Wavelength discrimination of the goldfish in the ultraviolet spectral range

Vision Res. Vol. 34, No. 11, pp. 151~1520, 1994 Copyright 0 1994 Elsevier ScienceLtd

Pergamon

Printed in Great Britain. All rights reserved 0042-6989/94$7.00+ 0.00

0042&89(93)EOO69-J

Wavelength Discrimination of the Goldfish in the Ultraviolet Spectral Range CHRISTINE Received

FRATZER,*

SASKIA DijRR,*

1 April 1993; in revised form 6 October

CHRISTA

NEUMEYER*t

1993

Wavelength discrimination ability of the gold&& was measured with a behavioural trahdng technique in the UV spectral range. First, spectral sensitivity was determined for the two iIsh to adjust the monochromatic lights (between 334 and 450 mn) to equal subjective brightness. The results of the wavelength discrimination experiment show that, independentof which wavelength the fish were trained on, the relative choice frequency reached values above 70% only at wavelengths longer than 410 nm. Wavelength discrimination between 344 and 404nm was not possible. Accordingly, the Al function increases steeply between 400 and 380 nm, with values between about 12 and 90 nm, respectively. Model computations indicate that the ArZfunction cannot be explained on the basis of the cone sensitivity spectra. Instead, inhibitory interactions have to be assumed which suppress the short wavelength flanks of the short-, mid-, and long-wavelength sensitive cone types in the W range. Colour

vision

Wavelength

discrimination

Goldfish

Curussius uurotlcs

INTRODUCTION

Ultraviolet

Training

experiments

wavelength discrimination for the goldfish (and not two as to be expected in the assumed case of a trichromatic Wavelength discrimination in goldfish was first colour vision): at 400, 500 and at 61Onm. Colour determined with behavioural training experiments by mixture experiments revealed that the high, unexpected Yarczower and Bitterman (1965) for the spectral range discrimination ability at 400 nm is not due to a side between 450 and 650 nm. In a re-measurement by band of absorption of long-wavelength cone photoNeumeyer (1986) performed with monochromatic light pigment, but to a separate UV cone type (Neumeyer, of equal fish-subjective brightness, adjusted according to 1985). Further additive colour mixture experiments a spectral sensitivity function measured under the same gave evidence that colour vision in goldfish is tetraexperimental conditions (Neumeyer, 1984), the range chromatic (Neumeyer, 1992). The UV cone type in was extended to between 400 and 720 nm. Here, as in goldfish was also shown by Hawryshyn and Beauchamp all previous measurements of spectral sensitivity by (1985) measuring spectral sensitivity in the UV range. other authors, it had been assumed that this range “Direct” measurements of the absorption spectra of the coincides for the goldfish with the visible range of the cone photopigments revealed the existence of the W electromagnetic spectrum. However, a look into the old cone type much later (Bowmaker, Thorpe & Douglas, literature with unbiased eyes would have taught us that 1991). cyprinid fishes can see UV light. Using a spectral apparTo represent the AIZ function for the entire visible atus designed for the experiments with honeybees by range of the goldfish, we now measured wavelength Alfred Kiihn (1927), Schiemenz (1924) and Wolff (1925) discrimination in the W range between 334 and 400 nm. had already shown that minnows (Phoxinus hvis) are If perceived hue is based on the ratio with which the able to discriminate monochromatic light even in the different cone types are excited, a relatively high wavenear-UV range. Wolff found that wavelength discrimilength discrimination ability should be found in the W nation was best in three spectral ranges: at 420-430 nm, range. This can be predicted because in this range all at 485 nm and at 590 nm. In between, at 440 nm and at four cone types are sensitive in part because of the 550 nm discrimination ability was rather poor. The AA /?-band of the absorption spectra of the photopigments. function obtained by Neumeyer (1986) in behavioural However, model computations of the A1 function in the training experiments also showed three ranges of best range between 400 and 720 nm showed that wavelength discrimination cannot be entirely predicted on the basis of the cone sensitivity spectra (Neumeyer, 1986). In+Institut fiir Zoologie, Arbeitsgruppe III, Johannes Gutenbergstead, it had to be assumed that the output of the UniversitBt, 55099 Mainz, Gennany. tTo whom reprint requests should be addressed. different cone channels is modified by inhibitory “a

3411I-O

1515

1516

CHRISTINE

interactions prior to the level at which the excitation ratios are determined. If similar modifying mechanisms are also active in the UV range, a rather poor wavelength discrimination ability should be found.

