Subliminal modulation of voluntary action experience: a neuropsychological investigation

51 52 61 Abstract Human voluntary actions are often associated with a distinctive subjective experience termed ‘sense of agency’. This experience could be a reconstructive inference triggered by monitoring one’s actions and their outcomes, or a read-out of brain processes related to action preparation, or some hybrid of these. Participants pressed a key with the right index finger at a time of their own choice, while viewing a rotating clock. Occasionally they received a mild shock on the same finger. They were instructed to press the key as quickly as possible if they felt a shock. On some trials, trains of subliminal shocks were also delivered, to investigate whether such subliminal cues could influence the initiation of voluntary actions, or the subjective experience of such actions. Participants’ keypress were always followed by a tone 250ms later. At the end of each trial they reported the time of the keypress using the rotating clock display. Shifts in the perceived time of the action towards the following tone, compared to a baseline condition containing only a keypress but no tone, were taken as implicit measures of sense of agency. The subliminal shock train enhanced this “action binding” effect in healthy participants, relative to trials without such shocks. This difference could not be attributed to retrospective inference, since the perceptual events were identical in both trial types. Further, we tested the same paradigm in a patient with anarchic hand syndrome. Subliminal shocks again enhanced our measure of sense of agency in the unaffected hand, but had a reversed effect on the ‘anarchic’ hand. These findings suggest an interaction between internal volitional signals and external cues afforded by the external environment. Damage to the neural pathways that mediate interactions between internal states and the outside world may explain some of the clinical


Introduction 1
Voluntary actions can be functionally defined by two key properties: they are internally-2 generated, as opposed to triggered by external stimuli, and they are often goal-directed 3 (Passingham, Bengtsson, & Lau, 2010). In addition, they are associated with two specific 4 subjective experiences: an experience of volition, and an experience of agency. The 5 experience of volition refers to pre-movement states and events such as desiring, intending, 6 trying and initiating, while "sense of agency" refers to the experience that one's voluntary 7 actions cause outcomes in the external world. 8 On one view, the experiences of volition and agency are post-hoc inferences, triggered by 9 monitoring one's actions and their outcomes. In this case, preparatory brain events that 10 precede action should not influence this experience (Wegner & Wheatley, 1999), although a 11 "prior conscious thought" about acting may be necessary to trigger such inferences (Wegner, 12 2003). Alternatively, experience of agency could depend on a readout of brain processes in 13 frontal (Fried, Mukamel, & Kreiman, 2011) and/or parietal areas (Desmurget et al., 2009) 14 that precede voluntary action. Importantly, these two views make different predictions about 15 how external stimuli might influence the experience of agency: If experience of agency is 16 merely a reconstructive inference, interventions which influence brain processes preceding a 17 voluntary action should have no influence on one's sense of agency, unless those 18 interventions generate some perceptual event which can figure in the inference. On the other 19 hand, if experience of agency depends on internal precursor signals that drive voluntary 20 action, any intervention that influences these signals may also affect experience of agency, 21 whether the intervention is consciously perceived or not. 22 In neuroscience, voluntary actions are often linked to a medial frontal pathway associated 23 with internally-generated movement, as opposed to a parietal-lateral frontal pathway for 24 reacting to external stimuli (Passingham et al., 2010). Human experiments drawing on this 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 5 though this approach has been criticised for lack of ecological validity (Schüür & Haggard, 27 2011). Intervening on volition in such paradigms is methodologically difficult, because the 28 experimenter cannot know when the participant will act. Further, any experimental 29 intervention on precursor processes should preserve the 'internally-generated' aspect of 30 voluntary action, rather than switching to a reactive mode of responding. Subliminal priming 31 offers one potential method for studying volition. For example, subliminal visual primes have 32 been used previously to manipulate the sense of agency by increasing the fluency of action 33 selection processes . Priming can "nudge" the brain towards 34 selecting one action rather than another (Eimer & Schlaghecken, 1998). Compatible priming 35 also increases sense of agency, as if the prime had made the action more strongly 36 intentional (Wenke, Fleming, & Haggard, 2010). However, subliminal visual priming 37 paradigms require a precise temporal relation between prime and a supraliminal 'go' signal. 