Ocellar spatial vision in Myrmecia ants

Insects use vision to detect relevant information from their environment to orient themselves, find conspecifics, forage, navigate, hunt and mate (Cronin et al., 2014). While the compound eyes have been studied extensively (e.g. Greiner, 2006; Land, 1989; Narendra et al., 2017; Warrant, 2008), many insects also possess a single-lens type eye known as an ocellus which has been relatively understudied (e.g. Mizunami, 1995; Ribi and Zeil, 2018; Warrant et al., 2006). Typically, three simple eyes are placed in a triangular formation on the dorsal surface of the head. Each ocellus consists of a lens, an iris, a corneageous cell layer, and a retina differentiated into dorsal and ventral retinae (Narendra and Ribi, 2017; Ribi and Zeil, 2018; Ribi et al., 2011). Almost all flying insects possess ocelli. Their functions have been best studied in dragonflies and locusts where the ocelli play a crucial role in horizon detection (Berry et al., 2007a; Stange et al., 2002) and attitude control during flight (Mizunami, 1995; Stange, 1981; Taylor, 1981; Wilson, 1978). In addition, input from the ocelli aids visually guided behaviours such as flight time initiation (Eaton et al., 1983; Lindauer and Schricker, 1963; Schricker, 1965; Wellington, 1974), optomotor responses (Honkanen et al., 2018) and phototactic responses (Barry and Jander, 1968; Cornwell, 1955). Ocelli are typically absent in walking insects; exceptions include the workers of desert ants of the genera Cataglyphis and Melophorus (Penmetcha et al., 2019; Schwarz et al., 2011a), and bull ants or jack jumpers of the genus Myrmecia (Narendra and Ribi, 2017; Narendra et al., 2011). Behavioural evidence shows that the ocelli of the desert ants derive compass information from celestial cues (Schwarz et al., 2011b,c), especially the pattern of polarized skylight (Fent and Wehner, 1985; Mote and Wehner, 1980).

Neuroanatomical studies have confirmed interactions between the ocelli and the compound eye, specifically between the large second-order neurons (L-neurons) that receive input from a large number of ocellar photoreceptors and the optic lobe where signals from the compound eyes are processed. For example, in honeybees and flies, efferent fibres run from the lobula plate into the ocellar tract (Strausfeld, 1976, see also Hagberg and Nässel, 1986). In the European field cricket, Gryllus campestris, fibres run from the medulla to the ocellar tract forming knob-like blebs (Honegger and Schürmann, 1975). In the Australian field cricket, Teleogryllus commodus, and house cricket, Acheta domesticus, neurons extend from the ocellar photoreceptors to the lobular layers of the compound eyes and additionally in the laminar layers of T. commodus (Rence et al., 1988). Physiological studies in T. commodus showed that the amplitude of the electroretinograms (ERGs) measured from the compound eyes was reduced by 20% following ocellar occlusion (Rence et al., 1988). However, it is unknown whether the visual information received by the compound eyes has an effect on the visual capabilities of the ocelli.

The capabilities of a visual system are determined by the extent to which it can discriminate between fine details of objects in a scene (spatial resolving power) and adjacent stimuli based on differences in relative luminosity (contrast sensitivity) (Land, 1997). The image quality of ocellar lenses has been estimated histologically, using the hanging drop technique (originally described by Homann, 1924) and modifications of this technique. Histological evidence suggests that in some insects the ocellar lenses likely produce under-focused images because the focal plane is located behind the retina in migratory locusts, Locusta migratoria (Berry et al., 2007b; Wilson, 1978), sweat bees, Megalopta genalis (Warrant et al., 2006), blowflies Calliphora erythrocephala (Cornwell, 1955; Schuppe and Hengstenberg, 1993), orchid bees Euglossa imperialis (Taylor et al., 2016) and Indian carpenter bees Xylocopa leucothorax, X. tenuiscapa and X. tranquebarica (Somanathan et al., 2009). In some insects the ocelli appear to be capable of resolving images with the plane of best focus located close to the retina as seen in European honeybees, Apis mellifera (Ribi et al., 2011, but see Hung and Ibbotson, 2014), paper wasps Apoica pallens and Polistes occidentalis (Warrant et al., 2006), and dragonflies Hemicordulia tau and Aeshna mixta (Berry et al., 2007a,c; Stange et al., 2002). Additionally, the spatial resolution of the honeybee ocelli was estimated by quantifying the contrast in the image produced by the lens (Hung and Ibbotson, 2014) and that of the dragonfly ocelli by measuring the acceptance angles of the ocellar photoreceptors (Berry et al., 2007c). In honeybees provided with vertical and horizontal gratings, it was found that contrast information for high spatial frequency was transferred through the ocellar lenses better than low spatial frequency (Hung and Ibbotson, 2014). In dragonflies, the acceptance angles were obtained by ray tracing through anatomical models of the median ocellar lens and retina. The acceptance angle in elevation (5.2 deg) was half that in azimuth (10.3 deg) suggesting higher resolution in the vertical plane (Berry et al., 2007c).

