Olfaction and Vision in Fish Schooling

A fish school must be regarded as a group of individuals in dynamic equilibrium, with the sensory input of each fish providing a feedback of information affecting the maintenance and behaviour of the school.

Many previous studies of schooling have concentrated upon analysis of the reactions of individual fish to stimuli involved in the formation and maintenance of school structure, as distinct from the group behaviour of the whole school. The important sensory modalities involved in both school structure and group behaviour are vision, the acoustico-lateralis sense and olfaction, which all give fish information about stimuli at some distance from their source.

Most work on the sensory basis of schooling indicates that vision is the most important modality (Spooner, 1931; Shlaifer, 1942; Morrow, 1948; Keenleyside, 1955; Breder, 1959; Shaw, 1962; and many others). It does not appear from any of this work that either shape, colour, movement or any other particular aspect of the visual stimulus is specific for any species. Moulton (1960) suggested that sounds produced by swimming fish are stimuli used in the maintenance of school structure. Harris & van Bergeijk (1962) conclude from research on the lateral line that it could be used for the maintenance of school structure, and this view is supported by Dijkgraaf (1963).

Less attention has been paid to the possible role in schooling of the chemical senses, although the repellent odour of ‘Schreckstoff’ is well known (von Frisch, 1938, 1941 a, b; review by Pfeiffer, 1962). Olfaction is to be functionally distinguished from gustation and the common chemical sense in fish largely on grounds of its higher sensitivity, allowing perception of odours at some distance from their source. Wrede (1932) was the first to show a positive response in untrained fish to a species-specific odour, using minnows in a two-compartment choice situation. Gbz (1941) followed up this work by a further study of minnows using a training technique. He showed that not only can minnows distinguish between a variety of fish and amphibian species, but even between individuals of their own species. Keenleyside (1955) showed that temporarily blinded untrained rudd preferred that end of a tank where rudd odour was present.

In the present study the behaviour of roach in an odour gradient, and of roach, grey mullet and Chromis chromis, a pomacentrid species, to visual stimuli involved in schooling were investigated. Incidental observations were made on other aspects of schooling behaviour related to the main problem of contact between individuals. Nearly all workers on schooling have coined their own definitions of ‘school’ or ‘shoal’, but that of Keenleyside (1955) is followed in the present study. A school is defined by him as ‘an aggregation formed when one fish reacts to one or more fish by staying near them’. Excluded from such aggregations are cases where the groups so formed are clearly associated for fighting or courtship. It follows that two fish can constitute a school provided that they are aware of each other. The degree of parallel orientation is not thought to be important.

Roach, Rutilus rutilus L. (Cyprinidae), of 9–12 cm. length, were caught in Coe Fen Brook in Cambridge. Those kept socially were put in large holding aquaria filled with tap water in the same room as the experimental apparatus. Some fish were kept in isolation in conditions described in the relevant section. The fish were fed on ‘Bemax’ with occasional chopped earthworm and Tubifex, but the fish also browsed occasionally on the algal film inside the aquaria. All aquaria were permanently aerated with a jet of tap water. Kept in this way the fish appeared healthy and none died during the experimental period.

Chromis chromis L. (Pomacentridae), of 8–10 cm. length, and grey mullet, Mugil chelo Cuvier (Mugilidae), of 10–14 cm. length, were caught in the Bay of Naples by fishermen from the Stazione Zoológica, and kept separately after capture in large holding aquaria. Both species were fed on macerated fresh sardine flesh. Aeration was provided by a jet of sea water from the laboratory circulation. Initial mortality was restricted to those with fins damaged or scales removed during capture, the remainder stayed healthy for the duration of the experimental work.

The behaviour of Rutilus was studied at Cambridge using a horizontal linear gradient tank similar to that devised by Shelford & Allee (1913), and subsequently used by Shelford & Powers (1915), and Bowen (1931). The apparatus used is shown diagrammatically in Fig. 1. Mains tap water at a controlled rate of 2 l./min. was equally divided into the two inlet tanks, and water from these then entered the experimental tank through perforated glass pipes in each corner. Flow out of the experimental tank from the middle was via a large siphon made of Alkathene (black polyethylene) expanded into a funnel made of Perspex (clear acrylic plastic) filling the whole width of the experimental tank. All other piping was of glass and Polythene (polyethylene).