38 They therefore involve externally-triggered rather than internally-generated voluntary actions. 39 Here, we used a novel design with subliminal electrocutaneous stimuli as a probe to 40  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 8 TP is a 54 year old, right handed woman. She is a former secretary with 11 years of 107 education. Twenty-three months before the testing session, she had a ruptured aneurysm of 108 the right anterior cerebral artery, resulting in subarachnoid haemorrhage, involving the genu 109 and trunk of the corpus callosum. After embolization, she had a vasospasm of the right 110 middle cerebral artery. The most recent MRI (14 months before the testing session) showed 111 damage in the corpus callosum (genu, body and splenium) and in the right anterior frontal 112 and right basal frontal cortex, involving the anterior and middle cingulate gyrus ( Fig. 1 & 113 showed residual attentional deficits (in the subtests "alertness", "acoustical vigilance" and 116 "divided attention" of the Italian version of the Test of Attentional Performance: Zimmerman 117 & Fimm, 1992;Zoccolotti, Pizzamiglio, Pittau, & Galati, 1994), mild impairments in 118 perspective memory (Rivermead Behavioural Memory Test: Wilson, Cockburn, & Baddeley, 119 1985), abstract classification abilities (Wisconsin Card Sorting Test: Heaton, Chelune, 120 Talley, Kay, & Curtiss, 2000) and executive functions (Tower of London: Culbertson & 121 Zillmer, 2005;Phonemic and Semantic Verbal Fluency Test: Spinnler & Tognoni, 1987). No 122 additional impairments were found in working memory (sub-test "Working Memory" of the 123 Italian version of the Test of Attentional Performance: Zimmerman & Fimm, 1992;Zoccolotti 124 et al., 1994), long-term and short-term verbal memory (Buschke-Fuld Test: Buschke & Fuld, 125 1974;Spinnler & Tognoni, 1987, Digit Span: Orsini et al., 1987 and spatial memory (Rey-126 Osterrieth Complex Figure Test: Caffarra, Vezzadini, Dieci, Zonato, & Venneri, 2014;127 Osterrieth, 1944, Corsi-Block tapping test: Spinnler & Tognoni, 1987, or logical and 128 reasoning abilities (Raven's Progressive Matrices: Spinnler & Tognoni, 1987). In addition, TP 129 showed no sign of apraxia with either limbs or hands (Test of limb apraxia: De Renzi, Motti,  130  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 9 TP complained of her left hand behaving in an uncontrolled manner. For example, she 133 reported that her left hand threw a towel into the bathtub full of water or that she blew her 134 nose with a napkin that she held in her left hand, instead of using the handkerchief in her 135 right hand. The episodes of uncontrolled behaviour of her left hand occurred on a daily basis 136 in the first few months after the lesion and were characterized by groping movements, 137 grasping, subsequent inability to release the grip, utilization behaviour and the persistent 138 feeling of unresponsiveness of the left hand. At that time, TP also had frequent episodes of 139 mirror movements (i.e., the anarchic left hand reproduced the movement of the unaffected 140 right hand), and also reported the inability to coordinate simultaneous different movements 141 of the two hands. At the time of the testing session, the frequency of the episodes of 142 uncontrolled behaviour was reduced. Episodes of grasping behaviour and subsequent 143 inability to release the grip occurred almost once a week. TP complained of the persistent 144 feeling of unresponsiveness of her left hand and reported her attitude to restrain the actions 145 of her left hand, by using the other hand to prevent it from moving. Despite the lack of 146 voluntary control of her left hand, TP never denied ownership of the hand. 147 All experimental procedures were exactly the same in TP and healthy participants. The only 148 difference is that all the data were collected from the right hand of the healthy participants, 149 while for TP data was collected from both the right (healthy) hand and the left (affected) 150 hand, in separate sessions. 151

Experimental procedure. 152
