Several hundred reconstructed cells from the mouse medial entorhinal cortex filling the frame, coloured by cell type: blue pyramidal cells throughout, a band of mint green stellate cells across the upper part, pink interneurons and purple glia scattered among them.

connectomic reconstructionmouse mec

Medial entorhinal cortex (MEC)

A part of the brain that helps keeps track of where you are. We imaged a block of it in electron microscopy, and now it needs mapping. It is the newest volume open to EyeWire citizen scientists in EyeWire II.

62,631 nuclei marked 2.0 × 1.3 × 0.45 mm

The brain's map of space

Put a mouse in a room and record a single cell in its medial entorhinal cortex, and something strange happens. The cell fires when the animal stands in one spot, then fires again a fixed distance away, and again, and again, until the firing positions trace a hexagonal grid across the whole floor. These are grid cells. Edvard and May‑Britt Moser found them in 2005, and shared the 2014 Nobel Prize in Physiology or Medicine with John O'Keefe for the work.

Nobody knows how the grid is built. The leading explanations disagree about what the wiring should look like, which is exactly the kind of argument a connectome can settle. The medial entorhinal cortex is also the main gateway from the cortex into the hippocampus, so its wiring shapes what the memory system receives in the first place. It is one of the first regions to degenerate in Alzheimer's disease.

The region sits at the back and underside of the cortex, in the retrohippocampal area, pressed up against the hippocampus it feeds. It is a thin sheet of cells folded over on itself, which is why a block of it a couple of millimetres across contains every cortical layer at once. You can see exactly where it is in the interactive mouse brain.

The block, to scale
Wireframe of the imaged block with scale bars marked every 100 micrometres and labelled to 2000. Two reconstructed cells sit inside it, one green with a dense arbor and a long process, one pink and smaller.
Loading the block, about 7 MB of meshes

Two reconstructed cells inside the imaged volume, at true relative size and in the positions they actually occupy.

Open full screen

Dataset Stats

Every nucleus in the block has been marked with a point, assigned to a cortical layer, and sorted into a cluster by nucleus size. These are early numbers and will change as proofreading advances.

62,631
Nuclei marked in the block
12,742
Presumed pyramidal cells
1,370
Presumed stellate cells
48,519
Nuclei with no cell type yet

How many neurons? Nobody knows yet

Of the 62,631 nuclei, 14,112 fall into the two clusters that carry a cell type name, pyramidal and stellate. The other 48,519 sit in six unnamed clusters. Some of those are neurons, interneurons especially, and some are glia, blood vessels and other non‑neuronal cells.

So the number of neurons in this block is at least 14,112 and certainly a good deal higher, and pinning it down is open work. The clustering is by nucleus size, which correlates with cell type but does not determine it, so every type on this page is a presumed type until a human has looked at the cell's shape.

Nuclei by layer

PIAL SURFACE I Layer I few cell bodies, mostly fibres 7,871 41 per µm of depth II Layer II stellate cells, and the pyramidal islands 10,463 71 per µm of depth III Layer III pyramidal cells 14,203 103 per µm of depth IV Layer IV lamina dissecans, the cell poor band 4,200 38 per µm of depth V Layer V the densest layer in this block 16,215 149 per µm of depth VI Layer VI deepest layer, then white matter 9,679 64 per µm of depth WHITE MATTER NUCLEI MARKED

Layer boundaries are drawn roughly. They count every nucleus, not just neurons.

Nuclei marked per layer, 62,631 in total
LayerNuclei
I7,871
II10,463
III14,203
IV4,200
V16,215
VI9,679

Everything we have pulled out so far

A sample of reconstructed cells, each drawn where it actually sits in the block and coloured by type. This is a sample, not a census: the block holds tens of thousands of cells and these are the ones picked out so far.

Descending through the cortex one layer at a time. Watch layer IV. It holds 4,200 marked nuclei against 14,203 in layer III and 16,215 in layer V: this is the lamina dissecans, the cell poor band separating the superficial layers from the deep ones, and one of the features that makes entorhinal cortex recognisable down a microscope. The scale shows how many cells are drawn against how many nuclei are marked in each layer, because the sample is not proportional and the drawn counts on their own would give the wrong impression.

