The vertebrate visual system maintains spatial relationships from the retina to the downstream visual areas via topographic projections (Huberman et al., 2008; Seabrook et al., 2017; Cang et al., 2018). This retinotopic organization emerges early in development (Debski and Cline, 2002; McLaughlin and O’Leary, 2005; Cang and Feldheim, 2013), and provides a fundamental platform for visual processing, ensuring that nearby points in visual space are represented by neighboring neuronal populations. In mice, retinal ganglion cells (RGCs) project to >40 brain areas (Morin and Studholme, 2014; Martersteck et al., 2017), with nearly all axons passing through the same pathway – the optic nerve, optic chiasm, and optic tract (OT) – before diverging to targets, such as the superior colliculus and the lateral geniculate nucleus (LGN). While retinotopy in these downstream areas has been well characterized (McLaughlin et al., 2003; Piscopo et al., 2013; Molotkov et al., 2023), it remains elusive to what extent retinotopy is maintained in the long-range RGC axons themselves.

Previous anatomical and tracing studies have demonstrated that retinotopy is largely preserved in the optic nerve (Bunt and Horder, 1983; Reese, 2011; but see Horton et al., 1979). Axons arising from specific retinal quadrants occupy consistent positions within the optic nerve, forming a reliable mapping of the retinal geometry as they exit the eye. This topographic organization is, however, substantially disrupted in the optic chiasm (Marcus and Mason, 1995; Colello and Guillery, 1998), where axons from the two eyes intersect and distribute themselves into either ipsilateral or contralateral side. The optic chiasm has thus been considered as a sorting hub, where retinotopy is relaxed to permit reorganization based on molecular cues, midline crossing decision, and eventual target-specific routing (Jeffery, 2001). After passing through the optic chiasm, RGC axons reorganize and partially recover retinotopic organization in the OT via pre-target sorting (Simon and O’Leary, 1991; Plas et al., 2005; Sitko et al., 2018). This indicates the presence of coarse retinotopy in the OT before reaching the target; however, questions remain on the precision and functional consequences of this reorganization.

Using in vivo electrophysiological recordings, here we mapped visual receptive fields (RFs) of individual RGC axons in the mouse OT. We quantified the spatial organization of these RFs, and took a data-driven modelling approach to evaluate how faithfully visual space is represented in the OT. Our results revealed a moderate degree of retinotopy, but not a fine-grained one. This highlights a robustness of the visual system to imprecision in long-range wiring, with refinement of the topographic organization taking place in each target area.

In a brain region with retinotopy, by definition, neighboring cells have adjacent receptive fields (RFs), representing nearby points in visual space (Cang and Feldheim, 2013; Seabrook et al., 2017). To examine if the optic tract (OT) has a retinotopic organization, we thus performed in vivo extracellular recordings of retinal ganglion cell (RGC) axons in the mouse OT, and mapped their RFs using white-noise stimuli and reverse correlation (e.g., Figures 1A–E). As a positive control, recordings were also made from the dorsal lateral geniculate nucleus (LGN; e.g., Figures 1F–J), a major retinorecipient area well-known to have a retinotopy (Piscopo et al., 2013).

Using multichannel silicon probes (4 shanks, each with 8–16 recording sites; Figure 1), we simultaneously recorded 6 ± 4 single-units from the OT (mean ± standard deviation, 33 animals) and 17 ± 16 single-units from the LGN (17 animals). We found that the measured RFs of those single-units from the OT recordings were widely distributed across the visual stimulation area (73° and 44° in azimuth and elevation, respectively; RF size range, 1.3°–13.1°; e.g., Figure 1E). This was also the case even for those units that had the largest signals on the same recording site, hence were supposedly located near each other in close proximity to the recording site (e.g., Figures 1B–D). In contrast, the measured RFs of LGN cells were spatially more clustered (e.g., Figure 1J) and demonstrated substantial overlap, especially among neighboring cells (e.g., Figures 1G–I) as expected from known retinotopy (Huberman et al., 2008; Seabrook et al., 2017). Such qualitative comparison already suggested that the mouse OT lacks retinotopy.

For quantitative population-level data analysis, we examined how the RF distance relates to the physical distance between simultaneously recorded units (Figure 2). We first used the electrode distance as a proxy of the distance between the units, and found a significant positive correlation for the LGN (Pearson correlation coefficient R = 0.42, p < 0.001, N = 4,491 pairs; Figure 2C), but not for the OT (R = −0.01, p = 0.8, N = 756 pairs; Figure 2A). The RF size differences between the pairs were small (LGN, 1.4 ± 0.9°; OT, 1.1 ± 0.6°; median ± median absolute deviation), indicating that consistent populations were sampled in each group. We next made a comparison between nearby and distant cell pairs, where nearby cell pairs were defined as those single-units that had the largest signals on the same recording site (i.e., zero electrode distance), whereas distant pairs as those with the largest signals on different recordings sites (i.e., non-zero electrode distance). In the LGN, the measured RF distance was significantly shorter for nearby cell pairs (4.8 ± 2.3°, median ± median absolute deviation; N = 150 from 11 mice) than for distant cell pairs [14.0 ± 5.8°; N = 4,341 from 17 mice; p(U-test) < 0.001; Figure 2D]. In the OT, however, RF distances did not differ between nearby (21.3 ± 15.1°, N = 39 from 18 mice) and distant pairs [29.0 ± 12.7°, N = 717 from 33 mice, Figure 2B; p(U-test) = 0.10]. Thus, neighboring RGC axons in the mouse OT do not necessarily represent adjacent visual locations.

