Oh Oh!: Most Amygdala Studies of Emotions May be Wrong. The “Blobology” of fMRI Fatally Flawed?

Standard

“This discovery..casts doubt on previous findings on amygdala function that rely solely on fMRI as evidence…“amygdala activations” reported in typical fMRI studies are likely confounded by signals originating in the BVR rather than in the amygdala itself, thus raising concerns about many conclusions on the functioning of the amygdala that rely on fMRI evidence alone..”

A few years ago I met some of the best and brightest neuroscience grad students in Chicago and was surprised at the scorn they had for “imaging” as a research tool.  The following study seems an important nail in the coffin of the claims of localization of brain function.  It is also a good cautionary note on neuroscience technologies and popular tools for mistaken research.

This kind of flaw really pulls the rug out from studies of emotions/feelings, about time!

Take Aways:

  • This discovery has rather wide-ranging implications. Most immediately, it casts doubt on previous findings on amygdala function that rely solely on fMRI as evidence, in particular where reproducibility has been limited and efforts to confirm them have repeatedly proven difficult.

    This finding is confirmed here in both a conventional fMRI dataset as well as in information of meta-analyses, implying that “amygdala activations” reported in typical fMRI studies are likely confounded by signals originating in the BVR rather than in the amygdala itself, thus raising concerns about many conclusions on the functioning of the amygdala that rely on fMRI evidence alone

  • it is important to acknowledge that, while being a very convenient tool for measuring brain activity in vivo, fMRI also has severe limitations that need to be explored in detail, but are currently often ignored or downplayed.

  • The resting-state findings illustrate that the signals seen in the voxels we found activated in the emotional task in the amygdala region are most strongly correlated with signals in areas characterized by large vessels, such as around the brain stem and in the lateral fissure.

“The human amygdala is the target of a large number of imaging studies due to its [claimed] central role in emotion processing, emotional learning and its potential involvement in various psychiatric disorders….Still, there is notable heterogeneity and disagreement between fMRI studies of the amygdala, both in terms of activations in tasks and functional connectivity during rest.

Typical forms of disagreement between studies are the failure of newer studies to replicate results from earlier papers or to find any significant results in the amygdala at all. More subtle effects can be differences in lateralization between studies or unintuitive lateralization effects within a study… the paradigms and regions of interest studied, the amygdala activation is the least reproducible both at group and at single-subject level, and that reproducibility of amygdala results even decreased after physiological noise correction, suggesting that the most reproducible findings in the amygdala region might be due to physiological effects. In addition, repeatability was much lower at single-subject level than at group level, leading to a critical view on potential diagnostic uses of such data as opposed to group-level comparisons only.

Another contribution to the heterogeneity of fMRI results might be found in the type of stimuli used. Emotional faces are a typical cue to evoke amygdala activations, but difficulties arise when choosing an appropriate control condition: neutral faces are considered unreliable in this respect, so non-face control stimuli are more widely used as control condition, but bear the risk of mixing activations due to emotion with activations related to face recognition. More generally, in an early fMRI meta-analysis found that visual stimulation is more robust in inducing amygdala activation than auditory stimulation (note that none of the studies using auditory cues lead to activations in the amygdala), and that fear is the most robust emotion to evoke activation, with a much higher proportion of studies using fearful emotional cues yielding significant results in the amygdala than happiness, sadness, anger or disgust.

This difference in amygdala activation strength depending on the emotion expressed by the faces shown is corroborated by a later meta-analysis … More recent studies though found amygdala activations also with non-visual stimuli…Amygdala activations in response to emotion can be evoked in a wide variety of different ways, as shown by various studies employing different paradigms, including auditory, haptic and even intrinsic (e.g., memory recollection) stimuli.

Meta-analyses investigating the results of these studies have identified significant variability across different stimulation types, highlighting potential heterogeneity in amygdala activation patterns introduced by the paradigm design.

There exists evidence that visual stimuli are among the most robust in producing amygdala activations, with a meta-analysis on subliminal stimuli pointing into the direction that in this particular case, reproducible activations in the amygdala region could be found only in visual stimulation using faces, not with any of the other (somatic, auditory, lexical) paradigm types.

Still, even among visual stimuli, the most commonly employed type of paradigm, there is considerable variability in terms of the exact setup of the paradigm as well as in the results induced in terms of amygdala activation. The origin of heterogeneity across different types of paradigms is not yet understood…

Another issue that has been investigated as a potential source of inconsistencies is low signal-to-noise-ratio (SNR), which can be problematic in the amygdala region due to local magnetic field inhomogeneity. Indeed, time series SNR is low in many voxels in and around the amygdala and can vary greatly between left and right … Moreover, lateralization effects are only rarely tested for statistical significance and thus often represent only small, statistically insignificant differences unlikely to be replicated in later studies. For the same reason, even when a study reports amygdala activations to be significant only in one hemisphere, the difference between the left and right amygdala activations might in itself not necessarily be statistically significant, and such a result should therefore not be misinterpreted as evidence in terms of lateralization effects

Thus, in the present study, we investigated activations in the amygdala region to the presentation of emotional faces using these new techniques alongside more conventional Blood-Oxygenation-Level-Dependent (BOLD) EPI sequences to investigate the origins of signal fluctuations and their heterogeneity in this region and the effects this may have on fMRI research using standard scanning techniques.

