Double cones in the avian retina form an oriented mosaic which might facilitate magnetoreception and/or polarized light sensing

Abstract

To navigate between breeding and wintering grounds, night-migratory songbirds are aided by a light-dependent magnetic compass sense and maybe also by polarized light vision. Although the underlying mechanisms for magnetoreception and polarized light sensing remain unclear, double cone photoreceptors in the avian retina have been suggested to represent the primary sensory cells. To use these senses, birds must be able to separate the directional information from the Earth's magnetic field and/or light polarization from variations in light intensity. Theoretical considerations suggest that this could be best achieved if neighbouring double cones were oriented in an ordered pattern. Therefore, we investigate the orientation patterns of double cones in European robins (Erithacus rubecula) and domestic chickens (Gallus gallus domesticus). We used whole-mounted retinas labelled with double cone markers to quantify the orientations of individual double cones in relation to their nearest neighbours. In both species, our data show that the double cone array is highly ordered: the angles between neighbouring double cones were more likely to be 90°/−90° in the central retina and 180°/0° in the peripheral retina, respectively. The observed regularity in double cone orientation could aid the cells' putative function in light-dependent magnetoreception and/or polarized light sensing.

1. Introduction

Birds are highly visual animals and it is therefore not surprising that avian retinas are among the most complex vertebrate retinas. Apart from normal vision, bird retinas might also be involved in magnetoreception [1,2] and maybe even polarized light detection [3]. However, we still do not know which retinal cell types would be responsible for these additional sensory functions.

In vertebrates, there are two main types of photoreceptor cells—rods and cones [4]. They all have the same basic structure consisting of an outer segment (OS) containing membrane discs, filled with visual pigments, an inner segment (IS) containing mitochondria, a cell body with a nucleus and other organelles, and, finally, an axon terminal with its synaptic endfoot connecting the photoreceptors to second-order retinal neurons [5,6]. In addition, most bird cones have an oil droplet specific to each cone type [7,8]. Almost all bird species have one type of rod, four types of single cones and one type of double cone [9]. It is well established that rods are mainly responsible for achromatic low-light (scotopic) vision and that the four single cones are responsible for daytime (photopic) vision, including tetrachromatic colour vision [10].

Double cones have a similar structure to single cones but consist of two cells which are attached and might be electrically coupled to each other [6,11,12]. These two cells are distinct by shape and size and were described as the principal and accessory member of the double cone. Apart from rare exceptions, both double cone members in birds contain the same long-wavelength-sensitive (LWS) visual pigment (iodopsin) as is found in the LWS single cones [13–15]. Oil droplets can be of the same or different types in the two double cone members [10,16]. However, in some bird species, the oil droplet of the accessory member is either missing, consists of a diffuse deposit of carotenoids or is fractionated [6,17].

Double cones have been found in fish, reptiles, birds and monotremes, but are absent in placental mammals. In birds, they are the most abundant type of photoreceptor cell and usually make up 40–50% of all cones ranging from 30% in the red field of the pigeon retina [18] to 80% in the Magellan penguin retina (Spheniscus magellanicus) [19]. Despite their retinal abundance, the function of double cones is still under discussion. It has been shown that they probably do not contribute to colour vision [20,21]. The proposed functions include luminance detection [22], motion detection [20,21,23], polarization vision [22,24,25] and most recently, magnetic sensing [1,15].

The magnetic compass sense of night-migratory birds seems to be based on a light-dependent, radical-pair-based mechanism in cryptochrome proteins [1,2,26–37] located in both of the birds' eyes [38–41]. Among the six cryptochromes known from bird retinas [15,36,42–51], cryptochrome 4 (Cry4) seems particularly interesting because it binds the cofactor flavin, which is crucial for any magnetic sensitivity of any cryptochrome [36,46,52–54]. So far, there is no convincing evidence of flavin binding in other types of vertebrate cryptochomes, e.g. Cry1 and Cry2 [52,55]. Furthermore, Cry4 from a night-migratory bird forms magnetically sensitive, long-lived radical pairs which fulfil a number of the key chemical and physical requirements needed to act as a magnetoreceptor [36]. Within the birds' retina, Cry4 is primarily located in the outer and ISs of double cones and LWS single cones [15].

