Wednesday, May 27, 2009

Do mirror neurons exist in humans? A new study says 'no'

Already a new paper to appear in PNAS is generating a buzz in the press.

The study is by Alfonso Caramazza and colleagues who used an fMRI adaptation paradigm. Adaptation was assessed both for observing (O) and then executing (E) actions and executing and then observing (as well as O-O and E-E conditions). Assessing adaption in both directions, E->O and O->E, is critical because (i) if mirror neurons exist, adaptation should occur in both situations, and (ii) adaptation in the case of observing and then executing could be interpreted as motor priming during the observation event. The critical result was that in the regions they examined, fMRI adaptation was found for E-E conditions, showing that there is coding of information relevant to action execution, and also in O-E conditions suggesting prima facie that action observation and action execution are activating the same set of neurons in the ROIs. However, E-O trials did not exhibit adaption, which they should have if in fact there is a shared substrate for observation and execution (it shouldn’t matter what the order of presentation), and neither did O-O trials indicating that the ROIs were not coding perceptually driven information. This pattern of results can be explained if the ROIs are coding action execution information (E-E adaptation) and if observing an action that one might have to execute can prime these action coding regions (O-E adaptation).

This is a significant advance over previous attempts to find adaptation effects in the human mirror system because clear evidence of adaptation was identified, ruling out a power issue, and because they assessed observation-execution adaptation in both directions. This allows the authors to conclude with some degree of confidence that the direct matching hypothesis is incorrect.

So could it be possible that mirror neurons don't exist in humans? I have said that such an outcome would be surprising. But this new result makes me wonder whether there might be something funky about the training situation of macaques in which mirror neurons have been found that lead to the development of neurons with mirror properties. In other words, do mirror neurons even exist naturally in monkeys? ...


Tom said...

Hi Greg,

From what I've read elsewhere, the standard pre-emptive defence against this paper is along the lines of,

1) Who says mirror neurons ought to adapt?
2) Even if we accept that they ought to, we can't infer much from fMRI adaptation studies as we're still unsure of the neural mechanisms that generate the measured responses.

I'm not particularly pro or anti the mirror neuron theory, and I haven’t read the paper yet, of course, but from the information released, my primary worry would be arguing against the theory from a null result with only 12 subjects in the study. I'm not so sure there wouldn't be a power issue - I'd be more convinced if n was ~30. I look forward to reading the paper.

Having said that, I do like the final point you made about training effects. I hadn't thought about that before.

Tony said...

I'm certainly not a neuroscientist, so I'm unable to ascertain whether Iacoboni's gripe about the study is valid. Are there varieties of neurons that don't adapt?

Greg Hickok said...

Can fMRI prove that mirror neurons do not exist in humans? Probably not. However, null results like this add to the list of problems for mirror neuron theories of action perception:

1. There is no evidence in monkeys that mirror neurons are the basis of action understanding as opposed to a sensory-motor reflex that is correlated with action perception.

2. Physiological evidence for motor activity in response to action perception in the human has not ruled out (or even attempted to address as far as I know) the possibility that motor priming is driving the motor activity rather than action understanding.

3. Damage to the motor system does not preclude action understanding.

And now,

4. There is no direct evidence for the existence of mirror neurons in the human brain, although there is evidence from this new PNAS study that perceiving actions can prime cortical regions for actions in regions thought to house mirror neurons.

Should mirror neurons adapt? I think a better question is, what is the theory or physiological evidence that suggests that mirror neurons should NOT adapt?

Think about it... the PNAS study demonstrated adaptation *in canonical "mirror system" locations* for executing actions (executing and then re-executing an action), but these same ROIs failed to yield adaptation to executing and then perceiving an action. So the claim would have to be that within ROIs containing mirror neurons, pure more cells adapt, whereas pure sensory and sensory-motor cells (i.e., mirror neurons) do not adapt. Why should this pattern hold?

Are there neurons that don't adapt? Good question. I don't know. If anyone does, please fill us in!

Eric Raimy said...

Long time reader, not so often commenter...

FYI Ioacoboni is raising questions and making some 'interesting' accusations about this paper over in the comments section on Mindhacks. Not encouraging readers to leave this lovely blog but the question about mirror neurons appears to be getting rather peronsal...

Greg Hickok said...

Looks like the quasi personal attack was retracted by Marco. Let's move beyond that and stay focused on the scientific issues, otherwise we might end up being featured in the next issue of David P.'s "Nature People". :-)

In Marco's substantive comments on that blog he raises an interesting issue about the nature of the signal underlying BOLD subtraction versus adaption type paradigms. He suggests in particular that adaption methods are sensitive to synaptic inputs to an ROI and not to action potentials. He then correctly points out that mirror neurons are defined by action potentials. I'm going to have to go back and read the paper he cites in support of this position (a paper by Logothetis and others) to see how solid this is. In the end I'm not sure we know enough about what we are seeing in fMRI to make any firm conclusions one way or another.

So again I don't think fMRI can prove that mirror neurons do not exist in humans. We need to be focusing on the big picture: what are these cells proposed to be doing? and is there any evidence to support such a proposal? In this respect we have two proposals on the table. 1. Rizzolati's position that they support action understanding and hence are part of a semantic system, and 2. my view that they support plain old sensory-motor association and do not support semantic "understanding". I view #2 as the null hypothesis as we know sensory-motor association systems exist in cortex. So far this is nothing out there that leads me to reject this null hypothesis.

Zarinah Agnew said...

I haven't yet read this paper in great detail, but there are others that have used the same approach.

In Gallese's 1996 Brain paper which characterises the properties of macaque mirror neurones, it is stated that the visual properties of these mirror neurones were '.. highly consistent and did not habituate.'.

To me this indicates that perhaps repetition suppression is not an ideal manner for testing for the existence of mirror neurones. Despite being a bit of a 'mirror neurones explain everything' sceptic myself, I don't believe this null result should or can be taken as evidence against the existence of human mirror neurones, at least not until someone has demonstrated a cross modal repetition suppression effect in single cells in non human primates.

Greg Hickok said...

Thanks for pointing out Gallese et al.'s observation as this is certainly relevant.

I agree that the failure to find evidence for execution --> observation adaptation is not going to prove anything about the existence of mirror neurons, but it provides an important counter example to those studies that have used adaptation to argue for the existence of such cells in humans.

Anonymous said...

I read the paper and am curious to know if people consider the paired-item paradigm as one that unambiguously indexes repetition suppression mechanisms.
To illustrate, an area showing less activity to
might do so because of repetition suppression to the second ""A" in AA, change detection to the "B" in AB or both. In other words, an area showing AB > AA might not show repetition suppression in the classical sense of attenuated response to a second stimulus. So in a certain region, you could find AB > AA but not
A2 > A1 in a paired presentation of A1A2.