Discussion

The increased functional sensitivity of low-TR multiband BOLD EPI made it possible to show that major signal changes measured in the amygdala region in a typical emotional task is not, in fact, located in the amygdala itself. [Oops!]

Rather, these signal changes occur in the adjacent Basal Vein of Rosenthal (BVR) that drains large regions of the medial temporal lobe and has confluences from other large veins in the amygdala region, and are therefore largely unrelated to neuronal activity in the amygdala itself.

While the suggestion that fMRI being only an indirect measure of neuronal activity is not in the least a novel concept, the possibility of signal changes in veins in the ventral brain at such a large distance from the neuronal origin have not been demonstrated before.

One should thus be careful when making generalizations on the venous structures around the amygdala, as there exist large inter-subject heterogeneity. Despite this anatomical variability, the confounding effect of the BVR on the signal measured in the amygdala region is very consistent, as it appears in multiple independent datasets under different circumstances.

The resting-state dataset, acquired during the same scan session with the same protocol, confirms that the signal in the voxels that showed highest activation in the amygdala region were characterized by strong correlation to the signal from voxels following the course of the BVR further around the brain stem towards the vein of Galen, the most common variant of the draining of the BVR, and from other voxels in regions with major vessels, like the lateral fissure.

This discovery has rather wide-ranging implications. Most immediately, it casts doubt on previous findings on amygdala function that rely solely on fMRI as evidence, in particular where reproducibility has been limited and efforts to confirm them have repeatedly proven difficult

What can be said, though, is that one should be more wary of fMRI signal changes in this region instead of attributing them to neuronal activity in the amygdala without careful analyses of potential confounding contributions. Due to the confluence of the amygdalar vein into the BVR, it is impossible to say whether—or, to what extent—signal changes measured in BVR voxels may be due to neuronal activity in the amygdala, in more distant brain regions, or both. Furthermore, it is important to acknowledge that, while being a very convenient tool for measuring brain activity in vivo, fMRI also has severe limitations that need to be explored in detail, but are currently often ignored or downplayed.

Whether a particular finding in the literature reflects amygdala or BVR signal often remains unclear, especially since the figures shown in a paper typically show slices passing through the amygdala, but give no further indication on the extent of activations in other nearby slices that might give more hints at whether the activation pattern continues to follow the BVR. In addition, the ubiquitous preprocessing step of spatial smoothing helps to diffuse BVR signal changes along most of its course except in the immediate vicinity of the amygdala, i.e. in the region of the confluence of the BVR’s striatal into the anterior peduncular segment.

The resting-state findings illustrate that the signals seen in the voxels we found activated in the emotional task in the amygdala region are most strongly correlated with signals in areas characterized by large vessels, such as around the brain stem and in the lateral fissure. This highlights that signal changes in these areas are not likely to reflect neuronal activity at the location of the measured signal change, but rather (task-related) vascular effects (possibly related to neuronal activity in more distant brain regions).

A further question to ask here is what the mechanisms are that lead to the stimulus-correlated signal changes in these voxels. The most likely candidate seems that it originates from blood drained from other brain regions which are directly activated by the task. Among the regions identified in our datasets, the fusiform gyrus appears to be the region with the largest (both by magnitude and extent) activation observed among the regions which most likely drain to the BVR…It also corroborates the observation quoted above that lateralization differences in activations in the amygdala region might be related to differences in visuo-spatial processing demands of the paradigms used, and provides a possible explanation of the mechanisms causing it.

Furthermore, a large fMRI study including 215 subjects recently identified a network active in facial recognition regardless of emotional content—this network also consisted of the amygdalae and the fusiform gyri, suggesting a connection between the two in their activation to the presentation of pictures of faces.

If this hypothesis on the origin of the signal change is true, it would also suggest that an emotional task widely used to produce activations in the amydala region actually does not seem to involve the amygdala in a way robustly measureable by fMRI.

Beyond the implications for research on the amygdala itself, if the effect indeed originates in the fusiform gyrus, then our results demonstrate that BOLD signal changes induced by neuronal activity can occur in voxels much farther from the actual source of activation than previously believed. It is thus plausible that such effects might also occur in other areas of the brain characterized by the presence of large vessels and where fMRI currently often leads to ambiguous results—the coordinates for the parahippocampal gyrus also fall on BVR voxels in our analysis, and the insula might also be a worthwile target for similar analyses.

Another intriguing possibility is that the signal changes in the BVR might have no localizable origin at all…Nevertheless, the confluence of blood with task-related oxygenation-level changes from wide areas might lead to a larger net sum effect in the veins draining these regions, and thus to a detectable signal change in the veins despite the signal changes in the individual regions contributing to it being too small to be detected in the experiment. Further research is needed to pinpoint the exact origin and mechanisms of the BVR signal fluctuations and clarify whether a local (e.g., the fusiform gyrus) or a more global origin is more plausible.

While at first glance, the implication that past investigations of emotional processing pathways in the brain have been heavily confounded by a physiological artifact seems largely negative, but the flip side of this coin is that, with current methods, the identification of this artifact can be rather easy…

original study – Boubela RN, et. al. (2015). fMRI measurements of amygdala activation are confounded by stimulus correlated signal fluctuation in nearby veins draining distant brain regions. Scientific Reports, 5 PMID: 25994551

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s