The hanging drop technique and other histological methods do not consider the physiological properties of the photoreceptors or the ocellar second-order neurons, which are essential for determining the visual capabilities of the eye. Hence, intracellular electrophysiology has been used to infer the spatial resolution of the ocelli. In dragonflies, the spatial resolution of the ocelli was extrapolated by measuring the angular sensitivities of the median ocelli photoreceptors (van Kleef et al., 2005) and ocellar second-order neurons (Berry et al., 2006, 2007a) in response to green and ultra-violet (UV) LED arrays. Similar experiments were done on the second-order neurons in the lateral ocelli of locusts (Wilson, 1978) in response to a Xenon arc lamp. In dragonfly median ocelli, the acceptance angles of the photoreceptors were 15 deg in elevation and 28 deg in the azimuth, in the vertical and horizontal plane, respectively, indicating relatively enhanced spatial resolution in the vertical plane (van Kleef et al., 2005). These values were a factor of 2 or more larger than when obtained from the ray tracing method mentioned previously (Berry et al., 2007c). Additionally, although the acceptance angles were slightly higher (elevation 20 deg, azimuth 40 deg) when measured from the ocellar second-order neurons, the hypothesis of enhanced spatial resolution in the vertical plane remains unchanged, providing evidence that spatial resolution is conserved after convergence of photoreceptors onto second-order neurons (Berry et al., 2006). In locusts, angular sensitivities were measured only in the horizontal plane and it was found that the field widths measured at 50% maximum sensitivity showed considerable variation. However, the total extent of the field for all the cells showed less variation and was at least 130 deg, indicating that the spatial resolution was low in these neurons (Wilson, 1978). Therefore, while the spatial resolution of the ocelli has been estimated using various techniques, the contrast sensitivity of the ocelli has been neither estimated nor measured physiologically (but see Simmons, 1993, for intracellular responses of L-neurons to sinusoidally modulated light of varying contrasts in locusts).

Pattern electroretinography (pERG) has been used to measure simultaneously the spatial resolving power and contrast sensitivity of the compound eyes in ants (Ogawa et al., 2019; Palavalli-Nettimi et al., 2019) and honeybees (Ryan et al., 2020). The pERG consists of non-linear ERG components that are generated by second-order visual neurons, at least in vertebrates (Porciatti, 2007), in response to contrast-reversing sinusoidal gratings that are contrast-modulated patterned visual stimuli at constant mean luminance. In insects, when measured in the presence of ON and OFF light stimuli, ERGs from the compound eyes show the presence of ON transients (Coombe, 1986; Ryan et al., 2020) and OFF transients (Coombe, 1986; Palavalli-Nettimi et al., 2019; Ryan et al., 2020) arising from the second-order neurons in the lamina. Conventional ERGs recorded from the ocelli of cockroaches and dragonflies show the presence of Component 3, a hyperpolarizing post-synaptic potential (particularly prominent in dragonflies; Ruck, 1961a) and a sustained after-potential (shown only in dragonflies; Ruck, 1961b) arising from the dendrites of the ocellar nerve fibres (ocellar second-order neurons). This finding was predominantly based on the relative magnitudes of the ERG components recorded from the two ends of the ocellus: the photoreceptor layer and the ocellar nerve fibres (Ruck, 1961a,b). Alternatively, Component 3 can also be described as a small negative-going wave when measured at the cornea. Hence, the pERG technique is ideal as it allows us to simultaneously determine the spatial resolving power and contrast sensitivity.

Ants of the genus Myrmecia are unusual, as closely related congeneric species are active at different times of the day (Greiner et al., 2007; Narendra et al., 2011). Each species has evolved distinct adaptations in their compound eyes (Greiner et al., 2007; Narendra et al., 2011, 2017; Ogawa et al., 2019) and ocelli (Narendra and Ribi, 2017; Narendra et al., 2011) to suit the specific temporal niches they occupy. The ocelli of night-active Myrmecia ants tend to have larger lenses and wider rhabdoms to improve optical sensitivity (Narendra and Ribi, 2017). Among worker ants, Myrmecia have relatively large ocelli, which makes this group ideal to investigate ocellar physiology in day- and night-active ants.

