Environment geometry alters subiculum boundary vector cell receptive fields in adulthood and early development

Spatial cognition in the hippocampal formation relies on neurons tuned to position and orientation, including place, head direction and grid cells, jointly forming a cognitive map anchored by both self-motion and external landmarks. Environmental boundaries are salient, extended landmarks that constrain movement and anchor the cognitive map, improving stability and error correction. Behaviorally, boundary geometry (e.g., rectangular enclosures) strongly guides spatial memory and reorientation across species. Within the network, medial entorhinal cortex (mEC) border cells fire near boundaries, while subiculum neurons include cells consistent with the Boundary Vector Cell (BVC) model, signaling allocentric distance and direction to boundaries. Developmentally, boundary-related inputs influence early spatial coding: place cells are more stable near boundaries before grid cells mature, and mEC border cells appear by P17. However, how boundary coding develops in the subiculum, a key hippocampal output structure encoding diverse spatial variables, remained unknown. This study asks how subiculum BVCs are defined and tuned across development, how environmental geometry (square vs circle) shapes their receptive fields, and how they respond to inserted barriers in adulthood and early development.

Prior work established: (1) place, grid, and head-direction cells as core spatial coding elements; (2) boundaries as strong allocentric landmarks anchoring spatial codes and behavior; (3) mEC border cells firing adjacent to boundaries; (4) a BVC model predicting neurons tuned to allocentric distance and direction from boundaries; and (5) developmental precedence of boundary-related stability before grid emergence, with mEC border cells present from P17. Subiculum encodes a wide range of spatial and task variables and routes information to multiple regions. Geometry shapes spatial memory and grid symmetry; corners and straight walls can impose directional symmetries on neural tuning. Despite this, objective BVC definitions in subiculum were limited, and the developmental trajectory of subicular boundary coding in behaving animals had not been characterized. This study builds on these findings by formal BVC identification, comparing subiculum with mEC, and testing geometry-induced distortions.

Subjects: 17 developing male Lister hooded rats (P12–P25 during recordings) and 4 adult males (3–6 months). Developing pups remained with dams until weaning (P21). Only males were used. Housing under 12:12 h light:dark. Ethical approval per UK ASPA. Surgery/recordings: Chronic microdrives with 8 tetrodes targeted dorsal subiculum (developing: AP −4.4 mm, ML 1.3 mm; adults: AP −5.4 mm, ML 1.5 mm; DV 2.7 mm). Postoperative recovery (1 day pups, 1 week adults), then daily advancement until subiculum layer was identified (theta LFP and pyramidal spiking). Single units: excitatory pyramidal cells (waveform width >300 µs). Data via Axona DACQ; position and head direction tracked via LEDs. Units sorted manually (TINT). Electrodes moved ≥50 µm daily. Behavioral environments: Open field for foraging milk drops. Square: 62.5 cm sides, 50 cm walls, black floor, fixed distal lab cues. Circle: 80 cm diameter, 50 cm walls, within black curtains with a single white card cue and lamp; animals transported in a closed black box between arenas to minimize cue carryover. Sessions began with two square trials; one circle trial per session (either after squares or after barrier testing). Trials 15 min, with 15 min inter-trial rest. Barrier test: A straight barrier (most often 50×2.5×50 cm; some pups 40×5×50 cm) inserted centrally into the square, aligned NS or EW, typically testing both orientations per session. Rate map construction: Periods with speed <2.5 cm/s removed. Position rescaled to standard grid (square: 25×25 bins of 2.5 cm; circle: 32×32 bins, iterative center estimation), smoothing with 5×5 boxcar; trials with <80% spatial coverage excluded. BVC model fitting: Idealized BVC rate maps computed by integrating a Gaussian receptive field in polar distance (r) and angle (θ) to boundaries, with parameters: preferred distance d (0–32.5 cm in square; to 40 cm in circle, step 2.5 cm), preferred allocentric direction φ (0–354°, step 6°), and base radial width σ0 (6.2, 12.2, 20.2, 30.2 cm). σrad=(β+1)σ0 with β=183 cm and fixed σang=0.2 rad; integration step Δθ ≥5.7°. For each neuron, exhaustive search selected best-fitting model maximizing Pearson r between neural and BVC maps; r_max recorded. BVC classification and thresholds: Significance assessed via spike-time shuffled nulls (1000 temporal shifts per cell/trial). A neuron was a BVC if: (1) r_max exceeded the cell-specific 99th percentile of shuffled r_max (rate-map threshold) and (2) exceeded the 99th percentile population threshold for area and age, and (3) Spatial Information exceeded the 75th percentile of shuffled Spatial Information (age-specific thresholds provided). Classification could be met on either of the two baseline square trials. False positive rate per neuron estimated 1.49% assuming independence of criteria and trials. Alternative, more stringent nulls used a field-shuffle preserving spatial fields; and an additional exclusion if a radially symmetric Place Cell model (2D Gaussian of widths 7,9,11,13 cm) fit better than the BVC model (applied to CA1 and subiculum in controls). Developmental metrics: Spatial Information (bits/spike), intra- and inter-trial stability (Pearson r across bins between trials), and BVC fit r_max quantified. ANOVA and post-hoc Tukey HSD tested age effects. Directional tuning analyses: Distributions of d, φ, σ0 summarized per age. Four-fold clustering of φ assessed with Rayleigh test on quadrupled, wrapped angles. Proportion of φ within ±12° of cardinal directions compared to chance. Short-range (d≤10 cm) vs long-range (d>10 cm) BVCs compared. Circle vs square comparison: BVC tuning in circle analyzed similarly; ensemble rotations between shapes computed; to rule out session-wise misalignment masking 4-fold symmetry, φ were corrected by ensemble mean rotation and re-tested for symmetry. Barrier responses: For BVCs with clear NS/EW φ classes and d<15 cm, define distal and proximal 12.5-cm zones flanking the barrier; compute absolute and normalized rate changes between baseline square and barrier trials for each side. Mixed-design ANOVA tested Age×Side effects; Simple Main Effects for post-hocs. In adults, proximal inhibition analyzed across σ0 categories. Controls: Sub-sampling to match developmental differences in path length, running speed, mean firing rate, and cluster stability; matching circle r_max to square. Head direction modulation assessed and compared to distributive hypothesis maps to control for sampling bias. Simulations with real trajectories and Poisson spikes tested whether four-fold clustering could arise from behavioral biases; φ drawn from flat vs empirical distributions. mEC comparison: Re-analysis of published mEC dataset using identical BVC fitting and shuffles; border cells defined by border score and Spatial Information exceeding 95th percentile of age-matched shuffles. Overlap and developmental stability of BVCs and border cells assessed.