The three glial types are drawn as one colour here. The automatic type prediction calls glia by nucleus size, and when an expert checked two of them, both were wrong: the cell called an oligodendrocyte was an astrocyte, and the one called a microglia was an oligodendrocyte. Splitting them into three colours would be claiming something this project cannot yet support, so until they are proofread they are simply glia.

Meet the cells of MEC

Here are some of the most common cell types found in the medial entorhinal cortex. We will be exploring them. Every one below is a real reconstruction from this block, and each card shows the segment ID so you can open it yourself.

Loading the cells in 3D, about 10 MB. Drag any card to turn it.
Render pending
A stellate cell from layer two of the medial entorhinal cortex, rendered in mint green on black. Dendrites radiate outward from a bright cell body in every direction like a star, and a single axon descends below.

Layer II

Stellate cell

The grid cell candidate. Dendrites spread out from the soma in every direction, like a star, and the axon leaves for the hippocampus. Layer II of the medial entorhinal cortex is where grid firing was first recorded.

1,370 presumed in this block, 79% of them in layer II.

More on stellate cells →
Render pending
A pyramidal cell rendered in blue on black. A dominant dendrite rises from the top of the cell body and shorter dendrites spread around its base.

Layers II and III

Pyramidal cell

The other principal cell. One dominant dendrite rises from the top of the soma and a skirt of shorter ones spreads around its base. In the medial entorhinal cortex pyramidal and stellate cells grade into one another, and there is a real argument that they are a continuum rather than two types, so this one was chosen by eye as a clear example.

12,742 presumed in this block.

Render pending
An inhibitory interneuron rendered in pink on black, with a dense local arbor around the cell body.

Layers II and III

Inhibitory interneuron

Local cells that quiet their neighbours, and the reason this volume needs people. They come in several shapes with no single look, and nucleus size does not separate them from pyramidal cells, so an automatic pass cannot find them. Human eyes can.

480 labelled so far. How many there are is not known.

Render pending
An astrocyte rendered in lilac on black. A small cell body with many fine processes spreading in every direction into a bushy cloud.

Not a neuron

Astrocyte

A star-shaped support cell whose fine processes wrap synapses and blood vessels. It is easy to mistake for a small neuron at a glance, which is part of why the neuron count in this block is still open.

188 labelled so far. This one is still an automatic call, and the glial predictions have already been wrong once.

Render pending
An oligodendrocyte rendered in cyan on black. A compact cell body with many branching processes, several of which close into small loops where the cell has wrapped itself around an axon.

Not a neuron

Oligodendrocyte

The cell that makes myelin. Follow the processes out and you will find small closed loops: each one is this cell wrapped right around a myelinated axon. A single oligodendrocyte insulates many axons at once.

137 labelled so far, but the automatic labels for glia are not reliable yet.

Render pending
A microglial cell rendered in gold on black. A small cell body with fine, heavily branched processes reaching outward.

Not a neuron

Microglia

The brain's resident immune cell. A small body with fine, heavily branched processes that are constantly reaching out and sampling the tissue around them. Cells like this are part of why the neuron count is still an open question.

24 labelled so far. Intact examples are rare, because microglia merge easily with the cells around them.

Render pending
A small bipolar cell rendered in steel blue on black. A compact cell body with one process leaving the top and one the bottom.

Layer II

Bipolar cell

Two processes, one out of each end of the cell body, and very little else. The least certain type on this page: only five have been labelled in the whole block, so there is almost nothing to check an example against.

5 labelled so far.

Stellate cells

The cell type this region is known for, and the one most closely tied to grid firing. Here is where they sit and what is known about them in this block.

Stellate cells are the principal cells of layer II. Dendrites leave the soma in every direction rather than in one dominant trunk with a skirt beneath it, which is what separates them from pyramidal cells, and their axons carry the output of this cortex into the hippocampus. Grid firing was first recorded in layer II of the medial entorhinal cortex, and stellate cells are the population most often held responsible for it.

