Woodhead ZVJ, et al. J Neurol Neurosurg Psychiatry 2017;0:1–9. doi:10.1136/jnnp-2016-314621
To cite: Woodhead ZVJ,
Crinion J, Teki S, et al. J
Neurol Neurosurg Psychiatry
Published Online First: [please
include Day Month Year].
1Department of Brain Repair
and Rehabilitation, University
College London, London, UK
2Wellcome Trust Centre for
College London, London, UK
3Department of Experimental
Psychology, University of Oxford,
4Institute of Cognitive
Neuroscience, University College
London, London, UK
5Department of Physiology,
Anatomy and Genetics,
University of Oxford, Oxford, UK
Dr Zoe VJ Woodhead, Wellcome
Trust Centre for Neuroimaging,
University College London, 12
Queen Square, London WC1N
3BG, UK; z. woodhead@ ucl.
Received 05 August 2016
Revised 24 November 2016
Accepted 01 February 2017
Auditory training changes temporal lobe connectivity
in ‘Wernicke’s aphasia’: a randomised trial
Zoe VJ Woodhead,1,2,3 Jennifer Crinion,4 Sundeep Teki,5 Will Penny,2 Cathy J Price,2
Alexander P Leff 2,4
Introduction Aphasia is one of the most disabling se-
quelae after stroke, occurring in 25%–40% of stroke sur-
vivors. However, there remains a lack of good evidence
for the efﬁcacy or mechanisms of speech comprehension
Trial Design This within-subjects trial tested
two concurrent interventions in 20 patients with
chronic aphasia with speech comprehension
impairment following left hemisphere stroke: (1)
phonological training using ‘Earobics’ software and
(2) a pharmacological intervention using donepezil, an
acetylcholinesterase inhibitor. Donepezil was tested in a
double-blind, placebo-controlled, cross-over design using
block randomisation with bias minimisation.
Methods The primary outcome measure was speech
comprehension score on the comprehensive aphasia test.
Magnetoencephalography (MEG) with an established
index of auditory perception, the mismatch negativity
response, tested whether the therapies altered effective
connectivity at the lower (primary) or higher (secondary)
level of the auditory network.
Results Phonological training improved speech
comprehension abilities and was particularly effective
for patients with severe deﬁcits. No major adverse
effects of donepezil were observed, but it had an
unpredicted negative effect on speech comprehension.
The MEG analysis demonstrated that phonological
training increased synaptic gain in the left superior
temporal gyrus (STG). Patients with more severe speech
comprehension impairments also showed strengthening
of bidirectional connections between the left and right
Conclusions Phonological training resulted in a small
but signiﬁcant improvement in speech comprehension,
whereas donepezil had a negative effect. The connectivity
results indicated that training reshaped higher order
phonological representations in the left STG and (in
more severe patients) induced stronger interhemispheric
transfer of information between higher levels of auditory
cortex. Clinical trial registration This trial was registered
with EudraCT (2005-004215-30, https:// eudract . ema.
europa. eu ) and ISRCTN (68939136, http://www. isrctn.
Speech comprehension impairments after stroke
caused by damage to dominant temporoparietal
cortex (ie, 'Wernicke’s aphasia' (WA) and global
aphasia)1–3 are resistant to treatment by conven-
tional methods4; hence, there is a need for new
evidence-based therapies to improve auditory
comprehension in aphasia.
One popular target for WA therapy is the low-level
deficit in auditory phonological analysis.5–7 This
typically involves training auditory discrimina-
tion with phonemes, consonant-vowel-consonant
segments or longer sequences. Previous attempts
have met with mixed results: whereas two case
studies reported positive effects on auditory
discrimination and/or comprehension,8 9 a further
case study10 and case series of eight patients11 failed
to find any significant improvements. We tested the
efficacy of phonological training using Earobics
software12 in a larger sample (n=20).
We also tested whether donepezil, an acetylcho-
linesterase inhibitor, could improve behavioural
outcomes. Studies in rats and bats have shown that
modulating the cholinergic pathway from nucleus
basalis to auditory cortex enhances experience-de-
pendent plasticity,13 whereas suppression of the
cholinergic system reduces it.14 15 As the cholinergic
system is specifically involved in behaviourally
dependent learning,16 cholinergic drugs may be most
effective when paired with behavioural therapy. Two
clinical trials in aphasia by Berthier and colleagues17
18 administered donepezil in conjunction with
conventional speech and language therapy. The
drug improved aphasia quotient scores, particularly
on picture naming. While Berthier and colleagues
studies included patients with aphasia of all types,
we focused on Wernicke’s and global aphasia and
tested whether aphasia severity interacted with the
efficacy of both phonological and donepezil thera-
pies using a within-subject, cross-over design.
As well as behavioural outcome measures, we
employed functional neuroimaging to investi-
gate the neural mechanisms of the therapeutic
effect. Following Teki and colleagues,19 changes
in the auditory cortex’s sensitivity to phonological
contrasts were assayed using the mismatch negativity
response (MMN) recorded with magnetoenceph-
alography (MEG). Participants were habituated to
a repeated spoken syllable (the standard stimulus).
