Ecological-network models link diversity, structure and function in the plankton food-web

A planktonic food-web model including sixty-three functional nodes (representing auto- mixo- and heterotrophs) was developed to integrate most trophic diversity present in the plankton. The model was implemented in two variants - which we named ‘green’ and ‘blue’ - characterized by opposite amounts of phytoplankton biomass and representing, respectively, bloom and non-bloom states of the system. Taxonomically disaggregated food-webs described herein allowed to shed light on how components of the plankton community changed their trophic behavior in the two different conditions, and modified the overall functioning of the plankton food web. The green and blue food-webs showed distinct organizations in terms of trophic roles of the nodes and carbon fluxes between them. Such re-organization stemmed from switches in selective grazing by both metazoan and protozoan consumers. Switches in food-web structure resulted in relatively small differences in the efficiency of material transfer towards higher trophic levels. For instance, from green to blue states, a seven-fold decrease in phytoplankton biomass translated into only a two-fold decrease in potential planktivorous fish biomass. By linking diversity, structure and function in the plankton food-web, we discuss the role of internal mechanisms, relying on species-specific functionalities, in driving the ‘adaptive’ responses of plankton communities to perturbations.


Derivation of biomasses of Functional Nodes
The following simplified schematization of the overall amount of total carbon (integrated value, between 0 and -60 m) in our system at Green and Blue states was taken into account:

(Equation 1)
Total Carbon (mg m -2 ) = C-mesozooplankton + Particulate Organic Carbon (POC) + Dissolved Organic Carbon (DOC) (Eq. 2) POC (mg m -2 , in surface or deeper layer) = C-phytoplankton + C-mixotrophic protozooplankton + C-heterotrophic protozooplankton + C-bacteria + C-detritus POC concentration (expressed as mg m -3 , particle size > 0.7 µm and excluding planktonic animals) was estimated at LTER-MC at seven depths (0.5, 2, 5, 10, 20, 40, 60 m) on a weekly basis in the years 2007-2009. We took as reference for the global POC-budget data referring to years 2007 and 2009, during which the Summer-alternation between Blue and Green states was particularly evident 1 . The repartition of Total Carbon for the Green and Blue states in the surface and deeper layers in the abovementioned main components is shown in Table 1.

POC (z) = A • Chl a (z) + B
In which A • Chl a (z) is the portion of POC at depth z associated to phytoplankton and, partially (details in the following sections), to mixotrophic protists. A is the conversion factor, while B is the portion of nonphytoplankton POC, such as bacteria, heterotrophic protists and detritus. C-phytoplankton (z) was thus derived by multiplying Chl a by A. In order to obtain the conversion factor A, distinct regressions were carried out for the surface and deeper water-layers, with no distinction between Green and Blue states data. The latter option was taken based on two considerations: i) in distinct regressions on either Green or Blue data points, POC and Chl a were positively related, with similar trends, suggesting that Chl a was the main driver of POC increase at both oceanographic states; ii) due to the paucity of data, all available data were aggregated in a single plot to increase the statistical significance of the derived linear regression. The positive correlation for the surface dataset was significant (r=0.79; p=10 -13 ), and parameters for Equation 3 were: A = 94.25 and B = 122.55. A statistically significant positive correlation between POC and Chl a was detected for the deeper layer too (r=0.58; p=10 -8 ),for which A = 25.25 and B = 63.48.
The overall integrated amount of C-phytoplankton (expressed as mgC m -2 ) within the surface layer was then distributed into the phototrophic FNs (FN 1-12) using the following formula: Where B i is the integrated biomass (0-5 m) of a generic FN i, RB i is the Relative Biomass (as mgC m -3 ) of the generic FN as derived by multiplying the average concentration of FN present at LTER-MC at 0.5 m during the studied period at either Green or Blue states based on 1 by the average cellular biomass of the taxa included in the FN (according to [2][3][4]. The same groups of phytoplankton as in 1 were used in our model with the exception of Cyanobacteria and Prochlorophytes (FNs 1-2), which were introduced anew. In case of multi-specific groups, the average carbon per cell was estimated taking into account the quantitative proportion of each species in the group, based on not-aggregated abundance data at 0.5 m (not shown). The RB of prokaryotic photoautotrophs was estimated based on published works referring to GoN 5,6 . Overall, the estimation of RB i was independent from that of C-phytoplankton, while B i was a sort of integration between the two independent estimates. The multiplying factors in Equation 4 refer to the contribution of photo-autotrophy in mixotrophic FNs (see below). In the surface water layer, seven mixotrophic FNs were introduced, namely Mixotrophic nanoflagellates, Small dinoflagellates, Medium dinoflagellates, Myrionecta rubra, Tontonia spp. Laboea spp. and Strombidium spp., (i.e. . Being these organisms both autotroph and heterotroph, we considered one fraction of their biomass as 'phytoplankton' carbon (thus related to Chl a levels in the system) and a second fraction as included in the 'protozooplankton' carbon pool. The multiplying factors in Equation 4 represent the contribution of autotrophy to the metabolism of each FN (see Box 1). (FN 13) Since no information was available in the literature regarding growth and metabolism in mixotrophic nanoflagellates (FN 13, Ollicola vangorii), we assumed that 50% of the biomass production in this organism was carried out via photo-autotrophy and we thus set a multiplying factor of 0.5 in Equation 4. We believe that the latter was not a strong assumption, being the contribution of mixotrophic nanoflagellates in terms of biomass extremely exiguous in comparison with other mixotrophs. (FN 14-15) Small and medium mixotrophic dinoflagellates bear proper plastids, which are transmitted to their progeny after celldivision. Autotrophy could account for 59% of their growth rate and, consequently, of their metabolism (recalculations from data in Fig. 4 and Tables 3-4 in   7 ). (FN 16) Myrionecta rubra (= Mesodinium rubrum) is the most studied mixotroph in the plankton (e.g. [8][9][10] ). This ciliate performs raptorial feeding at expenses of small cryptophytic microalgae (a group of phototrophs of size < 10 µm, grouped in the FN 12) and, upon ingestion of the prey, it can retain plastids for photosynthesis (i.e. kleptoplastidity, 11 ). The potential contribution of ingested carbon to production in this ciliate was around the 10% 9,12 .