By equalizing the rates of mains inflow and the final outflow from the apparatus the water levels were kept constant. The experimental tank was much larger than those used by the above-mentioned authors, being 180 × 28 cm. with water to a depth of 22 cm. This fact, plus the relatively slow flow rate, reduced the possibility of the fish receiving rheotactic cues due to water movement through the tank.

A rectangular grid of approximately 15 cm. squares was marked on the floor of the experimental tank and the positions of fish were viewed in a mirror set at 45° above the tank. Illumination was from two 100 W. bulbs placed symmetrically above and in front of the tank in such a way as to avoid casting any shadows into the water. It was found necessary to place ground glass screens against the end walls of the experimental tank in order to prevent the fish seeing its own reflexion. The experimental tank and the mirror were surrounded by a black screen forming part of a fight-tight canopy extending in front of the apparatus. The observer sat at a table outside the canopy and observed the fish through a small peephole.

The experimental tank used in Naples for the study of Mugil and Chromis was 100 × 30 cm. with water to a depth of 20 cm. The bottom of the tank was marked with a rectangular grid of 16·6 × 15 cm. rectangles. Observation was again by a mirror at 45° above the tank, the observer being seated behind a screen with a peephole. Illumination of the tank was by north daylight.

Prior to each olfactory experiment the experimental tank of the gradient apparatus used for the work on Rutilus was filled and allowed to stand to reduce water movements. Preliminary tests undertaken with potassium permanganate crystals held stationary at the surface of the water showed the descending columns of coloured solution to be always vertical forty minutes after filling. No olfactory experiments involving fish were undertaken until at least 1 hr. after filling of the experimental tank, when it was assumed that the water mass was still. All visual experiments were commenced at least 15 min. after filling when slight water movements still occurred. These were not considered to be significant.

At the start of olfactory experiments using attractant odour four Rutilus were transfered to one of the inlet tanks with no flow through the system. After 1 hr. flow of water was started, and after a further 5 min. to establish an odour gradient in the experimental tank, a single fish caught at random in the holding aquarium was transferred to the odour end of the experimental tank, i.e. the end with the inlet from the tank containing four Rutilus. Observation was begun within a few seconds. No attempt was made to analyse the odour which it was assumed the fish secreted or excreted at a constant rate. The technique was standardized to result in a concentration of four ‘fish-hour units’ of odour in the inlet tank at the start of an experiment. Control experiments differed only in that tap water flowed from each end, with no fish in either inlet tank.

A short series of negative olfactory experiments involving the ‘Schreckreaktion’ was carried out. The skin from one flank of a freshly killed Rutilus (an area of 8–10 cm.²) was ground in a mortar. The resulting paste was shaken up with water and filtered. Half the filtrate was added to one of the inlet tanks and the other half was set to run slowly out of a burette (into the same inlet tank) during the course of an experiment.

Fish were transfered between the holding aquaria and the experimental tank in a special net comprising a short tube of soft Nylon netting terminating in a cornerless Polythene bag. The fish were thus contained in 200–400 c.c. of water after the method of Verheijen (1953). The use of a normal net in which the fish flapped about when being transfered, was shown to cause a ‘Schreckreaktion’ on some occasions. This was also reported by Külzer (1954) in work on tadpoles.

Two sources of visual stimulus were used in experiments on Rutilus, Mugil and Chromis, either ‘mirror’ or ‘caged fish’. In the former, the ground glass screen at one end of the experimental tank was replaced by a mirror, and in the latter a watertight transparent box made of Perspex and measuring 28 × 15 cm. was placed at one end of the experimental tank. This box contained a single fish.

The position of the experimental fish was observed and plotted by hand on a grid drawn to correspond with that on the bottom of the tank, at 5 sec. intervals. These recorded positions were numbered and a new recording was begun every 5 min. Thus the results of a complete experiment, which usually lasted 20 min., comprised four grids each with the numbers from 1 to 60. All interpretations of the change of behaviour with time were based on the observed differences between successive 5 min. recording periods. The preference of the fish for different regions of the experimental tank was given by the distribution of frequencies of positional recordings with respect to the length of the tank. This regional preference is shown in the results by histograms each equivalent to a 5 min. recording period, drawn as if the odour or visual stimulus were to the left. The intensity of the preference of fish for the region of the tank where stimulation was maximal or minimal was given by the response strength. This is a measure of the time that a fish spends at one end of the experimental tank in excess of that predicted were its behaviour entirely random. The response strength is thus defined numerically as the sum of the deviations from random of the frequencies of recordings in the three regions of the tank nearest the stimulus.