In this study, we determined the contribution of the compound eyes to ocellar spatial vision using the pERG technique. We measured contrast sensitivity and spatial resolving power of the ocellar second-order neurons in the diurnal-crepuscular Myrmecia tarsata under different visual system occlusion conditions. Next, we measured the ocellar visual properties in four Myrmecia species to explore the effect of activity time on ocellar spatial vision. Lastly, to identify the effect of locomotory style on spatial vision, we compared the ocellar visual properties of diurnal walking Myrmecia ants with those of the diurnal flying honeybee, Apis mellifera.

We studied the physiology of the median ocelli in workers of four Myrmecia ant species: the diurnal-crepuscular Myrmecia gulosa (Fabricius 1775) (Sheehan et al., 2019) and Myrmecia tarsata Smith 1858 (Greiner et al., 2007); and the strictly nocturnal Myrmecia midas Clark 1951 (Freas et al., 2017) and Myrmecia pyriformis Smith 1858 (Greiner et al., 2007; Narendra et al., 2010). To identify whether ocellar spatial vision is influenced by locomotory differences, we compared the spatial properties of the day-active worker ants that have a pedestrian lifestyle with those of the workers of the day-active European honeybee Apis mellifera Linnaeus 1758 that predominantly fly. The animals were captured from multiple nests at the following locations: M. midas, M. tarsata and A. mellifera from the Macquarie University campus, North Ryde, NSW, Australia (33°46′10.24″S, 151°06′39.55″E); M. gulosa from Western Sydney University Hawkesbury campus, Richmond, NSW, Australia (33°37′46.35″S, 150°46′04.47″E); and M. pyriformis from the Australian National University campus, Acton, ACT, Australia (35°16′50″S, 149°06′43″E). Working with insects requires no ethics approval in Australia.

To perform electrophysiological recordings, the ants were anaesthetized by cooling them in an icebox for 5–10 min before removing their antennae, legs and gaster. Each animal was then fixed horizontally onto a plastic stage with its dorsal side facing upwards using beeswax. The orientation of the median ocelli varies slightly between species (Narendra and Ribi, 2017). For instance, in M. midas and M. pyriformis, the median ocelli are upward facing, whereas in M. tarsata and M. gulosa, the median ocelli are forward facing. In addition, median ocelli are overall more upward and forward facing than the lateral ocelli in ants (Narendra et al., 2011; Penmetcha et al., 2019). Hence, in our preparations we oriented the head to ensure that only the median ocellus faced the stimulus. To place an active electrode on the median ocellar retina, a sharp blade was used to thin the cuticle immediately posterior to the median ocellar lens until the retina was visible. Vaseline (Unilever) was placed on the thinned cuticle to prevent dehydration and a layer of conductive gel (Livingstone International Pty Ltd, Mascot, NSW, Australia) was added. The honeybees were prepared similar to the ants: after anaesthesia, their wings and legs were removed. The hair around the median ocellus was removed using sharp forceps for easier access to the retina and placement of the electrode. In honeybees, manual thinning of the cuticle was not required as it is sufficiently thin that the electrode can receive signals from the retina.

The animals were mounted within a Faraday cage wherein electrophysiological recordings were carried out on the median ocellar retinae. An active electrode of 0.25 mm diameter platinum wire with a sharp tip was placed at the point where the cuticle was thinned, posterior to the median ocellar lens in the ants. In the honeybees, the active electrode was placed on the cuticle, posterior to the median ocellar lens to access the retina underneath. The active electrode was immersed in the conductive gel in both cases. A 0.1 mm diameter silver/silver chloride wire was inserted into the mesosoma of the ants and the thorax of the honeybees, which served as an indifferent electrode.

To reduce any effects of circadian rhythms on eye physiology, the experiments were conducted at the activity time of each species, i.e. from 1 to 6 h post-sunset for nocturnal species and 2 to 8 h post-sunrise for the diurnal species.

Using conventional electroretinography we measured the ON–OFF responses from the median ocelli in M. tarsata (n=5) to confirm the presence of extracellular potentials produced by the ocellar second-order neurons in ERGs.

A white LED light source (5 mm in diameter with irradiance of 5.81×10⁻⁵ W cm⁻², C503C-WAS-CBADA151, Cree Inc., Durham, NC, USA) was used as a stimulus and placed 15 cm from the ocelli. The LED flickers at 1 kHz and is well beyond the temporal frequency of the species tested (flicker fusion frequency of compound eyes of Myrmecia ants is between 125 and 185 Hz, unpublished data). The ants were dark-adapted for 5 min prior to stimulation. Ten consecutive trials were carried out, each trial lasting for 10 s with the ON–OFF LED light stimulus presented after an initial delay of 0.5 s and for a total ON duration of 5 s using a custom-written script in MATLAB (R2015b, Mathworks, Natick, MA, USA). The signals were filtered between 0.1 Hz and 100 Hz and amplified at ×100 gain (AC) using a differential amplifier (DAM50, World Precision Instruments Inc., Sarasota, FL, USA).