BOX 1. Definition of autotrophy/heterotrophy in the FWM
(  The oligotrichous ciliates Laboea spp., Tontonia spp. and Strombidium spp. engulf microalgae and retain their plastids but are not able to maintain plastidial-division 12 . Aplastidic cells were rarely reported in these genera, which relied on photosynthesis to cover a large part of the respiration demand and showed growth limitation in absence of plastids 12 . On average, the overall contribution of photosynthesis to production/biomass was around 55% 12 . A meta-analysis based on published data 13 estimated that, close to light-saturation (200 microE, comparable with the light amount in the surface layer of the GoN during summer), the carbon-fixation rate in plastidic oligotrichous ciliates had an average value of 5 fg C per µm -3 of ciliate-biovolume per hour, and this rate was independent from the size of the ciliate. Based on these assessments, the fraction of biomass obtained via photosynthesis at saturating light was 0.81, 0.64 and 0.83 in Strombidium, Laboea and Tontonia (FNs [17][18][19], respectively, considering a summer day-length of 14 hours.

Carbon biomass of mixo-and heterotrophic protozooplankton in the surface layer
The overall amount of Carbon in protozooplankton integrated within the surface layer was estimated indirectly using the formula: (Eq. 5) C-protozooplankton (in surface layer, mg m -2 ) = C-phytoplankton (in surface layer) • C-protozooplankton (at 0 m) • C-phytoplankton -1 (at 0 m) where C-protozooplankton (at 0 m) and C-phytoplankton (at 0 m) are the sum of RBs within protozooplankton and phytoplankton, respectively, and are expressed as mg m -3 . The derivation in Eq. 5 stems from the assumption that protozooplankton at 0 m and the overall amount of this carbon pool in the whole layer (0-5 m) was in the same ratio than phytoplankton at 0 m and the total phytoplankton in the whole surface layer (0-5 m) (based on observations in 14 ). RB i values were obtained by multiplying the carbon per cell in each FN based on ( 4 ) and the abundance of the FN estimated by microscope counting (data in 1 ). Being some FNs made of multi-specific groups, the average carbon per cell in each FN was estimated taking into account the quantitative proportion of each species in the group based on non-aggregated abundance data at 0 m (not shown).
Carbon allocated in a generic heterotrophic protozooplanktonic FN i within the surface layer (FN 20-25) was estimated as:

Carbon biomass of unicellular organisms in the deeper layer
The total amount of phytoplankton in the deeper layer (C-phytoplankton (in deeper layer)) was obtained by converting direct measurement of Chl a across that water-layer. Yet, protozooplankton biomass in the deeper layer was inferred following the same criterion used for the surface layer: where C-protozooplankton and C-phytoplankton are expressed as mgC m -3 . We assumed that, in the deeper layer, the ratio between phyto-and protozooplankton was the same as in the surface layer.
In the deeper layer, we assumed that the concentrations of unicellulars detected at one single depth were representative of the whole water layer, stemming from the observation that the water below the seasonal thermocline was permanently mixed in the GoN during summer 14 . The repartition of biomass among unicellulars in the deeper layer was carried out using the following formulas: (Eq. 9) Referring to Equations 9-11, the relative biomass (RB) of the two prokaryotic FNs in the deeper layer (FN [26][27] was calculated by multiplying the abundances (derived from 5 ) by the carbon cell quota (the same as for FNs 1-2). The RB of nanoplankton (including only eukaryotic cells) at station LTER-MC during summer were derived from 15 . Within the total nanoplankton concentration, 30% was represented by coccolithophores (FN 29; same carbon-per-cell quota as for FN 11), while diatoms were the least abundant group, accounting for about 5% of the total (FN 30; carbon-per-cell quota as the median value of FNs 4-10).
Only three categories of mixotrophs were introduced in the deeper layer (FNs 31-33; same carbonper-cell quota as for , with the exclusion of plastidic ciliates. It is plausible that the latter occur mainly in the surface mixed-layer of marine systems, being them i) limited by plastid-providing food and light and ii) not able to slow down their respiration rates in absence of light ( 12 and references therein).
The definition of biomass of Myrionecta rubra (FN 16) needs a specific discussion. This species was also excluded from the deeper layer biomass budget, based on the fact that M. rubra is able to migrate daily for several tens of meters along the water column and to set in the surface mixed-layer at mid-day ( 12 and references therein). Since sampling at LTER-MC was usually carried out at that time of the day, we hypothesized that sampled M. rubra populations at our long-term station were at the same concentrations at which they could be found at depth during other periods of the day. Thus, contrarily to other protists, which were divided into two distinct populations (namely, surface and deep), despite the time of the day, we considered M. rubra as the only unicellular organism with a biomass distributed along the whole watercolumn, alike for mesozooplankters (see next sections).
Abundances of mixo-and heterotrophic dinoflagellates in the deeper layer (FN 32-33 and FN 35, same cell carbon as for  were derived from 15 , who reported that dinoflagellates accounted for about 15% of the total abundance at 25 m in the GoN during summer. We sub-divided this amount among the three categories of dinoflagellates already considered in the surface layer (two of them being mixotrophic, one being heterotrophic), but proportionally to the abundance ratios among these three categories in the surface layer. RB for these FNs was derived as for the other FNs, thus multiplying abundance by carbon-cell quota.
The abundances of other heterotrophic protists in the deeper layer (FNs 36-39, same cell carbon as in  were derived using the abundance of heterotrophic dinoflagellates (FN 35) as a reference. Heterotrophic dinoflagellates in the deeper layer below the seasonal thermocline accounted for <15% of the total of dinoflagellates at 25 m in the GoN during summer 15 . Since data were not available for the other heterotrophic protozooplankters, their abundances were mathematically derived assuming that the abundances of heterotrophic protozooplankon in the deeper layer were in the same proportions as the abundances of these organisms in the surface layer. Once estimated the concentrations, RB for these FNs was derived as for the other FNs.

Carbon biomass of heterotrophic bacteria in surface and deeper layers
Heterotrophic bacteria were divided into surface and deeper layers' populations (FN [40][41]. These quantities were derived by regression from Chl a data, according to 6 , since a strict positive co-variation was described between these two POC compartments in the GoN. Bacterial cells concentrations were converted in units of biomass (mg m -3 ) by multiplying them by 0.020 pg C, i.e. the carbon-per-cell-quota for marine bacteria 16 . The discrete values were then integrated through the surface and the deeper layers taken separately, to get absolute units of biomass (mgC m -2 ).