Occasionally, syllables that differed acoustically or
phonologically from the standard stimulus were
presented (the deviant stimuli). The surprise elic-
ited by these unanticipated deviant stimuli results
in a stronger evoked neural response, known as
the MMN.20 By comparing MMN responses with
acoustic and phonemic deviants, we could reveal
when and where neurons were sensitive to the
phonological content of the stimuli.
JNNP Online First, published on March 4, 2017 as 10.1136/jnnp-2016-314621
Copyright Article author (or their employer) 2017. Produced by BMJ Publishing Group Ltd under licence.
2 Woodhead ZVJ, et al. J Neurol Neurosurg Psychiatry 2017;0:1–9. doi:10.1136/jnnp-2016-314621
Dynamic causal modelling (DCM)21–23 was used to investi-
gate how the MMN arises from interactions between multiple
neuronal sources in the temporal lobes. Teki and colleagues
showed that in controls the difference between acoustic and
phonemic deviants was reflected in stronger local adaptation of
representations in the bilateral primary auditory cortex (Heschl’s
gyrus (HG)) and the auditory association cortex (superior
temporal gyrus (STG)). Patients with aphasia showed no such
adaptation in the left HG or STG, but instead showed stronger
feedforward connectivity on the right hemisphere. This study
involved the same participants, MMN paradigm and analysis,
but employed a longitudinal design to test the impact of phono-
logical training and donepezil on connectivity. We hypothesised
that an effective therapy would increase updating of phonolog-
ical representations on the left hemisphere and decrease reliance
on the right hemisphere regions.
MATERIALS AND METHODS
Our objectives were to investigate (1) whether phonological
training improved speech comprehension, (2) whether donepezil
facilitated phonological training effects and (3) the impact of the
therapies on directed connectivity within the auditory network.
A randomised, double-blind, placebo-controlled cross-over
design was used (figure 1). Each participant received four 5-week
blocks of treatment:
►drug only (5 mg daily dose of donepezil)
►drug and Earobics (10 mg daily dose of donepezil, plus
two 40 min daily sessions of Earobics)
►placebo and Earobics.
Participants were assessed at baseline, then before (1) and
after (2) phonological training, in both drug (D) and placebo
(P) conditions, leading to five distinct assessment time points:
baseline, D1, D2, P1 and P2. This design allowed a factorial
analysis of behavioural and effective connectivity outcome
measures, comparing performance before (D1 and P1) versus
after phonological training (D2 and P2), and when participants
were receiving donepezil (D1 and D2) versus placebo (P1 and
P2). There were no predetermined criteria for termination of the
trial or the definition of outliers.
Participants were assigned to one of two cross-over groups
(drug then placebo or placebo then drug). Block randomisation
with bias minimisation was used: a computer algorithm gener-
ated a random number to allocate each new participant to a
group. An independent researcher ensured that the cross-over
groups did not become unbalanced by more than four patients.
Participants, caregivers and researchers were blind to block allo-
cation. Randomisation codes were held by the study pharmacist.
Tablets were supplied in bottles labelled with the block number
when they should be taken (1–4). Each bottle contained 35 indis-
tinguishable lactose encapsulated tablets of placebo, 5 mg done-
pezil hydrochloride or 10 mg donepezil hydrochloride.
Outcome measures were prospectively selected. The primary
behavioural outcome measure was score on the comprehensive
aphasia test (CAT)24 speech comprehension scale (including
word, sentence and paragraph comprehension; maximum
score=66, aphasia cut-off=56), which was chosen for its high
ecological validity. Other CAT measures (written comprehen-
sion, speech repetition, naming, reading and writing) and a
sustained attention to response test (SART)25 were the secondary
outcomes. The primary neuroimaging outcome was the effect of
phonemic deviants on effective connectivity within the auditory
The participants were patients with chronic post-stroke aphasia.
As the study required approval from the Medicines and Health-
care Products Regulatory Agency, which deemed the study a
clinical trial of an investigational medicinal product, the power
calculation was based on the drug effect. The best evidence
available at the time of designing the study (from Berthier and
colleagues, who used donepezil to treat post-stroke aphasia),17
indicated that 20 patients would be required to detect a 5%
difference given a 7% within-person SD, 5% significance level
and 90% power. Data collection occurred from November 2006
to August 2009 and ended when 20 patients had completed the
protocol. Twenty-seven patients were enrolled, three withdrew
after the baseline time point due to the trial’s time demands and
a further four were excluded from the analysis as extensive left
auditory cortex damage made them unsuitable for the DCM
analysis (figure 2). We report the results from the remaining 20
Figure 1 Study design showing order of the ﬁve time points (baseline, D1, D2, P1 and P2) for the two cross-over groups.