One of the key requirements of the radical-pair mechanism is that the receptor molecules must have restricted mobility [31,56]. Furthermore, correlated responses based on multiple aligned receptors would improve the signal-to-noise ratio [57,58]. This could be achieved in the many parallel disc membranes present in the OSs of photoreceptor cells and/or in the cylindrical cell walls of the photoreceptor cell ISs [1]. In this respect, it could be important that Cry4 might anchor itself to iodopsin inside the double cones [59].

In order to separate differential responses of the photoreceptors to magnetic stimuli from differential responses to variations in light intensity and light-polarization angle, a comparison of the magnetic activation of receptors with very specific orientations in neighbouring cells would probably also be required [1,58]. Likewise, to detect the angle of maximum polarization, perpendicularly oriented receptor molecules are needed to separate changes in light intensity from changes in polarization angle [60–63]. Therefore, the orientation in regular and repetitive mosaics could be very advantageous for the sensitivity of avian double cones to magnetic fields and polarized light.

Photoreceptor mosaics occur in many animals. Humans and other primates have a regular-distance mosaic of S cones, whereas the M and L cones are distributed randomly [64]. In the chicken retina, the double cones and four single cone types all tile the retina in five spatially independent, self-organizing regular-distance mosaics [8]. In numerous fish retinas, the double cone mosaics do not only show a consistent distance to other double cones, but their mosaics involve very specific relative cell orientations [65–71]. The double cones can be oriented parallel or antiparallel to each other, they can form a square mosaic in which the clefts of two adjacent double cones are oriented at about 90° angles to each other (e.g. [67,68,72]), or they can generate hexagonal arrays [66,70]. In some cases, two or more types of mosaics are found within the same retina [67,72]. Simple forms of oriented row mosaics are also found in some reptiles [73]. However, so far, we have found only one mention of the possible existence of an oriented square double cone mosaic in a bird, the great tit (Parus major), but the pattern was never quantified systematically [74].

In the present study, we therefore wanted to know whether there are consistently oriented mosaics of double cones in the retina of night-migratory European robins (Erithacus rubecula) and/or non-migratory domestic chickens (Gallus gallus domesticus) which could aid a magnetic sense and/or polarization vision. To make a systematic quantification of how ordered the double cone patterns are, we developed a method that uses the orientation angles of neighbouring cells as an input. It allowed us to visualize and compare the cell orientation patterns in a systematic way. We also analysed the density of double cones in the retina and the distribution of Cry4 in the OSs of the two members of the double cones.

2. Methods

2.1. Experimental animals

In this study, we used the retinas from 15 birds: nine domestic chickens (Gallus gallus domesticus) and six European robins (Erithacus rubecula). Chickens were bred and raised in the animal facility of University of Oldenburg (Lower Saxony, Germany). European robins were caught by using mist nets in the vicinity of the Oldenburg University campus. Bird catching was done based on a permit from the Lower Saxony State Department for Waterway, Coastal and Nature Conservation (NLWKN, D7.2220/18). Chickens that were used in the experiments were 3–5 weeks old, all European robins were adults. Prior to the experiments, all birds were kept indoors under the natural photoperiod. European robins were sacrificed by decapitation; domestic chickens were sacrificed by decapitation if they weighed less than 250 g, otherwise, they were sacrificed by an overdose of Narcoren. All animal procedures were performed in accordance with local, national and EU guidelines for the use of animals in research and were approved by the Animal Care and Use Committees of the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit (LAVES, Oldenburg, Germany).

2.2. Tissue preparation and immunohistochemistry

The eyes were enucleated and dissected by removal of the cornea, lens and vitreous body. The dissected eyecups were transferred to a Petri dish containing Ringer's solution (containing in mM: 100 NaCl, 6 KCl, 2 MgSO₄, 1 CaCl₂, 1 NaH₂PO₄, 30 NaHCO₃ and 50 glucose). Each eyecup was cut into four equal pieces: the dorsal, ventral, temporal and nasal retinal regions. In the ventral region, a small fragment containing the pecten was removed. In each piece, the retina was separated from underlying tissues, including the pigment epithelium, and mounted on filter paper (0.8 µm pore size; AABP; Merck Millipore, Schwalbach, Germany) with the photoreceptor side facing up. The retinas were fixed in 2% paraformaldehyde in 0.1 M phosphate buffer (PB) for 30 min and washed in 0.1 M PB three times for 30 min.