After filling the consent form, the general experimental procedure was explained for the 153 participants. Non-painful electrocutaneous shocks were delivered from a programmable 154 10 cables, respectively. Intensity of the shocks depended on the trial type (see later). The 158 duration of each shock was set at 10 ms. 159 The behavioural task started after setting up the electrodes. Each experimental session 160 consisted of three tasks: First, a detection task was used to detect the lowest threshold level 161 at which participants were able to detect the shocks. Then, supra-and subliminal levels of 162 shock were calculated from the threshold measure and a signal detection task was 163 administered to confirm perception of the shock stimuli. Participants who did not pass the 164 signal detection task were excused and did not proceed to the next step. Finally, participants 165 performed the 'intentional binding' task, which has been widely used as a proxy measure of 166 sense of agency (for a review, see Moore & Obhi, 2012). 167

Threshold detection task. 168
An ascending staircase approach was used to detect the lowest levels at which participants 169 were able to detect the shock (Moore, Ruge, Wenke, Rothwell, & Haggard, 2010). Shocks 170 started at 0.1mA and increased in steps of 0.1 mA until the shock was detected, and then 171 decreased in steps of 0.05 mA until the shock was missed, and then increased again in 172 steps of 0.01 to find the detection threshold. A tone was played at the time of each shock 173 and participants were asked to report if they felt a shock at the time of the tone or not. In this 174 and all the later tasks, participants were instructed to report feeling a shock when they felt 175 any kind of stimulus, not simply a painful shock. The level for supraliminal shock stimuli was 176 set at 130% of the threshold level. The subliminal level was determined by reducing one step 177 (0.01 mA) from the threshold (e.g., if the detection threshold was 0.45 mA, the subliminal 178 level would be 0.44 mA). This strategy was chosen to ensure that subliminal shocks had 179 sufficient energy to influence brain processes, while remaining imperceptible (see below).

11
The estimated supra-and subliminal shock levels were then validated in a signal detection 182 task. Each signal detection task consisted of four types of trials in a randomised order: 20 183 subliminal shock trials, 20 subliminal catch trials (with no shock), 20 supraliminal shock trials 184 and 20 supraliminal catch trials (with no shock). In each trial participants heard two tones, 5 185 s apart. They received a supraliminal shock at a random time between those two tones in 186 supraliminal shock trials. No shock was delivered in catch trials. In subliminal trials, a train of 187 subliminal shocks were delivered every 1 s starting from the first tone and ending with the 188 second tone. At the end of each trial participants were asked to report if they felt any shock 189 between the first and the second tone or not. At the end of the task, participants' responses 190 were used to estimates the sensitivity index (d') for the supra-and subliminal shocks. To 191 proceed to the next step, participants were required to obtain a d' value within the range of 192 0.5-1.5 for subliminal shocks and a d' of >= 3 for the supraliminal shocks. The relatively high 193 sensitivity index for subliminal shocks means that participants could sometimes detect the 194 shock. We wished to ensure that the subliminal shocks were strong enough to influence 195 brain processes. Those subliminal shocks that were detected by participants during the main 196 task were discarded (see section 2.5). If their d' did not match this criteria, the threshold 197 detection task was repeated to find a new threshold followed by a signal detection task. If the 198 desired d' was not achieved after four attempts, participant was excused and did not 199 proceed to the intentional binding task. 200

Intentional binding task. 201
We used intentional binding paradigm as an implicit measure of agency. The task was based 202 on previous studies (Haggard et al., 2002), and was programmed in LabVIEW 2012 (Austin, 203 Texas). Participants viewed a clock hand rotating on a computer screen, located 60cm in 204 front of the participants in a quiet room. The initial clock position was random. Each full 205 rotation lasted 2560 ms. Participants made voluntary keypress by pressing the enter key 206 with their right index finger. Participants chose for themselves when to make the voluntary 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 12 made a time judgement according to condition (see later). Each experimental session 209 consisted of two conditions, presented in separate blocks. At the beginning of each block, 210 brief instructions for the relevant condition were displayed on the screen. In the baseline 211 condition, participants had to press the enter key at a time of their own free choice. The 212 clock hand stopped after 1500-2500ms (at random), and participants then judged the clock 213 hand position at the time of their keypress. In this