Woodhead ZVJ, et al. J Neurol Neurosurg Psychiatry 2017;0:1–9. doi:10.1136/jnnp-2016-314621
patients. Baseline data from all 20 participants were previously
reported by Teki and colleagues.19
All participants had left hemisphere stroke and normal
hearing (three female; mean age (range)=62.4 (43–90) years;
average time since stroke=3.3 (0.6–8.6) years; average lesion
volume=127.3 (24.2–403.6) cm3). The demographics, lesion
details and baseline behavioural scores are shown in table 1. The
behavioural inclusion criteria were speech production impair-
ment in CAT repetition and/or CAT naming, and speech compre-
hension impairment in CAT spoken word comprehension, CAT
spoken sentence comprehension and/or a custom-made vowel
During recruitment, it was assumed (wrongly as it turned out)
that the participants’ scores on the main outcome measure would
be normally distributed. It became apparent after the study
closed that they fell into two groups (moderate and severe), as
discussed in a previous report by Schofield and colleagues.26
None of the participants had the sample mean as their baseline
score. A cluster analysis confirmed this bimodal distribution. We
then treated the two groups as a factor in all further analyses,
including when testing the main hypothesis. Group membership,
moderate or severe, is listed in table 1.
The participants were classified into aphasia subtypes
depending on their speech production abilities. Those whose
object naming and repetition total scores placed them in the
bottom quartile for subjects with aphasia in the Predicting
Language Outcome and Recovery After Stroke (PLORAS) data-
base27 were classed as global (G); otherwise we classified them
as Wernicke’s aphasics (W). The majority of severe patients (six
out of eight) had global aphasia (see table 1). A χ2 test showed no
significant differences between the two therapy groups in terms
of aphasia subtype (c2(1, n=20)=0.74, p=0.65).
All participants provided written informed consent according
to the Declaration of Helsinki. Data were collected at the Insti-
tute of Neurology, University College London, and the study
was approved by the Joint Research Ethics Committee of the
National Hospital for Neurology and Neurosurgery and the
Institute of Neurology, University College London.
The phonological training software was Earobics version 1
for adolescents and adults.12 Earobics cycles through six inde-
pendently adaptive tasks: (1) ‘memory matrix’, environmental
sound-to-picture matching to train auditory short-term memory;
(2) ‘sound check’, two-alternative forced-choice task to probe
grapheme-to-phoneme mapping; (3) ‘get rhythm’, non-speech
and speech sounds to probe auditory segmentation; (4) ‘connec-
tivity’, compound word-to-picture matching to train phonological
blending; (5) ‘rhyme time’, rhyme detection task using sequences
of increasing numbers of words and (6) ‘same-different’, auditory
discrimination using phoneme and word pairs. The participants
were asked to complete 10 hours of training per week over each
5-week training block and record training duration in self-report
diaries with some help from their partner or carer.
Figure 2 Consolidated Standards of Reporting Trials (CONSORT) trial ﬂow diagram.
4 Woodhead ZVJ, et al. J Neurol Neurosurg Psychiatry 2017;0:1–9. doi:10.1136/jnnp-2016-314621
Drug administration and monitoring
Donepezil dosage was determined according to the British
National Formulary guidelines, starting at 5 mg for the first
5-week block. If this was tolerated, dose was escalated to 10 mg
for the second block. All patients except one (patient 17) esca-
lated to the higher dose. Adverse effects were monitored at every
time point using a checklist of all possible adverse events (see
online supplementary table).
Behavioural outcome measures
The CAT24 was administered at every time point as shown in
figure 1. Only data from D1, D2, P1 and P2 time points were
entered into the statistical analysis. Domain-general (non-verbal)
effects of donepezil were tested using a modified version of the
SART, a Go/No-Go task loading on sustained attention,25 presented
with E-Prime28 and administered at baseline, D2 and P2 sessions
only. The SART outcome measures included average reaction time
for correct Go trials, accuracy of response inhibition for No-Go
trials and post-error slowing. We also collected carer-reported
measures of activity and participation from the Social Commu-
nication and Daily Planning from the Functional Assessment of
Communication Skills for Adults (ASHA FACS)29 at all time points.
A T1-weighted structural brain image with 1 mm3 voxel size was
acquired with a Siemens Sonata 1.5T MRI scanner at baseline.
The Automated Lesion Identification toolbox30 for SPM831 iden-
tified lesion location and volume.
MEG and electroencephalogram data acquisition
The evoked response potentials were collected at D1, D2, P1 and
P2 time points. MEG was used in 18 subjects, on a CTF systems
274-channel whole-head MEG scanner with third-order axial
gradiometres and a sampling rate of 480 Hz. Electroencepha-
logram (EEG) was used in two subjects (patients 8 and 19) due
to MEG artefacts. EEG data were acquired with a high-density,
128-channel Bio-Semi headcap system, with a sampling rate of
MEG/EEG experimental paradigm and stimuli
Spoken auditory stimuli were presented binaurally in a passive
oddball paradigm, with the standard stimulus ‘Bart’, an acoustic
deviant, and two phonemic deviants ‘burt’ and ‘beat’. The deviant
stimuli were created by varying the frequencies of the standard’s
first and second formants, as described by Teki and colleagues.19
At each time point, four stimuli blocks were presented, each
containing 120 presentations of the standard and 30 presenta-
tions of each deviant in a pseudo-randomised order. Stimulus
onset asynchrony was 1080 ms. Stimulus amplitude was initially
set at 60 dB/sound pressure level and adjusted prior to scanning
to a comfortable level.
