The angular separation between the temporal edge of the optic disk and the foveal center is 15° in rhesus monkeys (Macaca mulatta), as measured by Weiskrantz and Cowey (1967) and de Monasterio et al. (1985). The corresponding distance in our sample eye was 3.18 mm. Using these data under the assumption that M. fuscata and M. mulatta have similar gross retinal structures, the conversion of retinal distance to visual angle is 212 μm/°. The examination area was located 3.00–3.25 mm temporal to the foveal center, and the center of this area was ~15° away from the foveal center.

The top-view distribution of 3159 rod spherules and 237 cone pedicles (Figure 1) was reconstructed from electron micrograph prints acquired at 4000× (10× enlargement of 400× negatives). This survey area, which formed a rough parallelogram of 73.5 × 224 μm with an irregular contour located 3.00–3.25 mm temporal to the foveal center, was measured to be 0.01684 mm² (Image-J; NIH, USA). Half of the cells that extended across the edge of the parallelogram were subtracted from the total number for density measurements. Thus, the corrected total numbers of rods and cones in this region were 2889 and 212, respectively, and the corresponding densities were 172 × 10³ spherules/mm² and 12.6 × 10³ pedicles/mm², respectively. The density ratio of rods to cones was 13.6.

The retinal neural circuitry is thought to be constrained by the density ratios of constituent neurons which in turn depend on the retinal eccentricity from center to periphery. Wässle et al. (1989, 1990) showed that ganglion cells and cone pedicles had almost the same density at 3 mm (~15°) eccentricity in the temporal retina of M. fascicularis. This locus is equivalent to our present sample obtained at the same eccentricity. Boycott and Hopkins (1991) showed that some of the invaginating (ON) and flat (OFF) midget bipolar cells had synapses with two adjacent cones throughout the retina from 2.5 mm to at least 10.0 mm in macaque and vervet monkeys (M. mulatta and Cercopithecus aethiops). We occasionally observed such midget bipolar cells in our series and also found that a midget ganglion cell collects input synapses from 2 to 4 midget bipolar cells in either ON or OFF pathway (data not shown). These preliminary surveys show that the examination area is not foveal but peripheral in nature.

We used a series of the electron micrograph prints acquired at 12000× (4× enlargement of 3000× negatives) for 3D reconstruction. This format provided enough open space in the sectional images to label unique letters for identification. To identify the continuity of a target process from one section to the next, we used the overlap area between adjacent cell profiles of this particular process and also all surrounding processes that were uniquely identified; we regarded the entire spatial relationship of the process with its neighbors as a specific property at each locus in every section. This method compensated for a disadvantage of our use of serial section transmission electron microscopy: successive sections could not be as completely physically registered as in serial block-face scanning electron microscopy (Helmstaedter et al., 2013). Human observers easily identified the correspondence between two characteristic patterns composed of many process profiles; some processes changed, but others remained the same. Observers also employed additional cytological clues to identify correspondence.

For automated 3D reconstruction (Mishchenko, 2009), the overlap area between adjacent cell profiles of the target process is thought to be critically important. The overlap area is largest when the process runs at a right angle to the section plane, but it becomes smaller and finally disappears as the encompassing angle approaches 0°, and thus this clue is completely absent in some longitudinally sectioned processes. To avoid this difficulty as much as possible, thinner sections (such as 20–30 nm) are required in Mishchenko's automated reconstruction approach.

However, when we expanded the observation area around the target process, because the sectional area of the target process became wider, clues about the relationship of the target process with its neighbors became rich enough to compensate for the loss in overlap area. In addition, human experts were able to evaluate the multitude of gray intensities caused by densely stained membranes in adequately thick sections. Many clues, such as the beginning or end of the process and the heading of the process, were contained in gradual changes in the finely graded images in specific places and directions. We traced every neuronal process while marking synapses and other features with colored pens on transparent sheets, drew the cell contours, and saved digitized contour lines on a personal computer using the Intuos-4 digitizer (Wacom, Saitama, Japan) and TRI/3D-SRF-R graphics software (Ratoc Systems International, Tokyo, Japan). Even after review of the graphics, precise areas of interest could be re-photographed whenever necessary by imaging the preserved sections with an electron microscope at different magnifications and tilting angles. Although the above procedure required a long time to obtain sufficient numbers of reconstructed neurons, it enabled us to determine all synaptic contacts of a given neuron as precisely as possible. Here, we present 3–5 reconstructed cells for each cell type. Furthermore, we compare our electron microscopy-based analysis with previous light microscopy-based studies of large numbers of cells at different retinal eccentricities (e.g., Wässle et al., 2009) to complement each other.