To visualize the double cones in fluorescence microscopy, the retinas were stained with antibodies against calbindin, because it is a reliable marker for double cones in the bird retina [15,75]. The fixed and washed whole-mounted retinas were blocked with 10% normal donkey serum in 0.1 M PB containing 0.5% Triton X-100 and 0.05% NaN₃ at 4°C overnight. After blocking, the samples were incubated with the primary antibodies (Calbindin D-28 k rabbit (Swant), 1 : 500 dilution) at 4°C for 5 days. To visualize Cry4 localization, the robin retina was stained with the Cry4-3c2 monoclonal antibody, previously used and tested in Günther et al. [15]. Additionally, we used a commercially available antibody OPN1SW-P13 (goat, polyclonal, Santa Cruz) raised against a C-terminal peptide of human OPN1SW to visualize shortwave-sensitive cones. This antibody was tested and used before [43], and the staining procedure was identical to the one described in Bolte et al. [43]. After washing with 0.1 M PB three times for 10 min, the retinas were incubated overnight with Alexa 568 or Alexa 488 donkey anti-rabbit antibodies (1 : 500) for the calbindin staining. For the calbindin-Cry4 double stainings, we used Alexa 568 donkey anti-rabbit (1 : 500) for calbindin and Alexa 488 donkey anti-mouse (1 : 500) for Cry4. For calbindin-OPN1SW-P13 double stainings, we used Alexa 568 donkey anti-rabbit (1 : 500) for calbindin and Alexa 488 donkey anti-goat (1 : 500) for OPN1SW-P13. This was followed by washing with 0.1 M PB three times for 10 min. In the next step, the retinas were mounted on microscope slides, then covered with Aqua-Poly/Mount medium (Polysciences, Inc., Warrington, PA, USA) and a coverslip.

2.3. Image acquisition and measurements

Images of the immunohistochemically stained retinal whole-mounts were taken with a Leica TCS SP8 confocal microscope using a HC PL APO CS2 40×/1.30 OIL objective.

For the visualization of Cry4 distribution in OSs of double cones, images were taken at a resolution of 1024 × 1024 pixels and physical size of 48.5 × 48.5 µm as a z-stack (23.19 µm thickness, step size 0.39 µm). The brightness and contrast were adjusted in ImageJ only for presentation purposes.

Images for the analysis of double cone orientations were taken at a resolution of 1024 × 1024 pixels and physical size of 96 × 96 µm as a z-stack (10–20 µm thickness, step size 0.45 µm). For each retina piece, three regions at the periphery and three regions in the centre were scanned (figure 2g). In addition, we also scanned a different piece of retina being double stained for calbindin and OPN1SW-P13 to visualize the relative patterns of double cones and shortwave-sensitive cones. We scanned one region in the central and one region in peripheral retina and used the same settings, size and resolution as above.

Figure 2.

Confocal microscope pictures of whole-mounted retinas from European robins (a,d) and domestic chicken (b,e). Equivalent pictures from rainbow trout retinas reproduced from Hawryshyn et al. [67] with permission from the Journal of Experimental Biology are shown for comparison (c,f). Notice the square mosaic relative orientation of the double cones in the central region (a–c) and row mosaic orientation in the periphery (d–f) in all of these retinas. The double cone cells were labelled with calbindin antibodies (magenta) in the bird retinas and with osmium tetroxide (OsO₄) in the trout retinas. Note that double cones become larger in the periphery as they become less dense. White arrows in (a), (b), (d) and (e) show the orientations of each double cone pair. The arrowheads point in the direction of the accessory member of the double cone. Red arrows represent typical orientation patterns. The black arrows in the trout pictures point to short-wavelength-sensitive (blue) cones, black arrowheads point to ultraviolet-sensitive (UV) cones. Scale bars 20 µm. The squares in G indicate the retina sampling locations for European robin and chicken retinas from which we analysed samples. The pecten is depicted as a grey ellipsoid. H, I: Illustration of how the angles between a reference cell and its 1st, 2nd and 3rd nearest neighbours were calculated. The minimum angular difference clockwise or anticlockwise between the two angles was recorded as the result

The principal and accessory members of double cones are easily distinguishable, since the principal cone is significantly larger than the accessory cone at the bottom of the OSs and at the upper end of the ISs. Thus, it was easy to calculate the orientation vector of each double cone, based on the position at which the accessory member was attached to its corresponding primary member (figure 2).

