Introduction and history

Among mammals, simian primates have the best-developed color vision. Humans, apes, and Old World monkeys show routine trichromatic color vision based on receptors sensitive to short, medium, and long wavelengths of light. We know from molecular biology that the short-wave sensitive (S) cones diverged from other cone types relatively early in the mammalian evolutionary history, whereas distinct long-wave (L) and medium-wave (M) pigments under the genetic control of X-chromosomes diverged in different primate species, at different times, during the course of evolution (Shyue et al., Reference Shyue, Hewett-Emmett, Sperling, Hunt, Bowmaker, Mollon and Li1995; Nathans, Reference Nathans1999). The bases of these discoveries are reviewed elsewhere in this volume (Hunt & Peichl, Reference Hunt and Peichl2013). As we note above (see Abstract section) and below, nearly everything we know about S-cone signals in the subcortical visual system comes from the study of visual systems in cats and primates (monkeys), and in this review, we concentrate on the results from macaque and marmoset monkeys, which are the species where most is known.

Early electrophysiological studies of color vision in central pathways began before or concurrent with significant discoveries on color vision evolution and genetics. The earliest experiments demonstrated color opponent properties in single cells but did not identify cone inputs or characterize spatial properties (DeValois, Reference DeValois1960; Hubel & Wiesel, Reference Hubel and Wiesel1960). Early recorders of single unit activity in monkey (usually macaque) retina and lateral geniculate nucleus (LGN)–treated S, M, and L cone inputs to receptive fields as potential inputs to a multidimensional system, where opponent inputs in different combinations (e.g., +S/−M; −S/+L; +M/−L) were equally likely (DeValois et al., Reference DeValois, Abramov and Jacobs1966; Wiesel & Hubel, Reference Wiesel and Hubel1966; Gouras, Reference Gouras1968; DeMonasterio et al., Reference DeMonasterio, Gouras and Tolhurst1975; Gielen et al., Reference Gielen, van Gisbergen and Vendrik1982; Creutzfeldt et al., Reference Creutzfeldt, Lee and Valberg1986; Crook et al., Reference Crook, Lee, Tigwell and Valberg1987). It soon became apparent that among spectrally opponent cells, a class called “blue-on/yellow-off” (hereinafter referred to as blue-on) was more commonly encountered than its symmetric counterpart “yellow-on/blue-off” (hereinafter referred to as blue-off); this was especially true in the retina (Malpeli & Schiller, Reference Malpeli and Schiller1978; DeMonasterio, Reference DeMonasterio1979; Valberg et al., Reference Valberg, Lee and Tigwell1986). A second concept derived from early work was that specific color-coding receptive fields exist that show spatially overlapping spectrally opponent “Type II” organization (Wiesel & Hubel, Reference Wiesel and Hubel1966; DeMonasterio & Gouras, Reference DeMonasterio and Gouras1975; Dreher et al., Reference Dreher, Fukada and Rodieck1976). This assumption turns out to be reasonable in the case of blue-on cells (Tailby et al., 2008a,Reference Tailby, Szmajda, Buzás, Lee and Martinb; Crook et al., Reference Crook, Davenport, Peterson, Packer, Detwiler and Dacey2009), but the spatial and chromatic organization of red–green opponent cells turned out to be more variable and complex than expected (Lee et al., Reference Lee, Virsu and Elepfandt1983, Reference Lee, Shapley, Hawken and Sun2012; Derrington et al., Reference Derrington, Krauskopf and Lennie1984; Rodieck, Reference Rodieck, Valberg and Lee1991; Reid & Shapley, Reference Reid and Shapley2002; Diller et al., Reference Diller, Packer, Verweij, McMahon, Williams and Dacey2004; Martin et al., Reference Martin, Blessing, Buzás, Szmajda and Forte2011). A final assumption of early work was that all color signals pass through the dorsal parvocellular layers of the LGN en route to the granular layers of primary visual cortex. This assumption is likely to be correct in the case of red–green opponent cells but as shown below is unlikely to be correct in the case of blue-on and blue-off cells.