The participants were instructed to attend to an incidental
visual detection task while ignoring the auditory stimuli. Pictures
of outdoor scenes were presented for 60 s, followed by a circle
(92% of trials) or square (8% of trials) for 1.5 s. The partici-
pants responded by button press to the circle and withheld their
response for the square. Average response accuracy was 87%,
Table 1 Demographics, lesion details and baseline behavioural performance of the patients with aphasic stroke
ID Group Sex
% of ROI
damaged Speech comprehension Speech production
1 1 M L M G 69.6 1.0 I 69.5 1.8 0.0 29 28 27 28 6
2 2 M R M W 62.7 1.2 I 42.4 7.2 21.1 30 22 38 34 38
3 1 M R M W 63 8.6 I 429.3 81.0 43.3 27 21 17 50 22
4 1 M L M W 61.5 7.6 I 314.0 52.9 0.0 23 25 16 38 12
5 1 M R M W 67.8 7.4 H 64.3 8.1 30.5 29 21 37 50 30
6 2 M R M W 60.5 5.6 I 161.1 52.7 9.3 27 30 34 54 27
7 1 M R M W 64.9 1.2 I 171.1 39.7 17.9 27 21 38 60 33
8 1 M R S G 61.4 1.9 I 242.9 52.6 52.6 24 6 16 0 0
9 2 F R S G 66.5 5.3 I 195.2 42.3 21.4 21 15 13 23 12
10 1 M R M W 61.5 3.4 M-L* 1.6 0.0 0.0 26 26 40 63 36
11 1 M R S G 63.3 0.6 I 69.0 8.3 33.5 20 13 12 0 0
12 2 F R S G 43.5 1.3 I 69.7 68.7 46.1 14 17 17 18 14
13 2 F R S W 46.3 0.7 I 31.7 60.9 15.3 26 12 26 26 41
14 2 M R S G 71.1 5.1 I 151.1 79.5 29.8 20 14 7 0 0
15 2 M R S G 62.4 3.7 I 168.6 20.2 0.0 22 11 21 30 0
16 1 M R M W 60.9 3.0 H 136.6 11.7 1.5 28 26 40 68 37
17 1†M R M G 45.4 3.7 I 61.2 5.8 16.7 27 21 39 31 18
18 2 M R M W 74.7 0.6 I 41.1 7.3 2.9 24 26 24 45 14
19 2 M R M G 50.2 1.8 I 280.5 29.5 0.9 25 26 39 28 2
20 1 M R M W 90.3 3.7 I 62.9 14.2 0.0 27 21 34 68 43
Behavioural measures included CAT spoken word comprehension (score cut-off for impaired performance=25), CAT spoken sentence comprehension (cut-off=27), vowel
identiﬁcation (36), CAT total repetition score (67) and CAT object naming (43). Values highlighted in bold indicate scores below the threshold for normal performance.
*Patient 10 had four small lacunes in the left hemisphere: (1) superior and lateral occipital lobe (I); (2) deep to the superior frontal sulcus (I); (3) superior longitudinal fasciculus
(L) and (4) inferomedial thalamus (L).
†Patient 17 failed to escalate to 10 mg of donepezil in the second drug block and remained on 5 mg for both blocks.
CAT, comprehensive aphasia test; I, infarct; H, haemorrhagic; M-L, multilacune; ROI, region of interest; STG, superior temporal gyrus.
Woodhead ZVJ, et al. J Neurol Neurosurg Psychiatry 2017;0:1–9. doi:10.1136/jnnp-2016-314621
with no significant differences in the number of errors (false-pos-
itives or false-negatives) as an effect of time (main effect of
chronological order of testing sessions) or due to drug adminis-
tration (main effect of drug vs placebo).
Preprocessing of MEG data in SPM831 included the following:
high-pass filtering (1 Hz), eye-blink artefact removal using
multiple source eye correction,32 epoching (−100 to 500 ms
peristimulus time), baseline correction (−100 to 0 ms prestim-
ulus time window), low-pass filtering (30 Hz), merging the four
blocks of each dataset, robust averaging and low-pass filtering
again. Preprocessing of EEG data in SPM1231 33 included the
following: high-pass filtering (1 Hz), epoching (−100 to 500 ms
peristimulus time), merging the four runs of each dataset, identi-
fication of blinks from vertical electro-oculography data, correc-
tion of blink artefacts using single value decomposition and
signal-space projection, robust averaging and low-pass filtering
(30 Hz). (Note: Low-pass filtering at 30 Hz would have removed
any time-locked effects in the gamma range from our analyses.)
Source localisation using the variational-Bayesian equivalent
current dipoles34 is described in the online supplementary material.
In brief, this compared different source models, containing five
possible sources identified from previous auditory MMN analyses
(left and right HG, left and right STG and right IFG) in different
combinations. The winning model, containing left and right HG
and STG only (p=0.88), was used as the spatial model for the DCM
analysis. All patients had at least 10% intact cortex within each
anatomical region of interest. All fitted sources fell within intact
cortex, not lesion (see online supplementary material for detail).