condition, participant's actions produced 214 no sensory outcome and they received no shock. In the agency condition, participants were 215 again asked to press the key at a time of their own free choice. However, this time each 216 keypress produced a pure tone (1000 Hz, 100 ms duration) after 250 ms and they 217 sometimes received a mild shock on their right index finger before pressing the key. At the 218 end of each trial, participants made two subjective reports. First, they reported the clock 219 hand position at the time of their keypress. Second, they reported whether they had felt a 220 shock or not. Each block in the agency condition consisted of two types of trials in a 221 randomised order: in two thirds of the trials a single supraliminal shock happened at a 222 random time, drawn from an exponential distribution (min = 1 s, max = 10 s, mean = 5 s) 223 ( Fig. 2A, B). In the other one third, a 1 Hz train of subliminal shocks occurred starting from a 224 random time within 500 ms from the beginning of the trial and continuing for 10 s (Fig. 2C). 225 The train ensures that any keypress occurs within 1 s of a shock. In all trials of the agency 226 condition, participants were asked to press the enter key whenever they felt like but to press 227 the key 'as quickly as possible' if they felt a shock. There were two possible outcomes in 228 trials with a single supraliminal shock: either participants waited long enough, received the 229 supraliminal shock and reacted (Fig. 2B), or they voluntarily pressed the key before the 230 occurrence of the supraliminal shock, in which case the supraliminal shock was cancelled 231 ( Fig. 2A). The former trials were categorised as 'reactive' trials, if participants accordingly 232 reported feeling the shock, and the later trials were categorised as 'voluntary' trials, if 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 13 Trials containing a train of subliminal shocks were also divided into two categories. First, if 235 the participant reported perceiving any shock, the trial was discarded. If the participant did 236 not report perceiving any shock, the trial was categorized as a 'primed-voluntary' trial. 237 The baseline condition was tested in two separate blocks of 15 trials each, at the beginning 238 and end of the experiment. The agency condition was tested in four blocks of 40 trials each 239 between the two baseline blocks. 240

Data analysis. 241
In signal detection task the proportion of hits, correct rejections, misses and false alarms 242 were calculated separately for supra-and subliminal shocks. These measures where then 243 used to compute the sensitivity index (d'). 244 In the intentional binding task, judgment error was defined as the difference between the 245 judged clock hand position and the actual time of the keypress on each trial. A positive 246 judgement error indicated a perceptual delay; a negative judgement error an anticipation. 247 The mean and standard deviation of the judgement errors across trials were then measured 248 for each trial type. Action binding was defined as the shift of reported time of action towards 249 its outcome, and was calculated by subtracting each participant's mean judgement error in 250 the baseline condition from that in the agency condition. Thus, perceptual association of an 251 action with a subsequent tone would produce a positive value for action binding. We then 252 used repeated-measures ANOVA and paired-samples t-test to compare action binding in 253 voluntary trials with action binding in primed-voluntary trials. Multilevel models were used 254 when comparing trial types with unequal sample size, using the lme function in R (R Core 255 Team, Vienna, Austria). The main purpose of having supraliminal shocks was to establish a 256 stimulus-response association between the shock and the action. We reasoned that this 257 makes the shock meaningful for action, and therefore more likely to prime action processing. 258 Finally, a Crawford test (Crawford, Garthwaite, & Porter, 2010) was used to compare TP' s 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 14 action binding scores from the healthy and affected hand with the action binding data in 260 healthy participants. 261 We additionally checked whether subliminal shocks could influence behaviour, as well as 262 sense of agency. The latency of each keypress from the immediately preceding subliminal 263 shock was measured. These latencies were averaged across all primed-voluntary trials 264 within each participant. We tested the null hypothesis that the action latencies in primed-265 voluntary trials are from a population with uniform distribution by using a separate Anderson-266 Darling test for each participant. 267

Experience of agency in healthy participants. 269