Records of the position of the fish, in a numbered succession at fixed intervals of space and time, allowed the calculation of speed and activity. Activity is defined as the total recorded distance total ÷ recorded time, i.e. cruising speed including stops. The, calculation of speed and activity is more fully discussed in another paper (Hemmings, 1966).

A total of forty experiments on the responses to Rutilus attractant odour was performed, and the mean result of these is given in Fig. 2a. There was a clear preference on the part of individual fish for that end of the tank where Rutilus odour was present.

Also shown in Fig. 2b is the mean result from seven experiments in which the repellent ‘Schreckstoff’ was used as the olfactory stimulus. Comparison of the two histograms shows that the positive reaction to the odour from undamaged fish was rather less strong than the ‘Schreckreaktion’ or negative response to the odour of damaged fish. This difference occurred because all the fish when tested showed a strong response to the ‘Schreckstoff’, but some fish showed very little response to the attractant odour. It might be expected that the ‘Schreckreaktion’ would be a more fixed behaviour pattern than the attraction to species odour. In the case of the attractant odour experiments, an inverse correlation occurred between the response strength and the activity of the fish, that is, those showing a lower preference for the odour were more active. A Spearman rank correlation coefficient (see Siegel, 1956) gave r_s = 0·53. Although this is not a high degree of correlation, it is highly significant, P < 0·0005.

Experimental fish were always introduced into the odour end of the tank in order that the small quantity of water in which they were carried should not taint the ‘pure ‘end of the gradient. If the result of the positive olfactory experiments could in some degree be accounted for by a tendency for fish to prefer the end into which they were first placed, then the results from control experiments arranged with the ‘introduction end ‘to the left should indicate this tendency. A 𝒳² test on the data from fifty control experiments gave a probability of being random of P = 0·95.

Two classes of decline in response strength occurred : first, the proportion of time spent at the stimulus end of the odour gradient decreased during the experiment; and secondly, the longer the fish had been in captivity in the holding tank the lower was their response when tested. These two changes are referred to below as the short-term and long-term declines in response strength.

The short-term decline is shown in Fig. 3 where the mean response strength from thirty experiments using attractant odour is plotted against time. A 𝒳² test on the frequencies from successive 5 min. periods gave a probability of being from the same population of P ≪ 0 001. One series of ten experiments was continued for 1 hr. and at the end of this time there was no significant departure from random, i.e. a response strength of zero. Although there was a marked decline in the response strength during the experiment, the response remained greatest at the odour end of the gradient. The ‘Schreckreaktion’ also shown in Fig. 3 does not show the same decline in response strength, but increases during the first three recording periods. A 𝒳² test on the frequency data gave a probability of being from the same population of P = 0·01. This suggests that the slow rise and sharp fall are significant. It may be that the break in response strength of both attractant and repellent at about the same time, is associated with a change in the equilibrium of the odour gradient. The curve for activity of fish during the positive olfactory experiments in Fig. 3 clearly shows the inverse relation between response strength and activity.

The long-term decline is illustrated in Fig. 4, showing the mean result from three series of ten experiments performed after the fish had been different periods in holding aquaria. A 𝒳² test on the original frequency data gave a probability of their being from the same population of P ≪ 0·001. A fourth series of experiments using freshly caught fish on the third day after capture failed to confirm the expected high response strength, but this was probably due to the high level of sexual activity shown by these fish in the holding tank. There is no evidence that olfactory stimuli play any part in the courtship behaviour of Rutilus. The ‘Schreckreaktion’ also shows a similar decline in response strength (Pfeiffer, 1962). These results may be evidence for a general lowering of responsiveness of fish kept in captivity.

Learning of odours plays some part in the social behaviour of fish (Kiihme, 1962), and latent learning of possibly significant odours could occur in the holding aquaria. To investigate this possibility a series of ten control experiments was carried out using recently caught fish as experimental animals. The odour was derived for half the experiments from fish that had been in the same holding aquarium as the experimental fish for the 3 days after capture, and half from fish that had been in a different holding aquarium for over a year. The results showed no suggestion of a difference in response, a 𝒳² test giving a value of P = 0·98.

