Keywords

1 Introduction

Neural connection development during critical periods in the visual system has profound effects on visual processing in the adult. In the early visual system, retinal ganglion cells send axons to the lateral geniculate nucleus (LGN) of the thalamus, which sends inputs to the primary visual cortex (V1). Proper formation of this circuit relies on normal visual experience in mammals, and the precise wiring within this system will not properly mature if vision is disrupted during the critical period. In the 1960s we learned from Hubel and Wiesel that disrupting visual experience during a critical period produces irreversible damage in visual processing (Wiesel and Hubel 1963a, b, 1965a, b). They monocularly deprived kittens of visual experience by suturing one eyelid closed during the first few months of their lives and found a dramatic change in neural responses in V1. Specifically, the distribution of ocular dominance shifted; that is, very few cells could be driven by the deprived eye. Moreover, when they performed the same experiment in adult cats, no such shift was observed (Wiesel and Hubel 1963a, b). This was the first evidence that normal sensory development depended on processes that occur during a critical period and has been the basis for examining plasticity in the visual system. Here we describe the development of the early mammalian visual system with a focus on the role of the thalamus in experience-dependent plasticity. Although the perception of the thalamus as more than a relay center is now commonly recognized, recent studies have expanded our understanding of the essential role of the thalamus in visual development. While the majority of literature concerning developmental plasticity largely focuses on primary visual cortex, we will also highlight the role of the LGN during sensitive periods of visual development. Notably, recent observations in the retinogeniculate pathway underscore the importance of earlier synapses in the visual system and this chapter will review these data within the context of what we already understand about the early visual system (Fig. 1).

Fig. 1
figure 1

Organization of the mature lateral geniculate nucleus (LGN) in mouse, ferret, cat, and primate (Rhesus macaque). (a) The mouse LGN is divided into the shell (blue) and core (red). The shell and core both receive inputs from the contralateral eye. A small medial portion of the core receives input from the ipsilateral eye (purple). (b) The ferret LGN is divided into laminae A, A1, and C (blue). The C layer is further subdivided into C, C1-C3. Laminae A and C and C1 receive projections from the contralateral eye and laminae A1 and C2 receive projections from the ipsilateral eye. Layer C3 does not receive retinal afferents. Horizontal plane is shown for visualization of laminae. (c) The cat LGN, like the ferret LGN, is divided into laminae A, A1, C, and C1-C3. Laminae A, C, and C2 receive projections from the contralateral eye. Laminae A1 and C1 receive input from the ipsilateral eye. Layer C3 does not receive retinal afferents. (d) The macaque LGN is divided into two magnocellular layers (M, layers 1–2, red), four parvocellular layers (P, layers 3–6, blue), and six koniocellular layers (purple). Mouse, cat, and macaque LGN are shown in the coronal plane. References: Mouse: (Godement et al. 1984; Grubb and Thompson 2004); Ferret: (Zahs and Stryker 1985); Cat: (Hickey and Guillery 1974); Macaque: (Malpeli and Baker 1975; Connolly and Van Essen 1984); D dorsal, V ventral, M medial, L lateral, A anterior, P posterior

Full size image

2 Gross Organization and Development of LGN

The visual input from the brain arises from retinal ganglion cells, the only input from the retina to the brain. These retinal ganglion cells are divided into separate classes that tile the retinal surface and convey different types of information to the brain (Sherman and Spear 1982; Livingstone and Hubel 1987; DeYoe and Van Essen 1988; Maunsell 1992; Hendry and Reid 2000; Kawasaki et al. 2004). Modern genetic methods have identified 46 distinct classes of retinal ganglion cells in mouse (Tran et al. 2019), 18 such classes in macaque monkey (Peng et al. 2019), and 12 classes in human (Yan et al. 2020).

The LGN begins as separate progenitor neurons representing central vision and peripheral vision and divide and mature in a gradient. This allows for earlier development of cells representing the central visual world, subsequently followed by cells carrying information from the peripheral visual world (Wiencken-barger and Casagrande 2002). As the visual system develops, a feed-forward pathway is formed from the retina to the lateral geniculate nucleus (LGN) to the primary visual cortex (V1). This pathway consists of two well-defined synapses. The first synapse, the retinogeniculate synapse, describes the connection between retinal ganglion cells (RGCs) and the thalamocortical cells of the LGN, while the second, the thalamocortical synapse, refers to the connection between LGN cells and cells in V1. RGCs encode visual information and send their axons topographically to the LGN, which is located in the thalamus adjacent to the optic tract. The optic tract is a thick bundle of axons that provides the major feed-forward input from the retina.

In many mammals, the LGN exhibits clear lamination patterns that can be observed with Nissl staining, but the number of layers and their organization varies from species to species. In primates, most individual layers receive retinal input from a single eye (Kaas et al. 1972). For example, the LGN of the macaque monkey is divided into six layers. Layers 1 and 2, the magnocellular layers, and layers 3–6, the parvocellular layers, can be further characterized by the eye-specific inputs they receive. Layers 1, 4, and 6 receive input from the contralateral eye, while layers 2, 3, and 5 receive input from the ipsilateral eye. A third distinct set of layers, the koniocellular layers, are interspersed between the magnocellular and parvocellular layers.

Cells in each LGN layer often exhibit similar functional properties. For example, macaque magnocellular layers can be characterized by their transient responses, shorter response latencies, and lack of chromatic selectivity (Sherman et al. 1976; Kaplan and Shapley 1982; Yeh et al. 1995; Usrey and Reid 2000; Levitt et al. 2001). In contrast, parvocellular layers are distinguished by their smaller receptive fields, sustained responses, and chromatic contrast sensitivity (Xu et al. 2001). Much less is known about the koniocellular layers but their cells feature a broad set of characteristics, suggesting that there are multiple classes of cells (Irvin et al. 1986; Hendry and Reid 2000; Xu et al. 2001).

