1. IMPORTANCE OF AN EYE FOR DETAIL

A. Diurnal Primates Have High-Acuity Spatial Vision

The fovea is an anatomically tiny but functionally dominant feature of the visual system in diurnal (day-active) primates. Figure 1 shows a highly schematic drawing of the output neurons [retinal ganglion cells (RGCs)], which are housed in the eye and project to the brain. The cone photoreceptors are very tightly packed in the center of the fovea (the foveola), where ganglion cells are very scarce [1–3]. As described below, the cones and RGCs are functionally connected via the extended processes of cones (Henle fibers) and by bipolar cells [4,5]. The result is that the fovea sends a high-resolution “megapixel” neural representation of the visual world lying within 5 deg diameter of the visual axis (red shaded area, Fig. 1).

 

Fig. 1. Simplified schematic drawing showing the main projection from the eye to the brain of a diurnal primate. Retinal ganglion cells (RGCs) in the back of the eye, and thalamocortical relay cells in the dorsal lateral geniculate nucleus (LGN) are shown as gray circles. The outlines of the eye and LGN are drawn approximately to scale. Red shading in the eye indicates the central-most 5 deg of visual angle, and in the LGN represents the region in which each cell has a receptive field within the same region. For simplicity, only the RGCs in the retina, and only one parvocellular and one magnocellular layer of the LGN are drawn.

Download Full Size | PDF

Over 80% of RGCs in macaque monkeys project to the dorsal lateral geniculate nucleus (LGN) in the thalamus [6,7]. The LGN in carnivores and primates is a prominently laminated structure, where each lamina gets dominant excitatory input from one eye, and relay cells in each lamina send an axon carrying monocular signals to the primary visual cortex [8–11]. I return to the question of two-eye inputs to LGN cells later in this review; for now, the LGN can be simplified to two main divisions comprising ventral magnocellular (M) (also referred to as MC where necessary in this paper to reduce ambiguity) and dorsal parvocellular (P) layers. The great majority of LGN cells are responsive to visual stimuli located within the central-most 5 deg, and P cells are ∼four times more numerous than M cells. These facts mean that a single-cell recording electrode placed randomly in the LGN will most likely encounter a P cell receiving (via a chain of synaptic connections) dominant excitatory input from one, or a small number of, cone photoreceptors located in the fovea.

B. Midget and Parasol Cells Form the Main Output Pathways of Primate Retina

Apart from the relatively small size of the marmoset eye, the retina and early visual system in marmosets show high similarity to that of humans and other Old World primates [9,12–15]. Figure 2(a) shows a whole-mount preparation of post-mortem retina from a human eye donor. The retina has been flattened from its roughly hemispherical shape by making relieving cuts. The major landmarks comprising the fovea and optic disk are shown in the outline drawing in Fig. 2(b). The image and drawing serve to emphasize the tiny physical size of the fovea in relation to its functional importance. The retina of a common marmoset Callithrix jacchus (the species from which most of the data shown in this review were drawn) is shown in Fig. 2(c) at the same scale. The left panel of Fig. 2(d) shows three Golgi-impregnated RGGs in a vertical (radial) section of a macaque retina drawn by Polyak [1]. The cells in the left panel were named midget cells for their very small dendritic tree, which ramifies in the inner plexiform layer. The cell on the right was named parasol cell for the appearance of the dendritic tree, which resembles “an open chinese [sic] umbrella or parasol” [1]. Midget and parasol cells project to the P layers and M layers of the LGN, respectively [6,7,16,17]. P LGN layers are likely the only central site where midget cell axons project. One early report described midget cells labeled following injection of a retrograde tracer [horseradish peroxidase] into the inferior pulvinar nucleus [18], but on retrospective analysis, this label could have been a result of uptake from optic tract fibers severed by the injection electrode as it passed close to the LGN. By contrast, the parasol cells project not only to the LGN but also to the superficial layers (stratum opticum) of the superior colliculus and possibly to the pretectum [19,20].