We reconstructed a total of 20 OFF bipolar cells from the macaque retina and 17 from the mouse retina, which included 3–5 cells of each type. A serial number was assigned to each cell that included the cell type (for example, cell DB3b-2 is the second diffuse bipolar type 3b cell included in the analysis). We acquired a dataset of four morphological variables from cells in the inner plexiform layer: (1) the level of stratification of the bipolar cell axon terminal, measured as the distance from the most distal axon terminal to the GCL, (2) the stratification thickness of the axon arbor, measured as the distance from the branching point of the axon cylinder to the axon terminal, (3) the axon arbor area projected on the horizontal plane, and (4) the number of axon synaptic ribbons. The mean and standard deviation for every variable obtained from each OFF bipolar type group are listed in Table 1. The series for mouse T1-3 and T3a-3 cells lacked a part of the axon arboreal processes (they extended in the horizontal direction beyond the series range), and thus these cells are omitted from the data in columns 3 and 4 of Table 1. The values for variables (1) and (2) from these cells are included in Table 1 because we consider them to be accurate.

Cluster analyses (Ward's joining method) using Statistica 06J (Statsoft Japan, Tokyo, Japan) were applied to 18 macaque cells and 15 mouse cells. Macaque cell DB1-2 extended a single tiny process from the axon terminal toward the GCL, and cell FMB-2 had an axon terminal with no descending projections due to spatial hindrance by the surrounding cells. Data from these two cells severely skewed the distributions of variables (1) and (2), and so these data were omitted from the cluster analysis for macaque. Likewise, cells T1-3 and T3a-3 were omitted from the cluster analysis for mouse. We also used Statistica 06J for the Pearson's correlation test where differences for which P < 0.05 were considered significant.

The DB3b type of OFF bipolar cells formed connections with both rods and cones in the macaque retina. For example, cell DB3b-1 made contact with eight rods and 11 cones (Figure 7A). Each of DB3b cells 1-4 made contact with 4.0 ± 2.7 rods and 9.0 ± 1.4 cones. Cell DB3b-1 made contact with each rod via a single basal contact, but with each cone with 1–19 basal contacts (Figure 7B). Each DB3b cell made contact with a single cone via 9.4 ± 6.3 basal contacts (n = 36). It was evident that not all rods formed this type of connection with DB3b cells. To assess the proportion of rods that made contact with DB3b cells, we examined an area where the dendrites of four DB3b cells covered the area occupied by contiguously distributed rods (Figure 7C). In this area, every rod was in close proximity to at least one of the four DB3b cells. However, only 16 rods made contact with one of the four DB3b cells, whereas the remaining 224 rods did not. Thus, 6.7% of rods formed a basal contact with an OFF bipolar cell in this locus of the macaque retina.

We also examined the photoreceptor connectivity of the OFF bipolar cell dendrites in the mouse retina and found that T3a, T3b, and T4 cells formed basal contacts with both rods and cones. Every cell of the three OFF bipolar cell types that were examined (3–4 cells per type) formed such contacts with variable numbers of rods and cones. The most representative case for each OFF bipolar cell type is displayed in Figures 8A–C; however, the dendrites of 4/11 examined cells were only partially reconstructed within the series. Therefore, we generated no mean values of the numbers of connected rods and cones for each OFF bipolar cell type. Instead, we focused our analysis on a certain area in which the dendrites of T3a, T3b, and T4 cells intermingled and covered all contiguously distributed rods (Figure 8D). Inside the dotted contour of Figure 8D, four cone pedicles (P9, P10, P11, and P14) were in contact with these three types of OFF bipolar cells, indicating that rod spherules in the same area were also in close proximity to these three types of OFF bipolar cells. Thirty rods contacted one of the OFF bipolar cells, and five rods contacted two OFF bipolar cells of different types; the remaining 157 rods had no contact with any OFF bipolar cell. Thus, 15.6% of rods formed a basal contact with an OFF bipolar cell, and 2.6% of rods made contacts with two OFF bipolar cells of different types. In total, 18.2% of rods had direct connections to OFF bipolar cells.

