Results

Relationships between size distribution and depth

We observed a significant increase (p-value < 0. 001) in fish size (total length) with depth at the community level (Table I). The median individual size increased consecutively between the epipelagic, upper mesopelagic, and lower mesopelagic depth layers (median individual size equal to 7.0, 9.0, and 10.6 cm respectively; Figure 2). Median individual size then decreased slightly between the lower mesopelagic and bathypelagic layers, with a median individual size of 10 cm in the bathypelagic layer. Furthermore, while the first three depth layers had an unimodal distribution, the bathypelagic layer presented a bimodal distribution with a peak of around 8 cm and another around 13 cm.
The relationship between size distribution and depth was also tested at the species level (Table I). Only four of the 12 species showed a significant relationship. Melanostigma atlanticum, Lampanyctus crocodilus , and Xenodermichthys copei showed a significant increase in individual size with depth. Myctophum punctatum was the only species showing a significant decrease in the individual size with depth.
Lampanyctus crocodilus showed an increase in individual size between the upper and lower mesopelagic layers, from a median size (total length) of 10.0 cm to 12.0 cm (Figure 3). This was followed by a stabilisation between the lower mesopelagic and bathypelagic layers with the same median size of 12.0 cm. Melanostigma atlanticum showed a continuous increase in the size of individuals with depth, with median individual sizes of 6.0, 7.45, and 9 cm respectively.Xenodermichthys copei showed a maximum median size in the lower mesopelagic layer (= 10.2cm total length). Myctophum punctatumshowed small differences in individual size between depth layers, with a median size of between 6.6 and 7.0 cm. Relationships for other species are available in Appendix 3.

Relationships betweenδ 15N values and size

δ15N values were determined for seven migratory and five non-migratory or short migratory species, for a total of 682 individuals (= isotopic dataset). Mean δ 15N values ranged from 9. 49 ± 0. 57‰ for Serrivomer beanii to 12. 36 ± 0. 33‰ for Aphanopus carbo (Table II). For this dataset, mean values of standard length ranged from 6. 3 ± 1. 7 cm for Argyropelecus olfersii to 77. 3 ± 10. 8 cm for Aphanopus carbo , withSerrivomer beanii having the widest size range (45.0 cm between the minimum and the maximum length) and Lampanyctus macdonaldithe narrowest (3.3 cm).
The relationship between δ 15N values and individual size was first investigated at the community level (Figure 4). The linear model results showed a significant increase inδ 15N values with individual size. However, theR 2 was very low (= 0.01), indicating high variability in the values.
The relationship between δ 15N values and size was then investigated at the species level (Figure 5A). The results of the linear models showed that six species had a significant increase ofδ 15N values with increasing individual size:Myctophum punctatum , Melanostigma atlanticum ,Lampanyctus crocodilus , Stomias boa , Serrivomer beanii, and Aphanopus carbo . Only Arctozenus risso had a significant decrease of δ 15N values with increasing size.
The other five species had no significant relationship between the two variables (Figure 5B). The δ 15N values were stable regardless of the size increase. However, the coefficient of variations allowed to distinguish species with low interindividual variability (e.g., N. kroyeri ) from species with high interindividual variability (X. copei ).

Variance partitioning

Results of the variation partitioning analyses showed that five species (the same as above) had their δ 15N values significantly influenced by individual size (Figure 6, NB: S. boaand A. carbo not considered in these analyses due to small depth range). L. crocodilus had the highest proportion of variation inδ 15N values explained by size (25.7%), followed by M. punctatum (24.7%), M. atlanticum (10.1%),S. beanii (13.2%) and A. risso (5.8%). Alternatively, inA. olfersii and X. copei , variations inδ 15N were significantly explained by depth, at a proportion of 5.7% and 5.5% respectively.

Summary of relationships at specific and community levels

According to the different relationships, diverse species patterns can be described (Table III). First, two species (L. crocodilus andM. atlanticum ) showed an ontogenetic change both in their vertical distribution and theirδ 15N values. The largest individuals of these species were caught at greater depth and theirδ 15N values increased with individual size. Four other species showed an ontogenetic change in theirδ 15N values but no change in their vertical distribution (from trawling data): A. risso , S. beanii ,S. boa, and A. carbo . Among the species that did not show any significant change in their trophic ecology (according toδ 15N values) with increasing size, differences appeared in the dispersion of δ 15N values. Some species such as N. kroyeri and L. macdonaldi had a restricted range of δ 15N values whatever the size of individuals (i.e. variation coefficients = 2.15 and 2.68, respectively) whereas some species such as X. copei showed a high dispersion of δ 15N values (CV = 6.57). Finally, three species showed no relationships among the variables tested:N. kroyeri , L. macdonaldi , and S. koefoedi . However, for L. macdonaldi , a particularly low size range was sampled for isotopic analysis (= 3 cm between the minimum and the maximum individual size). At the community level, an increase in individual size with depth was observed, as well as an increase in δ15N values with increasing size (Table III).