 

Fig. 2. Retinal landmarks and major cell types. (a) Post-mortem human donor retina prepared as a flat mount. (b) Outline drawing to show retinal landmarks. Circle indicates the optic disk. Star indicates the fovea. (c) Flat-mount preparation of marmoset retina, at the same scale as in (a). (d) Drawings from vertical (radial) sections of Golgi-impregnated retina of macaque (adapted from [1]).

Download Full Size | PDF

C. Fovea Is Linked to M–L Color Signals

Figure 3 illustrates schematically the link between high-acuity vision and M–L color vision in the fovea via the midget-P pathway in trichromatic diurnal primates. Polyak ([1], p. 418) recognized this link, and noted that in the fovea, “The midget ganglion cells greatly predominate … this is precisely the region where both the acuity and the color perception is the best.” The long (L) and medium (M) wavelength-sensitive cones make up over 90% of photoreceptors in trichromatic primates [21–26], and within the central few degrees, each cone provides dominant excitatory input to one ON-type and one OFF-type midget bipolar cell, which in turn each contact one midget ganglion cell [Fig. 3(b)] in “private line” pathways [1,5,27–29].

 

Fig. 3. Random retinal wiring for cone opponent receptive fields (a) Schematic drawing of a diurnal primate eye. Ganglion cells are shown as gray circles. (b) Schematic drawing of the midget (private line) pathway near the fovea. Each L and M cone provides excitatory input to two midget ganglion cells via two midget bipolar cells. (c) Inhibitory inputs to private line pathway. Excitatory (+) inputs from cones to bipolar cells and from bipolar cells to ganglion cells are opposed by inhibitory (–) inputs from horizontal cells and amacrine cells; the inhibitory spatial pool (surround mechanism) in both cases is larger than the excitatory (center) mechanism.

Download Full Size | PDF

The private line pathways mean that the receptive field center of foveal-supplied midget ganglion cells will thus inherit the spectral signature of either an L or M cone, as well as preserving the high-acuity spatial signals from the tightly packed cone array. The excitatory receptive field center input is counteracted by antagonistic surround signals that originate from a spatially wider pool of photoreceptors [Fig. 3(c)], and are conducted through the lateral processes of horizontal cells in the outer plexiform layer and amacrine cells in the inner plexiform layer [30–33]. The receptive field surround thus is fed by many more cones than feed the center. Because M and L cones are randomly positioned in the cone mosaic, the M to L weight in the surround will depend on the number of contributing cones and the local proportions of M and L cones. In the example shown in Fig. 3(c), a single L cone dominates the ganglion cell center input and the surround comprises a mixture of signals from L and M cones. Thus, the M cone input to the surround is greater than the M cone input to the center and yields spectral opponency superimposed on center-surround opponency. This “random wiring” model established to explain M–L opponency in Old World trichromatic primates [34,35] also can explain cone-opponent signals in the polymorphic color vision system of New World monkeys, as first synthesized by Mollon and colleagues [36]. They wrote: “Ganglion cells that draw their field-centre input from only one cone … will vary in their spectral sensitivities … the later stages will be able to extract the spectral information that is present in the retinal signal.”

D. Private Line Wiring in the Marmoset Fovea

Figure 4 shows evidence that the same kind of private line wiring in the midget-P pathway as described above for Old World monkeys and humans is also present in the central retina of marmosets. The spatial density of cones and ganglion cells in the central human retina [Fig. 4(a)] shows a sharp density peak of cone photoreceptors at the foveal center and a broader and lower density peak of ganglion cells near 1 mm eccentricity [2,3]. As noted above, the ganglion cells are displaced (predominantly by Henle fibers) to greater eccentricities than the cones that feed them. The peak ganglion cell density is much lower than that of cones, but the area of the retina containing these ganglion cells is much larger than the central bouquet area containing the cones that feed them (the increase in area is roughly proportional to the square of distance from the fovea). This fact means that within the central retina, the number of ganglion cells is easily more than double the number of cones that feed them. With increasing eccentricity, ganglion cell density falls more rapidly than cone density, and there is numerical convergence of cones to ganglion cells at eccentricities greater than ${\sim}{3}\;{\rm mm}$.