In our recent preliminary report (Tsukamoto and Omi, 2013), which was conducted independently of Puthussery et al. (2013), we identified a new type of DB OFF bipolar cell in addition to the three previously described DB cell types in macaque retina, and demonstrated that this new type of OFF bipolar cell makes basal contact with both rods and cones. The shape of the axon terminals of this new type was similar to that of DB2. We previously called this new cell type DB2b, but here we have adopted DB3b as its name throughout this article. When Boycott and Wässle (1991) morphologically classified macaque bipolar cells, they reported cell densities of 2935/mm² for DB2 cells and 968/mm² for DB3 cells. Thus, DB2 cells were three-fold more numerous than DB3 cells, probably due to a difference in cell size and/or a mixture of multiple types. It was highly likely that the newly discovered type of DB cells had been combined with DB2 cells as a single type in Boycott and Wässle's classification. Puthussery et al. (2013) named the newly discovered type DB3b, probably because the axon terminals of these cells were located almost at the same level as those of DB3 cells. Our observation that macaque DB3a and DB3b cells are homologous to mouse T3a and T3b cells, respectively, supports the use of this nomenclature; the axon arbor areas of “3a” cells are approximately double those of “3b” cells in both species.

Previous morphological and immunohistochemical analyses, the recent discovery of the DB3b type, and our current comparative morphometric study indicate that the mouse and macaque monkey retinas possess five homologous and distinct OFF bipolar cell types. In their respective species, cell types DB1 and T1 had the highest axon stratification level and the second-largest axon arbor area, whereas types DB2 and T4 had the lowest stratification level and the second-smallest axon arbor area (Figure 2; Table 1). If we tentatively eliminate types FMB and T2 from the classification samples, a sequence of DB1, DB3a, DB3b, and DB2 for macaque and a sequence of T1, T3b, T3b, and T4 for mouse reflect axon-to-GCL distance in descending order. These orderly distances may indicate the homologous correspondence between types DB1 and T1, DB3a and T3a, DB3b and T3b, and DB2 and T4. The remaining cell types in each species, FMB in macaque and T2 in mouse, therefore may correspond to each other.

FMB cells had the smallest axon arbor area in macaque, whereas T2 cells had the third-largest axon arbor area, which was more than twice as large as the smallest axon arbor area of T3b cells in mouse. However, Ghosh et al. (2004) suggested that macaque FMB cells and mouse T2 cells are homologous, because both displayed recoverin immunoreactivity (Milam et al., 1993; Euler and Wässle, 1995; Haverkamp et al., 2003a,b). In this case, morphological classification could be complemented by immunohistochemical approaches. It is conceivable that FMB cells drastically change their morphology from the common ancestral type shared by T2 cells in order to adapt to high-acuity vision under bright daylight conditions, with no critical changes in immunochemical properties. In this context, although our side-view profile of a T2 cell (Figure 2B) may not exactly resemble that displayed by Ghosh et al. (2004), possibly due to inherent morphological variability, it is very similar to that depicted by Ivanova et al. (2006). The axon arbor profiles of T2 cells and FMB cells were similar in appearance from the top view (Figure 4); at the center of the axon arbor area of both species, a round wide area (denoted by arrows) was seen. Similarly, a skirt-like stout portion was evident at the branching point of the axon arbor in both cells in the side view (Figure 2). This particular feature was not identified in other OFF bipolar cell types, which instead exhibited a few slender processes that diverged from the branching point of the axon. All these findings are consistent with the interpretation that the mouse T2 cell is homologous to the macaque FMB cell.

Based on the present results, we now add macaque monkeys to the list of mammals, such as mice, rats, cats, and rabbits, that harbor the third rod signaling pathway. However, we note that the pathway that connected rods to OFF bipolar cells appeared to be less distributed over the macaque retina. In mouse, three types of bipolar cells (T3a, T3b, and T4) were involved in this pathway, whereas in macaque, only one type of bipolar cell (DB3b) made contact with rods. In accordance with this difference, ~18% of mouse rods formed contacts with OFF bipolar cells, compared with only 7% of macaque rods.

In the mouse retina, every T7 ON bipolar cell that we previously examined (Tsukamoto et al., 2007) made direct synaptic contacts with at least two rods, and on average 1–2% of the rods had direct contacts with T7 ON bipolar cells. In contrast, in the present investigation we identified no contacts between the dendrites of one ON bipolar cell of each type and rods in the macaque retina (data not shown). Therefore, it is likely that the proportion of rods that directly contact ON bipolar cells is exceptional, if such direct contacts exist. These two species-specific differences suggest that, compared to the macaque retina, the mouse retina is better adapted to nocturnal conditions through its implementation of more channels for rod signals.