MATERIALS

AND METHODS

Animals

Two goldfish (size 67 cm, normal shape) were kept separately in two tanks (25 x 40 cm, 27 cm high) made from UV transmitting Plexiglas (Roehm, GS 2458). During the measurements, the experimental room was illuminated to 20 lx by two fluorescent tubes (Osram L 36W/l9, daylight 5000 de luxe) which were used to stimulate all four cone types about equally strongly. Otherwise, the room was illuminated by natural daylight. Apparatus and training procedure (a) Measurement of spectral sensitivity. To be able to present the different monochromatic lights at equal fish-subjective brightness, the spectral sensitivity function was measured for each subject with entirely the same method as described in detail in Neumeyer (1984). The goldfish had to discriminate between two test fields (2 cm diameter), one was dark (unilluminated), and the second one was illuminated with monochromatic light of different wavelengths (between 334 and 461 nm) at different intensities. The fish were trained to swim to the dark test field by giving small amounts of a food paste delivered from thin plastic tubes ending at the test field. For comparison, monochromatic light was given at the second test field. [Training on the dark test field has the advantage of training on the constant stimulus with the comparison field varying in wavelength and intensity. Furthermore, the spectral sensitivity function obtained in this way is very different in shape and absolute values from the function when trained on the illuminated field. It seems that the fish use different criteria in the two cases: a “colour” or “hue” criterion when trained on the dark field, and a “brightness” or “light” criterion when trained on the illuminated field (see Neumeyer, Wietsma & Spekreijse, 1991).] The test field was illuminated by a Xenon lamp (Osram, 150 W) via UV transmitting fibre optics (LF-UV, 1 m long, 6 mm diameter, Schott & Gen). To swap the position of the illuminated field from left to right and vice versa, the fibre optics were turned by a rotating device. Interference filters were used to obtain quasi-monochromatic light (334-388 nm: type UV-DIL, half-band width 4.5-6.8 nm; 371 nm: UV-DEPIL, halfband width 7 nm; and 404-461 nm: DIL; all Schott & Gen). Light intensity was attenuated by quartz neutral density filters (Balzers) or by glass neutral density filters at wavelengths > 400 nm (type NG, Schott & Gen). Test field illumination was measured radiometrically in ,uW at the diffusing screen seen by the fish as testfield (EG & G, radio/photometer, 550-1, calibrated also for the UV range), and transformed into quanta/cm2 sec.

FRATZER

et al

Training proceeded from the highest available intensity of the monochromatic light shown at the comparison test field, and continued by attenuating intensity in steps of 0.5 log units. Relative choice frequency of the two test fields was determined in single “measurements”, each consisting of 100 choices recorded in succession. The threshold criterion was set at 75% choice frequency. (6) Measurement of wavelength discrimination. The procedure is described in detail in Neumeyer (1986). For illumination of the two test fields, two slide projectors (Prado-Universal, 250 W) were used for wavelengths >371 nm, and the Xenon lamp as one of the two light sources for shorter wavelengths. The interference and neural density filters were inserted in the projector and in the Xenon-lamp, respectively. The position of the test field illumination was exchanged by rotating the fibre optics. The intensity of the monochromatic light was adjusted according to the spectral sensitivity values, with an amount of quanta/cm2 set 1 log unit above the value at which 75% choice frequency was reached. At the beginning of the experiment, both fish were reinforced for swimming to 404 nm, with 450 nm shown for unreinforced comparison. After relative frequency with which 404 nm was chosen reached about 90%. other longer wavelengths (434 and 416 nm, respectively) were given for comparison. Then the choice behaviour on 404 nm was tested against shorter wavelengths (334, 348, 358, 371 and 388 nm) in succession. After the entire series was completed, the fish were re-trained on a shorter wavelength (388 instead of 404 nm), and again tested against longer and shorter wavelengths. In the same way, the same fish were further re-trained on 371, 358, 348 and 334 nm, and tested against longer wavelengths only. To obtain the AA values, the criterion “just able to discriminate” was set at 70% choice frequency (not at 75%, because the linear range of the curves is here shorter as in the measurement of spectral sensitivity), and the wavelength value at which this was reached was read out of Fig. 2. The difference between this and the training wavelength 1+ then gave AA. As the values can be very different if measured towards longer or shorter wavelengths, respectively, AI was plotted against A+ + AJ.12. (c) Model computations. In model computations the A.1 function was calculated on the basis of the cone sensitivity functions. The method is described in detail in Neumeyer (1986). It is based on considerations by Hermann von Helmholtz (1891) stating that the “justnoticeable-difference” between two hues is represented by a constant distance in colour space. In the calculation, the distances between the relative excitation values (UV, X, y, Z, with uv + x + y + z = 1) of the four cone types caused by the monochromatic light were used. As a distance measured we used the Euclidian distance RI. At the beginning, we calculated the distance R for the wavelength pair 501 and 515 nm, for which a relative choice frequency of 70% (corresponding to the justnoticeable-difference) was found experimentally (in