Of the 47 recruited participants, 27 met the d' criteria of the signal detection task and went 270 on to do the intentional binding task. Four participants did not finish the intentional binding 271 task because their detection threshold was unstable during the task. Therefore, the final 272   (Fig. 3A). This suggests that experience of 288 agency towards an action and its effect is associated with precursor brain signals for that 289 action.  -116 -19]). This reversal of 295 intentional binding for responses to a supraliminal stimulus has been reported previously: 296 (Waszak et al., 2005). This finding suggests that the increase in action binding for primed-297 voluntary compared to voluntary trials could not be merely explained by the presentation of 298

shocks. 299
Finally, to make sure that unbalanced number of trials is not confounding the results, 300 participants' action binding data in each trial type was weighted by the ratio of number of 301 trials in that condition to total number of trials. The significant difference between the 302 conditions (t(22) = 2.25, p = 0.03, d z = 0.47, 95% CI [1 16]) suggests that action binding is 303 significantly stronger in trials with a subliminal shock even after controlling for unbalanced 304 number of trials. 305 If subliminal shocks influence brain processes during action preparation, we might expect to 306 find the effects not only on experience of agency but on some other behavioural measure 307 such as action initiation. We therefore tested the hypothesis that the subliminal shocks 308 influenced the latency of keypresses, by using the Anderson-Darling test to compare 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 16 latency distribution was significantly non-uniform in seven participants (supplementary table  311 3). The null hypothesis that this many tests being significant could happen by chance alone 312 was examined using a binomial test. By the binomial distribution, the probability of getting 313 seven significant non-uniform action latency distributions in a sample of 23 by chance is B 314 (0.05, 7, 23) = 0.00009401 (Fig. 3B). This suggests that subliminal shock has some 315 influence on behaviour. However, the presence and pattern of this effect differed across 316 participants. While in some participants subliminal shocks facilitated action initiation, in 317 others it delayed the time of the action (supplementary figure 1). 318

Experience of agency in TP vs. healthy participants. 345
Finally, we tested whether subliminal shock effects on action binding were significantly 346 different in TP and healthy participants, using Crawford test. This method tests whether a 347 single patient's score differs significantly from that in a control group, and also provides a 348 point estimate of the separation between the patient's score and the control group (Crawford 349 et al., 2010). The effect of the subliminal shocks was measured by subtracting each   17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 18 Healthy subjects and an individual with anarchic hand syndrome were exposed to subliminal 363 electrocutaneous stimulus during the precursor period before performing internally-364 generated actions that produced an external outcome. We used an established implicit 365 measure based on time perception to measure sense of agency. The perceived time of an 366 action has been found to shift towards its outcome for voluntary actions but not for 367 involuntary movements (Haggard et al., 2002). Using this 'intentional binding' index, we 368  were not. The direction of the effect, shifting action perception towards the subsequent 381 outcome, rules out explanations based on P-centre phenomena (Morton, Marcus, & 382 Frankish, 1976), or anchoring effects of the preceding shocks on time perception. Further, as 383 participants could not feel the subliminal shocks, this difference is unlikely to reflect a 384 conscious decision to control actions in a different way. Most importantly, the difference in 385 action binding between trial types could not easily be explained by a purely post-hoc 386 inference account of sense of agency, since the events perceived are identical in both 387 conditions. 19 Previous studies showed that explicit agency judgements could be modulated by using 389 visual subliminal priming Chambon, Sidarus, & Haggard, 2014;390 Haggard & Chambon, 2012). Participants reported stronger experience of agency over 391 action effects when the subliminal prime was compatible, compared to incompatible, with the 392 selected action . In those studies, as in our experiment, the prime 393 influenced a stage of action preparation that necessarily precedes both action and its effect. 394 This suggests that sense of agency cannot be purely retrospective. Rather sense of agency 395 must depend, at least in part, on signals arising during action preparation. Of course, this 396 does not rule out a further contribution from retrospective inference. 