Tsukamoto et al. (2001) showed that the rods in the mouse retina that contacted OFF bipolar cells were coupled with several neighboring rods by gap junctions, possibly to integrate rod signals. In the macaque retina, Hornstein et al. (2005) reported rod-rod tracer couplings with Neurobiotin, as well as rod-rod electrical couplings; those gap junctions were located in regions containing rod somas and axons. We observed that gap junctions between rod spherules were rare in the OPL of the macaque retina, confirming the findings of Raviola and Gilula (1973). Our further investigation into both the OPL and the ONL of the mouse retina revealed that rod-rod gap junctions were rare in the OPL but occurred frequently between adjacent somas, somas and axons, and adjacent axons, which were distributed from the outer limiting membrane to the ONL-OPL border, confirming and extending the previous findings (Tsukamoto et al., 2001). All the above findings strongly suggest that rod-rod gap junctions are localized in the ONL in both macaque and mouse retinas. Rod-rod gap junctions participate in all three rod signaling pathways as a kind of preprocessing of the output from rod (chemical and electrical) synaptic terminals. Where the third rod signaling pathway is concerned, a small portion (~18% for mouse, ~7% for macaque) of rods may electrically couple with other surrounding rods before directly contacting OFF bipolar cells. Despite the low percentage of rods that directly contact OFF bipolar cells, transmitted signal may be effectively driven by a much larger portion of the rods under certain mesopic conditions. Therefore, we cannot rule out the possibility that the third rod signaling pathway is a necessary device in an overall, intricate architecture that remains to be further elucidated.

Three known rod pathways are thought to differ in temporal signaling characteristics as well as in light sensitivity. Response speed is constrained by the number of synaptic transmission steps between the photoreceptor and bipolar cells. Ignoring rod-rod coupling, the first pathway consists of three steps: two chemical synapses from rod to rod and bipolar cells to AII amacrine cells, and a third chemical synapse from AII cells to OFF cone bipolar cells (or one parallel electrical synapse from AII cells to ON cone bipolar cells). The second pathway consists of two steps: one electrical synapse from rod to cone and one chemical synapse from cone to OFF (or in parallel ON) cone bipolar cells. The third pathway consists of only one step, a chemical synapse from rod to OFF cone bipolar cells. Soucy et al. (1998) first suggested the existence of direct synaptic contacts between rods and OFF cone bipolar cells in order to explain the fast response of ganglion cells to OFF stimuli in the coneless mouse. In human psychophysical and electroretinographic studies, a loss of flicker visibility and electroretinographic amplitude at frequencies near 15 Hz were shown to result from cancelation between sensitive “slow” and insensitive “fast” rod signals, with a time difference of ~33 ms (Stockman et al., 1995; Sharpe and Stockman, 1999). The sensitive “slow” rod signals could be unequivocally ascribed to the first pathway. The insensitive “fast” rod signals were thought to occur through the second pathway (via rod-cone gap junctions), according to the knowledge at that time. However, there remained a puzzling phenomenon: this “fast” signal was present even in an achromatic observer that lacked functioning cones (Stockman et al., 1991). It seems unlikely that a full range of the “first” signal can be solely ascribed to the low percentage of rods contacting DB3b cells in the third pathway. Nevertheless, these psychophysical and electroretinographic phenomena should be reconsidered in a more comprehensive framework that includes all three pathways.

Electrophysiological experiments have demonstrated that the third pathway is functional, with sensitivity and dynamics that are distinct from the other two pathways in the mouse retina (Field and Rieke, 2002; Pang et al., 2010). This pathway is thought to exhibit a light sensitivity that is intermediate between those of the first and the second pathways. The first and most sensitive pathway may work alone at the low scotopic level, whereas the second and third pathways may be activated in sequence as the light intensity increases. However, these multiple pathways appear to operate in parallel, according to a recent study by Ke et al. (2014); in the mouse retina, under a high mesopic level in which the second pathway (via rod-cone gap junctions) is fully activated, the role of the first (rod bipolar cell to AII amacrine cell) changed from encoding the absorption of single photons to encoding contrast modulations. Under the incessantly changing environmental light conditions at the scotopic and mesopic levels, the third pathway may also work simultaneously with the first and second pathways, but with different dynamics. This circuitry of multiple channels with different temporal characteristics for rod signals exemplifies the functional significance of parallel processing.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.