40/35
35.2
± 6.5 [20; 50]
39.1
± 8 [24; 58]
-
(1
,73) 5.49
0.02
Matrix [1; 19]
36/37
13.1
± 2.4 [7; 18]
9.9
± 2.5 [5; 15]
-
(1
,71) 31.63
<0.001
Benton (/54)
39/38
46.9
± 3.5 [39; 52]
45
± 3.2 [40; 53]
-
(1
,75) 5.96
0.017
VOSP SDS (/20)
-/38
-
19.7
± 0.6 [18; 20]
-
-
-
VOSP PD (/20)
-/38
-
19.7
± 0.6 [18; 20]
-
-
-
VOSP NL (/10)
-/38
-
9.1
± 0.9 [7; 10]
-
-
-
LARS [-15; 15]
-/37
-
-11.8
± 2.6 [-15; -7]
-
-
-
Disease duration (years)
-/40
-
9.7
± 5.3 [1; 20]
-
-
-
Worst affected side (L/R)
-/40
-
17/23
-
-
-
LEDD (mg/day)
-/40
-
1043.2
± 454.2 [250; 2355]
-
-
-
UPDRS III
1
(/108)
37/32
-
11.9
± 8.9 [1; 33]
30.5
± 12.6 [14; 62.5]
(1
,67) 50.31
<0.001
Hoehn & Yahr
1
(/5)
38/39
-
1.2
± 0.7 [0; 3]
2.1
± 1 [1; 5]
(1
,75) 23.43
<0.001
N = data available for each group: healthy participants/PD patients.
1
number of patients assessed during the ON and OFF DRT states: ON/OFF. VOSP SDS, PD and ND = Shape Detection Screening, Position Discrimination
and Number Location subtests from the Visual and Object Space Perception battery (VOSP). df = degrees of freedom. For the participants who could not be
assessed on the Matrix test, a minimum score of 130 on the Mattis Dementia Rating Scale [
41
] was nonetheless used to ascertain the absence of cognitive
dysfunctions.
doi:10.1371/journal.pone.0160329.t001
Fig 2. Decoding accuracy scores of the healthy controls and the PD patients whatever the emotion
displayed (all) and as a function of emotion. Mean
± standard error and boxplot.
doi:10.1371/journal.pone.0160329.g002
Facial Mimicry in Parkinson's Disease
PLOS ONE | DOI:10.1371/journal.pone.0160329
July 28, 2016
6 / 20
Confounding emotions
When the emotional nature of the error made was examined, analyses showed that the PD
patients provided globally a similar pattern of confusion than the HC. However, quantitative
and qualitative differences appeared for some aspects (
Fig 3
). For angry avatars, the confound-
ing emotion was mostly surprise, then fear and disgust, in both groups. For happy avatars, the
confounding emotion was quasi-systematically surprise in both groups. For neutral avatars,
the confounding emotion was sadness in both groups, then surprise among the PD patients.
The group x non-target emotion interactions were statistically significant for happy (
χ² =
12.06, df = 5, p = 0.034) and neutral (
χ² = 25.21, df = 5, p<0.001) avatars but not for angry
avatars (
χ² = 2.6, df = 5, p = 0.76). Surprise was more often selected by the PD patients than
the HC for happy avatars (
χ² = 15.33, df = 1, p<0.001) and neutral avatars (χ² = 21.25, df = 1,
p
<0.001). For neutral avatars, the PD patients also selected more often sadness than the HC
did (
χ² = 17.06, df = 1, p<0.001).
Facial reactions
A statistically significant group x emotion x muscle x interval interaction effect was found (
χ² =
431.65, df = 76, p
<0.001). In the HC, comparisons across emotions evidenced specific varia-
tions in response to the emotion for each muscle (
Fig 4A
,
S2 Table
). From 400 ms after stimu-
lus onset, corrugator activity decreased in response to expressions of joy whereas it increased in
response to angry faces and showed an intermediate pattern in response to neutral expressions.
Conversely, from 500 ms after stimulus onset, zygomaticus activity increased in response to
expressions of joy whereas it remained quite stable or decreased slightly in response to angry or
neutral avatars. The same applied to the orbicularis muscle from 700 ms after stimulus onset.
Whatever the recording interval, the variations of these two muscles in response to angry ava-
tars were not different from those observed in response to neutral expressions. In the PD
patients, comparisons across emotions evidenced specific variations in response to the emotion
Fig 3. Emotional nature of the misidentified expressions as a function of group and emotion displayed. Mean
± standard errors. Confusion
percentage among the HC are shown in white; confusion percentage among the PD patients are shown in grey.
doi:10.1371/journal.pone.0160329.g003
Facial Mimicry in Parkinson's Disease
PLOS ONE | DOI:10.1371/journal.pone.0160329
July 28, 2016
7 / 20
for the corrugator muscle alone (
Fig 4A
,
S3 Table
). From 700 ms after stimulus onset, corruga-
tor activity decreased in response to expressions of joy, increased slightly in response to angry
faces and showed an intermediate pattern in response to neutral expressions (there was no sig-
nificant difference between neutral and angry avatars except from 1700 to 1900 ms). Among
the PD patients, the responses of the zygomaticus and the orbicularis muscles did not vary
across emotions whatever the interval. Comparisons between muscle responses confirmed
these specific variations of muscular activity observed in response to emotions (
Fig 4B
; see
S2
Appendix
,
S4
and
S5
Tables).