They are where the textbook says they should be. Of the 1,370 presumed stellate nuclei in this block, 1,085 are in layer II. That is 79%, and it comes out of a clustering by nucleus size that never saw the layer labels, so the layer and the type agree without either being told about the other.

Presumed stellate nuclei per layer, 1,370 in total
LayerStellate
I42
II1,085
III223
IV8
V10
VI2

They avoid the pyramidal patches. Layer II is not a uniform sheet: pyramidal cells gather into islands and stellate cells occupy the water between them. Measured here, a stellate cell is 27 standard deviations less likely than chance to sit within 40 µm of a pyramidal cell, and that shortfall runs out to 271 µm. It is set out in Islands and ocean below.

What is still uncertain. "Stellate" here means a nucleus-size cluster, about 84% stellate where it has been checked against labelled cells. It is the cleanest of the automatic type calls on this page, and it is still a prediction rather than a cell anyone has looked at. The example in the gallery above is one we would particularly like proofread.

Islands and ocean

Layer II of this cortex is not a uniform sheet of cells. Pyramidal cells gather into patches, and stellate cells sit in the water between them. You can see it in the block before a single synapse has been traced, using nothing but the positions of the cell nuclei.

Three panels. Left, layer two seen face on: a blue density map of pyramidal cell nuclei showing distinct bright islands, with mint coloured stellate cell nuclei scattered in the darker gaps between them. Middle, the same numbers of each drawn at random from layer two, whose density map is almost flat and whose points show no such arrangement. Right, a graph in standard deviations from chance: pyramidal cells near other pyramidal cells run far above chance at short range and fall to zero by about 380 micrometres, while stellate cells near a pyramidal cell run far below chance and cross zero at 271 micrometres.
Layer II seen face on, looking through the cortical surface. Blue is the density of pyramidal cell nuclei, mint dots are stellate cell nuclei. Middle panel is the control. Both maps share their axes and their colour scale, so the comparison is fair.

The control is the argument. Comparing against points scattered evenly in a box would find structure in any dataset, because layer II is a curved sheet inside a block and not a box. So every comparison here is against the same number of nuclei drawn at random from layer II itself, two hundred times over, which holds the sheet's shape and its density fixed.

Pyramidal cells sit near other pyramidal cells far more often than chance allows, 51 standard deviations above it at 40 µm. That excess then dies away and reaches zero by about 380 µm, which is what a patch of finite size looks like rather than a general tendency to clump.

Stellate cells do the opposite. Near a pyramidal cell there are far fewer of them than chance allows, 27 standard deviations below at 40 µm, and the shortfall persists out to 271 µm before turning into a mild excess beyond it. The two cell types are not merely clustered. They are arranged around each other.

None of this is new biology. Calbindin positive pyramidal cells in layer II of the medial entorhinal cortex were shown to sit in a patchy, grid like layout by Ray and colleagues in 2014, with reelin positive stellate cells occupying the space between the patches, and the arrangement was tied to grid activity by Naumann and colleagues in 2018. What is worth something here is that a reconstruction of this block recovers that layout independently, which is a good sign about the data. Zhihao Zheng spotted it in this volume straight away.

What this does not say. "Pyramidal" and "stellate" here mean nucleus size clusters, not cells anyone has looked at. The stellate cluster is about 84% stellate where it has been checked and sits almost entirely in layer II, where stellate cells belong. Mislabelled cells would blur an arrangement like this rather than create one, so the direction holds even though the labels are not exact. And none of it is about wiring yet. It says where cell bodies sit. The point of tracing the connectome is to find out whether the wiring respects it.

The numbers (JSON)
Layer II from the pial surface, looking straight down. Only the 108 layer II neurons reconstructed so far: 46 stellate and 62 pyramidal. Seen from the side, layer II is a band that hides its own arrangement; seen from above, the arrangement is the picture. Pyramidal cells are drawn in gold here rather than their usual blue, because beside the mint of the stellate cells the two are almost impossible to tell apart, and telling them apart is the entire point of this one.

How to help

Open the volume in EyeWire II and start looking.

MEC is now in the dataset switcher in EyeWire II. Open the app, click Data in the top bar, and pick Medial Entorhinal Cortex. You will need to sign in with Google, twice, because the volume and the annotations live behind separate services.

