2023 Frontal Cortex Research Review
Honestly, this year has been both exciting and disappointing.
I've been following the field of frontal cortex research since I started my PhD in 2009. Fifteen years now.
Back then, our lab was still using a 3T Siemens Trio, and achieving 3mm spatial resolution was considered pretty good. And now? 7T has become widespread, and layer-specific fMRI is no longer anything new. But technological progress aside, there weren't really many breakthrough studies that truly advanced our understanding of the prefrontal cortex in 2023.
Let me start with a few noteworthy developments
Neuralynx's New Electrode Array
I personally handled the prototype of Neuralynx's Atlas E series electrodes released this year at the SfN satellite meeting in San Diego in March. 128 channels, each contact 25 micrometers, impedance controlled at 200-400kΩ. Their sales rep told me the production version would ship in September, but it got delayed until late November. Typical neurotechnology company behavior.
What's the selling point of this system? The geometric arrangement specifically optimized for primate dlPFC recording. Traditional Utah arrays have 400-micrometer contact spacing, which is too sparse for the small pyramidal cells in layer III of the prefrontal cortex. The Atlas E compresses the spacing to 150 micrometers, sacrificing some coverage area, but the single-cell isolation rate has indeed improved. Our lab's postdoc Xiao Li has been using it for two months, and the number of isolated units increased by about 40% compared to before.
Atlas E Series Specifications
It's not cheap. A complete system, including amplifier and acquisition card, is quoted at $47,000
That's more than $10,000 more expensive than similar products from Blackrock. Is it worth it? Depends on what research you're doing. If you mainly work with rodents, it's not necessary. If you're doing working memory research in monkeys, it's worth considering.
The Controversy Sparked by MIT's Nature Paper
I have to say a few words about Earl Miller's lab's paper in Nature in October. Using wide-field calcium imaging combined with Neuropixels, they claimed to have discovered directional asymmetry in gamma rhythms between the prefrontal and parietal cortex. Prefrontal to parietal is gamma, parietal back to prefrontal is beta. This conclusion itself isn't new—Engel and colleagues proposed a similar framework back in 2001. But Miller's group's data volume is indeed substantial: 12 monkeys, over 2000 recording sessions.
Where's the problem? Their analysis pipeline. I carefully reviewed the supplementary materials, and when they calculated phase-locking values, they used a 500ms time window. For gamma rhythms, this window is too long. I asked Miller this question in person at Gordon Conference last year, and he said they had done sensitivity analysis and the results were robust. But this data wasn't included in the supplementary materials.
I'm not saying this paper is wrong. Miller is a colleague I greatly respect—I read his classic work on prefrontal cortex working memory from twenty years ago over and over again when I was in grad school. But regarding this specific paper, I remain cautious.
Clinical Translation
Neuromod Technologies TMS
Specifically targeting treatment-resistant depression, with the target being the left DLPFC using the Beam F3 localization method. FDA approved this year.
Flow Neuroscience Home tDCS
This has been selling in Europe for three years and finally passed FDA this year. Priced at $399, used in conjunction with their app.
Personally, I've always had doubts about tDCS. It's not that it's completely useless—there are indeed some positive results. But the effect size is too small. If you look at Horvath's 2015 meta-analysis, the effect size for cognitive enhancement is around Cohen's d of 0.2. How clinically significant is that? Hard to say. I bought a Flow device and tried it for two weeks—didn't feel any difference. But I don't have depression in the first place, so that doesn't prove anything.
Computational Models
This year saw an explosion of papers using Transformer architectures to model prefrontal cortex function. I counted, and searching PubMed for "prefrontal cortex" AND "transformer" AND "2023" returns 47 papers. 2022 only had 12.
Frankly, I'm not optimistic about most of these. Many people just take GPT-2 or BERT, run some simple cognitive tasks, and then say "look, the hidden layer activation correlates with fMRI signals." Correlation coefficients between 0.15 and 0.3, with p-values pushed out by subject counts. Publishing one or two papers like this is fine, but now they're everywhere—it's a bit much.