Each image was analysed manually using ImageJ software ([76], Fiji). In some cases, to increase the visibility of cells, the built-in function enhance contrast was used. In each scanned z stack, one image was selected where most of the double cones were clearly identifiable. The cell orientations were evaluated using the arrow tool. An evaluator drew an arrow through the central line of each double cone, starting from the edge of the primary cone and ending on the edge of its accessory cone partner in a way that split both into two equal parts (figure 2). The evaluation was done in a blinded manner by three different people, and the average of their results was used. Arrows were treated as ROIs (regions of interest, i.e. selections). If the orientation of a given cell could not be clearly identified, it was evaluated by using images from the stack directly above or below the selected plane. The location of each double cone pair was defined to coincide with the central XY-coordinates of the arrow drawn through it. These XY coordinates served as the cells’ coordinates and the angle served as the cells' orientation vectors.

2.4. Data analysis and statistics

The angles between the nearest neighbour cell vectors were analysed with a custom-written R script (R-script available from the electronic supplementary material) using the spatstat package [77]. For each cell in the image, the angle between it and its nth nearest neighbour were calculated, where n is a number limited by the total number of cells in the image minus one. The angles were calculated as the minimal angular difference (values between −180° and +180°) between the vector of the reference cell and a vector of nth nearest neighbour of this cell. Examples of angle calculations for four cells are presented in figure 2h,i.

The total data presented are accumulated from several images from each region (dorsal, ventral, temporal and nasal) of the eyecup. Then, the distribution of these angles was plotted in a circular diagram (blue colour, figures 3 and 4). For comparison, we also generated a randomized dataset, with the same cell coordinates for each image, but using randomized angles (repeated 50 times). The equivalent circular distributions of these angles were plotted into the same circular diagrams (red colour, figures 3 and 4).

Figure 3.

The circular distributions of angles between nearest neighbours in the retina of rainbow trout (a,d), European robin (b,e) and domestic chicken (c,f). Red bars—randomly simulated uniform distribution, blue bars—real data. n.n.1 = nearest neighbour 1, n.n.2 = nearest neighbour 2 = the 2nd nearest neighbour, etc. 0° defines the orientation of the investigated cell. A value of 180° in n.n.1 means that the accessory cone of the nearest neighbour of the investigated cell was oriented in the opposite direction than the accessory cone of the investigated cell. A value of 0° means that the nth nearest neighbour is oriented in the same direction as the reference cell. The analysis represents the average result based on analyses of the nearest neighbours considering each and every cell in the retina as the reference cell one by one. N: the number of experimental animals. Asterisk—Kuiper's circular uniformity test significance.

Figure 4.

The circular distributions of angles between nearest neighbours in the retina of European robin (a) and domestic chicken (b) from a 0.3 mm² area. The orientational correlation in the periphery of the retina extends very far (beyond the largest piece of retina we imaged [i.e. the 2000th nearest neighbour in the chicken and the 7000th nearest neighbour in the robin]). Red bars—randomly simulated uniform distribution, blue bars—real data. n.n.1 = nearest neighbour 1, n.n.100 = nearest neighbour 100 = the 100th nearest neighbour, … , n.n.7000 = 7000th nearest neighbour, etc. A value of 180° in n.n.1, means that the accessory cone of the nearest neighbour of the investigated cell was oriented in the opposite direction of the accessory cone of the investigated cell. A value of 0° means that the nth nearest neighbour is oriented in the same direction as the reference cell. The analysis represents the average result based on analyses of the nearest neighbours considering each and every cell in the retina as the reference cell one by one. N: the number of experimental animals.

To get a statistical measure of how far the non-random mosaic extends, we used the two-way Kuiper's test of circular uniformity for 1 to 50th and 90th nearest neighbour relative angle distributions for each species. Each real data distribution was compared to the randomly generated uniform distribution of the same size and the test statistic and p-value were calculated. This procedure was repeated 50 times, each time with a newly generated uniform distribution. After that, the mean value of the test statistic and the p-value were calculated and used as final results.

The double cone cell density was calculated as the number of cells per square millimetre for every image used in the orientation dataset. A Wilcoxon rank-sum test was used to compare the data between different regions (centre and periphery) within one species and between different species within one region.

The datasets supporting this article have been uploaded as part of the electronic supplementary material.