Distribution of cone-opponent signals in LGN

Retrograde labeling studies in macaque and marmoset monkeys showed that the small bistratified (blue-on) cells project to the LGN (Rodieck & Watanabe, Reference Rodieck and Watanabe1993; Dacey et al., Reference Dacey, Peterson, Robinson and Gamlin2003; Szmajda et al., Reference Szmajda, Martin and Grünert2008), and one prescient single-cell study in macaque LGN indicated blue-on cells were more commonly encountered in the deeper P layers than in the superficial P layers (Schiller & Malpeli, Reference Schiller and Malpeli1978). Here, we focus first on results from recordings in marmoset monkeys, since the koniocellular layers (where, it turns out, many blue-on receptive fields are found; see below) are relatively broad (Walls, Reference Walls1953; Kaas et al., Reference Kaas, Huerta, Weber and Harting1978) and targetable with an electrode (Martin et al., Reference Martin, White, Goodchild, Wilder and Sefton1997). Fig. 1A compares the cone sensitivity in macaque monkey with examples of a male and female marmoset (Callithrix jacchus). Marmosets in common with most other New World monkeys show sex-linked polymorphism of color vision, whereby genes coding medium- and long-wavelength-sensitive pigments are present as allelic variants at a single locus on the X chromosome (reviewed by Mollon, Reference Mollon1989; Jacobs, Reference Jacobs2008). Males are thus obligate dichromats, whereas females expressing different alleles on each X chromosome show red–green cone opponent receptive fields (Yeh et al., 1995b) and can make trichromatic color discriminations (Tovée et al., Reference Tovée, Bowmaker and Mollon1992). The differences between LGN responses in dichromatic and trichromatic marmosets are detailed elsewhere (Yeh et al., 1995b; Kilavik et al., Reference Kilavik, Silveira and Kremers2003; Martin et al., Reference Martin, Blessing, Buzás, Szmajda and Forte2011). Fig. 1B shows a drawing from a coronal section through marmoset LGN. The vertical line shows the path of an extracellular recording electrode, which passed through the parvocellular (P), koniocellular (K), and magnocellular (M) layers. Colored symbols mark the positions of three example receptive fields recorded on this path, short horizontal lines mark the positions of other recorded receptive fields. Peristimulus time histograms (PSTHs) and contrast-response functions of the three example cells are shown in Fig. 1C–1E. The P off-centre cell (Fig. 1C) shows a vigorous response to achromatic contrast decrement (left PSTH) but no response to S-cone contrast increments (center PSTH), indicating no functional input from S-cones to this receptive field. Responses to drifting (5 Hz) low spatial frequency achromatic grating presented at different contrasts were subjected to Fourier analysis, the resulting amplitude component at stimulus frequency (right column) shows the expected low gain and linear contrast-response function (Derrington & Lennie, Reference Derrington and Lennie1984; Kaplan & Shapley, Reference Kaplan and Shapley1986; Crook et al., Reference Crook, Lee, Tigwell and Valberg1987; Croner & Kaplan, Reference Croner and Kaplan1995; Kremers et al., Reference Kremers, Weiss and Silveira2004).

Fig. 1.

Identification and distribution of short-wave sensitive (S) cone signals in the LGN. (A) Sketches of spectral sensitivity of cone photoreceptors in macaque monkey (left panel) and two color vision phenotypes of marmoset monkey (center and right panels). Numbers on wavelength axes show peak spectral sensitivities (nm). All monkeys express S-cones with peak sensitivity near 420 nm. Macaque monkeys show routine trichromatic color vision, expressing ML sensitive pigment with maxima near 530 nm (medium) and 561 nm (long). Marmoset monkeys show polymorphic color vision with three allelic variants of ML pigments with peak sensitivity near 543, 556 (not shown), and 563 nm. All males and females homozygous for the opsin alleles are dichromats (center panel), whereas heterozygous females express two ML pigments (right panel). (B) Drawing of a coronal section through marmoset LGN showing parvocellular (P, light shading) and magnocellular (M, dark shading) layers and interspersed koniocellular (K) layers. Path of an extracellular recording electrode is shown as a vertical black line and short horizontal lines show locations of recorded cells. Symbols show position of the cells with responses illustrated in C, D, E. (C) PSTH of P-off cell response to temporal square wave modulation of a spatially uniform achromatic (left column) and S-cone isolating (center column) fields. Note tonic response to maintained achromatic contrast and negligible response to S-cone modulation. Right column shows contrast-response curve for sinusoidal modulation of the achromatic field. Points show amplitude of fundamental Fourier component (f1) at stimulus frequency. Note linear contrast-response relation. (D) Responses of a K blue-on cell in the same format. Note vigorous responses to S-cone modulation. Right column shows response to S-cone isolating modulation. (E) Responses of an M-off cell in the same format. Note phasic response to achromatic contrast. Right column shows response to achromatic modulation. Note saturating contrast-response function. Stimulus field diameter (deg)/receptive field eccentricity (deg): P, 8.0/1.5; K, 1.0/0.9; and M, 2.0/8.6. (F) Schematic drawing of marmoset LGN showing P and M layers (gray texture) and K layers (blue shading). (G) Reconstructed depth of cone-opponent cells. Each symbol represents one recorded cell. Shaded regions represent P and M layers. Note segregation of blue–yellow opponent cells to K layers and red–green opponent cells to P layers. Panels B–E modified from Cheong et al. (Reference Cheong, Tailby, Martin, Levitt and Solomon2011). Data for panel F from Martin et al. (Reference Martin, White, Goodchild, Wilder and Sefton1997, Reference Martin, Blessing, Buzás, Szmajda and Forte2011). Error bars show standard deviations.