To further evaluate this, we modelled the expected RF distance for nearby cell pairs under the assumption of retinotopy (see “Methods” for details). For LGN, Monte Carlo simulations yielded an estimated RF distance between 4.2° and 4.9° (95% confidence interval; median, 4.6°; Figure 2D), given the retinotopy gradient of 0.11°/μm (Piscopo et al., 2013) and a reliable single-unit recording range of 40 μm from an extracellular electrode (Anastassiou et al., 2015). This is consistent with the measured RF distance (median, 4.8°), validating our modelling approach. Applying the same logic to the mouse OT (diameter, ~300 μm; visual field extending ~135° in both azimuth and elevation; Paxinos and Franklin, 2001), our model predicted the median RF distance of 7.9°–10.7° for nearby pairs (median, 9.3°; Figure 2B) if RGC axons were arranged retinotopically. This disagreed with the measured RF distance (median, 21.3°), hence rejecting the presence of retinotopy imposed as a key premise of the model.

To what extent does the mouse OT retain topographic organization? To address this question, we extended our model by introducing additive Gaussian noise to the RF location of the simulated RGC axons (see “Methods” for details), where σ = 1 corresponds to the represented visual field (135°). We then used Kendall’s τ as a measure of spatial monotonicity to quantify the degree of retinotopy, where τ = 1 and 0 represent precise and no retinotopy, respectively. As expected, the retinotopy level τ decreased with increasing noise level σ, and the estimated RF distance of nearby RGC axons expanded (Figures 2E,F). From the intersection between the measured and simulated RF distances, we then identified that the mouse OT had a moderate level of retinotopy (τ = 0.49) with σ = 0.13, corresponding to a deviation of RF location from a retinotopically ideal position by ~18°, or equivalently, a displacement of RGC axons by ~40 μm (Figures 2G,H). This indicates that the mouse OT lacks local retinotopy, while maintaining topographic organization at a global level.

Here we provide in vivo electrophysiological evidence for the absence of local retinotopy in the mouse optic tract (OT). Both population-level analyses and data-driven models failed to identify a fine-scale topographic organization in the OT that would make neighboring retinal ganglion cell (RGC) axons encode adjacent locations in the visual field. In contrast, corresponding analyses on the dorsal lateral geniculate nucleus (LGN) supported the presence of retinotopy as expected from previous studies (Huberman et al., 2008; Seabrook et al., 2017), thereby serving as a control for our functional circuit characterization.

We acknowledge several caveats in our results. First, the OT is predominantly comprised of RGC axons, but also contains axons from other sources, such as the parabigeminal nucleus (Reinhard et al., 2019; Tokuoka et al., 2020) and the superior colliculus (Gale and Murphy, 2014). Our physiological criteria (i.e., robust visual responses with short latencies and well-defined, spatially-confined linear RFs) may not be sufficient to fully exclude those nontargets. Second, while the OT is largely intermingled to form a broad retinotopy, there is partial and graded segregation of RGC axons by the cell type and projection target (Erskine and Herrera, 2014; Robles et al., 2014). Our models did not incorporate such typology, and thus may underestimate the actual degree of retinotopy in OT. Another caveat is that the nodes of Ranvier from adjacent RGC axons can be offset by up to ~70 μm (Butt et al., 1994); hence, they may not be detected on the same recording site. Nevertheless, the units we defined as “nearby” should still be in close proximity to a given recording site, and thus near each other if not the closest. This guarantees the validity and robustness of our data/model analyses.

Our finding that RGC axons in the mouse OT retain only coarse retinotopy implies a fundamental constraint of early visual pathway organization: i.e., pre-target axon sorting cannot establish precise spatial relationships over long distances (Plas et al., 2005). This is consistent with known limits of axon guidance mechanisms, such as Eph/ephrin signaling and axon fasciculation, that establish broad topographic gradients but lack the resolution required for single-axon precision (McLaughlin and O’Leary, 2005; Cang and Feldheim, 2013). Thus, fine retinotopy must be re-established within each target via activity-dependent mechanisms—including retinal waves and Hebbian plasticity—to impose local precision on coarse inputs (Debski and Cline, 2002; McLaughlin et al., 2003).

Such a two-step architecture likely reflects an adaptive circuit organization principle: by relaxing long-range wiring constraints, the system reduces developmental cost while increasing robustness. Small positional deviations in the OT have then minimal impact because final mappings are refined locally in individual targets. Moreover, each area remains free to optimize the retinotopic map for its own computational needs (Knapen, 2021), rather than relying on the geometric arrangement of incoming axons themselves. Testing these functional implications in the visual system—and examining if they extend across the central nervous system—will be an important direction for future studies.

No statistical method was used to predetermine the sample size. The significance level was 0.05 in all analyses unless otherwise noted. All experiments were performed under the license 233/2017-PR and 220/2024-PR from the Italian Ministry of Health, following protocols approved by the Institutional Animal Care and Use Committee at European Molecular Biology Laboratory. The data analyses were done in Python and MATLAB.