44 The Oldest Map
How Comparative Neuroscience Reconstructs an Ancient Hippocampal System
44.1 Different brains, one historical problem
Chapter 43 introduced the hippocampal formation as part of an ancient system for learning relations among places, states, events, and transitions. That claim creates a historical problem. The mammalian hippocampal formation is laminated and curves along the medial hemisphere; in humans, much of it is folded into the medial temporal lobe. The avian hippocampal formation lies as a differently organized sheet along the dorsomedial forebrain. Nonavian reptiles possess medial and dorsomedial cortical fields rather than a mammalian-looking hippocampal curl. Amphibians have a comparatively simple medial pallium. In teleost fishes, the candidate hippocampal territory occupies a dorsolateral position in an everted forebrain.
No single adult shape runs through that series. Even the apparent position changes. A method that searched for a small mammalian hippocampus inside every other vertebrate brain would therefore fail before the comparison began.
The problem is broader than the hippocampus. Evolution rarely preserves an adult organ by copying its final geometry into every descendant. Developmental tissue can expand, fold, migrate, subdivide, and enter new circuits. A structure inherited from a common ancestor can become difficult to recognize, while unrelated structures can acquire superficially similar shapes or functions. Comparative neuroscience must therefore distinguish homology, similarity caused by common descent, from analogy, similarity produced independently.
The answer cannot come from one privileged feature. Adult morphology matters, but it can be transformed. Developmental position matters, but boundaries can shift. Gene expression identifies cell populations and territories, but genes can be reused. Connectivity and physiology reveal circuit organization, but similar computations can evolve in different tissue. Lesions show what a region contributes to behavior, but similar functions alone do not establish common ancestry.
The strongest reconstruction comes from convergence.
Across the bony-vertebrate lineages examined here, an ancient medial-pallial system can be recognized through developmental topology, molecular specification, cell types, connectivity, physiology, and behavior. Evolution has rearranged and elaborated that system differently in different lineages. Its clearest conserved contribution is flexible spatial and relational memory: navigation is an ancient expression of that capacity, not evidence that every nonspatial function was added later.
The broad regional correspondence is better supported than any complete translation of subfields. Mammalian, avian, reptilian, amphibian, and teleost forebrains contain related hippocampal or medial-pallial systems. The exact matching of every field is a different question. A teleost region can belong to an ancient hippocampal territory without being a miniature dentate gyrus, CA3, or subiculum. The chapter will keep those levels separate.
44.2 How homology is reconstructed
The methods introduced in Chapter 43 allow us to ask a more precise question here: which territory corresponds to which, and at what level? A whole developmental region, one adult subdivision, a neuronal population, and a circuit are different objects of comparison. Evidence for one does not automatically establish all the others.
Within the dorsal telencephalon, the pallium is regionalized during development. In the mouse and chicken comparisons considered below, the medial pallium can be recognized beside the dorsal midline, the roof plate, and the cortical hem, a signaling territory at the pallial edge. The question is whether tissue in another lineage occupies a corresponding developmental position and follows a related program, even when its adult shape differs [@abellan2014hippocampal].
Developmental topology concerns the relationships among neighboring tissues. It is more informative than adult compass direction alone. A territory can expand or be displaced while retaining evidence of where it arose and which structures bordered it. This distinction will become decisive in the teleost forebrain, where the ventricular surface has a very different relation to the adult pallium.
Gene-expression patterns must be interpreted within those relationships. The RNA and protein methods described in the overview reveal different aspects of a developmental program; neither turns a single marker into a certificate of ancestry. A gene expressed in mammalian dentate granule cells can also occur elsewhere. A stronger comparison asks whether combinations of markers, developmental boundaries, and cellular organization align.
Single-cell transcriptomics extends the comparison from a few selected markers to expression profiles across many genes in individual cells. Investigators group cells with related profiles, compare those groups across species, and use anatomical labeling or spatial transcriptomics to locate them. A similar cell population can occur within differently organized regions. Conversely, one anatomical field can contain several populations with different evolutionary relationships. Expression similarity becomes most informative when it agrees with developmental and anatomical evidence.