In carnivores, such as the cat and ferret, the LGN is divided into the A, A1, and C layers (Guillery 1969; Linden et al. 1981) (Fig. 1). These layers are innervated by retinal ganglion cell classes called X, Y, and W cells (Shapley 1984). Cells in LGN exhibit similar receptive field properties as these retinal ganglion cell classes and the LGN neurons are also divided into classes referred to as X, Y, and W (Howland 1984). Retinal ganglion X-cells drastically outnumber Y-cells and send eye-specific inputs to layer A or A1 of the LGN (Lennie 1980). Retinal ganglion Y-cells send their projections to all three layers, with those from the contralateral eye sending synapses to layer C or A, and Y-cells from the ipsilateral eye sending synapses to layer A1 (Hickey and Guillery 1974) (Fig. 1). W-cells send inputs to the C layer, and, like koniocellular cells, are a heterogeneous superclass that is comprised of many cell types, including cells with non-concentric receptive fields, longer latencies, and many other types (Stone and Fukuda 1974; Shapley 1984; Hendry and Reid 2000). The relationships between X-cells and Y-cells of carnivores and the parvocellular and magnocellular cells of primates are unclear, with some investigators citing the linearity of X and parvocellular cells and non-linearity of Y cells and magnocellular cells as evidence of homology, while others cite the common contrast gain of X, Y, and magnocellular cells and suggest that the color-sensitive parvocellular cells are a class unique to primates (Shapley and Hugh Perry 1986; Van Hooser et al. 2003).

In other species, such as rats and mice, the division of cells in LGN is more subtle. The mouse LGN is divided into the shell and core (Reese 1988). The shell and core both receive inputs from the contralateral eye while only a small population of ipsilateral RGC afferents project to the core (Fig. 1). The shell is innervated by a heterogeneous population of RGCs, including cells that are orientation selective or direction selective, while the core receives inputs from α RGCs. α RGCs can be classified based on their large soma and dendritic field size and are found among a wide range of mammalian species (Peichl et al. 1987; Peichl 1991; Sun et al. 2002).

3 Innervation from the Retina: Molecular Cues

The circuitry of the visual system is remarkably precise. Its anatomical and functional organization begins to emerge prior to the onset of visual experience and develops in a feed-forward manner. Before eye-opening, visual system development begins with an initial phase during which axon mapping and rearrangement of different classes of retinal ganglion cells (RGCs) depend on molecular guidance cues and spontaneous activity. Molecular cues act to establish the initial retinotopic mapping of the LGN, forming a basic framework for vision.

RGCs are first guided by the ephrin/Eph family of axon guidance molecules (Penn et al. 1998). Ephrins and Eph receptors, expressed in the retina and its central visual targets, act as graded guidance cues for sending RGC axons to their appropriate locations (McLaughlin and O’Leary 2005; Flanagan 2006). As evidence for the essential role of ephrins in formation of topographic maps, it has been shown that loss or gain of EphA/ephrin-A function in mice produces defects in the placement of layers within the dLGN (Huberman et al. 2005a, b; Pfeiffenberger et al. 2005). Nevertheless, division of axonal projections from the two eyes into layers persists, suggesting that another process must be responsible for this laminar segregation, as we discuss next.

4 Innervation from the Retina: Spontaneous Activity and Laminar Plasticity

Although RGCs guided by molecular cues are able to send axons to their correct targets, without coordinated waves of activity, RGC axons fail to form dense terminations (Grubb et al. 2003; McLaughlin and O’Leary 2005). Instead, they form diffuse axon arborizations that fail to convey a complete representation of the visual world. These coordinated waves, known as retinal waves, begin prior to visual experience and can be described as coordinated, spontaneous waves of action potentials among neighboring RGCs. Retinal waves are initiated randomly and propagate across the retina, stimulating correlated LGN neuron spiking activity during this process (Mooney et al. 1996). Although they are generated randomly, waves of spontaneous activity retain a high degree of structure. For example, studies have shown that activity from neurons in same-eye LGN layers and same-sign (On or Off) sublayers is more correlated than activity in opposite eye layers in opposite-sign sublayers (Weliky and Katz 1999; Huberman et al. 2008) (For more in-depth review, see Feller (1999)). Further, in mice it has been shown that retinal waves are biased to travel in a temporal to nasal direction, which mimics the front-to-back motion that the animal would experience if it were moving forward in its environment (Ge et al. 2021).

Across species, retinal waves can be grouped into three stages during which distinct mechanisms result in different activity patterns that provide diverse functions. Studies have shown dramatic effects of blocking spontaneous activity during Stage II and Stage III waves. In ferrets, for example, when stage II retinal waves are blocked with epibatidine in both eyes during the normal period of eye-specific segregation (P1-P10), eye-specific segregation does not occur. However, blocking only one eye’s wave activity still allows for retinogeniculate segregation, but this manipulation results in the normal eye expanding its thalamic axonal territory, while the blocked eye has a dramatically reduced projection territory (Penn et al. 1998). If retinal waves are completely inhibited in mice, using ephrin-A2/A3/A5 triple knockouts, there is a disruption in eye-specific segregation and layer placement in the LGN (Pfeiffenberger et al. 2005). This and other studies demonstrate that, together with molecular guidance cues, this correlated activity is necessary for the development of retinotopic maps and eye-specific segregation of LGN.

In primates, retinogeniculate eye-specific segregation of LGN layers begins before the onset of visual experience. Eye-specific connections form prenatally and undergo accelerated growth during the last two trimesters (Rakic 1976), beginning segregation during the second trimester in layers 5 and 6 of the LGN. By the third trimester segregation of the LGN is complete, and the magnocellular and parvocellular layers can be differentiated by their characteristic response properties by 1 week postnatal (Huberman et al. 2005a, b) (Fig. 2).