397 Additionally, given that subliminal shocks increased our measure of sense of agency, 398 external stimulation facilitated putative precursor signals during action preparation. At first 399 sight, this may seem paradoxical, given the traditional dichotomy between brain systems 400 underlying internally-generated and externally-triggered actions (Passingham et al., 2010). 401 However, substantial cross-talk between the two systems exists. In one study, the reaction 402 time to an external-trigger stimulus was reduced in the very final phases of preparation of a 403 voluntary action (Obhi, Matkovich, & Chen, 2009; also see: Hughes, Schütz-Bosbach, & 404 Waszak, 2011). We speculate that during action preparation, the subliminal shock is taken 405 as an additional environmental cue. The subliminal shock may "nudge" the signal that 406 generates voluntary action, facilitating a threshold crossing event (Schurger, Sitt, & 407 Dehaene, 2012). In our paradigm, participants also occasionally reacted to supraliminal 408 shocks. It remains unclear whether this prior association between shock and action is 409 essential for the subliminal priming we observed. We hope to investigate this point in future 410 experiments. Interestingly, we also found some statistical evidence for effects of shock on 411 action initiation. However, this effect was not present in all participants, and the pattern of 412 influence differed across participants. While in some participants subliminal shocks 413 transiently facilitated action initiation, in others it delayed the time of the action. We note that 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 20 widely reported (e.g., Blankenburg et al., 2003). We speculate that subliminal shocks may 416 not only sum with the precursor signals during action preparation but also change the 417 threshold for the initiation of the voluntary action. The precise moment of action initiation 418 thus depends on both signal amplitude and the current threshold. Interestingly, the significant interaction between hand and trial type showed that subliminal 428 shock enhances sense of agency similar to healthy participants, but only when applied to the 429 healthy hand. Subliminal shock had no statistical effect when applied to the affected hand. 430 We suggest that, for the affected hand, a mechanism that uses precursor signals of 431 voluntary action to compute sense of agency is now disrupted. The normal function of this 432 mechanism would include integrating signals from the external environment and from 433 internal states to construct a coherent subjective experience of action. 434 Normal behaviour is an outcome of active interplay between internal states and the external 435 environment. Successful interaction of these two components is crucial for goal-directed 436 behaviour and inhibition of unwanted responses. Patients with focal damage in medial frontal 437 cortex (though without signs of AHS) show disruption to automatic motor inhibition, as 438 evident in a reversal of the normal negative compatibility effect in a masked-prime task 439 (Sumner et al., 2007). Abnormal facilitation by priming, as well as the involuntary object -1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 21 reflecting damage to a brain system that normally inhibits excessive environmental reactivity. 442 Our results suggest a second aspect to AHS. The normal subjective experience of action is 443 altered in AHS, and in particular the capacity to feel a sense of agency for voluntary actions 444 that are appropriately interfaced to subtle cues in the external environment. 445 The brain lesions of TP mostly involved the right anterior cingulate cortex (ACC) and the 446 posterior part of corpus callosum (CC) (Figure 1 & Table 1). Lesions in these areas have 447 been previously reported in patients with AHS (Hassan & Josephs, 2016). One fMRI study 448 compared brain activity during alien hand and voluntary movements of a patient with AHS 449 (Assal, Schwartz, & Vuilleumier, 2007). While alien hand movements were associated with 450 isolated activity in contralateral motor cortex, voluntary movements of the same hand 451 activated extensive networks including the ACC, suggesting a possible role of ACC in 452 voluntary action control. Moreover, ACC has been shown to be active during self-and 453 external-agency attribution tasks (Fukushima, Goto, Maeda, Kato, & Umeda, 2013;Nahab et 454 al., 2011). Other case studies have associated lesions in the CC with volitional disorders of 455 AHS (Della Sala, Marchetti, & Spinnler, 1991;Feinberg, Schindler, Flanagan, & Haber, 456 1992). CC connects the frontal and motor areas of the two hemispheres. Specifically, the 