Connections and behavior provide further tests. A candidate hippocampal territory should participate in an appropriate circuit, and damage should impair a relevant capacity without merely preventing the animal from sensing or moving. These findings test function; they do not independently establish ancestry. The strength of the reconstruction lies in agreement among evidence gathered at several levels.
44.3 The medial pallium across tetrapods
The mouse–chicken comparison shows how a broad correspondence becomes a set of testable anatomical proposals. Antonio Abellán, Ester Desfilis, and Loreta Medina compared developmental expression of seven regulatory genes with the location of the tissue. Early medial-pallial ventricular zones in both species expressed Lef1, Lhx2, and Lhx9, without the Lhx5 expression found in some neighboring territories. Finer combinations suggested subdivisions within that shared territory [@abellan2014hippocampal].
The chicken’s V-shaped area illustrates why exact boundaries matter. Much of it, together with ventral hippocampal tissue, expressed Prox1 and was proposed as DG-like. Its dorsal tip lacked Prox1 and Lmo3, prompting a different, CA3-like comparison. Thus, even within this one proposal, “the V-shaped area is the dentate gyrus” loses an essential distinction. The evidence concerns developing chicken tissue, not a universal renaming of every bird’s adult hippocampal atlas.
Sara Jiménez and Nerea Moreno examined the frog Xenopus laevis and the turtle Trachemys scripta. They distinguished Prox1-positive medial territory from a more dorsal territory characterized by Er81/Lmo4. The former was compared with dentate gyrus and the latter with Ammon’s horn. In the turtle, Prox1 and Lmo4 were detected as proteins, while Er81 was localized as RNA. In Xenopus, Prox1 labeling appeared only at later larval stages. A negative result at an earlier stage would therefore not establish the absence of the corresponding territory [@jimenez2022medialpallium].
The principal comparisons can be kept explicit without making the proposed names official synonyms:
| Species and territory examined | Proposed mammalian comparison | Basis and limit |
|---|---|---|
| Chicken: much of the V-shaped area and ventral hippocampus | Dentate-like | Developmental position and a combination including Prox1; excludes the dorsal tip in the Abellán scheme. |
| Chicken: dorsal tip of the V-shaped area | CA3-like | Position and absence of Prox1 and Lmo3 in the examined stages; a proposed field correspondence. |
| Xenopus: medial versus dorsomedial pallial subdivisions | Dentate-like versus CA-like | Prox1 versus Er81/Lmo4 patterns, interpreted across developmental stages. |
| Trachemys: medial versus dorsomedial cortex | Dentate-like versus CA-like | Related marker distributions within a different adult architecture; not an identity of all cells or connections. |
These proposals come from the two developmental studies just described [@abellan2014hippocampal; @jimenez2022medialpallium]. Their value is that each names tissue, evidence, and a comparison that can be tested further.
Cell-type studies offer such a test. Maria Antonietta Tosches and colleagues profiled cells from Trachemys and the lizard Pogona vitticeps, comparing them with mammalian datasets. Their hippocampal analyses combined transcriptomic relationships with the localization of markers such as PROX1, ETV1, and ZBTB20. Medial and dorsomedial cortical populations could be compared with dentate and CA populations, respectively. Across the pallium more broadly, the study found conserved inhibitory classes alongside substantial diversification of excitatory populations [@tosches2018evolution]. The important advance is the cellular scale of the comparison: a similar regional program need not yield an identical mixture of adult neurons.
The broad tetrapod conclusion is stronger than any exhaustive field-for-field translation. A related medial-pallial territory occurs in amphibians and amniotes, supporting its presence in their common ancestor. Birds are one living branch within reptiles, not a separate evolutionary step between reptiles and mammals. Nor does a living frog display an ancestral stage before a turtle. Every species combines inherited features with changes accumulated along its own branch.
Developmental comparison therefore constrains a historical reconstruction without replaying history. The common ancestor need not have possessed the mammalian hippocampal curl, the avian V-shaped area, or the exact adult subdivisions of a modern turtle. It possessed an ancestral system from which those arrangements could diverge. We can identify that continuity while remaining uncertain about the number and boundaries of the ancestor’s fields.