• Detection and prevalence: Recorded 517 subiculum neurons (4 adults) and 1080 (17 pups, P16–P25). BVCs were significantly more frequent than chance at all ages (Binomial p<0.001). Counts: P16–18: 74; P19–21: 61; P22–25: 46; Adult: 187. Proportion increased post-weaning (Z=2.35, p=0.019) but remained below adult at P22–25 (Z=3.41, p=0.001).
• Development of specificity and stability: Spatial Information, BVC r_max, inter- and intra-trial stability all increased with age (ANOVA: SI F(3,364)=16.6, p<0.001; r_max F(3,364)=43.1, p<0.001; inter-trial F(3,363)=74, p<0.001; intra-trial F(3,364)=100, p<0.001). Most measures rose from pre- to post-weaning (except SI; Tukey HSD P16–18 vs P22–25: SI p=0.61; others p<0.001) and remained lower than adult at P22–25 (all p<0.001). Control sub-sampling for path length, speed, mean rate, and cluster drift did not alter trends.
• Tuning distributions in square: d skewed toward short distances with median d unchanged across age (Kruskal–Wallis χ²(3,364)=3.69, p=0.30). σ0 distributed across all tested widths with no age change (χ²(3,364)=3.91, p=0.27). φ showed strong four-fold clustering aligned with walls at all ages (Rayleigh on quadrupled angles: P16–18 z=9.6, p<0.001; P19–21 z=17.0, p<0.001; P22–25 z=11.1, p<0.001; Adult z=50.7, p<0.001). Proportion within ±12° of walls was above chance and stable across development (χ²=3.46, p=0.33).
• Short vs long-range BVCs: Angular half-height widths (φ0) larger for long-range than short-range BVCs at all ages (Wilcoxon p<0.001). Wall-aligned φ more prevalent for short-range BVCs in developing animals until P21 (χ² Age×Range=27.4, p=0.002; Z-tests: P16–18 p=0.002; P19–21 p=0.036; P22–25 p=0.32; Adult p=0.94).
• Circle vs square geometry: In the circle, φ showed no significant 4-fold or unimodal nonuniformity at any age (Rayleigh quadrupled: P16–18 z=0.5 p=0.63; P19–21 z=1.5 p=0.23; P22–25 z=0.6 p=0.57; Adult z=1.8 p=0.16; unimodal tests all p>0.09). BVC r_max was lower in the circle, but matching r_max across shapes did not restore 4-fold clustering. Ensemble φ rotations between shapes were approximately coherent within sessions (κ≈2.15–2.23), but relative offsets varied (median |offset|: pups 30°, adults 22°). After aligning φ by ensemble mean, circle tunings still lacked 4-fold symmetry (P16–25 z=0.2 p=0.86; Adult z=0.2 p=0.80). Within ensembles (≥5 BVCs), 4-fold symmetry magnitude (quadrupled Rayleigh vector length) was reduced in circle vs square (Wilcoxon p=0.001). Simulations using real behavior showed that flat φ inputs did not produce 4-fold clustering, whereas real φ distributions did, ruling out behavioral sampling bias.
• Barrier insertion responses: Distal-side firing increased at all ages (proportion with increased distal firing: P16–18: 79%; P19–21: 78%; P22–25: 93%; Adult: 92%). Absolute distal>prox differences reached significance from P19 (Mixed ANOVA Age×Barrier×Side F(3,137)=10.8, p<0.001; SME Dist vs Prox: P16–18 p=0.068; P19–21 p=0.033; P22–25 p=0.010; Adult p<0.001). Normalized changes were significant at all ages (F(3,137)=21.6, p<0.001; P16–18 p=0.012; others p<0.001). Distal responses in pre-weanlings were lower than adults; post-weanlings not different from adults.
• Inhibitory component (adults only): Proximal-side firing was significantly suppressed in adults (1-sample t-tests vs 0 change: absolute and normalized p<0.001), not in developing groups (p≥0.57). Proximal inhibition in adults increased with broader σ0 (Absolute proximal: F(3,79)=7.91, p<0.001; Normalized proximal: F(3,79)=3.41, p=0.022), significant for σ0≥20.2–30.2 cm categories.
• Robustness and CA1 control: Results held under field-shuffle nulls and after excluding units better fit by place-cell models. CA1 yielded far fewer BVC-like fits; many were elongated place fields better fit by place models.
• mEC comparison: Re-analysis showed mEC BVCs and border cells are adult-like by P17 with no developmental changes in Spatial Information, stability, model fit, or φ wall alignment. BVCs in mEC had longer d than border cells, but neither changed with age; φ aligned to walls at all ages in mEC too. Overall, subiculum BVCs are present early but develop precise, stable, and inhibitory-field properties slowly (>P25). Environmental geometry imposes 4-fold directional clustering in squares but not circles, evident from the earliest ages tested.