 

Fig. 4. Foveal topography and connections of midget ganglion cells. (a) Spatial density of cones and ganglion cells in human retina. Symbols show spatial density estimated from sum-of-exponential fits to cell density measures. Cone peak density is at foveal center; ganglion cell peak is broader and displaced ${\sim}{1}\;{\rm mm}$ from foveal center. (b) In marmoset retina, the cone density gradient is sharper than human near foveal center and shallower at greater eccentricities. GCL, ganglion cell layer; INL, inner nuclear layer; IPL, inner plexiform layer. (a), (b) Adapted from [37]; (c), (d) adapted from [38].

Download Full Size | PDF

The topography of the same two populations in a marmoset retina [Fig. 4(b)] is very similar [14]. In marmosets, however, the cone density gradient in the peripheral retina is shallower, and the ganglion cell density gradient in the central retina is steeper. In marmosets, ganglion cells outnumber cones within the central-most 1 mm, but there is already convergence of cones to midget bipolar cells at greater eccentricities [39]. Figure 4(c) shows a vertical (radial) section through the central retina in a marmoset, with OFF midget bipolar cells labeled using antibodies against the carbohydrate epitope CD15 [38]. Comparison of the density of OFF midget bipolar cells with those of cone photoreceptors (making allowance for the lateral spatial displacements in the fovea) shows that each cone within 1 mm radius of the foveal center makes synaptic contact with one OFF midget bipolar cell, midget bipolar cells make contact with one or two cones at 1 to 2 mm eccentricity, three to four cones at 3–4 mm, and five or more cones in the peripheral retina [14,38,39].

In sum, private line connections are present in the marmoset fovea but do not apply to greater eccentricities. By contrast, with L/M cone connections, the short-wave sensitive (S) cones make no or only sparse contact with midget bipolar cells in marmosets [40]: this is a point of difference between marmoset retina wiring and wiring in macaque and human retinas, where S-cones do contact OFF midget bipolar cells [41,42].

The next step in the midget-P pathway is the connection between midget bipolar cells and midget ganglion cells. Here too we see a similar (but not identical) pattern of connections in marmosets as described for human and Old World monkeys. Figure 4(d) shows a vertical (radial) section through the central marmoset retina, which is double labeled to show midget bipolar cells (green) in relation to midget ganglion cells (magenta), which were retrogradely labeled following injection of a biotin-based tracer into the P layers of the LGN [38]. The experiment shows a tight pattern of one-to-one connections between midget bipolar cells and midget ganglion cells in the marmoset fovea, with convergence beginning at 1–2 mm and increasing to ∼10 midget bipolar cells converging onto a single midget ganglion cell by 4 mm eccentricity. No difference in the pattern of connections of L and M cones to bipolar cells, or bipolar to ganglion cells, is evident on comparing dichromatic and trichromatic phenotype marmosets [38,43]. In summary, the data are consistent with one-to-one connections in the fovea as a sufficient (but not necessary) substrate for M–L opponent signals in the marmoset retina.

 

Fig. 5. Transmission of spatial and spectral signals by parvocllular (P) cells in marmosets. (a) Peristimulus time histograms (PSTHs) of responses to 0.5 Hz temporal square wave achromatic (ACH), L-cone isolating (L) and M-cone isolating (M) modulation. “Red-ON” receptive field. Note cone opponent L+/M− signature. (b) Simultaneously recorded non-opponent OFF receptive field shows no response to L or M stimuli. Red traces show action potential waveforms. (c) Responses of the opponent cell shown in (a) to chromatic red-green (closed red symbols) and luminance-modulated (open gray symbols) sine gratings. Note vigorous low-pass characteristic response to chromatic gratings. (d) Non-opponent cell shows negligible responses to chromatic gratings. Both cells show similar bandpass response to luminance modulation. (e) Receptive field radius measured from responses to luminance-modulated gratings in dichromatic marmosets. (f) Radius from responses to the same stimulus in trichromatic marmosets. Note heavy overlap of the data sets. (c)–(f) Adapted from [47].