Use a computer for this one. The phone build of EyeWire II does not carry MEC yet, so the button below opens the desktop app on purpose.

What works today: viewing the volume, loading cells, and proofreading them. If a reconstruction has two cells fused together you can split them, and if a branch has been cut off you can merge it back on. That is the core of the work.

What does not work yet: marking a cell complete, assigning a cell type, and anything that counts toward the leaderboard. Those write to annotation tables that have not been set up for this volume yet. We would rather tell you that plainly than have the buttons fail quietly. They are coming, and your proofreading in the meantime is real and is saved.

Open EyeWire II New here? Start with the guide Watch a tutorial

Gallery

Everything on this page is rendered from the reconstruction itself. No cell here was drawn, straightened or stood in for by a model.

Download the print render, 6400 × 3600, 11 MB
Through the middle of it. The camera weaves down the long axis of the block and the cells light up as it reaches them, so a trail builds behind it, and at the end it pulls back and they are all still there. Cells that come right up against the lens fade out instead of blocking it. The 244 cells the path passes within 150 µm of carry a mesh built at four times the face density of the rest: at that range the version used for the population shots is visibly faceted.
One cell at a time. A slow drift from one cell body to the next through a real lens, with the focus racking onto each in turn. The focus distance is measured to the cell body every frame, so what is sharp is sharp because of where it actually sits in the tissue. Everything out of the plane goes soft, which is the only way a single cell can be picked out of a volume this crowded without deleting its neighbours.
782 cells, arriving. One at a time to begin with, then faster, until the whole reconstruction is standing. They arrive in a scattered order rather than sweeping across, so what fills in is the volume and not a wipe.
Two millimetres to eight micrometres, without a cut. One continuous zoom from the whole imaged block down to a single dendrite and the spines along it. The approach is exponential rather than linear, so every factor of ten gets the same amount of screen time. It stops where it does because that is where the meshes stop being able to hold a spine: the thirteen cells the camera ends up among were rebuilt at 75 triangles per square µm of surface, about 0.16 µm between neighbouring points, and the last frame is roughly fifty of those across.
Everyone in the block, sorted, then put back. All 782 cells leave their true positions for a sphere where each class holds its own band, hold there long enough to be counted, and then return to where they actually are. The width of a band is how many cells of that class have been reconstructed so far, which is not the same as how many are in the tissue.
One slow revolution. Every cell in its own colour rather than its class colour, which makes this the one render on the page you cannot read anything off. It is here because the shape of the reconstruction is worth looking at on its own.
A sweep through the depth. A lit slab travelling along the sectioning axis, the thin one. The whole block is only 447 µm front to back against 2031 µm across, and the cells outside the slab are held faint rather than hidden so you can see where the lit sheet sits inside the whole volume.
Reconstructed neurons filling a wireframe box that represents the imaged block. Mint green stellate cells and pink interneurons sit in a dense band across the upper part, blue pyramidal cells spread below them, and purple glia are scattered through. Two long green axons run out towards the lower right.
Cells sit at their true positions and true relative sizes inside the imaged block. The lower part of the box is empty because it lies below layer VI, where no nuclei have been marked: probing it finds tissue, but almost all of it is still in small pieces rather than whole cells.
One stellate cell, turning. The grid cell candidate, at full mesh detail. Dendrites leave the soma in every direction rather than in an apical trunk and a basal skirt, which is what separates a stellate from a pyramidal cell, and the bumps along them are dendritic spines. One axon runs away downward toward the hippocampus.
One type at a time. Each class on its own, then all of them fading in together at the end.
A full turn. At the quarter turn you can see what the still image hides: the imaged block is a thin slab, 447 µm deep against 2031 µm wide.

Who made this

The volume was imaged, aligned and segmented by Eric Hammerschmith with Zhihao Zheng in the Seung Lab at Princeton. A team of professional proofreaders has been working through it ahead of this release. The reconstruction you see on this page came out of their work.

This page and the renders on it were made by Amy Sterling. Every image here is rendered from the reconstruction itself: no cell was drawn, straightened, or stood in for by a model.