Certain ocellar second-order neurons (L-neurons) are known to form synapses with the descending interneurons which receive input from the compound eyes (Strausfeld, 1976; A. mellifera: Guy et al., 1979; Schistocerca gregaria, Schistocerca americana: Rowell and Pearson, 1983; Calliphora erythrocephala: Strausfeld and Bassemir, 1985). Hence, we measured ERGs from the median ocelli retina of M. tarsata for individuals with both visual systems intact (E⁺O⁺; n=5), and again with their compound eyes occluded (E⁻O⁺, n=5). These two experiments were conducted on the same individual consecutively and in random order without disturbing the placement of the electrode. Occlusion was done by applying a layer of opaque black nail enamel (B Beauty, Wauchope, NSW, Australia), which could be peeled off completely without damaging the eye or leaving any visible residue.

We used pattern electroretinography to measure the spatial resolving power and the contrast sensitivity of the ocelli of Myrmecia ants and A. mellifera (n=4 for M. gulosa, n=5 for all other species).

The pERG visual stimuli were projected by a digital light-processing (DLP) projector (W1210ST, BenQ corporation, Taipei, Taiwan) onto a white screen (W51×H81 cm) at a distance of 30 cm from the animal. The stimuli were vertical contrast-reversing sinusoidal gratings of different spatial frequencies (cycles deg⁻¹) and Michelson's contrasts. They were generated using Psychtoolbox 3 (Pelli, 1997) and MATLAB (R2015b, Mathworks) and controlled via a custom Visual Basic Software written in Visual Studio (2013, Microsoft Corporation, Redmond, WA, USA). The mean irradiance of the grating stimulus was 1.75×10⁻⁴ W cm⁻², which was measured using a calibrated radiometer (ILT1700, International Light Technologies, Peabody, MA, USA). A temporal frequency (frequency at which different spatial frequencies were presented) of 4 Hz was used for all the stimuli.

Prior to the first recording, the animal was adapted to a uniform grey stimulus with the same mean irradiance as the grating stimuli for 20 min. To measure the contrast sensitivity function of the ocelli, nine spatial frequencies (0.05, 0.1, 0.15, 0.2, 0.26, 0.31, 0.36, 0.41, 0.52 cycles deg⁻¹) and five contrasts (95%, 50%, 25%, 12.5%, 6%) for each spatial frequency were presented. The spatial frequencies were first presented in decreasing order of frequencies (0.52, 0.36, 0.26, 0.15, 0.05 cycles deg⁻¹), skipping one frequency in between. In order to evaluate any degradation in response over time, the interleaved frequencies were then presented in ascending order (0.1, 0.2, 0.31, 0.41 cycles deg⁻¹). At each spatial frequency, all five contrasts were tested in decreasing order. Each combination of stimuli was recorded for 15 repeats, 5 s each and averaged in the time domain. The averaged responses were then analysed using a fast Fourier transform (FFT) in the frequency domain. As a control, the non-visual electrical signal (background noise) was recorded at two spatial frequencies (0.05 and 0.1 cycles deg⁻¹) at 95% contrast with a black board to shield the animal from the visual stimuli before and after running the experimental series. The maximum signal out of four control runs was used as the noise threshold.

To identify the origin of the neural responses recorded in the pERG, we compared the pERG response amplitudes in M. tarsata measured at each of the nine spatial frequencies (at 95% contrast) under different conditions: (a) both visual systems intact (E⁺O⁺) (n=5), (b) ocelli occluded (E⁺O⁻) (n=3), (c) compound eyes occluded (E⁻O⁺) (n=4) and (d) visual systems intact with incisions in the lamina (E⁺O⁺L⁻) (n=3). In condition d, we made a rectangular window on the dorso-frontal region of both compound eyes using a sharp blade. We then accessed the lamina and made incisions on the basement membrane on both sides of the retinae using a sharp blade. The windows were subsequently covered with Vaseline to prevent dehydration and the recordings performed. Following the electrophysiological recordings, we removed the cuticle and the brain was observed under a stereo microscope (Leica M205FA at 140×, Leica Microsystems Pty Ltd, Sydney, NSW, Australia). The incision was confirmed by two investigators. Occlusions were done as mentioned above. Condition d was performed to explore the effectiveness of our method of occlusion to block the compound eyes' input to the ocellar second-order neurons.

