The results show that rTMS induces a sensation of movement in the absence of sensory feedback when applied over M1 and PMd. A sensory and motor block was accomplished using INB and SB and the effect was present in both the arm and the lower limbs. The sensation of movement was not accompanied by movement and may therefore be characterized as a movement illusion. The illusion was present in the arm when with M1 and PMd stimulation, and additionally in somatosensory areas of the leg when the leg area of the cortex was stimulated with rTMS. Interestingly, the absence of sensory feedback did not affect the induced sensation of movement over PMd to the same extent as that for the movement illusion evoked by M1 stimulation.
At baseline, with normal sensory feedback, the movement illusion was not as strong for PMd stimulation compared with M1 stimulation. Therefore, it could be argued that the lack of change in the sensation evoked by the PMd stimulation could be explained by a lower sensitivity to the effect of INB for the weaker sensation. However, when the M1 stimulation intensity was reduced in control experiment 1 (see supplementary Text S1
's Supplementary control experiment section and supplementary Figure S1
) such that the induced sensation was similar to the sensation of movement induced over PMd, we found that even intermediate (subjective scale ~2–3) sensations of movements were sensitive to INB when induced over M1. The lack of change in movement sensation associated with PMd stimulation suggests that PMd has a different role in movement sensation than does M1. A study in monkeys showed that premotor cortex activity represented the perception of movement and not the actual movement 
. Theoretical work 
has suggested that efference copies are generated in the premotor cortex and that efference copies may be important for perception. We propose that the findings of the present study reflect the notion that PMd generates a corollary discharge when stimulated with TMS, and that this directly influences perceptual areas including somatosensory cortex and possibly also M1. The network of PMd, M1 and possibly also S1 
is responsible for generating a sensation of movement. Ellaway et al. 
showed that M1 did not produce such a corollary discharge but suggested that premotor cortex might, although the study did not test this possibility directly. We have shown that stimulation induced sensation of movement over PMd is less sensitive to sensory and motor deprivation than is M1, and as the results in control experiment 1 shows, it is not due to less sensitive subjective measures of movement perception for low intensity movements.
It has been suggested that the mirror neuron circuitry linking areas of the ventral premotor cortex with the parietal area PF and the superior temporal sulcus generates the necessary components in establishing inverse models, whereby observed actions are coded into motor plans, and forward models that predicts the sensory (visual) outcome of the movement 
. In the present study we did not engage our subjects in tasks where they were supposed to view their own limbs moving (with and without sensorimotor deprivation) so we cannot provide evidence of whether we in fact targeted mirror neurons, because our tasks were not specifically designed to target that question. However, we may have engaged similar networks, in particular the premotor stimulation, may have engaged mirror neuron related networks.
Haggard and Whitford 
showed that TMS stimulation over SMA prior to a voluntary movement removed the sensory suppression effect, which normally accompanies voluntary movements 
. Electrophysiological 
and neuroimaging 
evidence shows that SMA is involved in early preparatory components of voluntary movements, where one function may be to predict the sensory consequences of the movements as indicated by Haggard and Whitford 
. Using similar stimulation intensity over SMA, i.e. approximately resting motor threshold for evoking MEP after stimulation over M1, we were not able to induce a sensation of movement. This suggests that the central cancellation of predicted sensory feedback is not directly related to the sensation of movements although both mechanisms require top down modulation of sensory areas. This likely explains our negative finding of absent sensation of movement after rTMS over SMA. However, direct cortical stimulation of SMA has previously induced movements 
, so an alternative explanation could be that we were not able to efficiently stimulate SMA with TMS when the TMS was not accompanied by a voluntary effort, as was the case for the study by Haggard and Whitford 
The stimulation trains over PPC did not evoke any sensation of movement. We chose this location based on our previous observation that areas in the posterior medial part of the parietal lobule showed properties related to integration of visual feedback and motor commands 
. By choosing this area we could explore whether stimulation of the region would lead to induced sensations of movements, but this was not the case. Therefore the stimulation over PPC was used as a control for comparison to stimulation over the other regions. A recent study by Desmurget et al. 
showed that stimulation directly on the cortical surface of parietal areas gave rise to “an urge to move”. In three of the four patients that were stimulated, the areas that were stimulated, were located more lateral than any of the areas that was used as PPC in the present study. In one of the patients the location of Desmurget et al's stimulation site was close to where we stimulated, but the brain tumour that this patient suffered from was also located nearby and could have induced cortical reorganisation that may turn out to produce different results than what would have been found in healthy subjects. Finally, Desmurget et al. used 4 s stimulation trains where we only produced sensations of movement over M1 and PMd with 500 ms trains. These differences in location, stimulation and the fact that Desmurget et al. studied “an intention to move” rather than “a sensation of movement”, may explain why we did not observe induced sensations of movement when PPC was stimulated. Finally, the difference between direct cortical stimulation during surgery and the application of high frequency rTMS may be so different, and therefore give rise to the differences between our study and the study of Desmurget et al. 
and the findings of Fried et al. 
. One difference is also that direct cortical stimulation often is conducted using much higher frequencies 
. Therefore further studies are needed in order to clarify the functional differences between direct cortical stimulation and rTMS.
One explanation could be that rTMS induces more widespread activation due to the extent of the magnetic field and accompanied stimulation area compared with small surface electrodes used during surgery. Another explanation would be that the stimulation of motor areas creates feedback activation of somatosensory cortex, whereas direct somatosensory stimulation requires longer stimulation times in order to create reverberating circuits with other regions.
One surprising observation in the Desmurget et al. 
study is that their stimulation-evoked movements failed to produce any movement sensation despite the fact that the movement itself must have evoked some afferent traffic. In contrast, the present study reveals a significant correlation between the motor output (number of MEPs in the arm muscles) and the subjective sensation of the movement. This raises the possibility that the patients investigated by Desmurget et al. may have been severely affected by their tumours or, possibly, influenced by sedatives, although the latter is not described in the paper. Their finding is also very surprising, because awake subjects easily can detect single MEPs evoked by TMS, either due to a conscious percept of the motor command or the sensory feedback coming from the muscle twitch.
In the INB and SB experiment stimulation trains lasted 500 ms. The duration was inspired by the minimal duration of stimulation trains used by Libet et al. 
in order to produce perceivable sensory sensations. Stimulation trains lasting 500 ms or more evoked sensory sensations whereas shorter duration stimulation trains did not. However, in the second control experiment (described in the supplementary Text S1
's Supplementary control experiments section) we also found that shorter (150 ms) magnetic stimulation trains produced perceivable movements with and without actual movement (see supplementary Figure S2
). This further suggests that there is a qualitative difference between direct electrical cortical stimulation and rTMS as a mean to produce illusions of movement.
Changes in cortical representation of body parts affected by sensorimotor deprivation have been studied with PET 
showing increased cerebral blood flow during INB but only during rest. MEP amplitudes in muscles proximal to INB have also been studies with TMS 
, revealing that these were increased in a proximal muscle during INB. However, in the present study the movement sensation reported by subjects was constricted to finger and wrist movement for the INB experiment and ankle joint rotation for the SB experiment. Only one subject showed elevated levels of MEP in the biceps muscles proximal to the tourniquet.
We have shown that in the absence of sensory feedback and accompanying movements, caused by ischemic nerve block or spinal block, it is possible to induce a movement illusion using rTMS over M1 and PMd for both the arm and the lower limbs. We have shown that the sensation of movement evoked by rTMS when applied over PMd is not affected by the absence of sensory feedback, to the same degree as when applied over M1, suggesting that a corollary discharge evoked by TMS over PMd induces sensation of movement.