Comparative studies often use Lef1, Lhx2, Lhx9, Lmo3, Lmo4, and Prox1 as evidence about medial-pallial identity. Their value lies in combinations, position, and developmental timing. Prox1, for example, is central to mammalian dentate granule-cell identity and has helped identify DG-like territories in other vertebrates. It is not a universal stamp that independently converts any Prox1-positive field into dentate gyrus.
Experimental manipulation of Lhx2 shows why regulatory genes are more than passive labels. In mouse genetic mosaics and timed inactivations, Lhx2 acted during an early critical period to specify cortical identity and suppress alternative hem or antihem fates. Medial cells lacking Lhx2 adopted cortical-hem identity, and the resulting ectopic hem tissue could organize neighboring hippocampal fields [@mangale2008lhx2]. The experiment demonstrates a causal role in constructing the developmental boundaries used for comparative inference.
It does not follow that one gene creates “the hippocampus” by itself. The cortical hem, Wnt signaling, other transcription factors, cell interactions, migration, and later circuit development all contribute. The causal lesson is narrower and stronger: the molecular patterns used to identify pallial territories participate in making those territories.
44.4 Conserved territory does not mean identical circuitry
A human arm, a bat wing, and a whale flipper share an inherited skeletal plan, but their proportions, joints, muscles, and functions differ. Homology identifies a historical relationship; it does not make the structures interchangeable. The same rule applies to pallium. Shared regional ancestry does not require identical lamination, cellular proportions, local connections, or behavioral specialization.
Even developmental boundaries do not establish an evolutionary sequence of functions. Abellán and colleagues proposed a medial-pallial origin for mouse medial entorhinal cortex and a different caudolateral pallial origin for lateral entorhinal cortex [@abellan2014hippocampal]. That distinction cannot by itself show that a spatial “where” system evolved first and an object “what” system arrived later. A developmental comparison identifies related territories. Dating a computation requires a different argument.
44.4.1 Ecological specialization within an inherited system
Once regional identity has been established, variation becomes evidence about specialization. The food-storing birds introduced in Chapter 43 offer a useful example. Krebs and colleagues found that food-storing passerines had a larger hippocampal complex relative to brain and body size than nonstoring species in their comparison [@krebs1989specialization]. This is an association between ecological dependence on memory and investment in a homologous system. It does not identify how much any one bird remembers from the size of its hippocampus.
Three questions must remain separate. Differences among species can reflect evolutionary specialization. Differences that emerge while an animal develops can reflect the interaction of inherited programs and experience. Changes within an experienced adult address plasticity over a much shorter interval. The same anatomical measure does not make those three comparisons interchangeable.
Pravosudov and colleagues made the last distinction concrete. Experienced adult mountain chickadees maintained on limited, unpredictable food supplies showed better cache recovery and spatial performance than birds given unrestricted food. Yet the groups did not differ significantly in hippocampal volume or total neuron number [@pravosudov2002food]. Better performance need not be accompanied by a detectable enlargement. A null volumetric difference also leaves many possible cellular and synaptic changes unmeasured.
The evolutionary argument is therefore about the organization of a useful system within an animal’s way of life. A cache-recovering bird, a swimming fish, and a terrestrial mammal face different sensory and motor demands. Related pallial machinery can be modified to meet them. The next comparison shows how far the adult architecture itself can change.
44.5 Eversion rearranges the teleost pallium
Teleost fishes create the hardest case because their adult forebrain violates tetrapod expectations. The major dorsal telencephalic territory is called D, and its subdivisions are named largely by adult position: Dl is lateral, Dm medial, Dc central, and Dp posterior. Those labels describe a fish brain. They do not identify mammalian homologies.
In tetrapods, the paired telencephalic hemispheres develop through evagination. Paired outgrowths of the rostral forebrain form hemispheres around ventricular cavities, leaving the ventricular surface on their inside. The medial pallium remains associated with the dorsal midline and the ventricular system, although later growth can fold it deeply into the mammalian temporal lobe.
The teleost pallium has an everted organization. Traditional diagrams portray the dorsal walls as bending laterally outward, displacing tissue related to the medial pallium toward a dorsolateral adult position. That contrast captures the resulting topological problem, but it should not be mistaken for a literal movie of a completed pallial sheet turning inside out like a sock.