WAVELENGTH

Neumeyer, 1986). Then other wavelengths were selected, and for each wavelength the specific second wavelength was sought for which the distance of the colour loci was equal to R. The difference between these two wavelengths was taken as the calculated Ail. RESULTS AND DISCUSSION

(a) Spectral sensitivity

1517

DISCRIMINATION

[%I 100

(4 j

F

90 80

70 60

Figure l(A) shows the choice behaviour of fish F. Choice frequency increased within about half a log unit of the amount of quanta given at the comparison test field to values between 60% and 90% for each of the different wavelengths. The spectral sensitivity functions of the two goldfish K and F in the wavelength range between 334 and 461 nm are given in Fig. l(B). Maximal sensitivity was found in both fish at 358-371 nm, with a second peak at 404 nm. While the first maximum is at a similar position to the maximal absorbance of the W photopigment (Bowmaker et al., 1991), we do not have an explanation for the high sensitivity at 404 nm. Nevertheless, the functions were used to correct for equal brightness for each fish individually.

300

[%I

100 90 80 70 60 i 300

350

LOO

L50 h[nml

FIGURE 2. Wavelength discrimination in the range between 334 and 450nm for the two fish F (A) and K (B). Ordinate: relative choice frequency in % with which the training wavelength was chosen. Parameter of the curves: training wavelength h+. Each symbol represents 100 choices of the fish obtained in one measurement. Bars give SDS from three measurements performed for a few comparison wavelengths. Line at 70%: threshold criterion to determine Al.

(b) Wavelength discrimination

10"

300

400

[ quantakdsl

A Inml

FIGURE 1. (A) Choice behaviour of fish F. Ordinate: relative frequency in % with which the unilluminated training test field was chosen. Abscissa: amount of quanta/cm*sec impinging on the comparison test field. Each symbol represents 100 choices of the fish. (B) Spectral sensitivity functions of fish F (0) and fish K (a). Ordinate: amount of quanta/cm* set at which 75% choice frequency was reached [from (A)].

Figure 2 shows the choice behaviour of the two fish. Trained on wavelength 404nm, choice frequency increased steadily towards longer comparison wavelengths, reaching more than 90% at 450 nm. Tested against shorter wavelengths, fish K reached frequencies above 70% (with a rather high standard deviation) at 371 and 358 nm, which decreased again at shorter wavelengths, while with fish F choice frequency reached values between 65% and 50% only. Trained on the next shorter wavelength (388 nm), the choice behaviour showed a similar increase when tested against longer comparison wavelengths as when trained on 404nm. Also when tested against shorter wavelengths, choice frequencies below 65% were obtained with the exception of one value (about 70% at 348 nm for fish K). Especially in fish F it is obvious that, regardless which wavelength (between 404 and 334nm) the fish was trained on, choice frequency increased to values above 80% only at comparison wavelengths longer than 404 nm. Slope and position of the increase are very similar in all cases. At shorter wavelengths, choice frequency remained mainly below 65%. Only in fish K was choice frequency in some cases higher than 70%. The functions shown in Fig. 2 indicate that wavelength discrimination is possible from 400 nm on

1518

CHRISTINE FRATZER et al.

