Pl. III

Part III · Intelligence in Action · Chapter 3

Learning, Memory & Reversal Learning in Octopus

Octopuses learn by nearly every paradigm tested. Foundational mid-20th-century work by J.Z. Young, Brian Boycott, and Martin Wells at the Naples Zoological Station established that Octopus vulgaris readily acquires associative and operant discriminations: presented with an object plus food (reward) or a mild electric shock (punishment), animals learn within tens of trials to attack or retreat. Wells and Young dissected two anatomically separable learning systems (Wells & Young, J. Exp. Biol., 1960s). Visual discrimination (shapes, orientation, brightness, size) is handled by the optic lobes, encoding stimuli by the pattern of receptors excited; tactile discrimination (texture, but notably not shape or weight, which the arms cannot represent) resides in the inferior frontal/subfrontal lobe system, encoding the proportion of chemo-tactile receptors excited. The vertical lobe (VL) sits atop both systems and functions as a shared memory/consolidation station rather than a primary sensory analyzer.

Reversal learning demonstrates genuine cognitive flexibility beyond simple discrimination. Historically, Wells found VL-lesioned animals were impaired specifically at reversing an established rough/smooth tactile discrimination. Modern work is methodologically careful: *Bublitz, Dehnhardt & Hanke (2021, Front. Behav. Neurosci.)* trained O. vulgaris on a left/right spatial reversal task—animals completed 2–13 successive reversals, with the best performer reaching 13. Critically, octopuses given positive reinforcement only often failed to learn at all; introducing an explicit incorrect-choice signal (ICS) transformed performance, letting them solve the task in a few sessions and progressively reduce errors across reversals (serial reversal improvement). Bublitz et al. (2017, Front. Physiol.) earlier cautioned that many older "reversal" claims conflated methodology with cognition—so the flexibility is real but the classic literature is partly contested.

Spatial learning and navigation are well documented. *Boal, Dunham, Williams & Hanlon (2000, J. Comp. Psychol.)* gave Octopus bimaculoides one open escape burrow among six; animals learned the location and retained it for ≈1 week, and reduced movement over exposure consistent with exploratory/latent learning. Moriyama & Gunji (1997, Ethology) showed maze/detour solving, with animals shifting from inefficient tactile groping to efficient swimming. Field and lab work by Jennifer Mather established that foraging octopuses use route-based spatial memory and even show landmark use during homing.

Memory is biphasic. Sanders and Young's lesion studies showed a short-term phase and a distinct long-term phase. Sanders (1970) quantified long-term tactile retention: performance fell 25% by 8 days, 50% by 24 days, 75% by 53 days, and 90% by 96 days—true multi-month memory. VL removal spares acquisition and short-term recall but degrades long-term storage, dissociating the two systems.

The cellular basis of consolidation is the field's crown jewel, driven by Binyamin Hochner's Hebrew University lab. Using a VL slice preparation, *Hochner et al. (2003, J. Neurophysiol.)* found a robust, activity-dependent, vertebrate-hippocampus-like LTP of glutamatergic field potentials at the superior-frontal-lobe (SFL)→amacrine synapse—striking convergent evolution. Shomrat, Zarrella, Fiorito & Hochner (2008, Current Biology) linked this LTP causally to behavior via a passive-avoidance task (attacking a negatively reinforced red ball; paradigm from Sanders & Barlow, 1971): tetanizing the VL tract accelerated short-term learning while transecting it slowed learning, yet both manipulations impaired next-day long-term recall—proof the VL and its LTP are required specifically for consolidation, not acquisition. Surprisingly for a vertebrate parallel, this LTP is NMDA-receptor-independent; the Hochner group (Turchetti-Maia, Stern-Mentch and colleagues, ≈2019–2024) identified a novel nitric-oxide (NO)-dependent "molecular memory switch": activity persistently activates NO synthase, producing presynaptic facilitation of glutamate release—NOS inhibitors block long-term LTP expression.

Connectomics (Bidel, Meirovitch, Hochner et al., 2023, eLife) mapped the VL's ≈25 million neurons: 89.3% simple amacrine interneurons (SAMs, ≈22 million) plus a newly discovered ≈1.6% complex amacrine (CAM) class. Remarkably, each SAM receives only a single synaptic input on a non-bifurcating neurite—a massive 1:12 "fan-out" expansion unlike the convergent "fan-in" of the cerebellum or insect mushroom body, suggesting an independently evolved associative architecture.

Octopuses also show observational learning: Fiorito & Scotto (1992, Science) reported naïve observers, after watching a trained demonstrator, selected the same colored ball and did so faster than by direct conditioning—an early (if debated) claim of social learning in an invertebrate. Open questions: the reality/mechanism of observational learning, whether octopuses form spatial "cognitive maps" versus route memories, and how a lobe-based memory relates to distributed arm-nervous-system learning.

Striking / counterintuitive:

• Octopus vertical-lobe LTP is strikingly hippocampus-like yet NMDA-receptor-INDEPENDENT, instead relying on a nitric-oxide 'molecular memory switch' — convergent function, different molecular hardware.
• In reversal learning, octopuses given only positive reinforcement often fail to learn; adding an explicit 'wrong-choice' signal is what unlocks flexible learning (Bublitz et al. 2021).
• The VL connectome shows a 1:12 'fan-out' where each amacrine interneuron gets just a SINGLE input — the opposite of the convergent 'fan-in' seen in the cerebellum and insect mushroom body, implying independently evolved associative circuitry.
• Blocking OR over-driving (saturating) VL plasticity both wreck next-day memory while having opposite effects on same-day learning speed — a clean dissociation of acquisition from consolidation.
• Octopus arms cannot learn object shape or weight by touch — the tactile system encodes only the proportion of receptors firing, so it discriminates texture but is 'blind' to geometry.
• Tactile memory is genuinely long-term: measurable retention persists for months, decaying only ≈50% at 24 days and ≈90% at 96 days.