Download Full Size | PDF

E. Transmission of Spatial and Chromatic Signals in the Midget-Parvocellular Pathway

The anatomical results described above imply that in trichromatic marmosets, the midget-P pathway must carry neural signals serving high-acuity spatial vision as well as M–L opponent signals serving the red-green axis of color vision. It is known from recordings in retina and LGN of trichromatic Old World monkeys that midget-P pathway cells, when stimulated with intermediate-to-high contrast achromatic spots or gratings, exhibit the smallest receptive fields at a given eccentricity [44–46]. A natural question arising is whether the superposition of M–L chromatic signals onto a system serving high-acuity spatial vision has come at any cost for transmitting signals serving spatial vision. Dichromatic marmosets lack an M–L chromatic channel, thereby offering a chance to answer this question.

Figure 5 shows the answer, from extracellular recordings of single-cell action potentials in the P layers of the LGN [48,49]. Each marmoset was diagnosed as dichromatic or trichromatic by the presence of M–L opponent responses in P cells as well as (in nearly all cases) polymerase chain reaction amplification of M/L opsin-encoding genes. Responses of two P cells in a trichromatic female marmoset to achromatic (ACH, 80% contrast), L-cone isolating (L, 8% cone contrast) and M-cone isolating (M, 8% cone contrast) temporal square wave 0.5 Hz modulation are shown in Figs. 5(a) and 5(b). The red-ON P cell [Fig. 5(a)] shows a cone opponent signature, with opposite sign responses to L-cone increments and M-cone decrements. The non-opponent OFF cell [Fig. 5(b)] shows vigorous responses to achromatic decrements but negligible responses to cone-isolating modulation. The spatial frequency transfer functions for these cells are shown in Figs. 5(c) and 5(d). The opponent cell [Fig. 5(c)] shows a low-pass characteristic with high modulation amplitude for M–L chromatic gratings. When the M cone modulating and L cone modulating grating components are set to the same spatial phase, the grating has no M–L chromatic contrast (it appears yellow-black), and the spatial frequency transfer function shows a strong bandpass characteristic. These curves are consistent with center excitation dominated by L cone(s) and surround inhibition dominated by M cones. The non-opponent OFF cell responds only feebly to M–L chromatic gratings but shows a strong bandpass characteristic for luminance-varying gratings. Other measurements from this cell (not shown) showed L-cone inputs to both center and surround mechanisms underlie this non-opponent response signature.

These cells were recorded simultaneously through a single electrode; their receptive fields were separated by 0.42 deg. The experiment shows directly the presence of cells with chromatic distinct characteristic but indistinguishable spatial frequency transfer for luminance contrast. This conclusion is supported by the population data drawn from 15 dichromatic animals [${\rm n} = {143}$ cells, Fig. 5(e)] and 11 trichromatic animals [${\rm n} = {187}$ cells, Fig. 5(f)]. For example, the mean center radius of receptive fields within 8 deg eccentricity in dichromats (0.062 deg, SD 0.036, ${\rm n} = {112}$) is very close to that of trichromats [0.063 deg, SD 0.036, ${\rm n} = {155}$, ${\rm p} = {0.66}$, Wilcoxon non-parametric rank-sum test for independent samples (henceforth, WILC)]. Further, among trichromatic animals, the corresponding values for opponent receptive fields (0.059 deg, SD 0.031, ${\rm n} = {55}$) is very close to that of non-opponent receptive fields (0.066 deg, SD 0.038, ${\rm n} = {100}$, ${\rm p} = {0.33}$, WILC). Detailed quantitative comparisons of the proportion of opponent and non-opponent cells, and the opponency strength in different trichromatic phenotypes are given elsewhere [48–50].

These and similar experiments by Kremers and colleagues [51–53] support the view that M–L chromatic signals in P cells do not come at a cost for transmitting high-acuity spatial signals. The corresponding evolutionary interpretation is that red-green color vision in primates has evolved by taking a “free ride” on a pre-existing pathway specialized for high-acuity spatial vision. A detailed comparison of the relative merits of P pathway and M pathway cells as afferent channels serving achromatic spatial vision is given in the review paper by Lee et al. [54].