Carbon biomass of Appendicularia houses
In the course of their life, Appendicularia build organic (mucous) 'houses' external to the animal body 36 . Once got engulfed by filtering activity, house are discarded by the animal that eventually secrete a new house and start feeding again. Appendicularia houses are thus a form of detritus characterized by a high nutritional value, rich of undigested microbial food, and they are eaten by a particular form of copepods (FN 55). The house renewal rates is independent from food concentration and ranges between 4.90 and 11 times the animal biomass per day, at food-concentrations spanning 60-2400 µg C L -1 (data recalculated from 26 ), with the minimum values being in the range of those found in our system. To avoid overestimations, we kept the most conservative values and estimated the biomass of houses in our system by multiplying Appendicularia biomass by 4.90 for both Blue and Green states.

Carbon biomass of faecal pellets
The amount of faecal pellets, as units of carbon, is proportional to the assimilation efficiency of an animali.e. the fraction of the carbon ingested which is utilized for respiration and growth. The complement to 1 of the latter carbon represents the carbon 'egested' as faeces. We distinguished between eight different classes of faecal pellets, namely those from calanoid copepods, juveniles of calanoid copepods, cladocerans, doliolids, salps, Oithona spp. (no distinction between early and mature life-stages), Detritivora and Carnivora. These groups were further agglutinated into three groups, large and slow-sinking faecal pellets (Carnivora 37 ), large and fast-sinking faecal pellets (Salps 36 ) and small faecal pellet (including all the others) (FN 60, 59 and 58, respectively). Differently from other biomass quantities in our model, faecal pellets where linked to rates of ingestion and assimilation of animals (see next sections).
The amount of faecal pellets present in the system (expressed as as mg m -2 ) was calculated as: where α and ε were the consumption rate and the fraction of consumed biomass not assimilated by the animal, respectively (see next sections). Since the assimilation efficiencies of animals depends to some extent from the food available, the values for these ratios were not equivalent in the Green and Blue foodwebs and thus also the amount of faecal pellets present in the system was different between these states. Moreover, since α and ε were represented by a range of variability (and not by a single value, see next sections dealing with production and feeding rates) in each of the system states, in calculating faecal pellets amounts, we used a single value for these rates, intermediate within the range of variability indicated above. Finally, assuming a fast physical degradation of large but loose faeces of chaetognats (Carnivora), the calculated amount of Carnivora faecal pellets were reduced to the 10% percent and the remnant 90% was assigned to generic particulate detritus.

Carbon biomass of generic particulate detritus
This quantity (FN 61) was derived by subtracting biomasses of all faecal pellets and Appendicularia houses from C-detritus (derived from Eq. 2); the latter included all uneaten and dead organic carbon derived from FNs mortality. The flow-to-detritus matrices for Green and Blue states are shown in Dataset 3.

Dissolved Organic Carbon
This component was assumed to be the 13% of the phytoplankton biomass, according to recalculations from 38 (see also details in the next section). DOC was divided between surface and deep layers (FNs 62-63).

Derivation of Production rate per unit of biomass (µ)
Values of Max-Min µ before MCMC and the final post-MCMC value are presented in Dataset 1.

Total phytoplankton µ
Primary production data collected at the LTER-MC almost weekly during summer months of 1984-1988 were analysed. The sampling was carried out as illustrated in 14 , at 0.5, 2, 5, 10, 20, 30, 40, 50, 60 m of depth. The particulate primary production (PP) was estimated using a standard 14 C method. The integrated particulate primary production (PP) for the two layers (0-5 and 5-60 m in depth) was obtained by computing discrete values. The 14 C method estimated only PP but we inferred also the amount of exudates release (ER) to get the total primary production (TP), which is the sum of PP and ER. According to 38 , ER accounted for 13% of total fixed carbon, on average (i.e. including all aquatic systems -recalculated data). By using the linear regression included in 38 with parameters derived for marine systems, the total primary production in our system was gained by the formula: (Eq. 13) TP (in surface or deeper layers) = PP (in surface or deeper layers) + 10^(-0.88 + 1.03 · log 10 (PP (in surface or deeper layers)) µ for surface and deeper layers were obtained by dividing TP by the total phytoplankton biomass (TB) estimated from Chl a with methods described in section 1. Moreover, we assumed that µ for photosynthetic FNs was equivalent to individual growth rates. Since the latter showed considerable intra-specific variability and were not completely coupled to available resources, we set the same range of variability of µ for the Green and Blue food-webs, considering that only small differences in productivity were detected between these two states.

