The Oldest Map
How a Conserved Patch of Brain Became the Hippocampus
The Oldest Map
How a Conserved Patch of Brain Became the Hippocampus
In the last chapter we made a promise we have not yet kept. We claimed that the hippocampus is ancient — that it is not a mammalian invention and certainly not a human one, but a structure we share, in recognizable form, with birds and reptiles and amphibians and fish. We leaned on that claim. The whole argument of the unit rests on it: if memory is built on a spatial map, and the spatial map is the original and ancestral function, then the map had better be old enough to be ancestral. So now we have to show our work. How do we know the hippocampus is half a billion years old, when the brains that carry it look so different from one another that a nineteenth-century anatomist, laying a fish brain beside a human one, could be forgiven for seeing nothing in common at all?
This is the chapter where the methods you met in Unit 3 do their hardest and most rewarding work, so it is worth pausing to recall what those methods were and why they matter here.
What the bauplan taught us
In Unit 3 we used the vertebrate bauplan — the shared body plan of the brain — to disabuse ourselves of a tempting and wrong idea: that the human brain is a special kind of thing, built on a new principle. It is not. Every vertebrate brain begins as the same neural tube, with the same three primary swellings that become five, organized along the same rostral-caudal and dorsal-ventral axes, partitioned by the same molecular gradients into the same major territories — telencephalon, diencephalon, and the rest. Evolution did not invent new brains for new animals. It took the parts that were already there and modified them: stretched some, shrank others, elaborated a few past recognition. The human cortex is not a novel organ. It is the same pallium every vertebrate has, expanded and folded until it dominates the head.
And we drew one methodological lesson from all this that is about to become the engine of the entire chapter. Morphology lies. Molecules tell the truth. When you try to match brain parts across species by how they look — their shape, their position, their size — you are constantly fooled, because evolution distorts shape freely. The thing that stays constant is not the form of a structure but the genetic address at which it is built: the particular combination of regulatory genes a patch of developing tissue switches on, and the position it occupies relative to its neighbors and to the signaling centers that pattern the tube. Two structures can look completely different and still be the same structure — homologous, descended from a common ancestral part — if they are built at the same molecular address from the same developmental program. This is the key point Unit 3 left you with, and we are now going to use it to do something that looks, at first, impossible: to find the hippocampus in a fish.
The problem morphology hands us
Set four brains side by side: a mouse, a chicken, a turtle, a zebrafish.
In the mouse you can see the hippocampus with your eye, once you know where to look — the elegant curled structure, the seahorse it is named for, tucked into the medial temporal lobe, with its distinctive interlocking fields. In the chicken there is no such curl. The avian hippocampus is a flat sheet of tissue along the dorsomedial wall of the forebrain, with none of the mammalian folding; for a long time its identity was genuinely uncertain. In the turtle there is a region of simple three-layered medial cortex that nineteenth- and twentieth-century anatomists argued about for decades. And in the zebrafish the situation is not merely difficult but seems, topologically, backwards — for a reason we will come to that is one of the strangest facts in comparative neuroanatomy.
If you tried to match these four by appearance, you would fail, and you would be right to fail, because appearance is not conserved. What is conserved is the address. And so the project of this chapter is to stop looking at what these structures look like and start asking where, in the developing brain, each one is built — which sector of the embryonic pallium it grows from, adjacent to which landmarks, expressing which genes. When we do that, the four resolve into one.
The medial pallium: the address of the hippocampus
Recall from Unit 3 that the roof of the telencephalon — the pallium, the part that in mammals becomes cortex — is divided during development into a small number of longitudinal sectors running medial to lateral. The most medial of these, the strip of pallium lying right next to the roof plate and the specialized signaling tissue at the dorsal midline, is the medial pallium. And it is from the medial pallium, in every vertebrate that has been carefully examined, that the hippocampus is built.