Fig. 2
figure 2

Timeline of developmental events in the retino-geniculo-cortical pathway of the (a) mouse, (b) ferret, (c) cat, and (d) primate (Rhesus macaque) visual systems. (a) Mouse (teal). References: 1. (Chen and Regehr 2000); 2. (Hong et al. 2014); 3. (Jaubert-Miazza et al. 2005) 4. (Godement et al. 1984); 5. (Demas et al. 2003); 6. (Tian and Copenhagen 2003); 7. (Pfeiffenberger et al. 2005); 8. (Tschetter et al. 2018); 9. (Liang and Chen 2020); 10. (Seabrook et al. 2013); 11. (Thompson et al. 2016); 12. (Rochefort et al. 2011); 13. (Gordon and Stryker 1996); (b) Ferret (pink). References: 14. (Linden et al. 1981); 15. (Wong et al. 1993); 16. (Johnson and Casagrande 1993); 17. (Jackson et al. 1989); 18. (Herrmann et al. 1994); 19. (Clascá et al. 2012); 20. (Chapman and Stryker 1993); 21. (Li et al. 2006); 22. (Issa et al. 1999); (c) Cat (blue). References: 23. (Wiesel and Hubel 1963a, b); 24. (Hubel and Wiesel 1964); 25. (Allendoerfer and Shatz 1994); 26. (Hickey and Hitchcock 1984); 27. (Kalil 1978); 28. (Ghosh and Shatz 1992); 29. (Albus and Wolf 1984); 30. (Albus and Fries 1980); 31. (Hubel and Wiesel 1970); (d) Primate (Rhesus macaque) (purple). References: 32. (Wiesel and Hubel 1974); 33. (Rakic 1976); 34. (Huberman et al. 2005a, b); 35. (Rakic 1977); 36. (Rakic et al. 1977); 37. (Hatta et al. 1998); 38. (LeVay et al. 1980); Approximate ages of eye-opening for each species are indicated above timelines. Conception and birth are labeled as embryonic (E0) and postnatal day zero (P0). Outlined bars indicate period of less developmental plasticity and dashed line (purple) indicates minimal plasticity. LGN lateral geniculate nucleus, OD ocular dominance

Full size image

While the retina is producing spontaneous activity that propagates to LGN and cortex, the cortex is also being driven by intrinsic spontaneous activity. Imaging and physiological studies in the ferret have demonstrated that spontaneous visual cortical activity is already highly modular 10 days before eye-opening (Chiu and Weliky 2001; Smith et al. 2018) when visual input is just beginning to be able to stimulate cortical cells (Krug et al. 2001; Akerman et al. 2002). These modules are the future positions of orientation columns in the cortex (Smith et al. 2018), indicating that considerable functional network connectivity in cortex has been established even before the onset of signals from the LGN or retina. Activity in the cortex is necessary for the proper formation of these orientation columns, as silencing the developing cortex during this period prevents the emergence of orientation selectivity (Chapman and Stryker 1993). Another key step in the formation of orientation selectivity in the cortex is the connections with the cortical subplate. The cortical subplate is a transient cortical structure that exists early in development and disappears around the time of eye-opening, and LGN axons initially make synapses with subplate neurons before extending their axons to make connections with layer 4 neurons (Allendoerfer and Shatz 1994). If the subplate is ablated, orientation selectivity never forms in visual cortex (Kanold et al. 2003).

5 Innervation from the Retina: Plasticity with Visual Experience

While eye-specific segregation occurs before eye-opening, early visual experience plays an important role in the mature visual circuit and is required for synaptic plasticity. Visual deprivation during early vision leads to a profound change in retinogeniculate synaptic strength and number of cells innervating the postsynaptic cell (Hooks and Chen 2008; Hong and Chen 2011) which we will turn to shortly. Cell-class-specific changes can also be observed in LGN lamina that lack visual experience. Monocular deprivation by lid suture (MDLS) is a classical, reversable method for understanding experience-dependent changes during early critical periods. With MDLS, there is a shift in OD toward the non-deprived eye, and associated structural changes in synapse development, in addition to alteration of properties of cells themselves, suggesting substantial network rewiring (Wiesel and Hubel 1963a, b). For example, in studies of monocular deprivation in primate, magnocellular cells of the deprived eye were found to exhibit somewhat faster latencies to optic chiasm stimulation, slightly stronger receptive field surrounds, and lower responsivities (Levitt et al. 2001). Magnocellular and parvocellular cells of the deprived eye also exhibited lower nonlinearities.

In an anatomical study of MDLS monkeys, laminae corresponding to the deprived eye exhibited cell shrinkage compared to non-deprived laminae with early lid closure (Vital-Durand et al. 1978; von Noorden and Crawford 1978; Headon et al. 1979; Sherman and Spear 1982; Tigges et al. 1984; Levitt et al. 2001), and cells were shown to have elevated synaptic densities overall at the soma of deprived eye (Lachica et al. 1990; Wilson and Forestner 1995; Levitt et al. 2001). In MDLS cats, there is a specific loss of Y cells, and an X-cell-specific reduction in spatial resolution in the deprived laminae (Sherman et al. 1972; Lehmkuhle et al. 1980; Sherman and Spear 1982).

As an interesting aside, the choice of manipulation for monocular deprivation has profound effects on the changes in neural activity that are induced in LGN and cortex. A study of visual deprivation via eyelid closure or tetrodotoxin (TTX) inactivation of the retina at the peak of the visual critical period in mice (~P28) showed that while both manipulations eliminated visually evoked activity, they did not affect spontaneous activity and average firing rate of LGN neurons. Instead, it seems that these two different manipulations had two different effects. Eyelid closure led to a decrease in correlative firing between simultaneously recorded cells, while TTX inactivation resulted in an increase in thalamic bursting activity (Linden et al. 2009). Therefore, it is important to think of MDLS or MD via inactivation as manipulations that do not remove activity but instead substantially alter its quality, and in different ways.

6 Synapse Development and Convergence Between Retina and LGN

After an initial experience-independent phase of retinotopic mapping, retinogeniculate cells undergo fine-scale refinement. During this phase of development, RGC synapse elimination and strengthening relies on spontaneous activity and visually evoked activity (Hong and Chen 2011; Hooks and Chen 2006). At this critical period of heightened plasticity, the onset of visual experience drives the elimination of weak retinogeniculate inputs so that, at maturity, an LGN neuron only receives strong input from a few RGCs (Chen and Regehr 2000; Jaubert-Miazza et al. 2005; Hooks and Chen 2006) (For more in-depth review, see Huberman (2007)) (Fig. 3). This describes the characteristic functional convergence of RGCs. However, functional convergence/divergence and morphological convergence/divergence are considered to be distinct properties of the retinogeniculate synapse. Functional convergence/divergence describes the number of RGCs that synapse onto a single LGN neuron and the number of LGN neurons that receive synaptic input from an RGC, while morphological convergence/divergence refers to the number of terminal boutons or axon arbors. Thus, a single LGN neuron may receive input from many different boutons or arbors, but only a few may functionally drive that cell (Tavazoie and Reid 2000). Following eye-opening, the LGN begins to develop in an experience-dependent manner.