µ of strictly phototrophic unicellular organisms in the surface layer
The maximum in-situ growth rates for Synechococcus (FN 1) was 1.2 d -1 during August 39 . Concerning Prochlorophytes (FN 2), maximum in-situ and in-vitro growth rates accounted for 0.6931 d -1 and 0.85 d -1 ( 40 and Chisholm unpublished data). Growth rates for eukaryotic phytoplankters were taken from a recent review 41 . Phytonanoflagellates (FN 3) included mainly Chlorophyceae having a maximum growth rate of 1.9 d -1 . Many surface-FNs in our model were composed of diatoms (FN 4-10), having a maximum growth rate of 2.25 d -1 41 . Since maximum growth rate was inversely related to diatom biovolume 41 , maximum growth rates for diatom-FNs were obtained following the allometric relation: (Eq. 14) Log 10 (µ (FN diatom)) = -0.25 * Log 10 (biovolume (FN diatom)) +0.75 derived from data in figure 3D in 41 .
The maximum growth rates of surface Coccolithophores (FN 11) was 1.7 d -1 41 . Concerning Phytomicroflagellates (FN 12), a heterogeneous class including chlorophytes and cryptophytes, we chose as maximum growth rate an intermediate value (1.5 d -1 ) between those of these two groups according to 41 .
The minimum µ for a generic photosynthetic FN i in the surface layer (1-12) was chosen taking into account the minimum value of primary production detected in the surface layer and during the specific state of the system (i.e. either Green or Blue), using the following equation: In which Min and Max µ for all phytoplankton were values reported in section 2.1.

µ of mixotrophic unicellular organisms in the surface layer
Differently from 'phototrophic' protists, ciliates are hard to cultivate and their physiology is largely unknown. The Max µ for mixotrophic nanoflagellates (FN 13) was set as for Phyto-nanoflagellates (FN 3), due to scarce background knowledge. The Max µ for mixotrophic dinoflagellates (FN 14-15) was set using the regression relating growth rate to metabolic state and cell-size reported by 7 , resulting in values of 1.17 and 1.02 d -1 for smaller and larger mixotrophic dinoflagellates (FN 14-15), respectively.
Myrionecta rubra (FN 16) can build considerable biomass in coastal-transition systems mainly at low salinity (about 10 PSU, 10 ). In saltier waters, the maximum growth rate was 0.52 d -1 in culture at 31 PSU 42 . Since it is plausible that the inverse relation between the growth rate and salinity is not necessarily linear, we set the Max µ of M. rubra at 0.52 d -1 , assuming this was the highest growth rate in marine coastal waters, even though salinity in our system of study was usually higher (around 37 PSU).
Scanty information was available for mixotrophic oligotrichous ciliates in the genera Tontonia, Laboea and Strombidium (FN [17][18][19]. The maximum growth rates for these were calculated according to the regression deriving maximum growth for mixo-and heterotrophic ciliates as depending from temperature and cell volume 43 , considering a temperature of 25°C (average value for the surface layer in the GoN during summer). The Max µ were 1.33, 1.54 and 1.35 d -1 for Tontonia, Laboea and Strombidium (FN 17-19), respectively. These data were in line with those reported in specific studies (e.g., 44 ).
The Min µ values for these mixotrophic protists was calculated by applying Equation 15, assuming a positive relation between mixotrophic productivity and that of all phytoplankton.
As for HNF (FN 20), a Max in-situ growth rate of 1.76 d -1 was reported in nature 45 .
Using the regression for growth rate in dependence to cell-size presented in 7 , the value of Max µ for FN 21 (Equivalent Sphere Diameter, ESD = 22 µm) resulted 0.95 d -1 .
Growth rate data for prostomatids (FN 22, cell size > 20 µm) were scarce in the literature. The physiological and feeding response under a monospecific diet were studied for the prostomatid ciliate Tiarina fusus and values of Max growth rate at 19°C were reported to range 0.1-0.47 d -1 , depending on the typology of prey 46 . By applying a Q 10 of 2.8 -an average for ciliates, according to 47 -the Max µ at 25°C resulted 0.98 d -1 .
Being information in the literature scarce for this organism, the maximum growth rate for the heterotrophic oligotrich Strobilidium spp. (FN 23) was derived using equation indicated for heterotrophic oligotrichous ciliates by 43 . The Max µ at 25°C was 1.64 d -1 .
A wider literature was available for growth rates of Tintinnids (FN 24). In a two-years observation, the Max growth rate resulted 1.66 d -1 for the whole tintinnids' community during summer 48 .
Nanociliates (FN 25, cell size < 20 micron) were a wide group in terms of taxonomy, including both prostomatids and other ciliates. Information about their growth performances were scarce and we took data from 49 , which reported, for a nanociliate assemblage, under laboratory conditions and relaxed intraguild predation, Max growth rates up to 2.36 d -1 .
The Min µ for all FN within heterotrophic protozoplankton was calculated with Equation 15, assuming that production in these organisms was in direct relation with production in photosynthetic organisms.