The analyses also evidenced a statistically significant group x emotion x muscle interaction
effect (
χ² = 1127.07, df = 4, p<0.001). Overall, the increased activity of the corrugator muscle
in response to angry avatars was lower in the patients than in the HC (
χ² = 14.89, df = 1,
p = 0.001). More precisely, corrugator activity in response to angry avatars differed between
the groups from 1000 to 1200 ms and from 1900 to 2000 ms after stimulus onset (
Table 2
).
Likewise, the overall increased activity in both the zygomaticus (
χ² = 78.96, df = 1, p<0.001)
and the orbicularis (
χ² = 23.29, df = 1, p<0.001) muscles in response to happy avatars were
greater in the HC than in the patients for whom these responses were almost non-existent.
These significant inter-group differences appeared from 700 ms after stimulus onset for the
zygomaticus muscle and from 900 ms for the orbicularis muscle (
Table 2
). In addition, the
overall decrease in corrugator activity in response to happy expressions tended to be lower in
the patients compared to the HC (
χ² = 6.65, df = 1, p = 0.089).
Regarding the mean EMG amplitudes measured at baseline, analysis showed a statistically
significant group effect (
χ² = 5.49, df = 1, p = 0.019) as well as a statistically significant group x
muscle interaction effect (
χ² = 124.78, df = 2, p<0.001). The mean EMG amplitudes measured
at baseline were higher among the PD patients compared to the HC (mean ± standard error:
PD patients = 5.25 ± 0.08
μV; HC = 3.91 ± 0.05 μV) but this effect was carried by the zygomati-
cus activity (PD = 6.02 ± 0.18 vs. HC = 3.69 ± 0.1
μV; χ² = 16.38, df = 1, p<0.001) as no
statistically significant group difference emerged for the corrugator and orbicularis activities
Fig 4. EMG responses (relative to baseline) recorded on sequential 100 ms intervals of stimulus exposure
according to muscle, group and emotion factors. Mean
± standard error. (A) emotion-specific variations for each
recorded muscle; (B) muscle-specific variations in response to the emotion displayed.
doi:10.1371/journal.pone.0160329.g004
Facial Mimicry in Parkinson's Disease
PLOS ONE | DOI:10.1371/journal.pone.0160329
July 28, 2016
8 / 20
(corrugator: PD = 5.19 ± 0.1 vs. HC = 4.49 ± 0.09
μV; χ² = 1.16, df = 1, p = 0.85 and orbicularis:
PD = 4.56 ± 0.12 vs. HC = 3.54 ± 0.05
μV; χ² = 3.65, df = 1, p = 0.17).
Finally, none of the clinical characteristics of the patients (disease duration, worst affected
side, LEDD, Hoehn and Yahr stages and UPDRS III scores ON and OFF DRT) had a statistically
significant effect on muscle responses whatever the emotion or the period of stimulus exposure
at the adjusted significance threshold for multiple comparisons (3 muscles x 3 emotions x 4 peri-
ods, adjusted
α value = 0.001). The effects obtained at the significant threshold of 0.05 are none-
theless shown in
S3 Appendix
.
Facial reactions and emotion decoding accuracy
At the adjusted significance threshold for multiple comparisons (3 emotions x 4 periods,
adjusted
α value = 0.004), only corrugator responses had a statistically (or quasi-) significant
effect on decoding accuracy of expressions of joy: in both groups, for all the 500 ms recording
periods except the first, the probability of accurately identifying joy increased with the corruga-
tor relaxation (500
–1000: χ² = 8.14, df = 1, p = 0.004; 1000–1500: χ² = 9.21, df = 1, p = 0.002
and 1500
–2000: χ² = 7.5, df = 1, p = 0.006). At this threshold, we still noted a statistically mar-
ginal group x zygomaticus muscle responses interaction in the joy condition in the first 500 ms
period (
χ² = 5.84, df = 1, p = 0.016): unlike what was observed in the patients, the probability of
Table 2. Inter-groups comparisons of EMG responses recorded on sequential 100 ms intervals of
stimulus exposure.