Dynamic causal modelling
DCM was conducted on preprocessed evoked responses to stan-
dard and deviant stimuli at each time point separately (D1, D2,
P1 and P2). Evoked responses from 1 to 400 ms peristimulus time
were tapered with a Hanning window. The DCM spatial model
used Equivalent Current Dipole (ECD) sources from bilateral HG
and STG dipole locations (see online supplementary material). The
neural model included exogenous auditory stimulation to the left
and right HG nodes at 60 ms; endogenous connectivity evoked by
standard stimuli in two forward, two backward and four lateral
connections between HG and STG sources (diagonal connections
were not modelled because although heterotopic connections are
present in auditory cortex they are sparse)35 ; and three modula-
tory effects evoked by each deviant stimulus compared with the
standard stimulus. A large model space of 2^8 (256) models was
estimated, each with modulatory effects modelled in a different
combination of the eight independent inter-regional connections.
Modulatory effects on self-connections, modelling a region’s
sensitivity to inputs,36 were present in all models.
Bayesian model averaging
For each connection, Bayesian Model Averaging (BMA) with
random effects37 was used at the subject level to average the
modulation caused by each deviant versus the standard at each
time point (D1, D2, P1 and P2). The difference in modulation
by phonemic versus acoustic deviants (phonemic sensitivity)
was calculated by averaging the modulatory effects for the two
phonemic deviants and subtracting the modulatory effect for the
The phonemic sensitivity values were entered into a repeat-
ed-measures analysis of variance (ANOVA) for each connec-
tion. The ANOVAs had within-subjects variables time (before/
after phonological training) and drug (on drug/placebo), and
between-subjects variables severity (moderate/severe groups)
and cross-over order (drug/placebo first).
Baseline language performance and structural MRI
Exploratory analysis of the baseline CAT speech comprehen-
sion scores using a two-step cluster analysis identified a division
between participants with moderate (n=13) and severe (n=7)
impairments (figure 3). Independent samples t-tests confirmed that
moderate and severe subgroups differed in speech comprehension
ability (t(18)=7.77, p<0.001), as well as in written comprehen-
sion, repetition, naming, reading and writing abilities (all p<0.05).
The subgroups did not differ significantly in age (p=0.35) or time
since stroke (p=0.43). Lesion overlay maps (figure 3) showed the
subgroups shared broadly similar lesion distributions, centring on
the left perisylvian cortex and superior longitudinal fasciculus.
The lesion volume was not significantly different between groups
Phonological training dose
Out of 40 completed training blocks (2 blocks × 20 partici-
pants), 28 self-report diaries were fully completed (70%). Of
those 28 diaries, the average training dose was 36 hours 38 min
per block (SD=16 hours 8 min; range: 6 hours 20 min to 65 hours
5 min; median=36 hours 28 min). Fifteen of the 28 diaries were
from blocks 1 and 13 were from Block 2. The mean training
dose for block 1 and block 2 did not differ significantly (block
1: mean=38 hours 58 min; block 2: mean=36 hours 26 min;
Compliance and tolerability of cholinergic drug therapy
Patients were required to return their medication bottles after
completing each block. Eighty-eight per cent of medication
bottles were returned. In these bottles, only 2.5% of unused
The participants completed an adverse effects report form at
each time point (see online supplementary table). Three partic-
ipants had incomplete report forms at one time point; hence,
reports were analysed for 17 participants only. We compared
the total number of adverse events when the participants were
on drug (D1 or D2) or placebo (P1 or P2). The participants
reported 32 adverse events when on drug and 28 on placebo. A
non-parametric sign test showed no significant difference in the
frequency of adverse events on drug versus placebo (p=0.79).
The most frequent adverse events were insomnia, head-
aches, dizziness and muscle cramps. For each type of event, the
frequency of occurrence on drug or placebo was compared using
the McNemar tests with binomial distribution. No significant
differences were observed for any type of event (all p>0.2).
Therapy effects on behavioural outcome measures: primary
The effects of phonological training and donepezil on CAT speech
comprehension scores were assessed using a mixed effects ANOVA
with two within-subjects variables: (1) time (before/after phonolog-
ical training) and (2) drug (on drug/on placebo). Two between-sub-
jects variables were also included: (1) aphasia severity (moderate/
severe) and (2) cross-over group (drug first/placebo first). The
results (figure 4) demonstrated that phonological training signifi-
cantly improved speech comprehension, indicated by a significant
main effect of time (F(1, 16)=6.56, p<0.05). Conversely, there
was a significant main effect of drug, with (unexpectedly) lower
6 Woodhead ZVJ, et al. J Neurol Neurosurg Psychiatry 2017;0:1–9. doi:10.1136/jnnp-2016-314621
scores on drug than placebo (F(1, 16)=11.60, p<0.005). There
was no interaction between phonological training and donepezil:
the effects were independent and in opposite directions. Time by
severity and drug by severity interactions were significant (F(1,
16)=6.6, p<0.05 and F(1, 16)=4.9, p<0.05, respectively): both
therapy effects were larger in the severe patient group.