2. SUBCORTICAL SEGREGATION OF COLOR SIGNALS

A. S-Cone Signals Travel through K Layers

The LGN is the major dorsal thalamic access point for visual signals to reach the primary visual cortex (area 17, V1) via geniculocortical relay cells in the main dorsal (P) and ventral (M) layers. Until the 1980s, the midget-P pathway was described in textbooks as carrying all the neural signals used for color vision to V1. But two main lines of evidence then raised an alternative possibility. First was the demonstration that non-midget ganglion cells could be retrogradely labeled following injections including the main P layers in macaques [55,56]. Second was evidence, also from macaques, that some cells in the sparsely populated interlaminar/koniocellular (K) regions between the main P and M layers are geniculocortical relay cells [57]. Many of these K cells target the cytochrome-oxidase rich “blob” regions in supragranular layers of V1 [58–60], which other studies had linked to color-selective responses [61,62]. This combination led to the question as to whether some or all color signals could be traveling on non-midget P and/or K pathways.

The K layer located between the main P and M layers (K3) in marmosets is more prominent than in macaques, which means this question can be tackled by targeting extracellular recordings to the K layers in marmoset LGN. Figure 6 shows a drawing from a coronal section through marmoset LGN with the path of a high-impedance electrode penetration. The main P and M layers are shown respectively in dark and light gray shading; the white regions between the main layers indicate the intercalated/K layers. Positions of example P, K, and M cells are marked with circles, and positions of other recorded cells on the same track are marked with horizontal ticks. The peristimulus time histograms [PSTHs, Fig. 6(b)] show responses of these three example cells to temporal square wave (0.5 Hz) achromatic and S-cone isolating contrast modulation of spatially uniform fields encompassing the receptive field center and surround. The P-OFF cell and M-OFF cell show respectively sustained and transient responses to achromatic contrast, but are not modulated by the S-cone isolating field. The K cell shows a “blue-ON” response signature with vigorous ON-sign responses to the S-cone stimulus but negligible responses to achromatic modulation.

 

Fig. 6. Identifying S-cone recipient receptive fields. (a) Drawing from a coronal section through marmoset lateral geniculate nucleus. Parvocellular (P) and magnocellular (M) layers are shaded. The unshaded regions between P and M layers are the intercalated (K) layers. Vertical line shows electrode path. Symbols indicate positions of recorded cells shown in (b). (b) Responses of the indicated P, K, and M cells to achromatic (left column) and S-cone selective (right column) 0.5 Hz temporal square wave modulation. P and M cells show vigorous response to achromatic modulation and negligible response to S-cone modulation. The K cell shows complementary response signature. Red traces indicate action potential waveforms. Adapted from [63].

Download Full Size | PDF

Figure 7(a) shows a drawing based on a typical coronal section through the mid-posterior level of marmoset LGN. The leaf plots [Fig. 7(b)] show the collated positions of S-ON and S-OFF cells (left) and M–L opponent “red-green” cells as a function of depth relative to the dorsal surface of the LGN. Here it becomes clear that although the K layers make up less than 10% of the total volume of the LGN, almost all S-cone opponent cells are encountered in the K layers [64,65]. A complementary pattern of segregation is given by the M–L opponent cells [49,50]. A similar (but less striking) pattern of segregation was shown in later experiments to apply to the macaque LGN, where the K layers are less clearly defined [66]. With this limitation, it can be concluded that the S-cone opponent and M–L opponent signals travel on distinct branches of the afferent visual pathway. This segregation continues at least to the level of afferent relay cell terminals in V1, where K terminals and S-cone opponent responses are segregated to supragranular cortical layers, but M–L opponent responses are more commonly recorded in the granular layer 4Cb [67]. In conclusion, the answer to the question raised at the beginning of this section is that some color signals travel on K pathways.