Interval
CORRU
—Angry
ZYGO
—Happy
ORBI
—Happy
0
–100
(0.3) ns
(0.2) ns
(0.2) ns
100
–200
(1.5) ns
(0.1) ns
(0.3) ns
200
–300
(0.03) ns
(0.3) ns
(0.4) ns
300
–400
(0.08) ns
(0.2) ns
(0.1) ns
400
–500
(2) ns
(1.3) ns
(0.1) ns
500
–600
(4.4) ns
(4.5) ns
(1.2) ns
600
–700
(7) ns
(13.15) = 0.052
(3.6) ns
700
–800
(6.2) ns
(28.7)
<0.001
(6.4) ns
800
–900
(5.6) ns
(50.7)
<0.001
(10.3) ns
900
–1000
(7.4) ns
(73.2)
<0.001
(15.6)
<0.05
1000
–1100
(14)
<0.05
(97.3)
<0.001
(19.9)
<0.005
1100
–1200
(14.5)
<0.05
(107.1)
<0.001
(22.5)
<0.001
1200
–1300
(11.4) ns
(97)
<0.001
(28.7)
<0.001
1300
–1400
(12.4) = 0.079
(95.9)
<0.001
(34.2)
<0.001
1400
–1500
(7.8) ns
(94.3)
<0.001
(37.1)
<0.001
1500
–1600
(9.7) ns
(93.7)
<0.001
(34)
<0.001
1600
–1700
(10.3) ns
(73.1)
<0.001
(21.8)
<0.005
1700
–1800
(11.7) ns
(51.5)
<0.001
(13.9)
<0.05
1800
–1900
(10.8) ns
(53.6)
<0.001
(17.3)
<0.01
1900
–2000
(19.8)
<0.005
(56.1)
<0.001
(25.8)
<0.001
Only the inter-groups comparisons of EMG responses recorded for the corrugator muscle in response to
angry avatars (CORRU
—Angry), the zygomaticus in response to happy avatars (ZYGO—Happy) and the
orbicularis in response to happy avatars (ORBI
—Happy) are shown here. No other comparisons were
signi
ficant (all p>0.1). Test statistics (χ
2
) are shown in brackets. Figures in bold denote statistically signi
ficant
differences (p value
<0.05). ns = non statistically significant = p value>0.1
doi:10.1371/journal.pone.0160329.t002
Facial Mimicry in Parkinson's Disease
PLOS ONE | DOI:10.1371/journal.pone.0160329
July 28, 2016
9 / 20
accurately identifying joy appeared to increase with the contractions of the zygomaticus muscle
in the first 500 ms of stimulus exposure in the HC. The effects associated with a significance
threshold of 0.05 are nonetheless shown in
S4 Appendix
.
Discussion
The current study was designed to evaluate the role of facial mimicry in recognition of facial
emotion. For this purpose, we investigated for the first time EMG responses to facial expres-
sions among patients suffering from Parkinson
’s disease (PD) in a facial emotion recognition
paradigm. Three main results emerged from our analyses. Firstly, in accordance with the litera-
ture, the PD patients were less accurate in decoding facial expressions of joy and neutral faces
compared to the healthy controls (HC). Secondly, the facial expressions were mimicked as
expected. However, the emotion-specific EMG variations were disturbed in the PD patients
with weaker than normal corrugator reactions in response to angry faces and with almost no
reactions from the orbicularis and the zygomaticus muscles in response to happy avatars.
Thirdly, the analyses highlighted statistically significant effects of the facial reactions on emo-
tion decoding accuracy. Notably, the corrugator relaxation as well as the zygomatic contraction
in response to happy faces is correlated to the decoding of joy.
Many studies have demonstrated the negative impact of PD in the recognition of emotion
expressed on faces [
42
–
50
] but other studies have failed to confirm these observations [
51
–
54
].
The review by Assogna and collaborators [
19
] and the meta-analysis by Gray and Tickle-Deg-
nen [
20
] identified different factors to explain this discrepancy: the implementation of small
samples, inadequate control for demographic characteristics of the participants, the depression
status, the presence of cognitive or visuospatial deficits and the influence of dopamine replace-
ment therapy (DRT). In this study, we used large samples (n = 40 in both groups). We excluded
individuals suffering from depression, apathy, cognitive and visuospatial deficits or impair-
ments in facial processing. We also took into account the sociodemographic and neuropsycho-
logical characteristics of the participants in the statistical analyses. None of these potential
confounding factors had a statistically significant effect except for age. As age had the same
negative effect in the two groups and since this effect is reported in the literature [
55
], we did
not give further consideration to this point. Finally, we examined the effect of DRT and other
clinical features of the patients on emotion decoding accuracy.
Our results confirmed the negative impact of PD in emotion recognition for happy and neu-
tral faces. Conversely, the PD patients did not differ from the HC in the recognition of anger.