That is the answer, stated plainly. The hippocampus of the mouse, the hippocampus of the chicken, the medial cortex of the turtle, and the corresponding region of the fish are all derivatives of the same pallial sector — the medial pallium — occupying the same position relative to the same landmarks and built by the same conserved program of regulatory genes. They look different because the medial pallium, like everything else in the brain, has been reshaped differently in each lineage. But they are the same structure. They are homologous. The map is as old as the medial pallium itself, which is to say, as old as the vertebrate forebrain.
For most students that is the level of detail the argument requires, and you can carry it forward without losing anything: the hippocampus is the derivative of the medial pallium, and the medial pallium is conserved across vertebrates, so the hippocampus is ancient. But the actual evidence — the specific genes, the way the address is read, the places where the reading gets hard and experts still disagree — is genuinely beautiful, and for those who want it, it is laid out in the deeper-dive sections that follow. Nothing in the main argument depends on opening them.
How exactly is the medial pallium identified? Not by one gene but by a combination — a molecular signature read against topological position, which is the standard logic of modern comparative neuroanatomy.
The foundational comparative work here is by Antonio Abellán, Ester Desfilis, and Loreta Medina, who examined the expression of seven regulatory genes — Lef1, Lhx2, Lhx9, Lhx5, Lmo3, Lmo4, and Prox1 — across the developing forebrains of mouse and chicken. In both species, the medial pallium is defined as the pallial sector lying immediately adjacent to the cortical hem (a signaling center at the dorsal midline) and the roof plate, and showing moderate-to-strong expression, in its ventricular zone, of Lef1, Lhx2, and Lhx9 — but not Lhx5. That combination, in that position, is the address.
Reading it against position lets them identify, in both mouse and chicken, the same set of medial-pallial derivatives: the hippocampal formation proper (the dentate gyrus, the Ammon’s-horn fields, the subiculum), the medial entorhinal cortex, and part of the amygdalo-hippocampal transition area. The chicken’s flat dorsomedial sheet, which looks nothing like the mouse’s curl, is built at the same address from the same program. Morphology hid the homology; the molecules revealed it — exactly the lesson Unit 3 promised would pay off.
Finer combinations carve the sector into subfields. The combinatorial expression of Lef1, Prox1, Lmo4, and Lmo3 picks out dentate-gyrus/CA3-like, CA1/subicular-like, and medial-entorhinal-like sectors that are recognizably comparable between the two species — evidence that not just the structure but its internal organization is conserved. Prox1 in particular is the workhorse marker of the dentate gyrus, so reliable that in the adult brain it essentially stains “dentate” across species.
There is a striking causal confirmation that these genes are not mere labels but actually build the structure. Lhx2 is a master regulator of this whole territory: in the mouse, deleting Lhx2 early causes the hippocampal primordium to transform into cortical hem tissue instead, and deleting it later causes the hippocampus to shrink drastically. The gene does not just mark the address; it specifies it. Remove the molecular instruction and the structure fails to form.
One detail that matters more than it should: the “what” stream is a different sector
Before we leave the mammalian and avian case, there is one finding tucked inside the molecular work that deserves to be pulled out into the open, because it does real work for the argument of this whole unit.
You will recall from the overview that the entorhinal cortex — the great gateway into the hippocampus — comes in two parts: a medial entorhinal cortex that carries spatial information (the home of grid cells, the “where” stream) and a lateral entorhinal cortex that carries information about objects and content (the “what” stream). We treated that as a functional division. It turns out to be a developmental one. The medial entorhinal cortex is built from the medial pallium — the same sector as the hippocampus proper. The lateral entorhinal cortex is built from a different pallial sector entirely, a distinct caudolateral territory that does not share the medial pallium’s molecular address.
Sit with what that means. The spatial stream — the “where” — is, developmentally, the core hippocampal lineage; it grows from the same ancestral patch as the map itself. The content stream — the “what” — is, developmentally, an immigrant, arriving from a different pallial territory to feed information into a spatial machine that was already there. This is not decoration. It is anatomical evidence, written into the developmental origin of the circuit, that the system is built around a spatial core, with object information added in from elsewhere. The map is the foundation; the “what” is the tenant. We will see this priority again and again, but here it is, visible in the embryology.