Fate mapping over the first five days of zebrafish development showed a more complex morphogenesis. Posterior telencephalic tissue is displaced laterally, the pallial ventricular zone expands strongly along the anteroposterior axis, and the posterior wall bulges dorsally as the thin tela choroidea spreads over the pallial surface. These early observations did not support uniform lateral outfolding. They also did not track every territory into the adult; later differential growth, folding, and migration remain relevant [@folgueira2012eversion].
This distinction matters because a perfect mirror reversal makes predictions that the adult brain does not consistently satisfy. A partial-eversion model proposes that the adult arrangement cannot be read as a complete reversal of tetrapod pallial sectors. Transcriptomic comparisons in a cichlid support several of that model’s regional assignments [@hegarty2024teleostatlas]. This is evidence about correspondence in the resulting anatomy, not a direct observation of every developmental movement. The teleost pallium is transformed, not merely flipped.
Teleosts and tetrapods belong to lineages that diverged near the base of the bony-vertebrate radiation. Finding related pallial territories on both sides of that division supports the presence of a hippocampal or medial-pallial system in their common ancestor [@hegarty2024teleostatlas; @rodriguez2002conservation]. That is a deep evolutionary claim, but it is not unlimited. The present comparison does not by itself reconstruct the precise organization of the earliest vertebrate forebrain, nor does it establish that every feature of a mammalian hippocampus was already present before ray-finned and lobe-finned fishes diverged. The inherited element is the regional and circuit ancestry from which later systems were built.
44.5.1 A broad correspondence inside a heterogeneous Dl
Molecular studies in adult zebrafish divided the pallium into major territories using combinations of ascl1a, eomesa, emx1, emx2, emx3, and Prox1. That work proposed a hippocampal relationship for part of Dl and an amygdalar relationship for Dm. It also divided Dl into dorsal and ventral components rather than treating the entire lateral pallium as one field [@ganz2015zebrafishpallium].
The proposal converges with older lesion evidence, but the fine-grained assignments remain active. Teleost species have undergone extensive independent diversification, and their atlases do not use every label in exactly the same way. Dlv in goldfish, Dl-vv and Dl-g in cichlids, and Dld in zebrafish cannot be translated as though they were fields in one standardized brain.
Brianna Hegarty and colleagues combined single-nucleus RNA sequencing with spatial transcriptomics in an adult cichlid telencephalon. Cell populations in the ventral-most part of Dl, Dl-vv, showed strong transcriptional similarity to mammalian CA3 populations and to hippocampal-like populations in other vertebrates. Other Dl-v populations showed similarities to mammalian subicular populations. The same analysis did not support the expected one-to-one match between the proposed granular Dl-g territory and mammalian dentate gyrus. At the regional level, Dl-g was most similar to mammalian visual cortex, while cell-level comparisons also identified relationships to populations in retrosplenial/cingulate and other cortical regions [@hegarty2024teleostatlas].
That result does not erase the teleost hippocampal pallium. It changes the resolution of the claim. The broad correspondence between ventral dorsolateral pallium and tetrapod hippocampal or medial-pallial systems is supported by development, cell types, physiology, and behavior. The proposition that one specific teleost patch is “the dentate gyrus” while another is “CA3” remains unresolved. Nor does a transcriptomic resemblance to visual cortex establish that Dl-g performs the same visual computation. The result identifies a relationship to investigate.
44.5.2 Circuits and activity provide independent tests
Connectional anatomy offers another line of evidence. In the weakly electric gymnotiform fish Apteronotus leptorhynchus, subdivisions of lateral and dorsal pallium form recursive loops. Tract tracing identified a pathway from DL to the intermediate dorsal pallium, DDi, then to a magnocellular dorsal field, DDmg, and back to DL, with recurrence also within DDi. Elliott and colleagues proposed that this organization could support operations resembling pattern separation and pattern completion in mammalian hippocampal circuitry [@elliott2017hippocampallike].