Furthermore, the nature of the misidentified expressions was globally similar between the
groups. Only quantitative and qualitative differences emerged regarding the recognition of joy
and neutral expressions. Whatever the emotion displayed, surprise was the most frequent con-
founding emotion especially among the PD patients. Then, both the HC and the PD patients
confounded anger with other negative emotions and neutral with sadness. When the partici-
pants had to recognize joy expressions, surprise was the only source of confusion. Since one
can be positively surprised, negatively surprised and even neither positively nor negatively sur-
prised but just astonished, surprise is ambiguous. This confounding emotion highlighted diffi-
culties to recognize emotions among PD patients. This effect is supported by studies which
depicted surprise as a source of confusion and showed a specific deficit of surprise recognition
among PD patients [
23
,
45
]. Likewise, the similar nature of the misidentified expressions
between HC and PD patients depicted a normal but a noised process of facial emotions. These
findings fit with the presumed role of the basal ganglia-thalamocortical connections underlying
emotional processing described by Péron and collaborators [
56
]. According to the model pro-
posed by these authors, a dysfunction in the pathways involving the basal ganglia as it
’s
Facial Mimicry in Parkinson's Disease
PLOS ONE | DOI:10.1371/journal.pone.0160329
July 28, 2016
10 / 20
observed in PD could prevent correctly inhibiting the non relevant information and/or cor-
rectly activating the relevant information causing the emotional judgements to be disturbed.
The fact that the patients
’ abilities to recognize emotions tended to increased with the
LEDD is fully consistent with the purported role of dopamine in the perception of emotion
[
57
]. A relationship between DRT and emotion recognition is in accordance with both the
amygdala dysfunction hypothesis and the dopamine depletion hypothesis in PD. Experimental
data support these explanations: for example, the restoration of amygdala response in PD
patients perceiving facial emotions during a dopamine-replete state compare to an hypodopa-
minergic state [
58
] and the better performances in recognizing facial emotions in medicated
compared to unmedicated PD patients [
50
].
The absence of a negative impact of PD in the recognition of anger fits the idea of Lawrence
and collaborators [
48
,
59
] that DRT could mask any deficit present in PD especially for anger
recognition. In 2007, these authors assessed performances of emotion recognition in PD
patients withdrawn from dopamine replacement therapy and showed indeed an anger-specific
deficit of recognition among these patients with a spared recognition of other facial emotions
(disgust, fear, sadness, joy and surprise). Likewise, Dujardin and collaborators [
45
] highlighted
a deficit in anger recognition in PD patients who had not yet received any medication. Besides,
from a methodological point of view, our study had some characteristics which may play a role
in our results. First, we used dynamic expressions in order to provide more ecological stimuli.
Considering the importance of dynamic features of the stimuli in the emotion recognition pro-
cess, the used of static faces could artificially cause a deficit in emotion recognition in PD, or at
least widened it [
19
,
60
]. Like us, Kan and collaborators [
47
] did not highlighted a deficit con-
cerning the recognition of anger when exposing participants to dynamic facial expressions and
they showed that the performances of the patients were largely lower when they had to recog-
nize sadness, disgust and anger from static stimuli compared to dynamic expressions. At last,
another argument to explain this absence of group difference concerning the recognition of
anger could also reside in the clinical characteristics of the patients. Indeed, the patients
involved in the current study did not present any potential non-motor symptoms (cognitive
and visuo-perceptual impairment, face processing deficit, depression or apathy) which could
have interfered with the performances on the affective recognition test [
19
,
20
].
To explain the deficit of facial emotion recognition in PD, some authors have suggested
functional and/or anatomical dysfunctions in brain structures such as the amygdala, the basal
ganglia including the ventral striatum, the orbito-frontal cortex and the insula as well as the
impairment of dopamine transmission in the mesocorticolimbic pathway [
18
–
20
]. However,
other authors have argued that PD patients could experience deficits in recognizing emotion
because of a reduced ability to mimic the perceived emotion. In fact, the presence of a common
neural substrate
—as part of the mirror neurons system—underlying the ability to express emo-
tions and to recognize facial emotions expressed by others, suggests that facial amimia could
contribute to the deficit in recognition of emotion in PD [
21
,
61
].
With regard to the EMG responses, our methodology enabled us to evidence emotion-spe-
cific facial reactions 500 ms after stimulus onset as described in the literature: among the HC,
facial reactions to angry expressions were characterized by an increased activity of the corruga-
tor muscle
—an important muscle for frowning in expressions like anger or sadness—and a
slight relaxation of the zygomaticus and the orbicularis
—muscles involved in the production of
smiling expressions by raising the corners of the mouth and forming
“crow’s feet” on the outer
corners of a person
’s eyes [
62
]
—whereas facial reactions to joy are characterized by the reverse
pattern. These variations, which occurred only from the first 500 ms after stimulus onset, have
been widely highlighted in previous studies and are considered to reflect facial mimicry Dostları ilə paylaş: |