Fig. 3
figure 3

Development of the lateral geniculate nucleus (LGN). (a) Development of retinogeniculate projections and LGN structure from juvenile to adult. Initially retinal afferents are overlapping. Contralateral and ipsilateral projection targets and axons develop while some afferents are still overlapping. Finally, the mature LGN segregates into eye-specific layers. For simplicity, LGN diagram features a single ipsilateral and contralateral layer. (b) Model for developmental retinogeniculate refinement. An LGN neuron initially receives many weak retinal ganglion cell (RGC) inputs. With development, inputs are weakened or eliminated and only a few inputs are strengthened. (c) Spatial receptive field development. With development, an unstructured spatial receptive field structure decreases in size, acquires a concentric center-surround antagonistic organization and the ratio of amplitudes of the center and surround components increases. (d) Temporal receptive field development. With development, an LGN temporal response peak latency and duration decreases. (e) Timeline of peak and overall developmental periods of spatial and temporal receptive field development in mouse, ferret, cat, and primate (Rhesus macaque) LGN. Primate cell development is divided into magnocellular (M) cells and parvocellular (P) cells. Saturated bars indicate peak developmental time periods and outlined bars indicate period of initial plasticity or continued development. References: 1. (Tschetter et al. 2018); 2. (Tavazoie and Reid 2000); 3. (Cai et al. 1997); 4. (Kiley and Usrey 2017); 5. (Movshon et al. 2005). Additional: (C J Shatz and Sretavan 1986); (Hooks and Chen 2006). LGN lateral geniculate nucleus, RGC retinal ganglion cell, M magnocellular, P parvocellular

Full size image

While past studies have found low convergence of RGCs on single thalamic cells (Cleland et al. 1971a, b, Mastronarde 1992; Chen and Regehr 2000; Ziburkus and Guido 2006; Hammer et al. 2015), there is recent evidence for a higher number of retinogeniculate inputs than previously thought. A number of physiological and ultrastructural studies of retinogeniculate inputs in mice have all revealed evidence of numerous RGC inputs to single LGN cells (Hammer et al. 2015; Morgan et al. 2016; Weyand 2016; Rompani et al. 2017; Liang and Chen 2020). In addition, a study in cat that transiently blocked ON retinal ganglion cells with 2-amino-4-phosphonobutyric acid found that ON-center LGN cells suddenly became sensitive to OFF inputs in the center of their receptive fields, revealing a hidden input that was not appreciated earlier (Moore et al. 2011). It is now estimated that there is an average of ten functional RGC inputs converging onto a single thalamocortical cell but only a few of these connections provide strong functional input (Chen and Regehr 2000; Jaubert-Miazza et al. 2005; Hooks and Chen 2006).

What is the function of these additional, weaker inputs? Retinogeniculate convergence is thought to serve a number of purposes and in mice has been described as conveying information in three modes (Rompani et al. 2017). In relay mode, different RGC boutons converging on a single LGN neuron share similar functions. That is, they share the same tuning properties. Relay mode represents the classic description of LGN cells that was appreciated in the time of Hubel and Wiesel’s collaboration. In relay mode, thalamocortical cells may increase or suppress signals conveyed by the retina and relay them to cortex, but they do not greatly modify the information being sent. They effectively reflect the sum of their inputs. However, in combination mode, converging axons exhibit the same preference for one visual feature but have a diverse range of other tuning properties. In this case a single LGN neuron could be selective for more complex stimulus features. The third mode describes binocular integration. Like combination mode, a single LGN neuron can combine inputs from a number of RGCs but in this mode the afferents originate from both eyes (Rompani et al. 2017). Regardless, it seems that by combining inputs tuned for similar features, an LGN neuron can reduce noise and trial-to-trial variability from individual inputs, while increasing the signal-to-noise ratio and robustness of its response. In this way, weaker inputs that may not individually generate a response in the LGN neuron can be combined, resulting in activation of the postsynaptic cell (Litvina and Chen 2017). Moreover, the combination of temporally correlated inputs can increase transmission efficiency (Liang and Chen 2020).

The two phases of retinogeniculate development described so far detail an initial phase that requires no visual experience and a second phase that requires visually-evoked activity for fine-scale refinement of the retinogeniculate synapse. However, there is also a recently observed final phase of retinogeniculate refinement that depends on normal visual experience. During this final critical period in visual development, visual manipulations have a greater potential to remodel retinogeniculate synapses (Hooks and Chen 2008). After eye-opening there is continued growth of axon arborization size and complexity (Liang and Chen 2020) and this persistence of development leaves open a window for plasticity. Late dark-rearing of mice during this period (P20-P35) after the onset of visual experience (~P12) has revealed unexpected changes in retinogeniculate synaptic refinement. In normally-reared mice, although size and branching complexity remain stable (Kim et al. 2010) at this time, bouton size and distribution along the arbor change (Hong et al. 2014). The initial broad distribution of boutons along the terminal arbor transforms into distinct clusters. However, if animals are dark-reared during this later time period, there is a corresponding increase in retinogeniculate inputs accompanied by a decrease in RGC input strength (Hooks and Chen 2006) effectively leading to a regression of visual development to its state prior to eye-opening. Moreover, manipulations before or after this critical period do not have the same robust effect on plasticity. Chronic visual deprivation (dark-rearing mice from birth) leads to modest changes in synaptic maturation without significantly affecting retinogeniculate input number or strength. Additionally, short-term deprivation (<4 days), or visual deprivation beginning at P16 or P25 does not have the same effect on synapse development as late dark-rearing (Hooks and Chen 2008), indicating a distinct period for experience-dependent plasticity.

7 Receptive Field Development and Plasticity

A receptive field is characterized by the spatiotemporal aspects of a visual stimulus that causes a cell to fire action potentials. Across species, a period of receptive field refinement begins before the onset of visual experience and RF properties continue to develop after eye-opening. Accordingly, the immature state of the receptive field looks different than that of the mature adult (Fig. 3). However, visual processing persists throughout this process, demonstrating that there must be certain developmental changes that occur to allow for visual information to be conveyed at any point during this sensitive period of refinement.