457 body and splenium of CC, which are mainly damaged in TP, connect the premotor areas 458 (Berlucchi, 2012). Damage to this area could thus lead to loss of transcallosal motor 459 inhibition of the contralateral hemisphere (Kim, Lee, Lee, & Lee, 2014). Interestingly, Wolpe 460 et al. (2014) found a relation between CC white matter loss and abnormal intentional binding 461 in patients with alien limb due to corticobasal degeneration. This deficit was largely confined 462 to anterior parts of CC. 463 Patients with AHS commonly report that their hand is not under their control or being 464 controlled by an external agent (e.g., Assal, Schwartz, & Vuilleumier, 2007). Our work 465 suggests that this phenomenology may arise from two distinct sources. The first source, and 466 the only one recognised in the current literature, is the positive symptom of the affected 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 22 a second source of AHS phenomenology, namely a reduced sense of agency for one's own 469 voluntary actions. In the normal brain, voluntary actions do not come "from nowhere", but are 470 aligned to subtle action possibilities suggested by the environment, akin to subliminal 471 priming in laboratory experiment. Such priming increases explicit judgements of agency 472 , and increased intentional binding in healthy volunteers. However, this 473 mechanism was absent for the affected hand of our AHS patient. To our knowledge, this is 474 the first study to investigate a negative symptom of AHS by measuring the effect of the 475 external world on experiences of voluntary actions. 476

Sense of agency as a readout of internal volitional signals: a cognitive model. 477
Based on our findings from healthy participants and TP we propose a cognitive model of the 478 experience of voluntary action (Fig. 5). We suggest that one key input to the experience of 479 agency is a readout of an internal volitional signal that precedes endogenous actions. This 480 internal signal, however, could be influenced by externally-triggered signals from the outside 481 environment (affordances): volition is not independent of the current environment and 482 response space (Schüür & Haggard, 2011). In the case of healthy participants and the 483 unaffected hand of TP, this external signal is integrated into the internal volitional signal to 484 facilitate action preparation. Thus, the weak sensory evidence suggesting action that is 485 provided by a subliminal prime is summed with the intention or predisposition to act provided 486 by the task instruction. This integration is accordingly reflected in a stronger action binding 487 and an altered distribution of acting. Thus, suggestions of the external environment are 488 integrated with intentions, and the sense of agency depends partly on a metacognitive 489 readout from the output of this 'integrator' (Fig. 5, node 1) (Fleming & Frith, 2014). 490 This interface between the will and the external world is damaged in AHS (Fig. 5). Classical 491 descriptions of AHS suggest that intentional control no longer inhibits affordance-based 492 respondingresulting in compulsive or utilisation behaviours (Fig. 5, node 2)  23 patient's experience of actions is no longer driven by metacognitive readout of one's own 496 intentions, but is instead driven by experience of actual motor outputs triggered by 497 environmental stimuli. As a result, patients with AHS frequently describe movements of the 498 affected hand as involuntary, even when they are well-formed and co-ordinated. For 499 example, patients may report that their affected hand 'has a mind of its own', 'is being 500 'naughty ', 'doing what it wants, not what I want', etc. 501 This model contains the inhibitory link from the voluntary to the reactive motor system that is 502 classically associated with AHS (Fig. 5, node 2). Our results here suggest that the interface 503 also involves a second link, whereby the external environment, even in mild subliminal form, 504 can gently nudge volition (Fig. 5, node 1). This nudge can lead to changed behaviour, as in 505 subliminal priming (Eimer & Schlaghecken, 2002), but also changed experience of volition, 506 as in the altered sense of agency here. Damage to the interface area in AHS also weakens 507 this second facilitatory link between the external environment and volition, preventing the 508 normal subliminal facilitation of sense of agency. Taken overall, a healthy sense of agency 509 requires that the voluntary motor system be responsive to appropriate external suggestions 510 when these align with one's own wishes, while retaining the ability to suppress externally-511 driven actions when these are not desired. Our results suggest that the cingulate and the 512 callosum participate in this bidirectional interaction. 513