3. Results

To investigate the distribution of Cry4 in the two members of double cone photoreceptors, we labelled whole-mounted retinas of European robins for Cry4 and calbindin, a marker for avian double cones (figure 1). At the most distal part of the OSs, pairs of adjacent members of many double cones are visible, but their appearances are quite irregular because many photoreceptors are slightly tilted (figure 1b,c). At this z-stack level, the calbindin labelling colocalized with Cry4 labelling in both double cone members (figure 1c). In the middle of the OS (figure 1d–f), the colocalization of calbindin and Cry4 labelling in both members of each double cone is even more clear, but the appearances of the double cone pairs are still affected by photoreceptor tilting. More proximal, at the transition of the OS into the wider ISs in the principal member of each double cone, the OS of the accessory member still continues and has a much smaller diameter than the principal member (figure 1g–i). There is no apparent Cry4 labelling in the IS of the principal member, whereas in the smaller accessory member OS, calbindin and Cry4 labelling are still colocalized (figure 1i; figure 1j–l, arrowheads). The morphological appearance of the double cone at this z-stack level is very consistent, and we therefore chose to use this level for defining the orientations of the two members of the double cones relative to each other and relative to neighbouring double cones (figure 2a,b,d,e).

Figure 1.

The distribution of Cry4 (green) and calbindin (magenta) in the OSs of double cones in the whole-mounted retina of European robins (Erithacus rubecula). The images (a–i) are from the whole-mounted central retina and were taken at different levels of the OS: distal (a–c), middle (d–f) and proximal (g–i). Vertical retina section (k–m) labelled for Cry4 (k), calbindin, a marker for avian double cones (l), and overlay of this double staining (m). Notice the Cry4 expression in the OSs of both members of the double cone and that the OS of the accessory member (arrowheads) starts more proximally than that of the principal member (arrows). INL, inner nuclear layer; IS, inner segment; ONL, outer nuclear layer; OPL, outer plexiform layer; OS, outer segment. Scale bar 10 µm for (a–i), 20 µm for (k–m).

To analyse the relative orientation angles of neighbouring double cones, we generated datasets for all retinal regions in the chicken and robin retina (for more details, figure 2). In addition, two images of rainbow trout retina with examples of square and row mosaics were taken from [67] and also analysed (figure 2c,f). The typical patterns of the cell orientations in the central and peripheral bird retinas are presented in figure 2a,b,d,e and compared with the square and row retinal mosaic of a rainbow trout (figure 2c,f). In general, the cell orientation patterns in the central avian retina resemble the patterns of the fish square mosaic (figure 2a–c), and the patterns of the peripheral avian retina resemble the fish row mosaic (figure 2d–f). However, even with the naked eye, it is clear that the avian retinas, both at the periphery and the centre, are much less strictly ordered than the examples from the fish.

In every acquired image in each retinal region, the nth nearest neighbour's orientation angles relative to a given reference cell were calculated, where n is 1, 2, 3, 4, 7, 16, 20 and 90 (figure 3). In each round, one cell was chosen as a reference cell and the orientation angles of its neighbours were calculated. This procedure was repeated for every cell in the image and as a result, the average distribution over all cells in the image was generated. The distributions from the same nearest neighbour, region and species were combined into one graph. For simplicity, in figure 3, the average results are presented for the centre and the periphery of the retina for both bird species, respectively. The data for each analysed image of fish retina are presented separately. The data for different regions in the avian retina are presented in electronic supplementary material, figures S3 and S4.

The nearest neighbour analysis showed that the angles at which the double cone cells are oriented to each other are not randomly distributed. The angles between the first few nearest neighbours are more likely to be 90°/−90° in the central avian retina (figure 3b,c) and 0°/180° in the peripheral avian retina (figure 3e,f). Starting approximately at the 7th nearest neighbours, the most likely relative angular orientation in the central avian retina switches to 0°/180° bimodality, but, in general, the distribution is much more spread out, and, from around nearest neighbour 16–20, the angular distribution becomes almost uniform (figure 3b,c). In the peripheral retina of both bird species, there is a transition from a rather uniform 180° unimodal distribution when the 1st to 4th nearest neighbours are considered to a 0°/180° bimodality when nearest neighbours beyond the 4th nearest neighbour are considered (figure 3e,f). The distributions for the peripheral retina do not become uniform even when considering the first 90 nearest neighbours.

In general, the distributions for the square mosaic of rainbow trout and the avian central retina are similar. The same is true for the row mosaic of rainbow trout and the avian peripheral retina. However, in both cases, the distributions in trout are much more regular (figure 3a,d).