‘i

I

z a

so

(c) Model computations

La

To test the validity of our conclusion that only the UV cone type cont~butes to colour vision in the W range, we performed the same model computations as in Neumeyer ( 1986). First, we calculated the A1 function on the assumption that the cone sensitivity functions [Fig. 4(A), continuous lines] are equivalent to the absorption spectra of the cone photopi~ents, and that they determine wa~len8th discrimination. The function is shown in Fig. S(A). In comparison with the measured AL\1 function in Fig. 3, the calculated values are much too high between 430 and 440 nm, and too low between 530 and 55Onm, to mention only the largest discrepancies. At wavelengths shorter than ~nm, the calculated values show a steep increase between 370 and 350nm. In Fig. 5(B),

30

30

10

300

separation of the excitation ratios, and, thus, an even better wavelength discrimination ability is to be expected if the b-bands of the cone photopigments, as given for example in Stavenga, Smits and Hoenders (1993), and the spectral transmission of the optical media are taken into account (not shown in the following).

100

so0

600

Mnml

FIGURE 3. AL function of the goldfish. Al values of fish F (0) and K (0) from Fig. 2. A A A: data for three goldfish tested in the spectral range between 400 and 720 nm (from Neumeyer, 1986, Fig. 5).

towards longer wavelengths, but not between 300 and 400 nm. The values of AL plotted against d are shown together with the values obtained in Neumeyer (1986) in Fig. 3. The values increase continually and steeply between 400 and 380 nm where AAreaches values of 90 nm. The slope of the AA function in this range is similar, and almost a mirror image of the slope at the long wavelength side of the function, at wavelengths longer than 610 nm. The function indicates that in the W range colour discrimination on the basis of hue was not possible. The inability to discriminate wavelengths between 334 and 404 nm, and the coinciding slope of the choice frequency functions between 400 and 450 run (Fig. Z), indicate that in the spectral range below 400 nm only the UV cone type contributes to &our vision. At wavelengths longer than 400 nm, the short wavelength (S) cone type is excited in addition. However, the sensitivity functions of the short-, mid-, and long-wavelength (S, M, L) cone types [Fig. 4(A)] gradually decrease between 400 and 3OOnm with values between 10% and 30%. frhe values were extrapolated from the values given by Bowmaker et al. (1991) and do not take into account the #?-band as a side maximum. Thus, the assumed values are probably even too small.] Therefore, as can be seen in Fig. 4(B) which shows the X, UY,y-side of the te~~edron of the goldfish (cf. also Neumeyer, 1992, Fig. 8), each wavelength between 300 and 400 nm excites the four cone types in a different ratio. Thus, wavelength discrimination ability in the UV spectral range should be possible at least between 360 and 400 nm. An even better

,

r

I

1.0

0.5

LOO

300

500

500

700 Inml

0%

FIGURE 4. (A) Spectral sensitivity functions of the four cone types of the goldfish (after Bowmaker et al., 1991). Dashed and dot-dashed lines show short waveiengtb flanks of the S-cone type used in model ~mpu~tions. (B) Spectral toci of wavehmgtbs between 300 and 440 nm cah$ated on the basis of the cone sensitivity spectra shown in (A). Only the x, uo, y-side of the tetrahedron (corresponding to the relative excitation of the UV, short- and mid-wavelength cone types) is shown. The excitation of the long wavelength cone type with z-values around 0.1 were here neglected.

WAVELENGTH

(4

FIGURE 5. (A) Calculated M function on the basis of the cone sensitivity fictions [solid tines in Fig. 4(A)]. (B) Comparison of theoretical and experimental M functions in the violet and UV range. 0 0, measured values from Fig. 3; A, calculated values based on the cone sensitivity functions in Fig. 4(A) (solid lines); dashed and dot-dashed lines, calculated values on the basis of the modified S-cone flanks in Fig. 4(A) (broken lines).

the calculated and measured values in the W range are shown together for better comparison. According to the AAfunction calculated on the basis of the cone sensitivity functions (triangles), a rather good discrimination ability should be found down to about 370 nm, which, however, was not the case. Second, we posed the question how the cone sensitivity functions in the W range have to be modified to give a similar A12function as that found ex~~mentally.