To assess whether the pERG response signal at the second harmonic frequency (8 Hz) of the FFT response spectrum differed significantly from 10 neighbouring frequencies, five on either side, for each spatial frequency and contrast combination, an F-test was used. The pERG amplitude was labelled as a significant data point if it differed significantly from the 10 neighbouring frequencies, and was labelled as a non-significant data point if it was not different. Only significant points were used to measure contrast sensitivity and spatial resolving power. The point at which the interpolation of the pERG response amplitudes above and below the noise threshold intersects with the noise threshold is taken as the contrast threshold and the spatial resolving power. The contrast threshold at each spatial frequency of grating was used to calculate the contrast sensitivity at that particular spatial frequency, this being the reciprocal of the contrast threshold (see Ogawa et al., 2019, for more details).

We used a linear mixed-effects model in RStudio (http://www.R-project.org/) to test whether the spatial frequency of the stimulus and the treatment conditions (E⁺O⁺, E⁺O⁻, E⁻O⁺, E⁺O⁺L⁻) had an effect on the pERG response amplitudes (significant and non-significant points) in M. tarsata. The spatial frequency of the stimulus and the treatment conditions were the fixed effects, and animal identity was a random effect. The significance of the fixed effect terms was examined using Type III ANOVA. Both pERG response amplitudes and the spatial frequency data were log-transformed.

To assess the effect of treatment condition (E⁺O⁺, E⁺O⁻) and stimulus spatial frequency on the contrast sensitivity function in M. tarsata, we used a linear mixed-effects model in R. The condition and the spatial frequency were the fixed effects and animal identity was a random effect. Using a linear model, we tested whether the spatial resolving power and the maximum contrast sensitivity differed between the two conditions. The contrast sensitivity and spatial frequency data were log-transformed before the data analysis.

Lastly, we used a linear model to test whether the maximum contrast sensitivities differed between the four Myrmecia species, and the spatial resolving powers differed among the four Myrmecia species and A. mellifera. A linear mixed-effects model was used to assess the effect of species, their time of activity and spatial frequency of the stimulus on the contrast sensitivity function of the four Myrmecia species using a maximum likelihood (ML) estimation method. The same was done among diurnal Myrmecia ants and A. mellifera to assess the effect of the species and their mode of locomotion on the contrast sensitivity functions. Time of activity in ants, spatial frequency and locomotion were used as fixed effects, and animal identity nested within species was used as a random effect. The significance of the fixed effect terms was examined using the t-test with Satterthwaite approximation for degree of freedom (lmerTest package). The model also reflected the variability in the dependent variable (contrast sensitivity function) due to the random effects. The contrast sensitivity and spatial frequency data were log-transformed before the data analysis. In all instances mentioned above, the model assumptions such as linearity were tested by plotting the residuals against the fitted values of the model. Log-transformation and normality were checked by comparing the histogram of the log-transformed data with that of the non-log transformed data. All linear mixed effect models were carried out in the lme4 package (Bates et al., 2015) of R (https://CRAN.R-project.org/package=lme4) using lmer.

We first recorded ERGs to ON–OFF LED stimuli from the median ocelli of M. tarsata under two treatment conditions: E⁺O⁺ and E⁻O⁺ (Fig. 1, pink and black lines) to confirm the presence of the extracellular potentials originating from the ocellar second-order neurons. In the E⁺O⁺ condition, at the stimulus onset, we identified Component 3, a hyperpolarizing post-synaptic potential (Fig. 1B, arrow), and the sustained after-potential of Component 3 was also identified at the stimulus offset (Fig. 1C, arrow). However, in the E⁻O⁺ condition, Component 3 was identified at the stimulus offset (Fig. 1C, arrow) but was absent at the stimulus onset (Fig. 1B, dashed arrow).