To determine, how far the 0°/180° bimodal pattern spreads in the peripheral avian retina, the relative angle distributions were calculated in a much larger retinal area (0.3 mm²) in both bird species (figure 4). Even between the cells at the very edges of this larger piece (nearest neighbour 2000 in chicken and 7000 in European robin), there is still a 0°/180° bimodality, which suggests that the 0°/180° orientation of the double cone mosaic stretches across the entire retinal periphery in both domestic chickens and European robins.

In figure 3, the distributions that are significantly different from uniformity at p ≤ 0.05 (mean p-values of 50 runs of the test) are marked with an asterisk. The value of the test statistics is presented in figure 5. The exact p-values and test statistic values are listed in electronic supplementary material, table S1. The highest statistic test values are found in rainbow trout retinas (figure 5), in which double cones are much more regularly ordered than in both bird species. The statistic test values for bird peripheral retina are higher than in the central retina, which also supports the described differences in regularity. We also performed modality tests using the R package flexcircmix for selected nearest neighbours: 1st to 3rd, 7th, 16th, 20th and 90th. The results are presented in electronic supplementary material, figure S5 and further show the similarities in modality and preferential angles for the square mosaic of rainbow trout and bird central retina and for the row mosaic of rainbow trout and bird peripheral retina.

Figure 5.

The test statistic values of the Kuiper's test of circular uniformity for the circular distributions of angles between the nearest neighbours from 1 to 50 for all studied species. Abbreviations: ER—European robin, GG—chicken, OM—rainbow trout.

We also used the dataset from figure 3 to calculate cell densities in the central in peripheral retina in both bird species. The results are presented in the electronic supplementary material, figure S2. The cell densities in the robin peripheral and central retina were 19 124 ± 718 and 51 897 ± 2866 cells per mm², respectively. The cell densities in the chicken peripheral and central retina were 14 298 ± 664 and 35 092 ± 2791 cells per mm², respectively. All values are given as mean ± s.e.m. The cell densities were significantly higher in both the periphery and the centre of the robin retina compared to the equivalent regions in the chicken retina (Wilcoxon rank-sum test, periphery: W = 712.5, p-value < 0.001, n = 58; centre: W = 384.5, p-value < 0.001, n = 48). As expected, the cell densities within a species in the central retina were significantly higher than in the peripheral retina for both robins and chicken (Wilcoxon rank-sum test: robin W = 448, p-value < 0.001, n = 46; chicken: W = 806.5, p-value < 0.001, n = 60).

Finally, we looked at the relative distribution patterns of double cones and shortwave-sensitive cones in chicken peripheral and central retina for comparison with the fish data. The results are presented in electronic supplementary material, figure S1 and suggest that in the chicken retina, the mosaics of single cones are independent from the ordered double cone mosaic. This is in contrast with the square mosaics formed by photoreceptors in the trout retina. There, double cones are arranged around a central blue cone in a way that the membrane between the two double cones (called ‘partitioning membrane’) faces surrounding UV cones [25] and may act as a dichroic mirror, reflecting polarized light onto the transverse axis of UV cones [24,25]. However, as the bird retina lacks a similar arrangement, UV cones are most likely polarization-insensitive.

4. Discussion

Our results show that, in the retina of domestic chicken and European robin, the orientation of neighbouring double cones is not random. The orientation patterns of neighbouring double cones change from a ‘square’-like mosaic in the central retina where the angles between neighbouring cells are more likely to be 90°/−90°, to a ‘row’ mosaic in the retina periphery where the angles between cells are more likely to be 180°/0°. A square mosaic pattern has been anecdotally suggested for the great tit (Parus major) [74]. The row mosaic pattern has, to the best of our knowledge, not been described in birds before.

Both types of mosaic, square and row, are well described in the retinas of fish [65–71] and reptiles [73]. It is also known that different types of mosaics can be combined within one retina [67]. The regular occurrence of ordered mosaics in different taxa may suggest that birds could have inherited it from their reptilian ancestors. However, convergent evolution cannot be excluded.

Even though the retinal mosaics of birds are not as perfectly regular as those observed in fish, their mosaics could still be beneficial for their magnetic compass sense and/or polarized light vision. In a theoretical study, Worster et al. [58] investigated in detail how directional information derived from the Earth's magnetic field can be distinguished from the much stronger variations in light intensity and polarization occurring during the course of a day, evening and night. As the radical-pair mechanism is light-dependent, the amount of Cry4 molecules in a signalling state depends both on their alignment with the magnetic field vector and the ambient light intensity. Thus, the large variations in light intensity over many orders of magnitude occurring in nature could easily obscure the weak magnetic field effects on Cry4. In addition, if light-sensitive magnetoreceptor molecules are strongly aligned, they will preferentially absorb light of a certain polarization, a process called photo-selection [58,78]. Again, variations in light polarization could easily obscure the weak magnetic signals.