Responses of a blue-on cell recorded in koniocellular layer K3 (between the P and M layers) are shown in Fig. 1C. This cell shows negligible response to achromatic contrast but is strongly excited by S-cone contrast increments, indicating on-type excitation from S-cones to the receptive field (Derrington et al., Reference Derrington, Krauskopf and Lennie1984; Lee et al., Reference Lee, Valberg, Tigwell and Tryti1987; Yeh et al., 1995b). Consistently, the contrast-response function for drifting S-cone isolating gratings shows linear dependence of response on S-cone contrast (Fig. 1C, right column). In addition to blue-on cells such as the one exemplified here, other receptive fields encountered in marmoset K layers included blue-off cells (Szmajda et al., Reference Szmajda, Buzás, FitzGibbon and Martin2006), “suppressed by contrast” cells (Solomon et al., Reference Solomon, Tailby, Cheong and Camp2010), orientation-tuned cells (Cheong et al., Reference Cheong, Tailby, Solomon and Martin2013), cells that otherwise resembled P cells in functional properties (likely displaced constituents of the P pathway), and cells with poor responsivity to any visual stimulus presented (White et al., Reference White, Solomon and Martin2001).

On passing from K layers to the M layers (Fig. 1D), the most commonly encountered receptive fields showed the expected transient responses to achromatic contrast (left PSTH), little or no functional input from S-cones (center PSTH), high achromatic contrast gain and saturating responses to high contrast levels (right column) (Derrington & Lennie, Reference Derrington and Lennie1984; Kaplan & Shapley, Reference Kaplan and Shapley1986; Crook et al., Reference Crook, Lee, Tigwell and Valberg1987; Croner & Kaplan, Reference Croner and Kaplan1995; Kremers et al., Reference Kremers, Weiss and Silveira2004). First, these examples show that marmoset P and M cell responses are consistent with the properties shown in Old World monkey models for human vision (i.e., predominantly macaque monkeys) and, second, that blue-on responses are present in the K layers of marmoset LGN. Population data consistent with these examples, and showing segregation of blue-on responses to K layers, are shown in Fig. 1F and 1G. These data were accumulated from over 20 marmosets. The curved layers of the LGN (Fig. 1F) have been “straightened” (Fig. 1G) and the areas of shaded zones set proportional to the volumes of the P and M layers. Reconstructed depth of cone-opponent cells are shown as leaf plots, where each symbol represents one recorded cell. Blue-on and blue-off cells were present in both dichromatic and trichromatic marmosets (White et al., Reference White, Solomon and Martin2001; Szmajda et al., Reference Szmajda, Buzás, FitzGibbon and Martin2006; Tailby et al., 2008b). Red–green opponent cells were only present in trichromatic female marmosets; these receptive fields were segregated to the P layers. Responses of red–green opponent cells are not further considered in the present review but are reviewed elsewhere (Lennie & Movshon, Reference Lennie and Movshon2005; Jacobs, Reference Jacobs2008; Lee et al., Reference Lee, Martin and Grünert2010; Martin et al., Reference Martin, Blessing, Buzás, Szmajda and Forte2011).

In sum, these data show that signals serving distinct (blue–yellow and red–green) axes of color vision in marmosets leave the LGN on distinct anatomical lines. This conclusion is supported by evidence for distinct termination of S-cone signal carrying axons in superficial layers of macaque primary visual cortex (Chatterjee & Callaway, Reference Chatterjee and Callaway2003). There is more recent evidence that S-cone signals are segregated to the K system in macaque (Roy et al., Reference Roy, Martin, Dreher, Saalmann, Hu and Vidyasagar2009). In macaque, the segregation of K layers from P layers is substantially less marked than in marmosets, and some blue-on and blue-off cells are located in “bridges” of K cells spanning the inner P layers (Schiller & Malpeli, Reference Schiller and Malpeli1978), a result consistent with earlier recordings (Roy et al., Reference Roy, Martin, Dreher, Saalmann, Hu and Vidyasagar2009). Finally, three older and one more recent study of cat LGN showed that blue-on cells with properties homologous to those in marmoset and macaque LGN are segregated to the ventral (C, C1, and C2) layers (Pearlman & Daw, Reference Pearlman and Daw1970; Cleland et al., Reference Cleland, Levick, Morstyn and Wagner1976; Wilson et al., Reference Wilson, Rowe and Stone1976; Buzás et al., Reference Buzás, Kóbor, Petykó, Telkes, Martin and Lénárd2013). Receptive fields showing spectral opponent properties like blue-on and (less frequently) blue-off cells were recorded in ground squirrel optic nerve and LGN: Michael (Reference Michael1973) found that, in ground squirrel LGN, the blue-on and blue-off cells partially segregated to the outer shell of the nucleus (near the optic tract) but only a small number of cells was studied. Buzás et al. (Reference Buzás, Kóbor, Petykó, Telkes, Martin and Lénárd2013) characterized 14 blue-on cells in cat LGN and found that these cells showed blue-on/yellow-off responses as a result of on-type excitation from S-cones and off-type excitation from medium- to long-wavelength range (ML) cones. The S-on and ML-off components showed considerable overlap, implying that blue-on cells in cats are functionally very similar to blue-on cells in macaque and marmoset monkeys. Blue-on fields were segregated to the ventral (C complex) layers in cat LGN; these layers are rich in nonstandard or “W” type receptive fields (Cleland et al., Reference Cleland, Levick, Morstyn and Wagner1976; Wilson et al., Reference Wilson, Rowe and Stone1976). Taken together, these studies suggest that anatomical segregation of signals arising in S-cones is a common feature in the LGN of primate and nonprimate mammals. Surprisingly, there are to the authors’ knowledge no reports on central physiology or distribution of S-cone signals in the thalamus of other mammals.