We measured the pERG response amplitudes from ocellar second-order neurons in M. tarsata for each spatial frequency at 95% contrast. To confirm whether the ocellar second-order neurons receive inputs from the compound eyes in addition to the ocellar photoreceptors, we compared the response amplitudes in E⁺O⁺, E⁻O⁺, E⁺O⁻ and E⁺O⁺L⁻ individuals (Fig. 2). The spatial frequency of the visual stimuli and the treatment conditions had a significant effect on the response amplitude (type III ANOVA of linear-mixed effect model, parameter=treatment condition, d.f.=3, F=3.31, P=0.03; parameter=spatial frequency, d.f.=1, F=141.23, P<2.2e−16; parameter=treatment condition:spatial frequency, d.f.=3, F=19.51, P=4.06e−10). For E⁺O⁺, E⁻O⁺ and E⁺O⁻ individuals, the response amplitude decreased with increasing spatial frequency (E⁺O⁺ slope=−1.06; E⁻O⁺ slope=−0.62, E⁺O⁻ slope=−1.32, E⁺O⁺L⁻ slope=0.02) (Fig. 2A–C). The response amplitude for E⁺O⁺ individuals was significantly different from that for E⁺O⁺L⁻ individuals (linear-mixed effect model, t=6.24, P=9.5e−09) and E⁻O⁺ (t=2.7, P=0.008) individuals. Similarly, the response amplitude for E⁺O⁻ individuals was significantly different from those for E⁺O⁺L⁻ (t=7.1, P=2.34e−10) and E⁻O⁺ (t=3.85, P=0.0002) individuals. Additionally, the response amplitudes were significantly different between E⁺O⁺L⁻ and E⁻O⁺ (t=−3.37, P=0.001) individuals but not significantly different between E⁺O⁺ and E⁺O⁻ (t=−1.59, P=0.12) individuals. However, at the lowest spatial frequency (0.05 cycles deg⁻¹), a linear model of the pERG response amplitude for E⁺O⁺ and E⁺O⁻ individuals as a function of treatment condition showed that the pERG response amplitude was close to being significantly different between the two treatment conditions (F_1,7=5.51, P=0.05). The mean (±s.e.m.) pERG response amplitude at the lowest spatial frequency (0.05 cycles deg⁻¹) was 0.009±0.0015 mV for E⁺O⁺ (Fig. 2A), 0.003±0.0002 mV for E⁻O⁺ (Fig. 2B), 0.016±0.0018 mV for E⁺O⁻ (Fig. 2C) and 0.001±0.0005 mV for E⁺O⁺L⁻ individuals (Fig. 2D).

We measured the contrast sensitivity and spatial resolving power in E⁺O⁺ and E⁺O⁻ individuals (Fig. 3). At each spatial frequency of the visual stimuli, the amplitude of the pERG response decreased with decreasing contrast. The contrast sensitivity decreased as the spatial frequency increased under both treatment conditions (Fig. 3A, Table 2). The maximum contrast sensitivity attained at the lowest spatial frequency (0.05 cycles deg⁻¹) was 12.9±1.2 (mean±s.e.m.) in E⁺O⁺ individuals and 6.8±1.2 in E⁺O⁻ individuals (Table 1). The spatial resolving power was 0.29±0.02 cycles deg⁻¹ in E⁺O⁺ individuals and 0.21±0.01 cycles deg⁻¹ in E⁺O⁻ individuals (Table 1). The contrast sensitivity function, the maximum contrast sensitivity and the spatial resolving power were significantly higher in E⁺O⁺ individuals compared with E⁺O⁻ individuals (Tables 2, 3, 4).

We measured the contrast sensitivity of the ocellar second-order neurons in four Myrmecia species and in the European honeybee A. mellifera in E⁺O⁺ individuals (Fig. 4). The contrast sensitivity decreased as the spatial frequency increased in all species (Fig. 4A, Table 5). The maximum contrast sensitivity attained at the lowest spatial frequency (0.05 cycles deg⁻¹) was highest in M. midas at 16.0±1.2 (6.3%) (mean±s.e.m.) and lowest in A. mellifera at 9.2±1.3 (10.8%) (Table 1). Among the ants, a linear model of the maximum contrast sensitivity as a function of species showed that the maximum contrast sensitivity did not differ significantly between species (F_3,14=0.84, P=0.49). Additionally, the variation in contrast sensitivity function was explained by the spatial frequency of the gratings, but not by the species or their time of activity (Table 5).

As A. mellifera is a flying diurnal species, we compared its contrast sensitivity function with that of M. gulosa and M. tarsata, which are also active under bright-light conditions but use walking as their primary mode of locomotion. We found that the variation in contrast sensitivity function was explained by the spatial frequency of the gratings, but not by the species or their mode of locomotion (Table 6).

The spatial resolving power of the ocellar second-order neurons in the five species was measured in E⁺O­⁺ individuals (Fig. 4B). Among the five species, M. pyriformis had the lowest spatial resolving power at 0.25±0.05 cycles deg⁻¹ (mean±s.e.m.) and M. gulosa had the highest at 0.34±0.02 cycles deg⁻¹ (Table 1), but this variation was not significantly different between species (linear model: F_4,19=0.93, P=0.47).

To confirm the presence of extracellular potentials from the ocellar second-order neurons in our ERG recordings, we measured ERGs responding to the ON–OFF stimuli from the ocellar retina of M. tarsata for E⁺O⁺ and E⁻O⁺ individuals (Fig. 1A). Similar to the ocellar ERGs in other insects (Ruck, 1961a,b), we found a hyperpolarizing post-synaptic potential and a sustained after-potential identified as Component 3 at the stimulus onset (Fig. 1B) and offset (Fig. 1C), respectively, for E⁺O⁺ individuals. In E⁻O⁺ individuals, the sustained after-potential identified as part of Component 3 was seen at the stimulus offset (Fig. 1C). This confirms the presence of ERGs from the ocellar second-order neurons in our pERG recordings. As mentioned earlier, some L-neurons are known to interact with descending interneurons which receive input from the compound eyes (e.g. Strausfeld, 1976). ­We suspect that the presence of these descending interneurons along with the ocellar second-order neurons may explain the presence of C­omponent 3 at both offset and onset in E⁺O⁺ individuals. However, when the compound eyes are occluded (E⁻O⁺), the ocellar second-order neurons alone produce Component 3 at the stimulus offset, although it was absent at the stimulus onset.