Worster et al. [58] proposed constraints under which light-dependent magnetoreception can overcome both confounding effects (light intensity and polarization):

• (i) Cry4 molecules need to be anchored, e.g. to photoreceptor disc membranes, to restrict their movement. All Cry molecules within one cell should be aligned in a very similar orientation [58]. This might be achieved by the binding of Cry4 to LWS opsins or to other proteins of the phototransduction cascade in photoreceptor OSs [59]. Cry4 alignment will ensure that all FAD chromophores of one cell are oriented in the same way, thereby maximizing the magnetic signal [31,56,79]. Cry4 orientation should be neither perpendicular nor parallel to the axis of the incoming light (for details, see [58]).
• (ii) Photoreceptors should occur in pairs with a fixed, well-defined relative orientation inside the two partners [58]. Second-order neurons can then compute the output ratio from the two members of a photoreceptor pair, thus removing the dependence on light intensity because both members would be hit by a highly similar number of photons.
• (iii) Photoreceptor pairs should be oriented 180° opposite to each other around the axis of the incoming light to provide pure magnetic information. The photo-selection effect by polarized light would be the same in the two photoreceptors, but the yield of signalling molecules generated by the magnetic field would be different. This could be achieved if the FAD chromophores are oriented in a specific three-dimensional angle range in relation to the cell axis (figure 4 in [58]). In this scenario, the differences in the signal ratio between the two cells would arise purely from the direction of magnetic stimuli.
• (iv) If the two cells are oriented 90° to each other around the axis of the incoming light, the differential signals generated by polarized light primarily in the opsins, but also secondarily in the cryptochromes, will identify the axis of dominant polarization, and since there are expected to be many more opsins than cryptochromes in the OSs of double cones, the polarizing angle signal is likely to overwhelm the magnetic signals by several orders of magnitude [58].

Worster et al. [58] suggested that these constraints could be fulfilled either within individual double cone pairs or between neighbouring photoreceptors in the bird retina. Here, we investigated the orientation of neighbouring double cones while assuming that Cry4 molecules are aligned in the same way within an individual double cone (figure 6). In the following, we will relate our findings to the predictions of Worster et al. [58].

Figure 6.

Principle illustration of how the double cones and the yield of their signalling molecules could encode polarized light and magnetic field stimuli. The grey arrow represents the magnetic field vector (M). The yellow arrow depicts the e-vector of the polarized light (P). Yellow dots represent the number of signalling molecules produced by the cell, when the polarized light is parallel to the transition dipole moment of cryptochromes within the cell. Grey dots represent the additional yield of signalling molecules in response to the magnetic field when the cell orientation is 0°. The number of yellow and grey dots are not to scale—the polarized light effects would be much larger than the magnetic field effects, especially during the day when the opsin signals (not included in the figure) would greatly overwhelm the cryptochrome signals. In our schemes, we assume that the magnetic field vector is oriented so that it maximizes the yield of the signalling molecules if the cell orientation is 0° in relation to the magnetic field vector and that the magnetic yield is minimized when the cell orientation is 180° in relation to the magnetic field vector. When the cell is perpendicular to the polarized light vector, no radical pairs are generated and therefore we cannot expect that the magnetic field will modify signalling molecule yield. The bi-coloured outlines of the double cones make it easier to see the changes in double cone orientation.

We found two different types of photoreceptor mosaics in the bird retina. In the peripheral retina, neighbouring double cones were rotated approximately 180° relative to each other around the direction of the incoming light, generating a row mosaic. Following Worster et al. [58], this arrangement would be ideally suited to potentially separate the weak magnetic signals from the larger variations in light intensity and polarization (figure 6): neighbouring cells oriented 180° relative to each other will be exposed to a very similar amount of photons and would react identically to any given polarized light stimuli, but differently to a given magnetic field vector (please also figure 4 in [58]). Additionally, in the row mosaic of the peripheral retina, many double cones have the same orientation. This could lead to a high level of coordinated responses from many cell pairs oriented in the same direction and therefore reacting in concert to the stimulus. This should increase the signal-to-noise ratio during magnetic field detection substantially. This may be particularly important, when birds migrate at night and only relatively few photons reach the retina.