The circuit supplies a mechanism worth comparing: incoming activity can engage a recurrent population and return to modify activity in its input territory. But tract tracing does not demonstrate pattern separation or completion. Those are computational proposals to be tested with activity and behavior, and the CA3-like comparison for DDi rests on a different species and evidence base from the cichlid Dl-vv comparison. Similar labels cannot splice those atlases into one brain.
Physiology supplies a further test. Wireless recordings from DDi in freely swimming Gymnotus found sparse activity linked to movement and active electrosensory sampling. Many cells were especially active during backward swimming, a behavior that brings nearby features toward the fish’s electrosensory fovea and increases sensory sampling [@fotowat2019navigation]. The candidate hippocampal circuit was therefore engaged while the animal actively sampled its surroundings during navigation. The experiment did not establish a mammalian-style place-cell code, but it connected the proposed circuit to the behavior for which that code would be useful.
A different preparation reveals a population code for position. Tracking microscopy allowed brain-wide calcium imaging while larval zebrafish explored an arena freely. Cells meeting the study’s criteria for place encoding were strongly enriched in the telencephalon. Their collective activity could be used to decode the fish’s location, and the population representation became more orderly with experience. When visual and self-motion cues were manipulated, the cells combined multiple sources of information and could remap between environments [@yang2024zebrafishplace].
The anatomical resolution of that experiment matters. The recordings established a telencephalic spatial code in larval zebrafish; they did not assign each place-encoding cell to an adult Dl subdivision or demonstrate that every such cell belongs to a hippocampal homologue. The result is nevertheless important. A fish telencephalon can construct a distributed, experience-dependent representation of location rather than supporting navigation only through a fixed sensory-response rule. The comparative question shifts from whether spatial population coding exists in teleosts to how that coding is distributed across their transformed pallial territories.
The evidence now converges from several directions. Development shows why adult position cannot be interpreted using a tetrapod atlas alone. Molecular and cell-type comparisons identify hippocampal-like populations within Dl. Connectional work reveals recurrent pallial loops. Neural activity relates those loops to active sensing, landmarks, and location. Lesions provide the most direct test of what the system makes possible.
44.6 What damage removes: place, time, and relation
The logic of a lesion experiment depends on the comparison condition. An animal that fails to reach a hidden goal may have lost spatial memory. It may also be unable to see, swim, remain motivated, or learn any new response. A useful experiment asks whether the lesion selectively disrupts one way of organizing behavior while sparing another that uses the same sensory and motor equipment.
44.6.1 Place versus cue
Fernando Rodríguez and colleagues compared two ways of reaching a goal. In the place task, its location remained fixed relative to surrounding landmarks, and the animal approached from more than one start. In the cue task, a conspicuous marker indicated a goal whose location varied. Accurate choices could therefore depend either on environmental relations or on approaching a currently visible cue.
Turtles with medial-cortex lesions learned to reach the goal from the trained starting positions. The deficit appeared when the maze was displaced to require a new route, or when landmarks near the goal were concealed: lesioned turtles failed these tests, whereas sham-operated turtles could still locate the goal. Cue learning remained available. Goldfish trained before surgery showed a different initial outcome: lateral-pallium damage markedly impaired postoperative place performance as well as transfer to new starts. Medial- and dorsal-pallium lesions did not produce that deficit in this experiment, and cue learning was spared [@rodriguez2002conservation].
The turtle result is especially instructive. Successful training alone did not reveal how the goal was found. Changing the start or the available landmarks exposed the difference between a flexible environmental representation and a more restricted guidance strategy. The spared cue condition also argues against a general inability to perceive, move, or learn. It does not prove that every sensory or motivational process was unchanged.
These results connect turtle medial cortex and goldfish lateral pallium to flexible spatial behavior. They do not restrict either system to space.
44.6.2 Trace versus delay
A second comparison removes physical space from the center of the task. In delay conditioning, a conditioned stimulus remains present until an unconditioned event begins; the two overlap. In trace conditioning, the conditioned stimulus ends, an empty interval follows, and only then does the unconditioned event occur. The animal must preserve a relation between events that are separated in time.