In order to evaluate the contribution of the signals from the compound eyes and the ocelli to the ocellar second-order neurons, we measured the pERG response amplitude in E⁺O⁺, E⁺O⁻, E⁻O⁺ and E⁺O⁺L⁻ individuals of M. tarsata (Fig. 2). The mean response amplitude at the lowest spatial frequency (0.05 cycles deg⁻¹) was higher in conditions when the compound eyes were intact (Fig. 2A,C) compared with conditions when the compound eyes were occluded or their neural input was disrupted (Fig. 2B,D). This indicates that the compound eyes contribute highly to the response amplitude of the ocellar second-order neurons. Furthermore, although the response amplitude overall was not significantly different between E⁺O⁻ and E⁺O⁺ individuals, we found that at the lowest spatial frequency (0.05 cycles deg⁻¹) the mean response amplitude was higher in E⁺O⁻ individuals (Fig. 2C) than in E⁺O⁺ individuals (Fig. 2A), with the difference close to being statistically significant. This raises the possibility that lower spatial frequencies might trigger an inhibitory response by the ocellar photoreceptors onto the ocellar second-order neurons. As seen in locusts, light intensity-dependent control of phototactic tuning tendencies occurs as a result of photoinhibition through the ocelli and photoexcitation through the compound eyes (Barry and Jander, 1968). This has been suggested to be a form of central light intensity adaptation (Barry and Jander, 1968). However, further physiological, neuroanatomical and behavioural evidence is required to confirm any inhibitory mechanisms by the ocelli in the context of spatial vision and its possible ecological relevancy.

The compound eyes' contribution to ocellar second-order neurons in ants may be explained by the neural pathways discussed previously. In various insect species, certain L-neurons are known to interact with descending interneurons (e.g. Strausfeld, 1976). We suspect a similar pathway is present in the ants studied. Additionally, the scale of the contribution of the compound eyes (Fig. 2A,C) was unexpectedly higher than the contribution of the ocelli to the ocellar second-order neurons (Fig. 2B). This may simply be a consequence of the high light-capturing ability of the compound eye owing to its multi-lens structure with a large number of facets (M. tarsata: 2724±67 facets per eye) (Greiner et al., 2007) and therefore numerous rhabdoms. This is drastically different when compared with the ocellus, which consists of a single but large lens and has fewer rhabdoms (M. tarsata: 46.5±7 rhabdoms in the median ocelli) (Narendra and Ribi, 2017). Overall, insects with both visual systems appear to largely use their compound eyes to obtain sufficient spatial resolution and sensitivity in order to perform various visually guided behaviours.

Incidentally, the mean response amplitude at the lowest spatial frequency (0.05 cycles deg⁻¹) was significantly higher in E⁻O⁺ individuals (Fig. 2B) than in E⁺O⁺L⁻ individuals (Fig. 2D). These differences may be due to the disruption in the neural pathway in E⁺O⁺L⁻ individuals. Consequently, this method was more effective at blocking compound eye input than was occlusion of the compound eyes with black paint in E⁻O⁺ individuals.

Although the sample size for some of our treatment conditions was low, the s.e.m. of our dataset were also low, indicating low data spread, further suggesting that physiological processes are conserved across individuals.

Because of the extremely low pERG response amplitudes from the ocellar second-order neurons of the E⁻O⁺ individuals (Fig. 2B), it was not possible to directly measure the contribution of the ocelli to the contrast sensitivity and the spatial resolving power of the ocellar second-order neurons in M. tarsata. However, the high pERG response amplitude in the E⁺O⁺ individuals (Fig. 2A) and E⁺O⁻ individuals (Fig. 2C) enabled us to estimate the visual capabilities of the ocelli. We found that the contrast sensitivity function and the maximum contrast sensitivity of ocellar second-order neurons were significantly different in E⁺O⁺ and E⁺O⁻ individuals. The mean maximum contrast sensitivity was higher in E⁺O⁺ individuals [12.9±1.2 (7.7%), mean±s.e.m.] than in E⁺O⁻ individuals [6.8±1.2 (14.8%)], indicating a significant contribution of the ocelli to contrast sensitivity of the ocellar second-order neurons. This demonstrates that inputs from both the ocelli and the compound eyes contribute to the contrast sensitivity of the ocellar second-order neurons.