8 Spatial Processing in LGN

Some aspects of LGN receptive fields are firmly established at the onset of visual experience, while other aspects undergo large-scale refinement. Because some developmental changes are already complete before eye-opening, such as the segregation of eye-specific retinogeniculate inputs, some spatial receptive field properties are also fully developed at eye-opening and do not seem to require visual experience. This includes the degree of On/Off specificity which can be seen across species, in mice (Tschetter et al. 2018) and ferrets (Linden et al. 1981; Hahm et al. 1991; 1999; Akerman et al. 2002). On the other hand, receptive field size and shape undergoes a substantial change, converting broad and imprecise receptive fields into ones that are smaller, sharpened, or more temporally precise in a number of species, including cat (Wiesel and Hubel 1963a, b, Daniels et al. 1978; Tootle and Friedlander 1989; Gary-Bobo et al. 1995; Cai et al. 1997; Suematsu et al. 2013) ferret (Tavazoie and Reid 2000; Akerman et al. 2002; Davis et al. 2015), mouse (Tschetter et al. 2018), and primate (Blakemore and Vital-Durand 1986) (Fig. 3).

The decrease in receptive field size develops gradually with visual experience (Daniels et al. 1978). In cats, for example, this process begins at 1 week postnatally and can be observed progressively at 2 weeks, 1 month and 6 weeks postnatal (Kiley and Usrey 2017) (Fig. 3). During this period of refinement LGN receptive fields of X-cells and Y-cells may undergo changes if the animal is visually deprived, such as a reduction in spatial resolution (Lehmkuhle et al. 1978; Kratz et al. 1979; Mower et al. 1982; Sherman and Spear 1982; Yin et al. 2006). In mouse, a decrease in receptive field size occurs during the first week after eye-opening (Tschetter et al. 2018) (Fig. 3). Spatial receptive field shape modification in LGN during development by the convergence of RGC inputs has been classically demonstrated in monosynaptic paired recordings of RGCs and LGN cells. These recordings of RGC-LGN cell receptive field pairs have demonstrated remarkable similarity in receptive field sign, size, and spatial position (Levick et al. 1972; Mastronarde 1987; Usrey et al. 1999). Functionally, LGN cells’ receptive fields therefore largely reflect their retinal inputs. With a decrease in receptive field size there is also an increase in stimulus size suppression that develops early and an increase in peak spatial frequency. These changes are indicative of the development of visual acuity.

9 Temporal Processing in LGN

Temporal properties of individual neurons play an important role in the successful transmission of signals across synapses. RGC spike timing can determine whether there is a postsynaptic thalamocortical action potential. Additionally, timing of thalamocortical spikes can affect the synaptic conveyance of visual information to cortex (Mastronarde 1987; Usrey et al. 1999; Usrey 2002). Paired extracellular recordings of RGCs and LGN neurons have shown that even if there is a strong synaptic connection, a single RGC spike may not be sufficient to cause the postsynaptic cell to fire (Cleland and Lee 1985; Usrey et al. 1999; Usrey 2002). Instead, it seems that local activity, such as modulatory inputs and membrane potential, determines the response mode of LGN cells (whether they fire rapid bursts or tonic trains of spikes) (McCormick and Bal 1994; Usrey et al. 1998; Sherman and Guillery 2002; Wang et al. 2007; Litvina and Chen 2017).

The transformation of temporal processing over development may also be attributed to retinogeniculate convergence. By combining RGC inputs with similar tuning for visual stimulus features, a single LGN neuron will receive temporally correlated synaptic inputs that summate and jointly increase the efficacy of transmission of signals, causing the postsynaptic neuron to fire. It is well documented that paired-stimulus enhancement occurs when two spikes are fired by an RGC within 30 ms of each other, so that the second spike is much more likely to drive a response in the LGN cell (Mastronarde 1987; Usrey et al. 1998). Synchrony of timing early in development may provide the initial patterning of RGC synapses in LGN and synchrony in the adult may be important for efficient transfer of information across synapses. The importance of initial mapping of RGC synapses can be understood by the mature state of sustained and transient responses already at eye-opening (Tschetter et al. 2018).

The developmental time course of some temporal response properties extends for much longer, particularly in comparison with spatial response properties (Suematsu et al. 2013). One seemingly ubiquitous receptive field property seen in immature LGN cells across species is long-latency responses (Fig. 3). In primates, initial long-latency responses and a reduced range of temporal frequencies that elicit responses can be observed in infants at 5 weeks old (Blakemore and Vital-Durand 1986; Hawken et al. 1997; Movshon et al. 2005). Longer latency responses in young animals may be necessary in order to extend the time window for temporal summation of spontaneous and/or immature visually-evoked activity. This might allow for Hebbian mechanisms to strengthen correlated responses early in development. The decrease in response latency is thought to be due, in part, to the development of retinal axon myelination (Elgeti et al. 1976; Moore et al. 1976) and biophysical membrane properties (Monyer et al. 1994; Dunah et al. 1996; Wenzel et al. 1996; Ramoa and Prusky 1997; Liu and Chen 2008). For this reason, temporal response properties continue to develop with visual experience during the first few months of life (Stavros and Kiorpes 2008) (Fig. 3). In primates, at 1-month postnatal, temporal resolution is still less than one half the response of adult values. However, contrast gain and peak response rates for optimal stimuli have reached two-thirds of adult values, indicating modest maturation during early development. By adulthood (24 weeks), primates have fully developed responses to these stimuli (Movshon et al. 2005). In the ferret, LGN cell latencies exhibit a gradual reduction during the first 2 weeks of visual experience and continue to become slightly faster into adulthood (Tavazoie and Reid 2000) (Fig. 3).

10 Orientation and Direction Selectivity

Orientation and direction selectivity are well-documented receptive field properties. Universal among examined mammalian species, neurons in the primary visual cortex exhibit a preference for the orientation of a visual stimulus (Van Hooser 2007). That is, when an object in the visual field is oriented at a specific angle, a cell will increase its firing rate. In carnivores and primates cortical orientation selectivity is organized in columns, spanning the surface of the primary visual cortex, such that adjacent columns have similar orientation preferences that are slightly shifted (Hubel and Wiesel 1962; Thompson et al. 1983; Weliky et al. 1996).