Therapy effects on behavioural outcome measures: secondary
Among the other CAT measures (written comprehension, speech
repetition, naming, reading and writing), the only notable
effect was a significant time by severity interaction for written
comprehension (F(1, 16)=8.56, p<0.05), driven by larger bene-
ficial effects of phonological training in severe patients compared
with moderate patients. There was a non-significant trend towards
better naming on drug than placebo (F(1, 16)=3.64, p=0.075),
consistent with prior observations by Berthier and colleagues.17 18
There were no significant changes in the SART or ASHA FACS
scores. Repeated-measures ANOVAs demonstrated no signifi-
cant effects of time, drug, severity or cross-over group.
Therapy effects on effective connectivity
MMN responses to acoustic and phonemic deviants are shown
in the online supplementary figure 1.
Figure 3 Baseline Comprehensive Aphasia Test data and structural brain imaging. Top: average T-scores for severe (n=7) and moderate (n=13) patients.
Bottom left: speech comprehension T-scores, showing division of severe and moderate subgroups. Bottom right: lesion overlay maps for severe and moderate
Figure 4 Effects of Earobics and donepezil on speech comprehension in severe and moderate patient subgroups.
Woodhead ZVJ, et al. J Neurol Neurosurg Psychiatry 2017;0:1–9. doi:10.1136/jnnp-2016-314621
Modulation of connection strength by phonemic versus acoustic
deviants (phonemic sensitivity) changed as a result of phonological
training (figure 5). The main effects of phonological training were
significant in two connections (figure 5A): (1) the left STG self-con-
nection (F(1, 16)=5.30, p<0.05) and (2) the left HG to left STG
connection (F(1, 16)=11.3, p<0.005). Three connections showed
time by severity interactions driven by stronger training effects in
the severe group (figure 5B): (1 ) the left HG to left STG forward
connection (F(1, 16)=8.30, p<0.05) (2) the left STG to right STG
lateral connection (F(1, 16)=6.65, p<0.05) and (3) the right STG
to left STG lateral connection (F(1, 16)=8.64, p<0.01).
Only one connection, the left HG to STG, showed a main
effect of drug (F(1, 16)=20.70, p<0.001) due to stronger
phonemic sensitivity on drug versus placebo (figure 5C). A
drug by severity interaction (figure 5D) showed that this effect
was larger for severe than moderate patients (F(1, 16)=9.79,
We hypothesised that phonological training would improve
speech comprehension and that donepezil would facilitate this
effect by enhancing the neuroplastic response to training. Only
the first hypothesis was supported: patients showed significantly
better speech comprehension after phonological training, but
worse comprehension on drug than placebo. Both effects were
stronger in more severely impaired patients: the severe subgroup
responded better to training and worse to drug than the moderate
subgroup. The effect of phonological training on comprehen-
sion was significant but clinically small: the average CAT speech
comprehension score during the placebo block increased from
52.0 to 53.2 after 5 weeks of training. The average dose was
36 hours per block. Lower doses of phonological therapy in
previous studies (6–12 hours) may explain why significant effects
have not been consistently observed.10 11 Although the effect of
phonological training was small, it is of clinical importance as
therapeutic interventions in patients with aphasia with severe
impairments of speech perception have been largely written
off, both in textbooks (‘(global aphasia is) sometimes known as
irreversible aphasia syndrome’38) and in systematic reviews (‘No
evidence of benefit’39).
The behavioural effect of phonological training was gener-
alised in two ways: first, training with Earobics stimuli gener-
alised to different stimuli used in CAT speech comprehension
tests and second, written comprehension improved, at least for
severe patients. This latter result suggests that either auditory
training acted on amodal, higher level semantic representations
or that strengthening auditory phonological representations
also facilitates grapheme-to-phoneme translation required for
written comprehension. A third option, that training improved
non-linguistic, domain-general cognitive abilities, is less likely as
there were no improvements on sustained attention.
The surprising result that speech comprehension was worse
on donepezil than placebo contrasts with improved Western
Aphasia Battery Aphasia Quotients40 was reported by Berthier
Figure 5 Signiﬁcant changes in phonemic sensitivity between time points. (a) Red connections showed signiﬁcantly stronger phonemic sensitivity after
Earobics training (main effect of Earobics); (b) red connections showed signiﬁcantly stronger training effects on phonemic sensitivity in the severe patients
than the moderate patients (Earobics by severity interaction); (c) main effect of drug and (d) drug by severity interaction.
8 Woodhead ZVJ, et al. J Neurol Neurosurg Psychiatry 2017;0:1–9. doi:10.1136/jnnp-2016-314621
and colleagues.17 18 Berthier only observed task-specific improve-
ments for picture naming. We saw a trend towards better
naming on drug than placebo, perhaps suggesting that donepezil
is better suited to treating speech production than comprehen-
sion. Husain and Mehta41 observe that cognitive-enhancement
drugs can have opposing effects on different tasks as an inverted
U-shaped relationship applies between neuropharmacological
enhancement and performance. Hence, if acetylcholine stimu-
lation was already high in auditory cortex, increasing it further
with donepezil could have impaired performance.