The only inter-specific controls comprised a series of ten experiments using small perch, Perca fluviatilis in the inlet tank. The result of these experiments showed no significant difference from random, a 𝒳² test giving a value of P = 0·85. This result indicates either that there is no response by Rutilus to odour from Perca, or alternatively that Perca does not normally produce any odorous substance.

The well-tried method of observing the behaviour of fish near visual images allowed a comparison to be made of the effectiveness of the directional visual stimulus with the non-directional odour as attractants for schooling fish. In the case of Rutilus it was possible to use exactly the same apparatus and techniques. The mean result for Rutilus from twenty experiments using the caged fish stimulus, and from ten using the mirror, are shown in Fig. 5a, and for comparison the olfactory result is given again in Fig. 5b. The marked contrast in the shape of the two histograms reflects the difference in the mechanisms of orientation to directional visual images and non-directional odour sources. By grouping the frequencies to show only the preference of fish for either half of the tank—a lower level of analysis followed by Keenleyside (1955)—a more meaningful comparison of visual and olfactory attraction is possible. The following table shows the two classes of experiment with mean frequencies grouped into stimulus and non-stimulus halves of the experimental tank:

It is thus clear that in the conditions of the experiment Rutilus odour and the visual image of Rutilus were both effective attractants to individual fish.

The visual results using mirror and caged fish as stimuli are given separately in Fig. 6 for Rutilus, Mugil and Chromis, together with the mean result of control experiments for the same species. A 𝒳² test was used to compare adjacent results, i.e. caged fish with mirror and mirror with control. The results of these tests are given as the values of P in the figure. Rutilus and Mugil showed strong reactions towards both types of visual stimulus, but in both species the result from the mirror experiments was less marked than that from the caged-fish experiments. This was probably due to the optical difference of the stimulus situations; an experimental fish at the opposite end of the tank sees the caged fish through a water mass equal to the length of the tank, but its virtual image ‘behind the mirror’ through a water mass equal to twice that length. Thus not only the clarity of the image but the angular size at the eye will differ. Chromis showed quite a strong reaction to the caged fish but its response to the mirror did not significantly differ from the control case. However, underwater observations in the field show that the response of Chromis to mirrors is then rather greater. The change in response strength during visual experiments is shown in Fig. 7. It is clear that the marked decline in response that occurs with the olfactory stimulus does not occur in this case. Mugil does in fact show a slight rise in response strength during the experimental period.

The control histograms show marked differences which make the interpretation of the visual results more difficult. Rutilus behaved at random by swimming backwards and forwards along the tank, but not showing any marked preference for the ends. Mugil and Chromis spent much time swimming at the end walls of the experimental tank, as if in an effort to escape. Chromis often spent some time at the beginning of experiments backed tail first in a corner, a typical territorial defense attitude for this species.

In order to assist in the interpretation of the visual results a number of experiments was devoted to the study of two fish swimming freely together, which are considered as a ‘school*. An attempt was made to measure the distance separating two fish when swimming in the experimental tank for comparison with the caged fish result. It proved impossible to record positions simultaneously by hand, so the separation distances were judged in units of 15 cm. by reference to the background grid which consisted of 15 cm. squares. The results of ten such separation experiments are given in Fig. 8a. A direct comparison with Fig. 6a is possible as the same distance scale along the abscissa is used. It was striking that the two fish swimming free were never more than 75 cm. apart in 2400 recordings (ten experiments), whereas the caged-fish result in Fig. 6a shows eighty-four recordings out of the same total when the experimental fish was more than 75 cm. from the caged fish. Two individuals of Mugil and Chromis were observed to behave in very much the same way, that is, they stayed very much closer than the results from visual experiments with mirror and caged fish would suggest.

Long periods of observation of the behaviour of two individuals of each of the species suggested that a common mechanism was responsible for the close co-ordination of swimming. The behaviour could be adequately described by three motor patterns which it is suggested are of basic significance in the cohesion of a fish school. The three motor patterns are as follows :

1. Exploratory tendency—either fish would initiate swimming in any direction away from the other.

2. Returning tendency—if the fish swimming away was not followed, it would turn and swim back to the other.

3. Following tendency—usually when one fish swam away it would be followed by the second fish.

The first motor pattern appeared to be quite spontaneous and is referred to as an ‘action’ in this context, although it is probably caused by some unknown internal or external stimulus. The other two motor patterns clearly depended upon the perception by either fish of the increase in separation brought about by the original action, and are therefore termed ‘reactions’. The dynamic relationship between two individuals seems entirely explicable by these three motor patterns. Such interspecific differences as occur can be explained by differences in the relative strength of the three components, and the delay between the occurrence of the exploratory action and one of the reactions.