In some species, cortical neurons exhibit selectivity for stimulus direction in addition to stimulus orientation. Direction-selective cells are found in certain layers of the primate visual cortex (Orban et al. 1986), throughout all layers of visual cortex in cats (Gilbert 1977; Ohki et al. 2005) and ferrets (Weliky et al. 1996; Li et al. 2008), and in some layers of the mouse (Niell and Stryker 2008; Rochefort et al. 2011; Hoy and Niell 2015) and rabbit visual cortex (Zhuang et al. 2013, 2014). In other species, such as tree shrew and squirrel, there are few direction-selective cells in visual cortex (Heimel et al. 2005; Van Hooser et al. 2013). Many mammals, such as mice (Kim et al. 2008; Huberman et al. 2009), rabbits (Barlow and Levick 1965), and squirrels (McCourt and Jacobs 1984), have large populations of retinal ganglion cells that are direction-selective, but these direction-selective retinal cells are apparently very rare or absent in the carnivore (Cleland and Levick 1974) and primate (Dhande et al. 2019) retina.

In carnivores, cortical orientation selectivity begins developing before eye-opening but requires visual experience for full maturation (Chapman and Stryker 1993; Li et al. 2006). Experiments in ferrets have demonstrated that when animals are dark-reared and visual experience is entirely eliminated, they still develop orientation-selective responses still develop, but not to the same level as that of a normally-reared animal. Binocular lid suture before the onset of visual experience, which affects the pattern of visually-evoked activity, results in dramatic defects and orientation selectivity is largely eliminated. While binocular deprivation via lid suture still allows for light activation of the retina, only low spatial and temporal frequencies are conveyed to the visual system. Thus it seems that the pattern of visual experience is crucial to the development of orientation selectivity (White et al. 2001). Similarly, direction selectivity requires patterned visual activity for development and matures during the first 2 weeks of visual experience in carnivores (Humphrey and Saul 1998; Li et al. 2006) (Fig. 2). However, if the animal is dark-reared, direction maps do not form and V1 neurons do not develop direction selectivity. Moreover, reintroducing normal visual experience at P45 (~2 weeks after eye-opening) allows for the return of some V1 response properties, such as orientation selectivity and contrast sensitivity, but direction tuning cannot be rescued (Li et al. 2006). Although direction selectivity has not been examined in dark-reared primates, there is also evidence in primates that orientation selectivity is present at birth and that direction selectivity requires visual experience. Direction selectivity matures over the first 4 weeks postnatal in the macaque (Hatta et al. 1998) (Fig. 2). Thus, there seems to be a precise critical period during which visual experience is necessary to cortical direction selectivity development.

While there has been considerable research into the circuit origins of orientation selectivity (see Ferster and Miller (2000) and Priebe (2016) for reviews), the precise mechanisms underlying direction tuning in V1 are still an active topic of research. One set of hypotheses for this circuit involves direct inheritance of direction selectivity from feed-forward inputs from individual LGN cells. We now have evidence that both orientation and direction selectivity can be encoded in individual LGN cells in a number of species, particularly in mice (Marshel et al. 2012; Piscopo et al. 2013; Scholl et al. 2013; Zhao et al. 2013), rabbit (Levick et al. 1969; Swadlow and Weyand 1985; Hei et al. 2014), and more weakly in squirrel (Zaltsman et al. 2015), cat (Hubel and Wiesel 1961; Daniels et al. 1977; Levick and Thibos 1980; Vidyasagar and Urbas 1982; Soodak et al. 1987; Shou and Leventhal 1989; Thompson et al. 1994), and primate (Lee et al. 1979; Smith et al. 1990; Cheong et al. 2013). In addition, orientation and direction-selective retinal ganglion cells (DSGCs), among other complex properties, can also be found in rodents and rabbits (Barlow and Levick 1965; Gollisch and Meister 2010). Additionally, it has been shown in cat (Vidyasagar and Urbas 1982) and mouse (Scholl et al. 2013; Zhao et al. 2013) that inactivating V1 has no significant effect on orientation selectivity in LGN cells, suggesting that this feature selectivity does not rely on a corticothalamic feedback mechanism and that V1 could potentially inherit such properties from LGN cells. Genetic tools have also revealed in mice that these DSGCs do indeed send monosynaptic inputs to LGN (Huberman et al. 2009; Kay et al. 2011; Rivlin-Etzion et al. 2011) and that these signals are then conveyed to superficial layers of V1 (Huberman et al. 2009). These data suggest that, in mouse, some V1 cells may inherit direction selectivity from the retina (Marshel et al. 2012; Cruz-Martin et al. 2014; Sun et al. 2015; Hillier et al. 2017).

Another family of hypotheses suggests that direction selectivity may arise from the patterns of convergence of LGN cells onto V1 cells. One possibility is that “lagged” and “nonlagged” LGN cells contribute to the spatiotemporal offsets (Cleland et al. 1971a, b, Marrocco 1976; Adelson and Bergen 1985; Mastronarde 1987; Saul and Humphrey 1990; Moore et al. 2005; Piscopo et al. 2013). Lagged and nonlagged cells differ in timing at low temporal frequencies but not high temporal frequencies. Therefore, as temporal frequency responses develop, there are a wider array of response times available (Saul and Feidler 2002). It is also possible that sustained and transient properties contribute to the development of direction selectivity (Cleland et al. 1971a, b, Marr and Ullman 1981; Lien and Scanziani 2018). Early sustained and transient responses driven by retinal waves may also inform direction selectivity. Sustained units active at low spatial phases with transient units at high spatial phases with different decay constants could combine to produce the spatiotemporal offset necessary for direction selectivity (Lien and Scanziani 2018). It could also be that temporal dynamics of ON and OFF cells confer direction selectivity. If ON and OFF cells exhibited different impulse responses, then their early maturation modified by temporal processing maturation could result in a direction-selective V1 cell (Chariker et al. 2021). Finally, direction selectivity could be conferred by inputs within the cortex. In the retina, null direction inhibition is a key mechanism for providing direction selectivity (Barlow and Levick 1965; Euler et al. 2002; Briggman et al. 2011), and there is some recent evidence that the inhibition to excitation ratio of signals arriving at a V1 neuron is slightly greater from the null side than the preferred side (Wilson et al. 2018; Rossi et al. 2020) (but see Priebe and Ferster (2005)). There are models of intracortical direction selectivity that suggest that direction selectivity can be produced via receptive field asymmetries and intracortical interactions (Goodwin et al. 1975; Emerson and Gerstein 1977; Sillito 1977; Ganz and Felder 1984; Suarez et al. 1995; Maex and Orban 1996; Livingstone 1998; Li et al. 2014). For example, time delays within a cortical cell’s receptive field could be generated in cortex, or intracortical inhibition between cells with different spatiotemporal receptive fields could produce cortical cells selective for opposite directions.