The effective connectivity (DCM) analysis examined the inter-
actions between auditory areas. We predicted that improved
speech comprehension would result from a tuning of phono-
logical representations in auditory cortex. In DCM such tuning
is expressed as a region’s self-connection, which acts as a gain
mechanism36: that is, a post-therapy increase in a self-connec-
tion’s strength would mean that the region had become more
sensitive to phonological contrasts. The results confirmed our
hypothesis: improved speech comprehension after phonolog-
ical training was associated with stronger modulation of the left
STG’s self-connection by phonological but not acoustic devi-
Phonological training also strengthened phonological sensi-
tivity of the forward connection from left HG to STG, suggesting
that updating of higher order representations of speech sounds
in the left STG was driven by stronger feedforward prediction
errors. The observation that phonological modulation of this
connection was also affected by drug (which did not result in
a behavioural improvement) suggests that strengthening this
forward connection alone is insufficient to improve speech
Our findings were not confined to the left hemisphere. In
severe patients, training led to stronger phonemic sensitivity in
the STG interhemispheric connections. This implies that after
severe left temporal damage, the left STG requires more support
from its right hemisphere homologue to perform phonolog-
ical discrimination. The baseline data in these patients treated
as a single group19 showed that modulation strength of the left
to right STG connection correlated negatively with phonemic
discrimination ability; that is, there was a decoupling of these two
regions in patients with better speech perception. It is not clear
whether this decoupling represented (1) the best the damaged
system could currently manage, with the left and right STGs
maintaining their specialised functions but no longer working
in tandem or (2) a maladaptive response to left-hemisphere
damage that may serve to limit further recovery, as, presumably,
auditory representations in the right hemisphere do not share
the linguistic specialisation of the left hemisphere. The longitu-
dinal data presented here suggest a more complex interpretation
because severity effects indicated (1) moderately affected patients
do not have impaired interhemispheric connectivity and therapy
effects are seen within their dominant temporal lobe and (2) in
more severely affected patients interhemispheric processing of
phonemic stimuli is abnormal, but can still be improved by audi-
In conclusion, our results demonstrate that phonological
training worked at a behavioural level, but the underlying neuro-
anatomical mechanism varied with aphasia severity. The left STG
mediated therapy effects in both patient groups: in the moderate
group, this occurred independently from the influence of right
temporal regions, whereas in the severe group interhemispheric
interactions were also involved in promoting recovery of speech
Funding This work was supported by the Wellcome Trust and the James S McDon-
nell Foundation (conducted as part of the Brain Network Recovery Group initiative).
APL and ST were supported by personal fellowships from the Wellcome Trust
(ME033459MES and 106084/Z/14/Z, respectively).
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.
Open Access This is an Open Access article distributed in accordance with the
terms of the Creative Commons Attribution (CC BY 4.0) license, which permits
others to distribute, remix, adapt and build upon this work, for commercial use,
provided the original work is properly cited. See: http:// creativecommons. org/
licenses/ by/ 4. 0
© Article author(s) (or their employer(s) unless otherwise stated in the text of the
article) 2017. All rights reserved. No commercial use is permitted unless otherwise
1 Wernicke C. Der aphasiche symptomenkomplex: eine psychologische studie auf
anatomischer basis. Breslau: Cohen and Weigert, 1874.
2 Eggert GH. Wernicke’s works on aphasia: a sourcebook and review. The Hague:
Mouton Publishers, 1997.
3 Goodglass H, Kaplan E, Barresi B. The assessment of aphasia and related disorders.
3rdedn. Baltimore, MD: Lippincott Williams & Wilkins, 2001.
4 Brady MC, Kelly H, Godwin J, et al. Speech and language therapy for aphasia
following stroke. Cochrane Database Syst Rev 2012;(5):1–145.
5 Luria AR. Basic problems of neurolinguistics. The Hague: Mouton & Co BV, 1976.
6 Robson H, Grube M, Lambon Ralph MA, et al. Fundamental deﬁcits of auditory
perception in Wernicke's aphasia. Cortex 2013;49:1808–22.
7 Whitworth A, Webster J. Howard D. A cognitive neuropsychological approach to
assessment and intervention in aphasia: a clinician’s guide. 2ndedn. Hove andNew
York: Psychology Press, 2013.
8 Morris J, Franklin S, Ellis AW, et al. Remediating a speech perception deﬁcit in an
aphasic patient. Aphasiology 1996;10:137–58.
9 Tessier C, Weill-Chounlamountry A, Michelot N, et al. Rehabilitation of word
deafness due to auditory analysis disorder. Brain Inj 2007;21:1165–74.
10 Maneta A, Marshall J, Lindsay J. Direct and indirect therapy for word sound deafness.
Int J Lang Commun Disord 2001;36:91–106.