In the mouse, the medial entorhinal cortex expresses the medial-pallial signature and arises from that sector, contiguous with the hippocampal fields. The lateral entorhinal cortex, by contrast, arises from the dorsolateral caudal pallium (DLP) — a sector whose ventricular zone does not show the defining medial-pallial expression of Lef1. The same arrangement is identifiable in the chicken. The functional what/where division of the entorhinal cortex therefore maps onto a developmental division of pallial origin: the spatial gateway is medial-pallial, the non-spatial gateway is not. The implication the main text draws — that the spatial system is the developmental core and the object system is added in — is read directly off this difference in origin.
Down into the deep branches: amphibians and reptiles
Push the comparison below birds and mammals and the address holds.
In amphibians — the frog Xenopus is the workhorse — the medial pallium can be identified by the same program, expressing the same key regulators in the same position, and indeed the expression appears early, before the telencephalon has even finished its basic morphogenesis. This matters because amphibians sit near the base of the tetrapod radiation; finding the medial-pallial program there tells us it was present in the common ancestor of all land vertebrates. In reptiles — lizards and turtles — the medial cortex that anatomists argued over for a century is, on the molecular evidence, the medial-pallial derivative, the reptilian hippocampus. The hippocampal formation was, in all likelihood, already present and recognizable in the common ancestor of all amniotes — the ancestor that birds, reptiles, and mammals share.
So the structure runs continuously down the tetrapod line: mouse, chicken, turtle, frog, each carrying a medial pallium built to the same specification, each growing a hippocampus from it. The map is at least as old as the first land vertebrates. But the most interesting case — and the hardest — is the one that takes us off the land entirely, back into the water, to the ray-finned fishes. And here the brain plays a trick that is worth the whole chapter.
The fish that turned its brain inside out
Here is one of the strangest facts in all of comparative neuroanatomy, and it is the reason the fish hippocampus was so hard to find.
When the telencephalon develops in you, in a mouse, in a chicken — in every land vertebrate — it does so by a process called evagination. The walls of the neural tube bulge outward and then the tissue folds, so that the pallium curls over with its medial edge toward the midline and its ventricular surface on the inside. This is the arrangement Unit 3 taught you, and it is why “medial” means what it means: the medial pallium is the part nearest the middle.
Ray-finned fishes do the opposite. Their telencephalon develops by eversion — the walls fold outward and back, everting like a sock turned partly inside out, so that the whole pallium is splayed open with its ventricular surface facing outward, covered by a thin roof. The topological consequence is profound and deeply counterintuitive: eversion reverses the medial-lateral order of the pallium. The sector that is most medial in your brain ends up most lateral in the fish, and vice versa. The structure you would instinctively look for at the midline — by analogy with every land vertebrate — is, in the fish, pushed out to the lateral edge.
This is precisely the kind of trap morphology sets, and precisely the kind of trap the molecular method was built to escape. If you look for the fish hippocampus where the mammalian hippocampus sits — medially — you will not find it, and generations of anatomists were confounded for exactly this reason. But if you stop trusting position-as-it-appears and instead read the developmental address, accounting for the eversion, the hippocampal homolog reappears — not at the medial edge where intuition demands it, but laterally, in a region of the dorsal telencephalic area called Dl. The fish has a hippocampus. It is just hiding in plain sight, on the wrong side, turned out by the eversion of the whole forebrain.
And recall — this is the satisfying part — that this is the very same region whose ablation, in the goldfish experiments we met in the overview, selectively destroys allocentric spatial mapping while sparing simple cued approach. The molecular anatomy and the behavioral lesion converge on the same patch of tissue. The place the developmental address tells us is the hippocampal homolog is the place that, when removed, takes the map away. Two completely independent lines of evidence — where the structure is built and what its removal costs — point to the same conclusion. That convergence is how comparative neuroscience earns confidence in a homology.
The “simple eversion model,” dating back to studies from the late nineteenth and twentieth centuries, proposes that the medial-to-lateral sequence of pallial zones in tetrapods appears in reversed, lateral-to-medial order in teleosts. On this view the teleost Dl (dorsolateral zone) corresponds to the mammalian medial pallium (hippocampus), while Dm (dorsomedial zone) corresponds to the lateral/ventral pallium and its amygdalar derivatives — the reverse of what raw position would suggest.