Gómez and colleagues tested goldfish with ventral dorsolateral pallium lesions in an eyeblink-like classical-conditioning preparation. The lesions severely impaired trace conditioning while sparing delay conditioning [@gomez2022trace]. Preserved delay learning argues against a general failure to detect the events or acquire a conditioned response. The critical difference was the temporal separation of the events; the result does not identify the neural activity that bridges the gap.
The result does not demonstrate human episodic recollection in a fish. Trace conditioning can be acquired without a narrative self or conscious return to a past event. It does show that the teleost hippocampal pallium contributes when separated events must be related across time. A system described only as a spatial navigator is too narrow.
44.6.3 Trained pairs versus a novel relation
A third experiment asked whether goldfish could use overlapping learned relations to guide a choice they had never been taught directly. The fish learned four premise pairs: A was rewarded over B, B over C, C over D, and D over E. Each pair could be learned as a separate discrimination.
The critical test paired B with D. Those two items had never appeared together. Both had been rewarded in one premise pair and unrewarded in another, making this a more informative comparison than testing A, which was always rewarded, against E, which never was. An ordered representation of the premises predicts a preference for B. Sham-operated fish did so. Goldfish with hippocampal-pallium lesions learned the original premise pairs but failed the novel B-versus-D test [@soteloparrilla2025transitive].
Again, the spared learning defines the deficit. The lesioned fish could discriminate the stimuli and store the rewarded member of each trained pair. They were impaired when behavior required the flexible use of relations among those pairs.
The term transitive inference describes the behavioral structure of the test. It does not establish conscious syllogistic deduction or uniquely identify a linear map inside the brain. Learned values, training history, and relational representations can make competing predictions in such tasks. This lesion result supports a contribution of hippocampal pallium to the novel use of learned information; the internal code remains a separate question. Chapter 43 used the example to establish the breadth of the system. Here its force comes from the contrast between acquired premise discriminations and impaired transfer.
Together, the three lesion contrasts reveal different impairments while preserving other routes to performance. Place learning requires a location defined by relations among environmental cues. Trace conditioning requires a relation across a temporal gap. Transitive inference requires a novel choice based on relations among separately learned pairs. In each case, the hippocampal or medial-pallial lesion leaves some sensory, motor, motivational, and learning capacities intact while impairing the particular relational demand tested.
These experiments also illustrate how functional evidence enters a homology argument. The lesions do not prove that turtle medial cortex, goldfish Dl, and mammalian hippocampus descend from the same tissue. A similar behavioral deficit could, in principle, result from damage to independently evolved circuits. The anatomical and developmental comparisons establish the historical candidate. The selective behavioral deficits then test whether the candidate retained a related contribution. Agreement between those independent lines of evidence is more informative than either one alone.
Nor do the three tasks have to depend on one undifferentiated mechanism. Place learning, trace conditioning, and transitive inference differ in sensory demands, time scale, training history, and required output. The important result is not that every relation is processed identically. It is that damage to candidate hippocampal pallium repeatedly impairs behavior when separately encountered elements must be organized and used beyond the immediate cue.
44.7 What the oldest map lets us infer
Spatial navigation remains the clearest ancient window onto hippocampal function. Space provides an unusually tractable geometry. The animal occupies one location, faces one direction, encounters measurable boundaries, moves through identifiable transitions, and arrives at visible or hidden outcomes. Investigators can vary start points, rotate landmarks, move goals, and ask whether the animal has learned a place or merely a response.
That experimental clarity can be mistaken for evolutionary exclusivity. The place-versus-cue experiments show that flexible spatial organization is old. They do not show that ancestral hippocampal pallium represented space and nothing else. Trace conditioning and transitive inference show why a navigation-only description of living teleost hippocampal pallium is inadequate. Teleost pallium also contributes when relations must be maintained across time or recombined to support an untrained choice.
My reading of the comparative evidence begins with the operation rather than the textbook category. A hippocampal system differentiates elements while binding them into an organized context. It allows a cue to mean something because of where it occurred, what preceded it, what followed it, or how it relates to other learned items. It supports behavior when the answer is not supplied by one immediately available stimulus.