µ of unicellular organisms in the deep layer
The Max and Min µ for each unicellular FN, except diatoms, were derived using the following formulas, assuming that specific µ were correlated with µ of all phytoplankton in the deeper water-layer:  (FN 4-10)), respectively. Median µ for all phytoplankton community were taken from section 2.1.

µ of mesozooplankton
Life history of the cladoceran Penilia avirostris was extensively studied in the Mediterranean Sea and along Brazilian coasts [50][51][52] . This animal (FN 42 in our model) showed extremely high population growth pursued by parthenogenetic reproduction during summer 50,51 . Juveniles of cladocerans were mini-adults, occupied a niche comparable to the latter and grew very rapidly to the adult stage (<3 days). In our model, the Max-Min µ values for P. avirostris were set assuming that population increasing was dependent only on the animal birth rate and the main boost to population growth came from parthenogenetic eggs' hatching, instead of from somatic growth, alike in other planktonic animals (e.g. copepods). Since the relation between growth rates and food-environment was not studied in detail for this species, one and a single µ interval, derived from the above-mentioned studies, was considered for both Green and Blue states (growth rate ranging 0.10-1.37 d -1 ). Yet, scanty information was available in the literature for Mediterranean Evadne spp. and Pseudevadne tergestina, the other cladocerans included in our model (FN 43). We thus assumed that Min and Max µ for these animals were the same of those set for P. avirostris, in both Green and Blue states.
Differently from cladocerans, copepod life-histories included several different stages. Population/individual growth rate was a good proxy of these animals' productivity (e.g., 53 ). Copepods' µ was parameterized by the weight-specific fecundity, assuming that adult-female production was completely realized by means of egg production, without considering any changes in individual bodybiomass in adults at short-time scales 54 . Yet, juvenile-copepods' µ was represented by the weight-specific growth, since early stages underwent changes in individual biomass at short time-scales 54 . µ for both adults and juvenile copepods in our FMW were set according to an extensive synopsis of growth rates of copepods and by using mathematical relations presented therein and linking growth rates to i) bodybiomass of the species, ii) temperature and iii) food concentration 54 . Since copepods occurred all over the water column in our FWM, we set Max-Min µ values as the rates predicted for temperatures and food conditions present in the environment at surface-deeper layers in the Green and Blue states, respectively (T ranging 25-15°C and food ranging 203-21 and 37-17 mgC m -3 , respectively). Being Juvenile copepods in our model a mixture between the two genus Paracalanus and Clausocalanus, rates averaged from those referring to the latter were compiled.
Unlike cladocerans and copepods, pelagic tunicates, including Appendicularia, Doliolids and Salps (FN 50-52), reproduce only once in a lifetime (they are semelparous) and the organisms usually die after reproducing 36 , they have no larval stages and newly born individuals have feeding performances and ecological role comparable to adults and, in some cases, a relatively large body-size ( 36,55 and reference therein). Nonetheless, adults undergo an intense somatic growth, which could be anyway contemplated when setting the µ. The growth rate of pelagic tunicates has been represented by the so-called 'lifetime fitness', a parameter that in population ecology studies is commonly used as synonym of 'recruitment' 55 . Lifetime fitness expresses the maximum intrinsic rate of natural increase and it is determined as ln(egg produced per animal)/(development time from fertilization of an egg to spawning by the organism hatched from that egg). Thus, the lifetime fitness integrates both the timing of egg release and the somatic growth that precedes spawning. This parameter was, to our opinion, thoroughly suitable to model µ in pelagic tunicates at the time scales of our model.