 

Fig. 7. Segregation of cone opponent signals in the LGN. (a) Drawing from a coronal section through marmoset LGN. Cells in parvocellular (PC) and magnocellular (MC) layers are indicated by shaded circles. (b) Location of cone-opponent receptive fields relative to the dorsal border of the LGN. Note concentration of short-wave sensitive (S) cone opponent responses in koniocellular (K) layers 3 and 4, concentration of medium/long-wave sensitive (M/L) opponent responses in PC layers, and lack of opponent responses in magnocellular (MC) layers. Adapted from [47].

Download Full Size | PDF

B. S-Cone Recipient Receptive Fields Have Distinct Spatial and Temporal Properties

Retrospective analysis of the early literature shows a consistent presence of S-cone recipient receptive fields among the “type II” category introduced by Wiesel and Hubel [8,44,68–70]; these cells show large receptive fields with overlapping ON and OFF regions. Figure 8 compares the receptive field center radius of P cells with that of S-cone recipient S-ON and S-OFF cells in marmosets [71]. The S-ON receptive fields have typically four to five times greater radii than P cells at nearby eccentricities, and S-OFF receptive fields are ∼two-fold larger than S-ON cells. Further, the visually evoked onset latency of S-ON and S-OFF cells [Fig. 8(b)] is longer (by 8–25 ms) than that of P and M cells [72]. These results largely parallel results of the studies cited above and more recent studies in macaques [17,73], and are consistent with links between S-cone circuitry and melanopsin-expressing (intrinsically photosensitive) ganglion cell populations [74]. The question as to how much “cross talk” there is of S-cone signals to M and P pathways is treated elsewhere [71,75,76].

 

Fig. 8. Distinct receptive field properties of S-cone recipient cells in marmoset LGN. (a) Receptive field center radius of parvocellular (P) cells measured from responses to achromatic sine gratings and S-cone recipient (S-ON and S-OFF) cells measured from responses to S-cone isolating sine gratings. Bar plots in right panel show mean and s.e.m. for cells within 10 deg eccentricity. Note large center radius of short-wave sensitive (SWS) cone opponent cells. (b) Visual evoked onset latency for temporal square-wave achromatic pulses in magnocellular (M) and parvocellular (P) cells, and for S-cone isolating pulses in S-cone recipient (S-ON and S-OFF) cells. (a) Adapted from [71]; (b) adapted from [72].

Download Full Size | PDF

C. Resolving Mysteries of the S-OFF Pathway(s)

The anatomical discrepancy noted above (Section 1.D) whereby S-cones make contact with OFF midget bipolar cells in macaques but not in marmosets is supported by recordings of small-field S-OFF receptive fields in the central retina of macaque monkeys [42,77]. An early recording study in macaque LGN reported cells with strong inhibitory input from S-cones and large receptive fields, with responses that in retrospect are consistent with inputs from melanopsin-expressing ganglion cells. If S-OFF receptive fields in marmosets are based on private line connections in the central retina, then they would be closer in radius to midget-P pathway P cells than to S-ON cells. But as shown in Fig. 8(a), this is clearly not the case in marmoset LGN. The recordings that gave rise to the data presented in Fig. 8 were made with natural optics, that is, they were not corrected for chromatic aberration. The receptive fields of both S-ON and S-OFF cells would be enlarged by uncorrected chromatic aberration, but it is hard to understand how chromatic aberration could enlarge S-OFF fields to a greater extent than S-ON fields. Taken together then, there are clear links between S-cone signals and large receptive fields, likely corresponding to some or all melanopsin-expressing ganglion cell populations, in both macaque and marmoset monkeys. Electrophysiological evidence for S-OFF midget receptive fields in macaques is solid, and anatomical evidence for S-off midget circuits in macaque and human central retinas is also firm.

 

Fig. 9. Parallel subcortical visual pathways in highly simplified and schematic views. (a) Parvocellular (P) and magnocellular (M) layers get input from high-acuity retinal ganglion cell populations and transmit visual signals in tight topographic order to the granular layers of primary visual cortex (V1). (b) Koniocellular (K) layers get input from diverse populations of wide-field ganglion cells and transmit visual signals in widespread projections including superficial layers of V1 as well as extrastriate areas.