11 Woolf C, Panton A, Rosen S, et al. Therapy for auditory processing impairment in
aphasia: an evaluation of two approaches. Aphasiology 2014;28:1481–505.
12 Houghton Mifﬂin Harcourt Learning Technology. http://www. earobics. com/ demos/
13 Kilgard MP, Merzenich MM. Cortical map reorganization enabled by nucleus basalis
activity. Science 1998;279:1714–8.
14 Ji W, Gao E, Suga N. Effects of acetylcholine and atropine on plasticity of central
auditory neurons caused by conditioning in bats. J Neurophysiol 2001;86:211–25.
15 Conner JM, Culberson A, Packowski C, et al. Lesions of the basal forebrain
cholinergic system impair task acquisition and abolish cortical plasticity associated
with motor skill learning. Neuron 2003;38:819–29.
16 Ramanathan D, Tuszynski MH, Conner JM. The basal forebrain cholinergic system
is required speciﬁcally for behaviorally mediated cortical map plasticity. J Neurosci
17 Berthier ML, Green C, Higueras C, et al. A randomized, placebo-controlled study of
donepezil in poststroke aphasia. Neurology 2006;67:1687–9.
18 Berthier ML, Hinojosa J, Martín MC, et al. Open-label study of donepezil in chronic
poststroke aphasia. Neurology 2003;60:1218–9.
19 Teki S, Barnes GR, Penny WD, et al. The right hemisphere supports but does not
replace left hemisphere auditory function in patients with persisting aphasia. Brain
20 Näätänen R, Gaillard AWK, Mäntysalo S. Early selective-attention effect on evoked
potential reinterpreted. Acta Psychologica 1978;42:313–29.
21 David O, Kiebel SJ, Harrison LM, et al. Dynamic causal modeling of evoked responses
in EEG and MEG. NeuroImage 2006;30:1255–72.
22 Kiebel SJ, David O, Friston KJ. Dynamic causal modelling of evoked responses in EEG/
MEG with lead ﬁeld parameterization. NeuroImage 2006;30:1273–84.
23 Kiebel SJ, Garrido MI, Moran R, et al. Dynamic causal modeling for EEG and MEG.
Hum Brain Mapp 2009;30:1866–76.
24 Swinburn K, Porter G, Howard D. The comprehensive aphasia test. Hove: Psychology
25 Robertson IH, Manly T, Andrade J, et al. 'Oops!': performance correlates of everyday
attentional failures in traumatic brain injured and normal subjects. Neuropsychologia
26 Schoﬁeld TM, Penny WD, Stephan KE, et al. Changes in auditory feedback
connections determine the severity of speech processing deﬁcits after stroke. J
27 Price CJ, Seghier ML, Leff AP. Predicting language outcome and recovery after stroke:
the PLORAS system. Nat Rev Neurol 2010;6:202–10.
Woodhead ZVJ, et al. J Neurol Neurosurg Psychiatry 2017;0:1–9. doi:10.1136/jnnp-2016-314621
28 Schneider W, Eschman A. Zuccolotto A. E-Prime reference guide. Pittsburgh:
Psychology Software Tools Inc, 2002.
29 Frattali CM, Thompson CK, Holland AL, et al. Functional assessment of
communication skills for adults (ASHA FACS). Rockville, MD: American Speech-
Language-Hearing Association, 1995.
30 Seghier ML, Ramlackhansingh A, Crinion J, et al. Lesion identiﬁcation using
uniﬁed segmentation-normalisation models and fuzzy clustering. NeuroImage
31 . Wellcome Trust Centre for Neuroimaging; http://www. ﬁl. ion. ucl. ac. uk/ spm/
32 Berg P. Scherg M .A multiple source approach to the correction of eye artefacts.
Electroencephalogr Clin Neurophysiol 1994;90:229–41.
33 Litvak V, Mattout J, Kiebel S, et al. EEG and MEG data analysis in SPM8. Comput
Intell Neurosci 2011;2011:1–32.
34 Kiebel SJ, Daunizeau J, Phillips C, et al. Variational Bayesian inversion of the
equivalent current dipole model in EEG/MEG. NeuroImage 2008;39:728–41.
35 Lee CC, Winer JA. Connections of cat auditory cortex: II. Commissural system. J
Comp Neurol 2008;507:1901–19.
36 Kiebel SJ, Garrido MI, Friston KJ. Dynamic causal
modelling of evoked responses: the role of intrinsic connections. NeuroImage
37 Penny WD, Stephan KE, Daunizeau J, et al. Comparing families of dynamic causal
models. PLoS Comput Biol 2010;6:e1000709.
38 Crystal D. The Cambridge Encyclopedia of Language. 2nd edn. Cambridge:
Cambridge University Press, 2010.
39 Brady MC, Kelly H, Godwin J, et al. Speech and language therapy for aphasia
following stroke. Cochrane Database Syst Rev 2016;(6):CD000425.
40 Kertesz A. Western aphasiabattery. San Antonio, TXCorporation
41 Husain M, Mehta MA. Cognitive enhancement by drugs in health and disease.
Trends Cogn Sci 2011;15:28–36.