Open questions:

• Is Fiorito & Scotto's (1992) observational learning genuine social learning, or explainable by simpler stimulus-enhancement/local-enhancement mechanisms? It remains debated and imperfectly replicated.
• Do octopuses form true allocentric 'cognitive maps' during navigation, or rely on route memories and egocentric/landmark strategies?
• How does centralized vertical-lobe memory integrate with learning distributed in the arm/peripheral nervous system (which contains ≈2/3 of neurons)?
• What is the full molecular cascade of the NO-dependent LTP switch and how does it achieve months-long persistence without NMDA receptors?
• What is the functional role of the newly discovered complex amacrine (CAM) cell class in the vertical lobe circuit?
• How comparable are learning capacities and memory mechanisms across octopus species (most mechanistic work is on O. vulgaris) and between octopus, cuttlefish, and squid?

Key researchers/labs: Binyamin Hochner (Hebrew University of Jerusalem) — vertical lobe electrophysiology, LTP, consolidation, Graziano Fiorito (Stazione Zoologica Anton Dohrn, Naples) — learning paradigms, observational learning, Octopus vulgaris model, Tal Shomrat — VL LTP and behavioral consolidation studies, Martin J. Wells (Cambridge) — classic tactile/visual discrimination and lesion work, J.Z. Young & Brian Boycott (UCL) — foundational octopus brain and memory-system anatomy, Jennifer A. Mather (University of Lethbridge) — foraging, spatial memory, cognition and behavior, Frederike D. Hanke & Alexandra Bublitz (University of Rostock) — modern reversal and spatial learning, Jean G. Boal — cephalopod spatial learning and navigation, Yaron Meirovitch / Flavie Bidel — VL connectomics.

Key papers #

• Shomrat T, Zarrella I, Fiorito G, Hochner B (2008). The Octopus Vertical Lobe Modulates Short-Term Learning Rate and Uses LTP to Acquire Long-Term Memory. Current Biology — Causally links vertical-lobe LTP to memory consolidation: tetanizing or transecting the VL tract both impair long-term recall while oppositely affecting short-term learning rate.
• Hochner B, Brown ER, Langella M, Shomrat T, Fiorito G (2003). A Learning and Memory Area in the Octopus Brain Manifests a Vertebrate-Like Long-Term Potentiation. Journal of Neurophysiology — First demonstration of robust, activity-dependent, hippocampus-like LTP in an invertebrate brain slice — a landmark case of convergent evolution.
• Bidel F, Meirovitch Y, Schalek RL, ... Hochner B (2023). Connectomics of the Octopus vulgaris vertical lobe provides insight into conserved and novel principles of a memory acquisition network. eLife — Maps ≈25 million VL neurons and reveals a unique single-input, 1:12 fan-out feedforward circuit distinct from cerebellum and mushroom body.
• Boal JG, Dunham AW, Williams KT, Hanlon RT (2000). Experimental evidence for spatial learning in octopuses (Octopus bimaculoides). Journal of Comparative Psychology — Demonstrates that octopuses learn and retain (≈1 week) the spatial location of an escape burrow, evidence of true spatial learning.
• Bublitz A, Dehnhardt G, Hanke FD (2021). Reversal of a Spatial Discrimination Task in the Common Octopus (Octopus vulgaris). Frontiers in Behavioral Neuroscience — Shows serial reversal learning (up to 13 reversals) and that an incorrect-choice signal is critical — flexibility beyond mere discrimination.
• Fiorito G, Scotto P (1992). Observational Learning in Octopus vulgaris. Science — Controversial early claim that naïve octopuses learn a discrimination by watching trained demonstrators, faster than by direct conditioning.
• Sanders GD (1970). Long-term memory of a tactile discrimination in Octopus vulgaris and the effect of vertical lobe removal. Brain Research — Quantifies multi-month tactile memory decay and shows vertical-lobe removal selectively degrades long-term retention.
• Wells MJ, Young JZ (1960s). Centres for Tactile and Visual Learning in the Brain of Octopus; A Touch-Learning Centre in Octopus. Journal of Experimental Biology — Establishes anatomically separate tactile (inferior frontal/subfrontal) and visual (optic lobe) learning systems with the vertical lobe as shared memory store.
• Bublitz A, Weinhold SR, Strobel S, Dehnhardt G, Hanke FD (2017). Reconsideration of Serial Visual Reversal Learning in Octopus from a Methodological Perspective. Frontiers in Physiology — Critically re-examines classic reversal studies, arguing methodology inflated some cognitive-flexibility claims.
• Turchetti-Maia AL, Stern-Mentch N, Hochner B, et al. (2024 (bioRxiv)). A novel nitric oxide (NO)-dependent molecular switch mediating LTP in the Octopus vulgaris brain. bioRxiv preprint — Identifies NO/NOS as the presynaptic mechanism of VL LTP, explaining its NMDA-independence.