Spatial properties and anatomical implications

In the foregoing descriptions, blue-on and blue-off receptive fields were distinguished by their strong functional inputs from S-cones. Blue-on cells are excited by S-cone contrast increments; blue-off cells are excited by S-cone contrast decrements. In this section, we discuss the spatial properties of blue-on and blue-off receptive fields. This question bears on the anatomical substrates of S-cone signal-carrying cells in the retina, a topic also treated elsewhere in this issue.

Early studies of spatial properties used hand-held or projected stimuli; for spectral characterization, the illuminant was passed through broad-band (gelatine or polyester based) or narrow-band (interference) filters, and the stimuli were customarily delivered at high contrast on dark backgrounds. Stray light, under such conditions, can bias toward overestimating receptive field size, especially in the central visual field of monkeys where receptive fields are tiny. Despite this problem, there is consensus from early and later studies that blue-on receptive fields are larger than P cell receptive field centers at any given eccentricity. Fig. 2A and 2B compares blue-on, blue-off, and P cell field radius recovered from responses to drifting achromatic (for P cells) or S-cone selective gratings (for blue-on cells), using the venerable difference-of-Gaussians model (Rodieck & Stone, Reference Rodieck and Stone1965; Enroth-Cugell & Robson, Reference Enroth-Cugell and Robson1966; Croner & Kaplan, Reference Croner and Kaplan1995). The measurements are consistent with the demonstrated anatomical substrates, whereby blue-on cells get input from small bistratified cells and P cells get input from midget class ganglion cells. In trichromatic monkeys, many P cells (specifically, nearly all P cells near the fovea) show red–green opponent responses because the excitatory center and inhibitory surround mechanism get different medium-wave and long-wave receptor weighting. The question of cone selectivity in P cell fields is reviewed elsewhere (Conway et al., Reference Conway, Chatterjee, Field, Horwitz, Johnson, Koida and Mancuso2010; Lee et al., Reference Lee, Martin and Grünert2010).

Fig. 2.

Characteristics of blue-on and blue-off receptive fields in marmoset LGN. (A) Center radius from difference-of-Gaussians fit of responses to drifting S-cone-isolating gratings. Note radius of parvocellular (P) cells is smaller than that of blue-on and blue-off cells at a given eccentricity. (B) Mean radius of P cells, magnocellular (M) cells, and blue-on and blue-off receptive fields in central 10 deg. Note that radius of on and off cells is indistinguishable for that of P and M cells, whereas blue-off receptive fields are substantially larger than those of blue-on cells. (C, F, I) Contour maps showing the spike-triggered average stimuli for uniformly distributed, pseudo-randomly presented contrast steps on S-cone-modulating (left column), and ML cone-modulating (right column) axes. Cell class and receptive field eccentricity are shown at left of the S-cone plots. The P cell (C) shows weak inhibitory input from S-cones. Note good agreement in the spatial position but distinct response probability profiles of the cone opponent inputs for blue-on (F) and blue-off cell (I). (D, G, J) Spike-triggered average stimulus timecourse at the spatial location showing the highest ML-cone variance (D) of the highest S-cone variance (G, J). Vertical gray lines indicate peak times. Note that the ML timecourse is slower than the S timecourse for the blue-on cell (G) and blue-off cell (J). (E, H, K) Difference-of-Gaussian fits (lines) to fundamental Fourier component amplitude (points) of responses to drifting cone-selective gratings for these three cells. Note characteristic band-pass response of P cell (E) to ML cone isolating gratings (red line). Note the blue-on cell shows low-frequency roll off in the spatial tuning curves for S and ML cone-selective stimuli, and blue-off cell shows low-pass spatial tuning for both conditions. Note that the ML field for this cell is slightly smaller than the S field but on average across blue-on cells, the ML field is slightly larger than the S field. Panels A, B modified from Tailby et al. (2008b). Panels C–K modified from Martin, Solomon, Tailby, and Cheong (unpublished observations).