So what is the definitive mechanism and how does experience play a role? It’s complicated. A critical period of early visual experience appears to be required for the development of direction selectivity in carnivores and primates (Hatta et al. 1998). However, there is conflicting evidence describing experience-dependent plasticity of direction selectivity in rodents and rabbits. Some studies show that visual experience is not required for the development of direction selectivity in the retina or cortex and that, at eye-opening, DSGCs are already present and V1 cells are already tuned for direction (Chan and Chiao 2008; Elstrott et al. 2008; Chen et al. 2009, 2014; Sun et al. 2011; Wei et al. 2011). However, there is also evidence in these species for the requirement of normal visual experience in the development of direction tuning (Pearson et al. 1981). Direction selectivity in mouse layer 4 increases with visual experience, suggesting an experience-dependent influence (Hoy and Niell 2015). Research has shown that visual experience is required for the clustering of cells along the cardinal directions in mice DSGCs (Bos et al. 2016). Within one study, there was evidence that there are distinct retina-independent and retina-dependent computations for cortical direction selectivity (Hillier et al. 2017), which leaves open the possibility that some cortical direction selectivity is computed from retinal direction-selective sources and other cortical direction selectivity is computed via inputs that are not themselves direction-selective.

11 Clues from the Rapid Development of Direction Selectivity

Developmental changes can sometimes be used to tease apart circuit mechanisms. In the case of the ferret, cortical direction selectivity can be rapidly induced in naïve animals by providing 3–9 h of experience with a moving visual stimulus (Li et al. 2008; Van Hooser et al. 2012; Ritter et al. 2017; Roy et al. 2020; Li et al. 2008; Van Hooser et al. 2012; Ritter et al. 2017. Stacy et al. (2021) examined what changes occurred in LGN receptive fields during the rapid induction of direction selectivity and found that LGN cell latencies, sustainedness/transience, and orientation and direction selectivity were unchanged after 6 h of visual experience with a stimulus that causes a profound increase in direction selectivity in visual cortex. These results indicated that cortical direction selectivity can precede the maturation of LGN cell latencies (Tavazoie and Reid 2000), so it is unlikely that absolute LGN cell latencies are critical for direction selectivity in cortex. This lack of influence of short-term experience on LGN cells is consistent with another study that examined the impact of long-term dark-rearing on orientation or direction tuning of cat LGN cells and found no influence (Zhou et al. 1995). Consistent with these results, another experiment where optogenetic stimulation was provided to the cortex in lieu of visual experience showed that non-specific cortical activity was sufficient to cause an emergence of cortical direction selectivity (Roy et al. 2016), suggesting that cortical mechanisms are likely to be the main drivers of the development of direction selectivity in ferret V1.

12 Ocular Dominance and Its Plasticity

If the view through one eye is blurred during early development or if the alignment of the two eyes is compromised by strabismus, the impact of stimulation from that eye on the cortex will be greatly reduced, in a condition referred to as amblyopia (Hensch and Quinlan 2018). This condition is often modeled in animals by providing monocular deprivation through artificial lid suture (Wiesel and Hubel 1963a, b). The brain is only sensitive to this discrepancy in the quality of the input from the two eyes for a limited time after the onset of visual experience. This sensitive period has been measured carefully in animals such as the mouse (Gordon and Stryker 1996), ferret (Issa et al. 1999), and cat (Hubel and Wiesel 1970), and generally opens 1–2 weeks after the onset of visual experience and closes some weeks later, although some plasticity is possible in adult mice. In monkeys and humans, this critical period extends for several months or years (Lewis and Maurer 2005). Very recently, it has been shown that temporarily providing a complete inactivation of the dominant eye with tetrodotoxin can restore visual acuity through the weak eye even after the critical period has passed in both cats and mice (Fong et al. 2021).

Several cortical mechanisms have been identified that contribute to this phenomenon. After MDLS, there is a substantial weakening of the thalamic inputs from the deprived eye (Heynen et al. 2003). This weakening affects both feed-forward excitatory and inhibitory inputs, and does not, by itself, account for the reduced net drive of the deprived eye in cortical circuits (Miska et al. 2018). Intracortical inhibition is also increased, which serves to provide a net reduction of activity in response to input from the deprived eye (Maffei et al. 2006; Miska et al. 2018). These functional changes in feed-forward input reflect synaptic changes but occur well before the gross rearrangement of thalamocortical axons is a result of critical period plasticity (Antonini and Stryker 1996; Antonini et al. 1999; Silver and Stryker 1999; Trachtenberg et al. 2000; Trachtenberg and Stryker 2001). In a second step of plasticity, the overall drop in activity during MDLS activates homeostatic synaptic mechanisms and homeostatic increases in cortical cell excitability, which serve to potentiate overall responses from either eye (Kaneko et al. 2008; Hengen et al. 2013; Lambo and Turrigiano 2013). The net effect of these changes after several days is a net decrease in drive from the deprived eye, and a net increase in drive from the open eye.

Until very recently, ocular dominance plasticity was thought to be mediated exclusively through cortical methods and not through any changes in the lateral geniculate nucleus. When Hubel and Wiesel first described ocular dominance, their MDLS experiments in kittens provided the common notion that long-term visual deprivation leads to a decrease in thalamocortical axon size (Antonini and Stryker 1993; Antonini et al. 1999) and changes in LGN cell size were recapitulated in monkey LGN (Movshon and Dürsteler 1977), but that the functional properties of LGN neurons remained the same (Wiesel and Hubel 1963a, b). Further, because OD plasticity was assumed to result from a competitive process between the inputs from the two eyes, it was thought that OD plasticity was a distinctly cortical phenomenon, due to clear eye-specific segregation among LGN layers and what they thought were exclusively monocularly driven LGN cells (Wiesel and Hubel 1963a, b; Gilbert and Wiesel 1992). However, new findings in mice have challenged these ideas, both structurally and functionally.