1519

DIS~~MINA~ON

For this reason we modified the short wavelength flank of the S-cone type [broken lines in Fig, 4(A)], keeping the W cone type constant (the M- and L-cone sensitivity functions were set at zero). The result of the calculation is shown in Fig. S(B) for two cases: depending on whether the S-cone sensitivity flank starts to increase at 380 nm or at 400 nm, the M. function increases between 390 and 350 in, or between 410 and 360 mn, respectively. The latter calculated Ai, function comes close to the measured AI values [circles in Fig. 5(B)] (not taking into account the two lower values of fish K). The model calculations revealed that the poor wavelength discrimination in the W range is most probably due to the fact that only the W cone type contributes to colour vision, and that the contribution of the S-cone type comes into play only at wavelengths longer than 400 nm. Thus, wavelength discrimination cannot be understood directly on the basis of the cone sensitivity functions. Instead, it has to be assumed that neural interactions modify the sensitivity functions-a conclusion already drawn for the spectral range between 400 and 720 nm in Neumeyer (1986). As neural mechanisms we assume mutual inhibitory interactions between cone types of adjacent maximal sensitivities, which are also able to explain the spectral sensitivity function (Neumeyer, 1984; Neumeyer & Arnold, 1989). According to this hypothesis, the short wavelength flank of the S-cone type would become steeper reaching zero values between 300 and 400 nm if the W cone type inhibits the S-cone type. This type of inhibitory interaction could already exist on the level of the horizontal cells with “opponent” cell characteristics, for example in the tetraphasic cells found in another cyprinid fish (TriMdon hukonesis) by Hashimoto, Harosi, Ueki and Fukurotani (1988). So far tetraphasic horizontal cells have not been found in the goldfish. However, not only the short-wavelength flank of the S-cone type, but also those of the M- and L-cones have to be at zero in the UV range. This would be the case if the S-cone type inhibits the M-cone type, and the M-cone type inhibits the L-cone type, as already assumed in Neumeyer and Arnold (1989) to explain the spectral sensitivity function.

REEERENClEs Bowmaker, J. K., Thorpe, A. & Douglas, R. H. (1991). Ultraviolet-sensitive cones in the goldfish. Vision Research, 31, 349-352. Hashimoto, Y., Hbrosi, F. I., Ueki, K. & Fukurotani, K.-K. (1988). Ultra-violet-sensitive cones in the color-coding systems of cyprinid retinas. Neuroscience Research (Suppi.], 8, 81-95. Haw~hyn, C. W. & Beauchamp, R. (1985). Ul~a~olet photosensitivity in goldfish: An independent U.V. retinal mechanism. Vision Research, 25, 1l-20. Kuhn, A. (1927). Uber den Farbensinn der Bienen. Zeitschrtft fir Vergleichenaiz Physiologic, 5, 762-800. Neumeyer, C. (1984). On spectral sensitivity in the goldfish. Evidence for neural interactions between different “cone mechanisms”. Vision Research, 24, 1223-l 23 1. Neumeyer, C. (1985). An ultraviolet receptor as a fourth receptor type in goldfish color vision. Naturw~se~~~e~, 72, 162-163.

1520

CHRISTINE

Neurneyer, C. (1986). Wavelength discrimination Journal of Comparative

Physiology

in the goldfish.

A, 158, 203-213.

Neumeyer, C. (1992). Tetrachromatic color vision in goldfish: Evidence from color mixture experiments. Journal of Comparative Physiology A, 171, 639649.

Neumeyer, C. & Arnold, K. (1989). Tetrachromatic color vision in goldfish and turtle. In Kulikowski, J. J., Dickinson, D. M. & Murray, I. J. (Eds), .Seeing contour andcolour (pp. 617631). Oxford: Pergamon Press. Neumeyer, C., Wietsma, J. J. & Spekreijse, H. (1991). Separate processing of “color” and “brightness” in goldfish. Vision Research, 31, 537-549.

Schiemenz, F. (1924). Faber den Farbensinn der Fische. Zeitschrift ftir Vergleichende Physiologic, 1, 175-220.

FRATZER et al Stavenga, D. G., &nits, R. P. & Hoenders, B. J. (1993). Simple exponential functions describing the absorbance bands of visual pigment spectra. Vision Research, 33, 101l-1018. Wolff, H. (1925). Das Farbunterscheidungsvermogen der Ellritze. Zeitschrift fir

Vergleichende

Physiologic,

3, 279-329.

Yarczower, M. & Bitterman, M. (1965). Stimulus generalization in the goldfish. In Mostofsky, D. J. (Ed.), Stimulus generalizution. Stanford, Calif.: Stanford University Press.

Acknowledgements-We are very grateful to Neil Beckhaus for improving the English of the manuscript. Supported by Deutsche Forschungsgemeinschaft (Ne 215/9-l).