In contrast with the peripheral retina, the central avian retina shows a high proportion of double cones that are oriented at −90°/90° to their first nearest neighbour, leading to a square mosaic. This orientation is theoretically ideal for polarization vision. In this scenario, one double cone would be activated very strongly when the polarization axis of the polarized light coincides with the chromophore's preferred axis of absorption, while the other double cone would not (figure 6). Because photo-selection would almost certainly overwhelm any magnetic field effects, this scenario will not be suitable for magnetic field detection. This scenario might be relevant to behavioural studies suggesting that polarized light cues interact directly with the magnetic compass [80].

In summary, the row mosaic of double cones in the peripheral avian retina may favour magnetoreception whereas the square mosaic in the central retina may favour polarized light detection (figure 6). Interestingly, in rainbow trout (Oncorhynchus mykiss) which has been shown to be sensitive to both magnetic field direction and polarized light, double cone photoreceptors also form both square and row mosaics [67,81–83].

The findings of Worster et al. [58] are based on perfect angles (90°, 180°) between perfectly cylindrical sensory cells. However, the double cone orientations in the real bird retina occur within a certain range and most likely, photoreceptors are also not perfect cylinders. Therefore, theoretical calculations should be made to determine how much the observed biological deviations from the ideal modelling parameters would affect the detection of magnetic and/or polarized light signals in the avian retina.

The row mosaic in the peripheral retina shows a greater degree of order than the square mosaic in the central retina. This may suggest that the avian retina is better suited to extract magnetic than polarized light information. In fish, polarized light vision was suggested to be based on dichroic effects: the partitioning membrane of double cone photoreceptors may reflect polarized light onto the transverse axis of UV cones [24,25]. As fish UV cones have a highly stereotyped orientation in relation to double cones, this dichroic effect could make UV cones polarization-sensitive. In the avian retina, however, UV cones do not seem to have a fixed orientation in relation to double cones (electronic supplementary material, figure S1). This makes polarization sensitivity in avian UV cones rather unlikely.

Please note that the considerations above assume that Cry4 is oriented and aligned in the same way in the two members of any double cone. There are also other ways in which magnetic and polarized light sensitivity could theoretically be achieved. One such alternative option would require Cry4 to be oriented and aligned consistently within a given member of a double cone, but differently between the two members of a double cone. Thereafter, the same geometrical reasonings as above would be valid by comparing inputs between the two members of a double cone instead of comparing inputs between neighbouring double cones. However, since we do not know the exact arrangement of any cryptochrome molecules within any double cone, and the principle geometric arguments would ultimately be the same, we will not go into every option here.

To summarize, we observed that the double cones in avian retinas show a high degree of orientational order and regularity. We suggest that the square mosaic pattern in the inner retina could potentially help birds distinguish between different polarizations of light and that the row mosaic pattern in the peripheral retina could potentially aid a magnetic compass sensing. However, this is only correlative evidence and does not in itself prove that birds detect magnetic fields and/or polarized light in their double cones. To make such a conclusion, extensive research is still needed to gather direct molecular, anatomical, physiological and behavioural evidence for such mechanisms.

Ethics

All animal procedures were performed in accordance with local, national and EU guidelines for the use of animals in research and were approved by the Animal Care and Use Committees of the Niedersäcchsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit (LAVES, Oldenburg, Germany).

Data accessibility

The datasets supporting this article have been uploaded as part of the electronic supplementary material [84].

Authors' contributions

R.C.: data curation, formal analysis, funding acquisition, investigation, methodology, project administration, software, validation, visualization, writing—original draft and writing—review and editing; G.D.: investigation and writing—review and editing; L.S.: investigation and writing—review and editing; K.D.: conceptualization, funding acquisition, methodology, project administration, resources, supervision, validation, writing—original draft and writing—review and editing; H.M.: conceptualization, funding acquisition, methodology, project administration, resources, supervision, validation, writing—original draft and writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

The authors declare no competing interests.

Funding

This project was funded by the Deutsche Forschungsgemeinschaft (DFG; Grant SFB1372: Magnetoreception and Navigation in Vertebrates, Project 395940726 and DFG Research Training Group Grant GRK 1885; both to H.M. and K.D.). H.M. also received funding from the European Research Council (under the European Union's Horizon 2020 research and innovation programme, Grant 810002 (Synergy Grant: ‘QuantumBirds’)) and R.C. received funding from the DAAD - Graduate School Scholarship Programm, 2017 (57320205).