activity during a silent word generation paradigm, followed by overt pronunciation of the generated word. During silent word generation, an entirely cognitive task, the authors found a significant increase in α- and θ-power, but not in β- and γ-power. During overt speech, the recordings indicated a small, but insignificant decrease of β-power (Wojtecki et al., 2017).

To the best of our knowledge, an increase of β-power during the preparatory phase preceding movement has not been described before, neither in limb motor nor in speech control. The present study cannot provide definite answers regarding the functional significance of β-activity and its relationship to speech production, but may stimulate hypotheses for future research. Hyperactivity in speech motor areas, as demonstrated by PET and fMRI, has been interpreted as cortical mechanisms to compensate for basal ganglia dysfunctions. After all we know of the function of β-oscillations, an increase of β-power has an anti-kinetic effect (Jenkinson and Brown, 2011) and cannot be regarded as compensatory in nature.

It is important to note that the PD patients studied here were non-demented individuals with preserved verbal fluency. In addition, only fast, correct and intelligible responses have been included in the EEG analysis. Thus, the increase in β-power seen here does not necessarily cause speech production deficits. Moreover, the increase in β-power was seen in PD patients on medication, receiving a mean equivalent L-dopa dose of 677 mg/day. Importantly, the increase of β-band activity in our study was observed in the time window 200–400 ms after presentation of the visual cue, but not immediately preceding speech onset. Overt speech production is a multi-stage process that involves cognitive processes (mental retrieval of phonemes and combination of phonemes to syllables and words) and motor processes (developing an articulatory plan and execution of articulatory movements; Indefrey, 2011). Behavioral and electrophysiological studies delineated the temporal evolution of this cascade of events (Indefrey, 2011). The time window of the β-band increase observed here is dominated by a cognitive process, the retrieval of the phonological code. Planning of the articulatory movements starts later, about 150 ms before onset of articulation (Indefrey, 2011). This time course may explain why our patients had no symptoms of dysarthric speech with normal verbal fluency in the neuropsychological assessment and normal speech latency during the EEG recording. Our results would be compatible with an impairment of phonological processing, which has not been expected before and which was not tested explicitly in this study.

Our study has important limitations. In order to design a short experimental paradigm, feasible for patients with PD, we included only 25 trials per condition. In a future study, we would prefer 50 or more trials per condition to increase the signal-to-noise ratio of the EEG recordings. Moreover, a relatively high number of participants had to be excluded because of ongoing muscle artifacts. In a future study, we would take more time to familiarize the participants with the speech production paradigm and train them to relax their facial and jaw muscles. Regarding the analysis of EEG data, we decided to reference to the average activity of both mastoid electrodes. As the choice of reference may influence spectral power, the use of a neutral reference may have advantages (Yao, 2001; Yao et al., 2005).

In conclusion, the changes of oscillatory brain activity preceding limb movement and speech production differ in PD. While the physiological decrease of β-power preceding movement onset is delayed and smaller in PD patients off medication and normalizes under dopaminergic treatment, speech production is preceded by an increase of β-power in our patients who were on dopaminergic therapy compared to controls. This fundamental difference in rhythmic brain activity may contribute to the poor efficacy of dopamine and deep brain stimulation in PD speech disorder. Future research is warranted to determine the generators and the pathophysiological role of β-oscillations in PD speech disorder. First of all, a comparison of rhythmic brain activity between PD patients with and without speech impairment is needed to establish the relationship between behavioral speech performance and brain oscillations. Second, speech-related brain oscillations should be studied in patients on and off treatment (dopaminergic medication or deep brain stimulation) to gain further insights into the effects of antiparkinsonian treatment on speech-related β-power. Finally, a dedicated study on phonological processing in PD should find out whether the abnormal brain activity seen here influences cognitive processes in speech production and contributes to the pathophysiology of PD speech disorder. The results of these investigations may help to improve the currently often unsatisfactory therapy of this frequent and debilitating symptom of PD, e.g., through modification of deep brain stimulation protocols or further development of non-invasive stimulation techniques.