A study was made of the changes in reaction to visual and olfactory stimuli of four Rutilus kept in isolation instead of socially in a holding aquarium. A number of the resulting changes in behaviour give indications of possible sensory mechanisms that operate in typical schooling behaviour.

Two fish (A1 and A2) were isolated after 214 days living socially in a holding aquarium and two (B1 and B 2) were isolated immediately after capture. Each isolation aquarium measured 28 × 28 cm. with water to a depth of 32 cm. Conditions of feeding and aeration were exactly as for the social holding aquaria. The isolates thus experienced the following deprivations :

1. Visual contact with other fish. (The tanks were screened from each other with black paper and after a few days the growth of an algal film prevented the fish seeing their own reflexions.)

2. Olfactory stimuli from other fish.

3. Sounds, water movements and pressure changes associated with the presence of other fish.

In isolation all four fish were much less active than those living socially. Evidence for this comes not only from direct observation but also from the distribution of faecal pellets which collect in the corners of an aquarium when fish are active. The pellets in the four isolate aquaria were always scattered over the whole area of the bottom.

The same four fish were used for two groups of experiments : the first group comprised three trials per fish of each experiment type, control, visual and olfactory arranged in a randomized order; the second group comprised one trial per fish of control, olfactory and visual type, in that order. The fish had been in isolation for 171 days prior to the first group of experiments, and an interval of 132 days elapsed before the second group was started. An interval of 61 days then elapsed before a two-fish separation experiment and return to social life.

The control experiments showed no departure from random, a 𝒳² test giving P = 0·99. One interesting point was that all fish were much more active in the experimental tank than in their isolate aquaria.

Shown in Fig. 9 are the curves for the short-term changes in response strength of the two isolate series of olfactory experiments and their mean, and the mean social result for comparison. The results indicated that the total response of isolates was greater than the mean result for fish living socially, and that this was due to a less marked short-term decline in response strength. There was no significant difference between the results of the two A fish that had been a long period in captivity and the two B fish that were isolated immediately on capture. The long-term decline in response strength thus occurs in the isolates to a much lesser extent than in the socially living fish. The results of the second series of isolate experiments defy logical explanation but the two series together show that experiments must be conducted with larger numbers of isolate fish and with more intermediate testing.

The visual reaction of isolate fish was tested using only the caged-fish stimulus. The normally clear-cut visual response was completely upset by periods of isolation, so that the mean frequency distribution of position recordings would give no real information about the behaviour of the fish. The results could be simply categorized as follows: (1) positive reaction ( + )—attraction towards caged fish; (2) neutral reaction (N)—apparent indifference to the other fish; (3) negative reaction ( − )—avoidance of the vicinity of the other fish. The following table shows the results for each fish from each visual experiment, with the time interval in days between successive experiments:

The four symbols shown for each experiment represent four 5 min. periods into which each experiment was divided. Although the normally high response to visual stimuli was completely altered by isolation, there is some evidence for a return to the positive reaction occurring either within or between experiments.

After an interval of 61 days a short ‘two-fish separation’ experiment was carried out on three of the isolate fish which were then returned to the social holding aquarium. The experimental technique was exactly as for the socially living fish but here using one isolate fish and one social fish, and recording for only 5 min. The mean result is shown in Fig. 8i. The reason for the rather greater separation shown in this series of experiments can be described by using the ‘action and reaction’ system of motor patterns described above. On first being placed in the experimental tank both fish were very active, showing little response to each other, but quite quickly the social fish began to follow the isolate. This behaviour persisted until sometime later when the isolate fish showed returning and following reactions also; the returning reaction of the isolate fish was the last motor pattern to appear.

At the conclusion of each experiment the two fish were returned to a holding aquarium. After about 5 min., when co-ordinated swimming with the school was poor, the isolates became indistinguishable from other members of the school.

