Tuesday, June 2, 2009

Can fMRI adaptation demonstrate (or refute) the existence of mirror neurons?

A recent fMRI study published by Caramazza and colleagues in PNAS used an adaptation paradigm and found no evidence for the existence of mirror neurons in humans. Basically, in brain regions that are thought to house mirror neurons, executing the same action twice in a row resulted in an attenuation of the fMRI response (the so-called adaptation effect) whereas executing and then observing the same action did not result in adaptation. The latter finding was taken to indicate that cells in these regions are not coding for both action execution and action observation, as one should find if mirror neurons exist.

Marco Iacoboni questioned this logic by arguing with the source of the underlying signal in adaptation studies:

adaptation paradigms ... change synaptic efficacy, which is invariably associated with a decoupling between action potentials and local field potential. When action potentials and local field potential do not correlate, the fMRI signal correlates with the local field potential, not the action potential. This means that Caramazza is not imaging action potentials. And guess what? Mirror neurons are defined by patterns of action potential activity. (see the full commentary here)

Iacoboni cited a paper by Bartels, Logothetis, and Moutoussis (2008) to support his claim. I wouldn't argue with Iacoboni's general argument that because fMRI may not be measuring spiking activity we cannot conclude from the PNAS study that mirror neurons do not exist (way too many negatives in that sentence but you get the idea). I would argue however with his apparent confidence regarding what the source of the fMRI adaptation signal is.

What Bartels et al. argue (convincingly) is that the BOLD response is complex, being driven by some combination of spiking activity and dendro-somatic activity, the later thought to be reflected in local field potentials. These two types of activity can dissociate and it is possible (likely even) that much of the BOLD signal under some circumstances reflects primarily non-spiking activity. In a specific case where detailed single-cell neurophysiology is available, namely direction of motion specificity in area MT, Bartels et al. argue that adaptation effects found in MT for direction selectivity do not reflect adaptation in spiking activity in MT but rather are a downstream reflection of such adaptation which happens elsewhere. The more general point they make is this:

the presence or absence of adaptation in an area measured using fMRI therefore does not allow for the conclusive inference of either the presence or absence of the neural property in question

So from the Caramazza study we cannot conclude whether cells in their ROIs exhibited adaptation or not in their spiking patterns because the adaption that was observed (e.g., for executing an action twice, the E-E condition) might be nothing more than a downstream reflection (inputs from) the site of the actual adaptation.

Let's run with this possibility. Suppose some upstream area outside the presumed human "mirror system" is actually where the E-E adaptation effect is occurring (note that the human mirror system has been identified using fMRI with its uncertain signal source). In this unidentified area that is sensitive to action execution we might expect to find not only adaptation for E-E events but, if mirror neurons exist, also for E-O (execute-observe) events. Yet no such adaptation was found.

Put differently, Bartels et al. arguments apply to making inferences about the cell properties within the adapting region identified with fMRI, not about making inferences regarding the existence of a cell population somewhere that shows adaptation.

For this reason, I think it is a serious mistake to dub Caramazza and colleagues study as fatally flawed and disregard it completely. It failed to identify any evidence of neural adaptation between action perception and action execution within the human "mirror system". Assuming a "downstream" argument for adaptation effects, what this might mean is that the human "mirror system" is treating action observation and action execution as distinct types of events either because there are no cells in this system that respond to both or because inputs to these regions treated these events as distinct.

At the same time, I think it would be a serious mistake to take this study as conclusive evidence that mirror neurons do not exist in humans. In fact, given our limited understanding of the source of the fMRI signal, I have a hard time taking any single fMRI study as conclusive evidence for anything. One needs converging results from multiple methods to make any kind of strong conclusions.


BARTELS, A., LOGOTHETIS, N., & MOUTOUSSIS, K. (2008). fMRI and its interpretations: an illustration on directional selectivity in area V5/MT Trends in Neurosciences, 31 (9), 444-453 DOI: 10.1016/j.tins.2008.06.004


Tom said...

The paper is out now in the 'early access' section of the PNAS website. I've read it and, even if we accept (for the time being) the premise that mirror neurons do show adaptation effects as measured with fMRI, I'm afraid to say that I don't find it at all convincing. What kills it for me straight away is that in the observe-observe condion (i.e. repeated observation of an action) they only find adaptation effects significant at p<.05 in one ROI - left lateral occipital cortex - with no significant adaptation in homotopic right occipital cortex (See panel A of fig 2). How on earth can you draw any conclusions from an absence of adaptation for the execute-observe condition when you can't find expected adaptation effects in visual cortex for repeated presentation of the same visual stimuli? Unfortunately, this study is underpowered.

The Vlad said...

The 'mirror-neuron theory', and it's attendant link to ambiguous fMRI and physiological data, reminds me of the (potentially apocryphal) Pauli quote "It's not even wrong."

Greig de Zubicaray said...

I have read several comments across this and other blogs to the effect that Caramazza's study with a sample size of N = 12 lacked power to detect an effect. It is worth noting that the replication Experiment 2 of Chong et al. in Current Biology used a sample size of 9 to demonstrate adaptation using a cluster ROI derived from Experiment 1 (N = 17). Dinstein et al. (2007; J Neurophysiology) used N = 13.

Field strength/SNR considerations should also play a part in any critique in terms of a given fMRI study's capacity to detect an effect. Caramazza's study was performed at 4T, Experiment 1 of Chong et al. was performed at 1.5T while Experiment 2 was performed at 3T. Dinstein et al. performed their experiment at 3T.

As for whether mirror neurons show adaptation/habituation in monkeys it is worth noting that these studies typically habituate the monkeys to both observation and performance of experimental actions via training that can last as long as 2 months (e.g., Ferrari et al., 2005, J Cognitive Neurosci, 17, 212-226). Whether you can train an undergraduate student for this period of time remains to be demonstrated!!!