The growth performances of Appendicularia (FN 50) have been extensively explored at laboratory conditions. A life-cycle-based model was built for Oikopleura dioica -that is the main appendicularian species present in the GoN during summer 29 (plus unpublished data presented in 55 ): therein, lifetime fitness was strongly influenced by both temperature and food concentration. Appendicularia, and pelagic tunicates in general, are particularly sensitive to high food concentrations: being them strong filter-feeders, they easily undergo super-saturation/engulfment, with negative effects on reproduction. Food availability to Appendicularia in our system do not exceeds 200 mg C m -3 and these animals remain well below the saturation threshold (> 300 mg C m -3 ). By taking O. dioica as a reference 29  Among Doliolids (FN 51), the genus Doliolum was the most represented in the GoN during Summer and included mainly the species D. nationalis, which features a bi-phasic life-cycle in which asexual production of new individuals could strongly shorten the generation time 36 . Growth rates of doliolids was studied less extensively than Appendicularia. A lifetime fitness of 0.26 d -1 at 20°C was estimated for D. gegenbauri 55 , that also showed a limited dependence from the food-concentrations 56 . Assuming a linear dependence between temperature and growth rates in the range 15-25°C, this parameter was 0.20-0.33 d -1 between 15 and 25°C (based on Q 10 assumed for pelagic tunicates 57 ). This rate referred to the complete life-cycle of D. gegenbauri, encompassing different life stages (i.e. up to six). Yet, Doliolum nationalis was capable to run a further single-stage cycle thanks to asexual reproduction. Since no estimation of growth rates for D. nationalis has been done so far, to our knowledge, we decided to assign a Max µ of ~1.9 d -1 , assuming that, during the short cycle, Doliolids had growth rates similar to those of Salps, also capable of asexual growth 36 . Due to the uncertainties in the definition of the growth-rate range in Doliolids (FN 51), we assigned Max-Min µ values of 1.9-0.20 d -1 in both Green and Blue food-webs.
Information about growth rates of Salps (FN 52) was even scantier. For Thalia democratica, the most represented species in the GoN during summer, the lifetime fitness derived from different published works ranged 0.18-1.9 d -1 in a range of temperature of 14-22°C, with the lower values measured in the lab and the highest ones estimated from the field ( 55 and references therein). Nonetheless, a 1.9 d -1 could theoretically occur only in particular circumstances, such as massive salp outbreaks in nature ( 55 and references therein). Yet, lower values of growth rate were probably biased by the confinement of animals in a limited space in the laboratory. Considering these issues, we decided to not take any assumption in establishing salps' µ and we considered the whole range of variability of their lifetime fitness, setting Max-Min µ at 0.18-1.9 d -1 , for both Green and Blue states.
Meroplankton (FN 53) included planktonic larvae of benthic organisms with life-cycles longer than those of holoplankters in general. Assuming that eggs and larvae of benthic organisms were always present in the summer plankton of the GoN, we used the rate of somatic growth as the unique proxy of µ for meroplankton, since we were not interested in the all lifetime fitness (i.e. recruitment) of these animals. Growth rates ranging 0.025-0.15 d -1 were reported for larvae of the polychaete Polydora ciliata at 16°C in the laboratory 58 , with the highest rates gained at food saturation, i.e. at food concentration of about 1 mg C L -1 , a value about double the highest concentration of POC in our system of study. Applying the 2.8 Q 10 by 47 , the rates were ranging 0.07-0.42 d -1 . Due to the scarce knowledge on the relation between growth, food and temperature in meroplankton, we assigned an identical Min-Max µ range (0.025-0.42 d -1 , respectively) in both Green and Blue states.
No information was gathered for the growth rates of Carnivora mesozooplankton -i.e. chaetognats (FN 60). µ were interpolated by Ecopath based on these animals' consumption rates.