"Publishing one or two papers like this is fine, but now they're everywhere—it's a bit much."
— On the Transformer-prefrontal cortex modeling trendThere are a few that are done seriously. The paper from Wei Ji Ma's group at NYU, using recurrent neural networks to model prefrontal cortex decision-making processes under uncertainty, has very solid mathematical derivations. And Stanford's Surya Ganguli—their low-dimensional manifold analysis is genuinely enlightening for understanding prefrontal population coding. The others, honestly, I just skimmed the abstracts and figures and moved on.
Optogenetics Progress
Boyden Lab's multi-color optogenetics paper this year is worth mentioning. They developed three new opsin variants with excitation wavelengths at 470nm, 590nm, and 630nm respectively, enabling independent manipulation of three cell types in the same brain region. This is very significant for prefrontal cortex research. The prefrontal cortex has too many inhibitory interneuron subtypes—PV-positive, SST-positive, VIP-positive—with completely different functions. Before, you could only manipulate one or two types; now you can independently manipulate three simultaneously, greatly expanding experimental design possibilities.
About Connectomics
There's one more thing I want to mention this year. The Allen Institute updated their mouse connectome database in July, adding projection data for prefrontal cortex subdivisions. The resolution is 25-micrometer voxels. The data volume is huge—downloading the entire dataset took me three days, using the lab's gigabit network.
The data quality is impeccable, but there's an issue. Their prefrontal cortex parcellation uses the Paxinos and Franklin atlas, and this atlas's division of the mouse prefrontal cortex has always been controversial. Where exactly is the boundary between PrL and IL? Different labs have different standards. The Allen Institute chose one standard, but that doesn't mean everyone agrees. Be careful about this when doing meta-analyses with this data.
The Equipment Pricing Problem
Finally, a complaint about equipment prices. When I was preparing the lab budget at the beginning of this year, I discovered that Neuropixels 2.0 prices had gone up again. When version 1.0 first came out, it was about $1,000 per probe; now 2.0 costs $1,800. Intan's amplifier boards have also gone up 15%. Supply chain issues are part of the reason, but I think there's also an element of manufacturers taking advantage of the situation to raise prices.
Equipment Price Changes
"Doing systems neuroscience just burns money."
— A reality of the fieldDoing systems neuroscience just burns money. Our lab's annual equipment and consumables budget is around $180,000, not counting personnel salaries. For a PhD student doing a prefrontal single-cell recording project, from training animals to publishing papers, four to five years is normal. If things go smoothly, you might publish one or two good papers; if not, you switch projects and start over. This field is not for people who want quick results.
Manual Techniques
After discussing electrophysiology and imaging, let me briefly mention manual techniques.
For craniotomy surgery, there's nothing new this year. Kopf's stereotaxic instruments are still the same models that have been used for twenty years. Some labs have started using 3D-printed custom head frames—our lab's engineer Lao Zhang made a set too, with decent precision and much cheaper than commercial products.
| Equipment | Model/Brand | Price | Notes |
|---|---|---|---|
| Viral Injection | Nanoliter 2020 | ~$15,000 | Still mainstream |
| Budget Alternative | WPI UMP3 | ~$6,000 | Haven't used, heard it's okay |
| Vibratome | Leica VT1200S | $25,000 | Gold standard |
| Chinese Alternatives | Various | ~$6,250 | Good enough for general IHC |
For viral injection, Nanoliter 2020 is still mainstream, around $15,000. Some people are promoting cheaper alternatives, like World Precision Instruments' UMP3, around $6,000. I haven't used it, but I've heard it's okay.
For perfusion and sectioning, the Leica VT1200S vibratome remains the gold standard, but the $25,000 price tag puts many labs off. Some Chinese models cost only a quarter of that, and the slice quality is good enough for general immunohistochemistry. If you're doing patch clamp, you still need to use Leica.
End of 2023 Review