Navigation is one ecological expression of that operation. A route is a sequence of states connected by possible transitions. A place is defined by relations among boundaries, landmarks, directions, and goals. A shortcut requires the animal to use those relations in a combination it has not directly traversed. The spatial map is already a relational structure.
Temporal association and ordered inference reveal related pallial contributions under different task demands. They do not prove that all hippocampal functions were present in identical form in the common ancestor of teleosts and tetrapods. Human autobiographical memory requires capacities far beyond trace conditioning, including rich perceptual reconstruction, semantic knowledge, language, social identity, and a culturally extended sense of time. The continuity lies in an inherited way of organizing relations, not in a claim that a goldfish remembers a childhood afternoon.
Calling the system relational should not create a new psychological box and then assign that box to one brain structure. Relations are descriptions of task structure: among landmarks, between stimuli separated in time, or across overlapping premise pairs. Different relations can recruit different sensory systems, pallial partners, neuromodulators, and local hippocampal computations. The comparative evidence identifies a recurring demand that exposes the lesion deficit. It does not establish a unitary faculty of “relational memory” stored in one place. This distinction preserves the behavioral construct without reifying it as a self-contained neural module.
This interpretation changes how human memory should be introduced. We need not assume a pure navigation device to which evolution later attached a diary. Human episodes can be understood as an elaboration of an inherited system for binding, distinguishing, and using relations beyond the present stimulus. Lineage-specific changes in fields, cells, and connections expanded its capabilities; human cortical and cultural systems expanded the content available to it. The precise order in which spatial and nonspatial operations evolved remains unresolved.
I take spatial navigation to be an ancient expression of a broader relational capacity: organizing past experience so that its structure can guide the next action.
Map names the preservation of relations, not a picture projected onto an inner screen. Oldest names the deep ancestry of the system, not a claim that the first vertebrate hippocampus has been recovered as an unchanged object. The modern hippocampal formation is an evolutionary transformation of ancient medial-pallial circuitry.
44.8 Coda: from ancestry to mechanism
This chapter has established the historical identity of the system. Adult shape alone could not do that. Developmental topology linked differently arranged pallial territories. Gene expression and cell types strengthened the correspondence while revising some exact subfield assignments. Connectivity and physiology showed that candidate teleost territories participate in recurrent circuits engaged by active sensing and navigation. Lesions revealed a conserved requirement for flexible relations in space, time, and learned order.
Chapter 45 turns from ancestry to mechanism. Place, direction, boundary, grid, speed, and event-related activity will not appear as isolated labels pasted onto neurons. They will be treated as interacting population signals through which movement and sensation become organized structure. The question will no longer be how the hippocampal system can be recognized across vertebrates. It will be how that system constructs a world through which an animal can act.
Established findings. Developmental topology, combinatorial gene expression, cell-type comparisons, connectivity, physiology, and lesion effects support an ancient hippocampal or medial-pallial system across tetrapods and teleost fishes, placing that system deep in bony-vertebrate ancestry. Mammalian, avian, reptilian, and amphibian systems derive from comparable medial-pallial territories. Teleost development changes the relation between ventricular surfaces and adult territories; molecular and behavioral comparisons identify hippocampal-like components in dorsolateral pallium. Damage to reptilian medial cortex or teleost hippocampal pallium disrupts flexible place learning while sparing visible-cue learning. Teleost lesions also impair association across a temporal gap and novel use of overlapping learned relations.
The chapter’s organizing interpretation. Spatial navigation makes an ancient relational operation especially visible. Places, routes, events, and learned orders can all require elements to be bound according to context and transition structure, then used when one present cue does not specify the answer. Human episodic memory elaborates this architecture; it is not a separate faculty pasted onto a completed navigation module.
Open questions. The exact correspondence between teleost pallial subdivisions and mammalian dentate, CA, subicular, and retrosplenial fields remains unresolved. Developmental, transcriptomic, and connectional comparisons do not always produce one-to-one matches, and teleost species should not be forced into a single atlas. The available evidence also does not establish a strict evolutionary sequence in which spatial representation necessarily preceded every nonspatial relational operation. The strongest conclusion concerns broad regional continuity and selective impairments across several task domains, rather than one universal circuit or an experimentally recovered ancestral repertoire.