Although a clear picture is given by concordant anatomy and physiology in blue-on cells and P cells, the anatomical picture regarding blue-off cells is murky. Isolated examples of sparse (large-field) ganglion cells showing blue-off responses and large receptive fields were reported from in vitro studies of macaque monkey retina (Dacey & Packer, Reference Dacey and Packer2003). This result is consistent with recordings from LGN in marmoset and macaque monkeys, where blue-off cells show large receptive fields (Szmajda et al., Reference Szmajda, Buzás, FitzGibbon and Martin2006; Tailby et al., 2008a,Reference Tailby, Szmajda, Buzás, Lee and Martinb). However, the reported presence of blue-off responses in parvocellular giant/melanopsin (PGM) cells makes this picture more complex (Dacey et al., Reference Dacey, Liao, Peterson, Robinson, Smith, Pokorny, Yau and Gamlin2005). The accepted role of PGM cells in primates and other mammals is their contribution to “non-image-forming” visual functions such as circadian control (via projections to suprachiasmatic and pregeniculate nucleus) and pupillary control (via projections to the olivary pretectal nucleus; reviewed by Sand et al., Reference Sand, Schmidt and Kofuji2012). How off-type S-cone inputs to PGM cells (which means PGM activity would be reduced by short-wave light) relate to their non–image-forming role is not clear. On the other hand, the evidence that PGM cells do project to LGN (Rodieck & Watanabe, Reference Rodieck and Watanabe1993; Dacey et al., Reference Dacey, Peterson, Robinson and Gamlin2003, Reference Dacey, Liao, Peterson, Robinson, Smith, Pokorny, Yau and Gamlin2005; Szmajda et al., Reference Szmajda, Martin and Grünert2008) suggests that PGM-cell carried signals could reach the cortex and contribute to pathways serving conscious vision. This enigmatic dual role for PGM cells needs further clarification. Of course, any role for PGM cells in conscious vision must be limited by the very low density of PGM cells relative to P and M cells, especially in the fovea.

The question of anatomical basis of blue-off signals in primates is even further complicated by an electron microscopic analysis that reported off-type S-cone connections to five midget bipolar cells in the fovea of one macaque monkey (Klug et al., Reference Klug, Herr, Ngo, Sterling and Schein2003). This result predicts that there should be tiny (P-size) blue-off receptive fields in the fovea. Yet, such receptive fields have never been reported from over three decades of retinal or LGN recordings. For example, nearly all the blue-off fields within 5 deg of the marmoset fovea are substantially larger than the blue-on fields at the same eccentricity (Fig. 2A) (Szmajda et al., Reference Szmajda, Buzás, FitzGibbon and Martin2006; Tailby et al., 2008b). The subtle but important point here is that chromatic aberration cannot fully explain the large field size of purported “blue-off-midget” cells, for blue-on cell fields would likewise be enlarged by out-of-focus short-wave light. This outstanding puzzle is also taken up elsewhere in this issue (Dacey et al., Reference Dacey, Crook and Packer2013; Miyagishima et al., Reference Miyagishima, Grünert and Li2013).

Opponent inputs to blue-on cells

When analyzed using drifting gratings, the small bistratified cells recorded in retina show well-matched and overlapping blue-on and yellow-off receptive field components (Crook et al., Reference Crook, Davenport, Peterson, Packer, Detwiler and Dacey2009; Dacey et al., Reference Dacey, Crook and Packer2013). Spatiotemporal analysis of blue-on receptive fields using reverse correlation analyses in in vitro macaque retina, however, showed “hot spots” presumably corresponding to the location of individual S-cones feeding the receptive field (Chichilnisky & Baylor, Reference Chichilnisky and Baylor1999). Furthermore, this method as well as in vivo recordings from macaque retina and marmoset LGN showed that the yellow-off receptive field region is consistently slightly larger (by about 30% of the area) than the blue-on region (Solomon et al., Reference Solomon, Lee, White, Rüttiger and Martin2005; Field et al., Reference Field, Sher, Gauthier, Greschner, Shlens, Litke and Chichilnisky2007; Tailby et al., 2008b). Such deviation from the ideal “Type II” (Wiesel & Hubel, Reference Wiesel and Hubel1966) color-coding organization can have consequences for response of blue-on cells to achromatic stimuli. If the opponent inputs are balanced in functional weight and size, an ideal Type II receptive field will show zero response to achromatic contrast. But blue-on cells stimulated with drifting achromatic gratings can show complex properties, including orientation and direction selectivity as a result of different size, shape and response timing of the blue-on and yellow-off receptive field subunits (Forte et al., Reference Forte, Hashemi-Nezhad, Dobbie, Dreher and Martin2005; Tailby et al., 2008a). These effects are explained in detail elsewhere (Tailby et al., Reference Tailby, Dobbie, Hashemi-Nezhad, Forte and Martin2010), and for now, the important point is that an experimenter restricted to using achromatic stimuli might find the responses of blue-on cells indecipherable.