The findings demonstrate that subiculum BVCs encode allocentric vectors to boundaries but their receptive fields are modulated by environmental geometry. In square arenas, φ tunings align to wall orientations, producing a robust four-fold symmetry absent in circular arenas. This geometry-dependent modulation appears intrinsic to BVC coding rather than a byproduct of behavioral sampling or head-direction biases, and may reflect sensitivity to straight walls and corners, consistent with the strong influence of geometry on spatial memory. Inserted barriers evoke both expected distal excitatory fields and a previously uncharacterized proximal inhibition in adults, suggesting inhibitory inputs help sculpt BVC receptive fields and could support coherent, attractor-like population dynamics maintaining consistent directional relationships across environments. Developmentally, subiculum BVCs show a protracted maturation of spatial specificity, stability, and barrier-evoked responses compared to mEC boundary-responsive neurons, which are mature by P17. Nonetheless, the core geometric signature (wall-aligned φ in squares and uniform φ in circles) is present from the earliest ages tested, implying that geometry-dependent structuring of boundary-vector representations is an early-emerging feature of hippocampal coding. These results integrate with theories that environmental geometry anchors spatial representations and can distort downstream codes (e.g., grid shearing), and they predict behavioral consequences when switching between environments of differing shape, especially near corners where BVC representations change most between shapes.

This study provides a formal, objective characterization of subiculum Boundary Vector Cells in adult and developing rats, showing that: (1) BVC prevalence exceeds chance from P16 onward, but spatial precision and stability mature late (>P25); (2) environmental geometry strongly shapes BVC directional tuning—four-fold wall alignment in squares versus uniform tuning in circles—observable from earliest ages; and (3) barrier insertion elicits both distal excitation and adult-specific proximal inhibition, indicating inhibitory contributions to BVC field formation and potential attractor-like network dynamics. Re-analysis of mEC confirms early maturity of boundary-responsive coding, underscoring a sequential developmental progression along the hippocampal circuit. Future work should dissect circuit mechanisms mediating geometry sensitivity (e.g., roles of corner-encoding subiculum neurons), test how rearing geometry and experience shape BVC development, examine interactions with egocentric boundary codes, and evaluate behavioral impacts of geometry-induced BVC distortions, particularly near corners and during shape transitions.

• Subjects were all male rats; sex differences in development or geometry effects were not assessed.
• Animals were reared in rectangular cages, leaving open whether early experience with geometry contributes to observed wall-aligned φ; effects of circular rearing were not tested.
• The circle arena lacked common extra-maze cues with the square, leading to ensemble rotations; while corrected analytically, residual variability could influence comparisons.
• The arena size (62.5 cm square) may limit detection of very long-range tunings compared to larger arenas reported elsewhere.
• Barrier-induced proximal inhibition emerged only in adults; underlying synaptic circuitry was not directly measured.
• mEC analyses did not include circle tests in the re-analyzed dataset, limiting direct shape-comparison for mEC.
• Head-directional modulation was largely attributable to spatial sampling bias, but residual modulation could still contribute minimally to tuning estimates.
• Developmental electrophysiological stability (cluster drift) was greater in young animals; extensive controls suggest results are robust, but subtle effects cannot be fully excluded.