Molecular work supports a version of this. Analysis of conserved markers (ascl1a, eomesa, emx1/2/3, Prox1) in adult zebrafish divides the pallium into four main territories, with the dorsal subdivision of Dl proposed as homologous to part of the mammalian hippocampal formation, and Dm as homologous to the pallial amygdala. Expression of prox1 in a subregion of Dl has been read as marking a dentate-gyrus homolog.
But you should know that this homology is genuinely unsettled at the finest grain, and the chapter would be dishonest to present it as closed. A 2024 spatially-resolved single-cell atlas of a cichlid telencephalon found that the ventral-most pole of Dl (Dl-vv) bears strong transcriptional similarity to hippocampal CA3 — consistent with the standard view — yet found that the putative dentate-gyrus region (Dl-g) was not transcriptionally similar to mammalian dentate gyrus, resembling instead mammalian visual cortex. The deep homologies are real at the level of major territories; the one-to-one matching of individual subfields across the eversion remains a live research problem. This is exactly what an honest application of the molecular method looks like — it resolves the big homologies that morphology could not, and it is candid about the seams where the resolution is not yet fine enough.
Note also the methodological point underneath all this: identifying homology in the everted brain requires the developmental and molecular approach. Position alone is actively misleading here — it points to the wrong side of the brain. The fish is the strongest possible demonstration of Unit 3’s lesson, because it is the case where trusting morphology guarantees the wrong answer.
What the deep history means
Step back from the genes and the topology and see what we have established, because it is the foundation the rest of the unit stands on.
The hippocampus is not a recent structure. It is a derivative of the medial pallium, one of the founding sectors of the vertebrate forebrain, and it is identifiable as such in mammals, birds, reptiles, amphibians, and ray-finned fishes — across animals whose common ancestors swam in oceans hundreds of millions of years ago. It has been radically reshaped in each lineage: curled and folded in mammals, flattened in birds, splayed onto the wrong side of an everted forebrain in fishes. But beneath the reshaping it is, demonstrably, one conserved structure, built at one conserved address by one conserved developmental program.
And everywhere we can test its function in these deep branches — in the navigating goldfish, the place-learning lizard, the caching bird — it is doing the same kind of work: building and consulting a map of space in the service of finding what the body needs. This is the payoff of the comparative method for our thesis. We did not simply assert in the overview that navigation is the ancestral function of the hippocampus and that memory of events is a later passenger. We can now see why that ordering is forced on us by the evidence. The structure is ancient, it is spatial in the oldest lineages that possess it, and the elaborations that will eventually let a human remember a birthday are built on top of a spatial machine that was finding food in the water long before there were birthdays to remember.
This is also, exactly, the moral of Unit 3, now made specific. The human hippocampus is not a special organ built on a new principle for a new purpose. It is the same medial-pallial derivative every vertebrate carries, modified — elaborated, expanded, repurposed — but continuous with the oldest map in the brain. Evolution did not give us a new structure for memory. It took the ancient organ of navigation and let it grow rich enough to navigate other things.
Looking ahead
We have now established two of the three legs the unit stands on. We have the function — prediction, the use of the past to prepare for the future, argued in the overview. And we have the deep history — the medial-pallial map, ancient and conserved, argued here. What we still owe is the mechanism: the actual cells, in the actual hippocampus, that build and read the map. How does a sheet of neurons represent a place at all? What is a place cell, a grid cell, a head-direction cell? How does the static map become the dynamic engine that runs routes forward into the future? That is the subject of the next chapter, where we leave the deep evolutionary past and descend, finally, into the circuit — to watch the oldest map being drawn, one firing neuron at a time.
And only after that — once we understand what the map is for, where it came from, and how it works — will we be ready to meet the human case, and the famous patient whose lost hippocampus opens almost every other textbook and will, deliberately, close this unit rather than open it.