The results from these experiments agree with those from previous work on schooling behaviour that vision is the sense primarily involved. Little attempt has been made to analyse or even describe school cohesion with the exception of the work of Breder (1959), of Horstmann (1960) and of Shaw & Tucker (1965) who investigated the optomotor responses of schooling fish. It is suggested that the system of action and reaction tendencies described above is of value in describing school cohesion, and may even be of underlying significance. The graph in Fig. 10 shows non-quantitatively how the tendencies are related in the simple but representative case of a two-fish school. All schooling fish seem to show a strong exploratory tendency which is normally maximal when they are at the optimum separation. If fish ‘1 ‘swims away from fish ‘2’, the most likely occurrence is that fish ‘2 ‘follows it (R2), because the following tendency is observed to be greater than the returning tendency. Alternatively the fish ‘1 ‘may stop exploring and show a returning tendency (R1). The observations by Keenleyside (1955) and Steven (1959) could very easily be described by this system which might help in relating the optomotor responses to school cohesion. Many hours of underwater observation in the Mediterranean and in British waters suggests that this system of tendencies does have real significance. Characteristic changes in school shape, inter-individual distance and orientation associated with, for example, the onset and cessation of feeding in a wide variety of species can be explained by changes in the relative strength of the different tendencies.

It has often been suggested that the inter-individual distance in animal groups is brought about by the balance between attractive and repulsive ‘forces’ (for fish, Parr, 1927; for birds, Emlen, 1952; Crook, 1961). If the optimum distance in Fig. 10 is the result of attractive and repulsive forces then it is clear that the attractive force is visual; if one observes two fish on either side of a glass sheet, they lie close against the plane surface or try to nose their way through, but if one slides the glass sheet away they take up a normal inter-individual distance. It seems unlikely that there is any optical information which does not pass through the glass sheet, suggesting that the repulsive force could be the water movements caused by the swimming of each fish, perceived by the lateral-line sense of the other. The work of Harris & van Bergeijk (1962) shows the lateral line to be sensitive to near-field displacements at a distance not greater than would fit the requirement of a repulsive force.

It is commonly stated that fish schools break up at night because vision is the only sense used in schooling behaviour, but there is good field evidence that this is an over-simplification. There must occur conditions where light intensity or water turbidity preclude visual contact between individuals. In these conditions it is suggested that the response to an odour gradient is one mechanism by which total dispersal of the individual fish in a school is prevented. Naturally the structure of the school changes as the main modalities involved change. The typical pelagic school with all individuals parallel cannot be maintained by responses to non-directional stimuli such as odour gradients. Therefore loosening of the structure into what some authors have called an aggregation probably occurs. Continuous production of odour at the centre of the school would offset turbulent diffusion processes which would continually tend to disperse the odour and consequently the fish. Provided that there remained a high enough concentration of odour at the centre of a water mass to ensure a gradient adequate for orientation by fish on the fringes, the school would be maintained during the critical period when vision would be inoperative. There is some evidence that a passive drift of fish in a water mass under these conditions does occur in the open sea (Jones, 1962). It is useful to consider the curves in Fig. 10 extended beyond the limit of visual contact between fish. The following tendency depends upon intervisibility and must fall to zero, but the exploratory and returning tendencies may remain constant as a pattern of activity and turning that are a part of the mechanism of orientation to odour gradients described in another paper (Hemmings, 1966). It was shown that the response strength in an odour-gradient situation was inversely related to activity. The experimental method was not designed to show the cause and effect relationship in this correlation, but two alternatives are possible: either the fish becomes less active when it senses an odour of a particular concentration, or alternatively when some other factor lowers activity a greater response is then shown to odours. If the latter were the case in those species that show a diurnal activity cycle with lower activity at night, then it might be expected that the response to odours increases and so the school does not break up. The role of the lateral-line sense in conditions of low visibility would be an obstacle-avoidance sense preventing fish bumping into each other. This is still in effect a ‘repulsive force’ and quite within the known limits of the lateral-line sensitivity. Thus the change-over from vision plus lateral-line by day to olfaction plus lateral-line at night does not require any great change in the role of the lateral-line sense.

Financial support for this work which forms part of a Ph.D. thesis submitted to the University of Cambridge, was received from the Development Commission in the form of a Fishery Research Training Grant, and from the Royal Society and the Philosophical Society of Cambridge for table space at Naples. I am indebted to Prof. C. F. A. Pantin for accommodating me in the Department of Zoology and to Dr P. Dohrn and the staff of the Stazione Zoologica for their assistance. I would like to thank Dr H. W. Lissmann for his help and supervision during the work, and Dr Lissmann, Dr F. R. Harden Jones and Mr B. B. Parrish for suggestions at the manuscript stage.