In brief, the LGN of the mouse is divided into a shell and core (Clascá et al. 2012), which both receive inputs from the contralateral eye, and it was thought that only a small portion of the core received input from the ipsilateral eye (Fig. 1). However, newer data from adults and juveniles have shown that there may not be any cells that respond exclusively to ipsilateral eye stimulation. In addition, a large percentage of the LGN cell population is binocularly driven (Howarth et al. 2014; Rompani et al. 2017; Sommeijer et al. 2017), indicating that a fraction of eye-specific inputs are integrated prior to sending input to cortex. Furthermore, this evidence for binocular convergence was recapitulated in the primate, demonstrating that this finding is not a distinct feature of the mouse retinogeniculate synapse but may be common in mammals (Zeater et al. 2015).

While it has long been established that there are eye-specific laminae in the primate LGN, recent work characterizing the koniocellular layers has illuminated more complexity of individual cell responses than was originally appreciated. It was shown in marmoset that a relatively large percentage of the koniocellular layers (~30% of cells) are binocularly driven (Zeater et al. 2015). The presence of binocularly driven cells in the primate LGN has large implications for signal integration. These findings indicate that binocular signals may be integrated earlier in the visual system than previously thought. In cats there is also evidence for binocular interactions in the LGN, although the majority of binocular cells are suppressive in nature (Sanderson et al. 1971). Additionally, similar to the monkey, the total percentage of the LGN cell population thought to be binocularly driven is low (3% in monkey; 2–10% in cat) (Erulkar and Fillenz 1960; Bishop et al. 1962; Kinston et al. 1969; Dacey 1994; Cheong et al. 2013; Zeater et al. 2015; Dougherty et al. 2019). However, these data are relevant for understanding where in the visual pathway signals from two eyes converge, which is essential for understanding binocular vision.

New findings have also elucidated the potential for distinct mechanisms regulating thalamic plasticity and cortical plasticity. Data showed that, after a week of visual deprivation, LGN boutons that were monocularly driven exhibited a reduced or lack of response to the deprived eye, and often began to respond to the non-deprived eye (Jaepel et al. 2017; Huh et al. 2020). These responses were not propagated backward to the LGN from cortex, as silencing of cortex did not alter the changes in LGN axon plasticity with MDLS (Jaepel et al. 2017). In juvenile mice, early visual deprivation resulted in a change in LGN neuron response properties via depression of the deprived eye only. However, when they looked at cortex, they found that OD plasticity was the result of strengthening of LGN responses from the non-deprived eye (Rose et al. 2016; Sommeijer et al. 2017). These results suggest that changes in neurons in the LGN directly contribute to ocular dominance phenotypes in amblyopia.

13 Corticothalamic Development

There is a high likelihood that an interaction between retinogeniculate and corticothalamic refinement during development shapes mature LGN neurons. Although retinogeniculate inputs are the primary source of excitatory drive of thalamocortical cells, corticothalamic neurons make up a much higher percentage of synapses in the LGN compared to retinogeniculate afferents (Guillery 1969; Erişir et al. 1997a, b). In the mammalian visual system LGN cells project primarily to layer 4 and less densely to layer 6 of V1 (Hubel and Wiesel 1972; Gilbert and Kelly 1975; Hendrickson et al. 1978; Blasdel and Lund 1983; Ferster and Lindström 1983; Thompson et al. 2017). Layer 6 of the visual cortex in turn provides the larger part of corticothalamic feedback to the LGN (Gilbert and Kelly 1975; Katz 1987; Fitzpatrick et al. 1994; Briggs et al. 2016). Corticothalamic cells of the feedback pathway provide excitatory synaptic inputs to LGN directly in a retinotopic manner, in addition to sending synapses to other cortical layers that LGN projects to.

While much less is understood about the role of corticothalamic development, it is thought that corticothalamic feedback acts to modulate LGN activity, sharpening receptive fields of LGN cells in addition to enhancing signal transmission through LGN (Briggs and Usrey 2008). Like retinogeniculate development, corticothalamic development requires visual experience (Fagiolini et al. 1994; Gordon and Stryker 1996; Kang et al. 2013). However, corticothalamic projections develop later than retinogeniculate inputs and complete innervation after the onset of visual experience (Shatz and Rakic 1981; Jacobs et al. 2007; Brooks et al. 2013). Evidence from studies of layer 6 corticothalamic suppression suggests that during an intermediate window of maturation in the early visual pathway, corticothalamic innervation continues to influence the fine-tuning of the visual circuit (Thompson et al. 2016; Liang and Chen 2020). The enduring plasticity of retinal inputs to LGN at this time likely allows for their continued refinement via feedback from V1 and optimization of connectivity and complex feature tuning (Thompson et al. 2016). Thus, the timing of their developmental periods is essential for proper circuit formation.

14 Conclusions

Recent interest in characterization of the thalamic contribution to visual development and information encoding continues to evolve our understanding of the important role of the LGN in visual processing. In addition, growing evidence for signal integration and more complex feature selectivity of individual LGN neurons continues to alter not only our methods for understanding the retino-geniculo-cortical pathway, but also our interpretations of past findings, including what we thought we knew about ocular dominance plasticity (Figs. 2 and 3). Moreover, the advancements in technology development continue to allow for a more comprehensive depiction of the inputs and outputs of the LGN and the role of experience-independent and experience-dependent refinement. We describe a number of sensitive periods in visual development that importantly contribute to the refinement of the visual system and normal visual processing. Despite the anatomical differences in the organization of LGN across species, including mouse, ferret, cat, and primate, the changes in receptive fields of LGN neurons during development have several elements in common. Thus, findings from other species during their respective sensitive periods can aid in our understanding of general visual circuit wiring. Findings from typical development and experimental manipulations during these periods will continue to elucidate necessary properties of working neural circuits and differences that have evolved to maintain such robust organization across structures and species.