Download Full Size | PDF

3. BEHIND THE BLUE DOOR

A. K Pathways Are a Diverse and Ancient Set of Visual Pathways

The properties of receptive fields in the main P and M layers have been intensively examined for over five decades, but K pathways have remained largely unexplored. As shown schematically in Fig. 9, M and P pathways [Fig. 9(a)] are thought to have evolved recently (in common with the lemniscal somatosensory pathways and “direct” auditory pathway), and form tightly organized inputs to granular layer 4 of the primary visual cortex [78–80]. M and P pathways are concerned with high-acuity analysis of visual inputs. By contrast, K pathways [Fig. 9(b)] are part of an evolutionarily ancient group of thalamocortical pathways (including paralemniscal somatosensory and tegmental auditory pathways). These pathways project mainly to superficial or “supragranular” cortical layers, and make widespread connections within and across cortical areas [59,81–85]. Anatomical and physiological evidence further suggests that activity in K pathways could serve to coordinate or bind activity within and across cortical areas [82].

B. Color Unlocks the Door to Study of K Pathways

The fact that afferent nerve signals likely serving the S-cone opponent axis of color vision are associated with K pathways gives an experimental opportunity to understand other facets of K pathway function. This opportunity arises because the presence of S-ON responses means that the recording electrode is almost certainly in a K layer, allowing targeted recordings of nearby K cell receptive fields. This simple insight allowed our group to make a series of discoveries. The first was the presence of orientation selective cells in K layers, while orientation selectivity to that time was considered to first emerge in the primary visual cortex [63]. The second was delineation of a specific pathway linking a dedicated retinal bipolar and ganglion cell pathway (narrow thorny ganglion cells) to the K layers [86]. The third was the presence of a slow (multi-second period) intrinsic activity rhythm in K cells, linking the K pathways to brain circuits that generate rhythmic activity (“brain waves”) in sleep–wake cycles, anesthesia, and epilepsy [87]. Such interactions are manifest, for example, when repetitive visual stimuli induce epileptic seizures. These experiments revealed high diversity in geniculate afferent pathways and receptive fields in the dorsal thalamus with properties that, in primates at least, were customarily considered to be an emergent property of cortical circuitry.

C. Link between Color and Binocular Vision

During the experiments on orientation selectivity in K cells (Section 3.A), it also became apparent that some of the recorded K cells could be excited by stimuli delivered through either eye. This observation goes against the established wisdom that in diurnal primates, the LGN is a laminated structure where cells in each lamina get excitatory from one eye [78]. The final section of this paper describes further exploration of this property of receptive fields in K layers.

Recordings in K layers are challenging because K cells are small and K layers are thin. To increase the yield of recorded cells, we adopted semiconducter-based array recording technology, whereby the single recording tip of a high-impedance sharp electrode is replaced by 36 recording surfaces arranged on parallel shanks. The path of such an array electrode in the LGN is shown in Fig. 10(a). The recording surface configuration together with a single thalamocortical relay (TCR) cell in the LGN at the same scale is shown in Fig. 10(b).

 

Fig. 10. Array electrode recording in marmoset LGN. (a) Coronal section through marmoset LGN processed following array recording experiment. Vertical paths of dual-shank array electrode are visible. (b) Configuration of array electrode. Each shank is supplied with 16 recording surfaces. Red inset drawing shows a thalamocortical relay (TCR) cell at the same spatial scale. (c) Reconstructed locations of cells recorded in a single array electrode penetration passing through parvocellular (P) and koniocellular (K) layers (adapted from [88]).

Download Full Size | PDF

A sample of reconstructed positions of recorded cells [Fig. 10(c)] makes clear that a high density of recording can be achieved. Further, because typically five to 10 receptive fields are be recorded simultaneously, it becomes possible to disentangle idiosyncratic properties of individual neurons from drifts in physiological states that affect all LGN cells in the same way. As might be expected from the size of the TCR cell drawn in Fig. 10(b), the same receptive field can on occasion be recorded on more than one surface. This is, however, easily correctable because the recorded action potentials occur simultaneously on both surfaces.

