Thread by @BrainyActs

Our new Nature Communications paper was published today (open access), and it's a pretty big deal. I'll tell you why (Thread). @NatureComms nature.com/articles/s4146…

This paper addresses a number of foundational issues in the use of magnetoencephalography. One big one is the "inverse problem", which means basically that estimating brain activity, based on sensors outside the head, has infinite possible solutions.

The overall problem is similar to what was faced in WWII when German U-boats attacked Allied ships. By increasing the number of sensors (in this case, hydrophones), the estimates for the location of the subs were improved and countermeasures deployed.

MEG sensors are superconducting, quantum interference devices (SQUIDS) located in a fixed array within our CTF MEG scanner. SQUIDS and the associated amplifiers detect the tiny magnetic fluxes created when neurons are active. (pic source: mne.tools/stable/auto_ex…)

While there are 275 MEG sensors, there are ~86 billion neurons in the brain, with differences in their density across different brain regions, so pinpointing an active region in the brain is still an "underdetermined" problem with many possible solutions and uncertain resolution.

This is significant if you're using MEG to find areas of focal epilepsy in advance of surgery, or to determine important areas of cortex to spare. Poor targeting can be the difference between good (curative) and bad (debilitating) outcomes. (Pic source: pediatricneurosurgery.org/diagnosis/epil…)

So, how could we provide ground truth to MEG signals? We could have rigged up a model of an emitter in a saline bath. Problem with that is that the brain isn't homogeneous, and for this to be helpful to patients, the model needed to reflect the complexity of the human brain.

So we did two things: 1) We developed a non-human primate model for MEG; 2) we combined this with a very precise method for activating tiny regions in the brain.

This method of activation is called optogenetics. There's a lot about this method you probably already know. For the rest, you can ask @eboyden3 or @KarlDeisseroth - here's a nice brief explainer by Ed:

In our study, we transduced small regions in hippocampus and cortex with channelrhodopsin-2 during a procedure to implant a fiberoptic with an integrated recording lead. I'd done this sort of thing before, (pubmed.ncbi.nlm.nih.gov/9130682/) but optogenetics provides numerous advantages.

We used AAV2/10-CaMKIIa-ChR2-eYFP, which was provided by our colleague, @CarolineEBass. Because it uses YFP, we could actually see where the cells were that made the proteins.

After transducing the channelrhodopsin-2 in the target regions, we waited until it was expressed in the circuits, which we could verify by recording standard field potentials - the actual electrical signal.

While inside the MEG scanner, we the piped light (either pulses or even ramps and sinewaves) through the fiber, systematically and repeatedly activating the tissue.

So, we’ve got a brain that we can activate, a means of activating, and a detection array. The recordings were analyzed using synthetic aperture magnetometry, a ”beamformer” method that in essence triangulates signals and allows statistical assignment of the signals to 🧠regions.

So 1) what did we find, and 2) why is it important?

By pinging the brain with optogenetics, we removed the uncertainty about the source of neural signals - because we now controlled the source. And because the transduced ChR-2 was in a restricted pool of neurons, we could test the ability of the algorithm to verify the source.

We demonstrated that useful MEG signals could be recovered from relatively deep sources, something that is still a bit controversial. The reliability of detection is a function of distance from the sensors, and of noise. This should add data to this discussion.

By using the algorithm to construct virtual source series (basically a simulated brain signal derived from the sensor array) we could demonstrate that a optogenetically-elicited blip in brain activity could register as a blip in the source series. Basically, whole brain e-phys.

This is important because in standard brain imaging, activity must be correlated with specific behaviors or brain states, about which there is uncertainty. We essentially leap-frogged this natural activity and played in our own signals. I love behavior, but it can be messy!

Doing this with MEG also leverages the temporal advantages of MEG. you may know this but fMRI is in essence a "delayed" signal compared with MEG, which is a direct read out of magnetic flux due to neural activity (think Maxwell's equations).

In principle this approach could be used to optimize MEG detection methods, might allow us to map the brain by pinging areas and mapping the results, or influencing ongoing activity and behaviors - including the recording and playback of sequences.

And by varying the size and spacing of transduced fields, it should be possible to more precisely determine the detection limits of MEG.

This was a group effort. @DrG_MDPhD was the MD/PhD student who grunted through this, along with student turned postdoc turned faculty @neuroptics, MEG guru extraordinaire Jennifer Stapleton, surgeon @c_constan PhD student @ERogersPhD (more to come) and James Daunais.

Just a reminder that it is open access (expensive!), so please take advantage of it. There's more to come with this work, so stay tuned. /FIN

I should have mentioned that this work was supported by NIAAA, NINDS and the VA.

Inquiries should be directly to the authors. This thread is our press release. Thanks for all of the positive interest!

@NIAAAnews @NINDSnews @VeteransHealth