Advances in stimulus technology and data processing allow the spatiotemporal structure of receptive fields to be studied in detail, using reverse correlation analysis (Chichilnisky & Baylor, Reference Chichilnisky and Baylor1999; Field et al., Reference Field, Sher, Gauthier, Greschner, Shlens, Litke and Chichilnisky2007, Reference Field, Gauthier, Sher, Greschner, Machado, Jepson, Shlens, Gunning, Mathieson, Dabrowski, Paninski, Litke and Chichilnisky2010; Solomon et al., Reference Solomon, Tailby, Cheong and Camp2010). For example, Fig. 2C–2K shows that the contour maps for a nonopponent P on-centre cell (Fig. 2C), a blue-on cell (Fig. 2F), and a blue-off cell (Fig. 2I) show the spike-triggered average stimuli for uniformly distributed, pseudo-randomly modulated contrast on S-cone-modulating (left column), and ML cone-modulating or “silent S-cone” (right column) axes. As expected, the P cell (Fig. 2C) shows a small ML excitatory region and weak off-sign S-cone input. For the blue-on cell (Fig. 2F) and the blue-off cell (Fig. 2I), there is fair agreement in the spatial position of the cone opponent inputs, but the spatial response profiles are distinct. The spike-triggered average stimulus timecourse for S and ML isolating conditions (Fig. 2D–2J) shows the expected biphasic timecourse for ML excitation in the P cell (Fig. 2D). For the blue-on cell (Fig. 2G) and the blue-off cell (Fig. 2J), the ML timecourse is slower than the S timecourse (for the blue-on cell, the response peaks are separated by over 15 ms). Thus, there are spatial and temporal asymmetries in the opponent inputs to the blue-on and blue-off cells. That the ML input is often weaker than the S input to blue-on and blue-off cells is also shown in the responses to drifting cone-selective gratings (Fig. 2H and 2K). The S and ML inputs to blue-on cells in macaque retina and LGN measured in vivo show closely matched temporal properties (Yeh et al., 1995a; Solomon et al., Reference Solomon, Lee, White, Rüttiger and Martin2005; Tailby et al., 2008a), but ML delays of 6–20 ms relative to S inputs are reported from in vitro recordings (Chichilnisky & Baylor, Reference Chichilnisky and Baylor1999; Crook et al., Reference Crook, Davenport, Peterson, Packer, Detwiler and Dacey2009). Furthermore, unlike the situation in macaque retina (Crook et al., Reference Crook, Davenport, Peterson, Packer, Detwiler and Dacey2009; Dacey et al., Reference Dacey, Crook and Packer2013), the blue-on receptive fields in LGN commonly show spatial antagonism in the S-cone inputs, evidenced by a low-frequency roll-off in the spatial tuning curve for S-cone-selective stimuli (Fig. 2H). The centre-surround antagonism in S-cone inputs to blue-on and blue-off cells is however weaker than that exhibited by P cells (Fig 2E; Tailby et al., 2008a,Reference Tailby, Szmajda, Buzás, Lee and Martinb).

What is the source of yellow-off opponent inputs to the blue-on receptive fields? As detailed elsewhere in this issue (Dacey et al., Reference Dacey, Crook and Packer2013; Miyagishima et al., Reference Miyagishima, Grünert and Li2013)—in macaque retina, connections from diffuse off-type bipolar cells to the outer tier of small bistratified cells will show yellow-off responses. Inputs to off-bipolar cells are naturally biased against S-cones for the simple reason that S-cones make up less than 10% of the cone population (Marc & Sperling, Reference Marc and Sperling1977; Wikler & Rakic, Reference Wikler and Rakic1990; Martin & Grünert, Reference Martin and Grünert1999; Roorda & Williams, Reference Roorda and Williams1999). Furthermore, the connections of diffuse bipolar cells with S-cones are weaker than their connections with ML cones (Lee & Grünert, Reference Lee and Grünert2007). Contribution of outer retinal (horizontal cell) and inner retina (amacrine cell) inhibition to the blue-on receptive fields in primates is treated in detail elsewhere (Dacey et al., Reference Dacey, Crook and Packer2013; Miyagishima et al., Reference Miyagishima, Grünert and Li2013).

Selectivity and isolation of S-cone signals

So far we have discussed the responses of cells receiving strong inputs from S-cones. Blue-on and blue-off cells are straightforward to identify in a laboratory setting: blue on cells are vigorously excited and blue-off cells are profoundly inhibited, by almost any stimulus which is dominated by short-wave light. A different and more subtle question is cross-talk or “leakage” of S-cone signals to P and M cells (Lee, Reference Lee2008). Experiments designed to answer this question must take into account the details of spectral absorption in optic media as well as photoreceptor sensitivity and the contrast sensitivity of P and M cells. We explain the process here in some detail, so that newcomers to the field can see how to avoid pitfalls and interpret apparently conflicting results.

Fig. 3A sketches the stimulus-eye system. Before absorption by the receptors, (1) the light passes through (and is partially absorbed by) the lens and other optic media. (2) Light from stimuli close to the visual axis is imaged at the fundus (macula lutea, 3), which contains a variable amount of carotenin-based visual pigments (Wyszecki & Stiles, Reference Wyszecki and Stiles1982; Snodderly et al., 1984a,Reference Snodderly, Brown, Delori and Auranb). The optical density of macular pigment falls sharply with visual field eccentricity (Fig. 3B). The absorption of human S, M, and L cones is sketched in Fig. 3C–panel 1 (Smith & Pokorny, Reference Smith and Pokorny1975; Stockman et al., Reference Stockman, MacLeod and Johnson1993; Lamb, Reference Lamb1995). The lens and other optic media (shading, Fig. 3C–panel 2) effectively cut off wavelengths below 400 nm and reduce differentially the sensitivity of S-cones. The macular pigment (shading, Fig. 3C–panel 3) provides absorption in the 400–500 nm range. The effective photoreceptor absorbance (the “corneal sensitivity”), therefore, has a more complex shape than that dictated by the receptor sensitivity alone. Furthermore, every eye is unique: the absorbance characteristics of the lens change with age, the density and distribution of macular pigment show great individual variation, the peak sensitivities of cones can vary due to cone opsin polymorphism, and so on. Next, we consider the effect of this unavoidable variation on measurements designed to detect S-cone signals.