We speculate that in ants the descending interneurons receive information from the ocelli first and subsequently from the compound eyes. This is quite possible because of the fast transmission of signals via the L-neurons found in several insects (Guy et al., 1979; Mizunami, 1995). This could either modulate or gate the signals from the compound eyes (Guy et al., 1979). The input from the compound eyes could further increase the contrast sensitivity of the ocelli to enable efficient navigation (e.g. Fent and Wehner, 1985; Schwarz et al., 2011b). Increased contrast sensitivity of the ocellar second-order neurons based on the contribution of the compound eyes together with the fast transmission of signals through the L-neurons would be beneficial for navigation and other visually guided behaviours (e.g. Barry and Jander, 1968; Cornwell, 1955; Honkanen et al., 2018; Lindauer and Schricker, 1963; Schricker, 1965; Wellington, 1974).

The spatial resolving power of the ocellar second-order neurons in E⁺O⁺ individuals (0.29±0.02 cycles deg⁻¹) and E⁺O⁻ individuals (0.21±0.01 cycles deg⁻¹, Fig. 3B) was significantly different. Although a large number of ocellar photoreceptors converge onto very few ocellar second-order neurons (Berry et al., 2006; Chappell et al., 1978; Goodman and Williams, 1976; Guy et al., 1979; Mizunami, 1995; Patterson and Chappell, 1980; Toh and Kuwabara, 1974), in dragonflies the spatial resolution is conserved even after this convergence, indicating the possibility of local processing within the ocellar neuropil (Berry et al., 2006). Our results suggest that in ants, the ocellar second-order neurons are involved in processing such that they contribute to ocellar spatial vision to some extent but its functional significance remains to be investigated. It is likely that the ocellar spatial acuity is enhanced when there is input from the both the ocelli and compound eyes.

Nocturnal Myrmecia ants have larger ocelli and wider ocellar rhabdoms (Narendra and Ribi, 2017) compared with their diurnal relatives, indicating that the ocelli might have higher contrast sensitivity. Therefore, we compared the contrast sensitivities of the ocellar second-order neurons in E⁺O⁺ individuals of Myrmecia species active at discrete times of the day (Fig. 4). However, the contrast sensitivity was not significantly different between the species, indicating that the time of activity did not have an effect on this. This is consistent with our knowledge of the compound eyes, where the time of activity does not explain the difference in contrast sensitivity of the compound eyes of diurnal and nocturnal Myrmecia ants (Ogawa et al., 2019).

To identify whether ocellar spatial properties were affected by the mode of locomotion, we studied the diurnal flying European honeybee A. mellifera (E⁺O⁺) and compared it with diurnal Myrmecia ants. We chose A. mellifera because their ocelli have been well studied: interaction with the compound eyes has been mapped (Guy et al., 1979) and the plane of best focus is known to lie on the ocellar retina (Ribi et al., 2011). With an ocellar lens diameter of 294 µm, A. mellifera (Ribi et al., 2011) have distinctly larger ocelli than do Myrmecia ants (M. tarsata 129.2 µm: Narendra and Ribi, 2017). Our results showed that the species and their locomotion did not have an effect on their contrast sensitivity functions.

Based on our experimental paradigm, we suspect that the main function of the ocellar second-order neurons for E⁺O⁺ individuals in all five species is to detect overall bright and dim contrasts, indicating that the overall processing of visual information is similar in their peripheral neural pathways. However, it must be stated that the contrast sensitivity of species may change depending on the intensity of light present. Our results are reflective of contrast sensitivity at a particular light intensity and may differ at different light intensities. Future studies could also measure the response–stimulus intensity (V–log I) functions, through which the saturation and adaptation levels of photoreceptors can also be assessed.

The spatial resolving power of the ocellar second-order neurons of Myrmecia ants and that of A. mellifera for E⁺O⁺ individuals measured in this study was not significantly different. The neural pathways relaying information from the compound eyes to the ocellar second-order neurons are likely to be conserved across all five Hymenopteran species, leading to the lack of difference in spatial resolving power in all species.

In conclusion, our results provide physiological evidence that the compound eyes modulate the signals generated by the ocelli and, therefore, significantly affect the contrast sensitivity of the median ocelli and subsequently the spatial resolving power, with a noticeably larger improvement in contrast sensitivity than in spatial resolving power. The functional significance of this modulation, and how it affects visually guided behaviours, remains to be discovered.

We thank Ken Cheng and Chris Reid for their comments on an earlier version of the manuscript. We acknowledge the Wallumattagal clan of the Dharug nation as the traditional custodians of the Macquarie University land. We conducted research on the lands of the Dharug people and collected specimens from the lands of the Dharug and Darkinjung people.