Figure 11(a) shows two pairs of PSTHs from two simultaneously recorded K cells, each of which was excited by stimuli delivered through the dominant (Dom) or non-dominant (Non-dom) eye [88]. The non-dominant eye responses are (by definition) of lower amplitude than the dominant eye responses, and the response peak typically appears 10–20 ms after the dominant eye response peak. The overall shape of the PSTH is nevertheless maintained between dominant and non-dominant eye responses. These binocular responses are more likely a result of convergent direct input from the two eyes than a result of cortico-geniculate feedback pathways, because the feedback pathways typically have modulatory effects rather than high-amplitude “driver” modulation exceeding 15 spks/s.

 

Fig. 11. Binocular responses in marmoset LGN. (a) Peristimulus time histograms (PTSHs) of two koniocellular (K) cells to 200 ms temporal square wave S-cone isolating (left column, S-ON cell) and M/L-cone isolating (right column, suppressed-by-contrast cell) pulses. Upper row shows responses to dominant (Dom) eye stimulation, and lower panels show responses to non-dominant (Non-dom) eye stimulation. Note excitatory responses to stimuli delivered through either eye. (b) Scatterplots compare peak response amplitudes for stimuli delivered through dominant and non-dominant eyes. Parvocellular (P) and magnocellular (M) pathway cells respond only to dominant eye stimulation. Koniocellular (K) cells receiving S-cone input (“K-S”) and other K cells (“K-non-S”) show variety of excitatory responses to non-dominant (Non-dom) eye stimulation. Adapted from [88].

Download Full Size | PDF

The scatterplots in Fig. 11(b) compare peak response amplitude to dominant (Dom) eye stimulation to non-dominant (Non-dom) eye stimulation. Baseline (“spontaneous”) spike rates have been subtracted. As expected, the P and M populations show negligible responses to non-dominant eye stimulation. The K population by contrast shows variable and sometimes close to equal amplitude responses to non-dominant and dominant eye stimulation. The S-cone recipient (K-S) and other K cell receptive field subtypes (K-non-S) show a similar distribution of non-dominant eye influence, leading to the conclusion that the K pathways form a subcortical site of binocular integration in marmosets. Further details and comparisons of binocularity in primate, carvivore, and rodent LGN are given in [88].

Figure 12 shows a pairwise correlation matrix where the response amplitude in 10 ms bins during 400 following stimulus onset is correlated for all recorded K cells (${\rm n} = {19}$), with distinct K cell subclasses grouped. The observed correlations for each cell lie on the negative slope diagonal. Other points on the same row/column show synthetic correlation with every other cell (in most cases from separate recordings). Here the aggregation of positive correlations (warm colors) within each subclass indicates that the binocular convergence in K pathways involves selective preservation of response properties across the two-eye inputs rather than indiscriminate convergence. Such preservation may indicate that the same kind of developmental plasticity is at work in the K pathways as that underlying the matching of convergent two-eye inputs in the primary visual cortex.

 

Fig. 12. Pairwise correlation matrix. Each matrix point shows correlation of response amplitude in 10 ms bins for 400 ms following stimulus onset. Measured responses of 19 koniocellular cells appear on the negative slope diagonal. Synthetic correlation values from all other measured cells are shown on the same row/column. Adapted from [88].

Download Full Size | PDF

The foregoing observations allow the speculation that K pathways represent a primordial dorsal thalamic scaffold serving low-acuity visual functions including short-wave color vision and crude binocular vision. The high-acuity M and P pathways by contrast deliver monocular signals to the cortex, where fine-grain retinal disparity can be extracted to form the basis for true stereoscopic binocular vision.

4. CONCLUSION

Study of the subcortical visual system in marmosets has given small but significant advances to understanding the cell types serving primate color vision, and the way that the distinctive properties of M-L opponent and S-cone opponent color vision channels emerge in the retina and are transmitted through distinct subdivisions of the LGN in the dorsal thalamus. Comparison of structure and physiology of dichromatic and trichromatic marmosets supports the origin of M-L opponent color channels with the fovea and high-acuity vision on one hand, and between S-cone opponent color channels and evolutionarily primitive subdivisions of the visual system on the other hand.