Fig. 3.

Predicting and measuring weak S-cone inputs. (A) Sketch of the stimulus–eye system showing sites of wavelength-dependent absorption: (1) cone opsins; (2) lens and other optic media; (3) macular pigment and nonopsin retina structures. Cone opsin sensitivity modified by these passive filters is nicknamed “corneal sensitivity”. (B) Example of eccentricity dependence of macular pigment density in a human subject (modified from Hammond et al., Reference Hammond, Wooten and Snodderly1997). (C) Modification of spectral cone absorbance. Panel 1 shows receptor sensitivities. Panel 2 shows reduction of short-wave receptor sensitivity by lens and other optic media, which absorb light wavelengths below 450 nm (purple shaded region). Panel 3 shows reduction of sensitivity by macular pigments between 400 and 500 nm (cyan shaded region). (D) Principle of silent substitution. Two lights (λ ₁, λ ₂) to which medium-wave sensitive (M) cones are equally sensitive (left panel) are exchanged in sinusoid counterphase (right panel). The predicted modulation of M cone is zero, whereas the modulation of short-wave sensitive (S) cones will be substantial because S-cones are much more sensitive to λ ₁ than to λ ₂. (E) Hazard of silent substitution. If the M cone sensitivity to λ ₁ is greater than predicted, the exchange will no longer be silent for the M cone. (F, G) Example failure of silent substitution. Contrast-response curves showing saturating hyperbolic “Naka–Rushton” fits (lines) and fundamental Fourier component amplitude (points) of one magnocellular cell’s responses to nominally cone-selective gratings. Note high gain and saturating responses to ML cone modulation. Cone-selective gratings in (F) were produced using appropriate corneal sensitivities for the receptive field, which lay outside fovea. Gratings in (G) were produced using foveal corneal sensitivities. Note spurious response to S-cone-selective gratings in (G). Responses to maximum S-cone contrast (∼35 imp/s) are equivalent to ∼5% modulation of M cone.

Fig. 3D sketches a simplified “silent substitution” experiment. Here, our goal is to modulate S-cones but not M cones. We choose two wavelengths to which M cones are equally sensitive (λ ₁, λ ₂; Fig. 3D), make their intensities the same, then exchange them, for example, as out-of-phase sinusoids (Fig. 3D). Under this stimulus, the M cone excitation is not changed (the M cone is “silent”), but S-cone excitation will change appreciably because S-cones are much more sensitive to λ ₁ than to λ ₂. Modulation in a downstream component of the visual pathway should now indicate functional input from S-cones. (The principle shown here for a two-receptor system can be elaborated to three receptors by mixing three lights.) But, now consider the effect of deviation of M cone sensitivity from those predicted by the cone fundamentals (Fig. 3E). If the M cone is more sensitive to λ ₁ than to λ ₂, the exchange will no longer be silent for the M cone, and the activity in a downstream component of the visual pathway may be wrongly attributed to S-cone inputs.

The problems of silent substitution are most pervasive in recordings of high-gain cells such as magnocellular cells, which act as very sensitive amplifiers of cone contrast. A cautionary example is shown in Fig. 3F and 3G. These contrast-response curves were recorded from a magnocellular cell with a receptive field well outside the fovea, at 14 deg, eccentricity. The high-gain saturating response to ML cone modulation is similar to that shown in Fig. 1E. For properly constructed S-cone isolating gratings, this cell shows no response (Fig. 3F). But in a recording made from the same cell a few minutes later (Fig. 3G), we record a substantial response to S-cone isolating gratings. The difference here is that for Fig. 3G, the S-cone isolating grating was constructed using “wrong” receptor sensitivities (incorporating macular pigment absorption). The resulting deviation from ideal ML cone substitution conditions has caused a small degree of ML cone modulation, which lies at the root of the spurious response. Reported S-cone inputs to magnocellular cells in macaque LGN may be affected by this problem (Chatterjee & Callaway, Reference Chatterjee and Callaway2002), although specific S-cone signal injection at LGN level in macaque cannot be ruled out. Parasol cells in macaque retina show no sign of S-cone input when measured with a highly sensitive method, using fundamentals appropriate for the receptive field eccentricity of the cell (Sun et al., 2006a,Reference Sun, Smithson, Zaidi and Leeb). Likewise, the magnocellular cells in marmoset LGN show negligible S-cone input, that is, below the weight expected from random cone connections (Tailby et al., 2008b).

Notwithstanding the technical difficulties outlined above, there is consistent physiological evidence for weak S-cone inputs to the midget-P pathway. In macaque monkeys, functional input from S-cones to midget-P cells is undetectable in recordings from central retina (Sun et al., 2006a,Reference Sun, Smithson, Zaidi and Leeb). However, Field et al., (Reference Field, Gauthier, Sher, Greschner, Machado, Jepson, Shlens, Gunning, Mathieson, Dabrowski, Paninski, Litke and Chichilnisky2010) identified S-cone inputs to ganglion cells using in vitro recordings of peripheral macaque retina. They showed that there is reliably detectable (i.e., nonzero) excitatory S-cone input to off-midget cells (Field et al., Reference Field, Gauthier, Sher, Greschner, Machado, Jepson, Shlens, Gunning, Mathieson, Dabrowski, Paninski, Litke and Chichilnisky2010). On average, the functional weight of these S-cone inputs to off-midget cells was, however, even lower than the already low proportion (5–10%) expected on the basis of random cone connections. Furthermore, about 40% of S-cones make no contact at all with midget bipolar cells (Field et al., Reference Field, Gauthier, Sher, Greschner, Machado, Jepson, Shlens, Gunning, Mathieson, Dabrowski, Paninski, Litke and Chichilnisky2010); a result which is consistent with anatomical evidence for sparse connections of S-cones to midget bipolar cells in marmoset retina (Lee et al., Reference Lee, Telkes and Grünert2005). Accordingly, recordings from macaque and marmoset LGN show weak and variable inhibitory S-cone inputs to P cell receptive fields (Tailby et al., 2008b). In summary, both anatomical and physiological evidence show that there is functional segregation of S-cone signals toward blue-on and blue-off cells and away from P and M cells.

The pioneering advance made by Field et al. (Reference Field, Gauthier, Sher, Greschner, Machado, Jepson, Shlens, Gunning, Mathieson, Dabrowski, Paninski, Litke and Chichilnisky2010) was to individually stimulate and identify different cones with high spatial resolution stimuli in vitro. A conceptually related but technologically distinct approach is to bypass the spatial limitations imposed by natural optics in vivo. McMahon et al. (Reference McMahon, Lankheet, Lennie and Williams2000) used laser interferometry to study the fine structure of P cell receptive fields in the fovea. Roorda and Williams (Reference Roorda and Williams1999) and Sincich et al. (Reference Sincich, Zhang, Tiruveedhula, Horton and Roorda2009) used adaptive optics ophthalmoscopy to show that P cells can respond to stimulation of a single cone photoreceptor. These methods have not yet been combined with spectral analysis to identify cone classes but they have the potential to map chromatic inputs to receptive fields in vivo to the level of precision now achievable in vitro. A drawback is that both methods require advanced technology currently beyond the reach of most visual neuroscience laboratories.

Distribution of S-cone signals in subcortical and cortical areas

How widespread are S-cone signals among subcortical and cortical targets? The previous sections have shown that within the LGN of marmoset and macaque monkeys, two broad functional groups of on- and off-centre cells with strong S-cone input can been identified, and these cells likely form part of the primitive koniocellular division of the LGN (Hendry & Reid, Reference Hendry and Reid2000; Szmajda et al., Reference Szmajda, Buzás, FitzGibbon and Martin2006; Szmajda et al., Reference Szmajda, Martin and Grünert2008; Tailby et al., 2008a; Roy et al., Reference Roy, Martin, Dreher, Saalmann, Hu and Vidyasagar2009). Consistently, the likely homologous C complex layers of cat LGN are the target of blue-on cells (Cleland et al., Reference Cleland, Levick, Morstyn and Wagner1976; Buzás et al., Reference Buzás, Kóbor, Petykó, Telkes, Martin and Lénárd2013). However, in another primitive visual center, the superior colliculus (Fig. 4C), we found no evidence for cells with strong S-cone input. This result is in contrast to evidence for long-latency S-cone inputs to collicular cells recorded in awake macaques. Likewise, recordings from primary visual cortex in (anaesthetized) marmoset and (waking or anaesthetized) macaque consistently show a broader distribution of S-cone inputs to V1 cells than for P and M cells, yet no evidence for a distinct group of cells with S-cone input weight like blue-on and blue-off cells in LGN (DeValois et al., Reference DeValois, Cottaris, Elfar, Mahon and Wilson2000; Solomon et al., Reference Solomon, Lee, White, Rüttiger and Martin2005; Hashemi-Nezhad et al., Reference Hashemi-Nezhad, Blessing, Dreher and Martin2008). The question of cortical distribution of S-cone signals is taken up again in other chapters of this special issue (Xiao, Reference Xiao2013; also refer the Conway article in this issue). Here, we want to reinforce the fact that the good understanding of the distribution of S-cone signals among retinal and geniculate cells stands in contrast to the limited data available from other subcortical targets in primates.

Fig. 4.

Comparison of S-cone weights. Each histogram shows the functional weight of S-cone input as a proportion of all cone inputs to the receptive field. (A) Blue-on and blue-off fields in LGN. (B) Cells in primary visual cortex (V1). (C) Cells in superior colliculus (SC). (D) Parvocellular (P) cells. (E) Magnocellular (M) cells. Note lack of strong S-cone input to cells other than blue-on and blue-off LGN cells.