Derivation of Consumption rate (α) and diet
For each FN, Max-Min α ratios were selected, in each of the two states (i.e. Green and Blue), as derived from mass-specific ingestion rates taken and recalculated from the literature. In case of mixo-/heterotrophic unicellular organisms present in the model in two distinct populations, in either surface or deeper layers, Max-Min α were indicated for each layer. Diet-matrices were built considering information from the literature. Yet, since feeding preferences were important in defining biological differences among FNs, in our model, we set both the overall spectrum of food eaten and a tentative food-selectivity based on expert knowledge. We also set a % of variability to the strength of each trophic link depending on the strength of the specific background information. As for other trophic parameters, α values and trophic linkweights for the final input data of the model were defined by means of uncertainty analyses via a MCMC procedure performed at the level of the all web and considering all organisms involved. Values of Max-Min α before MCMC and final value after MCMC are presented in Dataset 1. Diet-matrices before and after MCMC are presented in Dataset 2.

α and diet of nanoflagellates
Few information were found about specific ingestion rates of mixotrophic flagellates, mainly Ollicola vangoorii (FN 13 and FN 31, in surface and deeper layers, respectively), present in our FWM. We assumed that Max-Min α in FN 13 and FN 31 were at least twice the Max-Min µ, at both surface and deeper layers. Bacterivory was the dominant feeding behaviour in this nanoflagellate 59 .
Few studies were available in the literature concerning feeding activities of Heterotrophic Nano-Flagellates (in our model, FNs 20 and 34, at surface and deeper layers, respectively). In our model, Max-Min α for HNF were set at 1.8-0.3 d -1 , based on Callieri et al. 60 . Since the latter study was conducted over a wide temperature range, i.e. 9-20°C, we assumed that the α-range above was valid for both surface and deeper layers. Concerning diet, HNF assumed 12-60% from heterotrophic bacteria, 0.6-19% from Synechococcus and 0.02-21% from Prochlorococcus 61 . Some HNF species also grazed pico-eukaryotes with a size <3 µm 62 . The latter were partially represented in our system of study within FNs 3 and 28. We included in HNF diet all these food-items, giving however much relative weight to trophic links involving prokaryotes. Since it was not possible to disentangle the effect of temperature on the ingestion rate from the information available in the literature, the α ranges described above were set for both water layers considered in our model.

α and diet of dinoflagellates
Concerning dinoflagellates' diets, these organisms ate within a wide array of prey, according to meta-analysis on published data 7,63 . The prey of small mixotrophic dinoflagellates (FN 14 and 32) spammed from bacteria and prochlorophytes (ESD < 1 µm) to other small-sized dinoflagellates (ESD within 10 µm; cannibalistic feeding) and diatoms. The prey-range of larger-sized mixotrophic dinoflagellates (FN 15 and 33), in turn, included all other unicellular organisms, except large ciliates, which showed a cell-size apparently out of the range of mixotrophic dinoflagellates' feeding (ESD > 40 µm, in our model). Heterotrophic dinoflagellates (FN 21 and 35) ate over a prey-range wider than that for larger-sized mixotrophic dinoflagellates. Dinoflagellates' diet in our FWM was further improved by using information about the optimal-prey size for these organisms in laboratory conditions 7 : i.e. mixotrophic and heterotrophic dinoflagellates showed better growth performances when predator-prey-size ratios ranged 0.8-5.8 and 0.4-5.3, respectively. Moreover, a further weight to trophic links was set based on the observation that the higher ingestion rates were got when predator to prey size ranged 0.8-1.4 and 0.9-2.6 for mixotrophic and heterotrophic dinoflagellates, respectively 7 . The identical diet was assumed for both surface and deeper layers in our FWM.

α and diet of mixotrophic ciliates
The feeding habit of mixotrophic ciliates (FN 16-19) was relatively well known. They all showed kleptoplastidity (i.e. the retention of chloroplasts from a prey and their utilization for photosynthesis) and were considered as 'obligate mixotrophic', since they needed to engulf small microalgae to support their photosynthetic activity 12 (further details in Box 1). As already discussed above, mixotrophic oligotrichous ciliates  were introduced in the sole surface layer, while M. rubra population (FN 16) was widespread all over the water column. Thus, for each of these organisms, only one range of Max-Min α values was set.
In our FWM, we assumed that Myrionecta rubra (FN 16) behave mainly as a phototroph, being summer a favourite period for growth and cryptophytes always available in the GoN. The maximum ingestion rate reported for M. rubra accounted for 13.1 % of the ciliate body-mass per day at 18°C (recalculated data from 42 ). Thus, a Max α of 0.131 d -1 was set, despite the difference of temperature between the experimental conditions at which it was measured (i.e. 18°C) and the temperature at which M. rubra was supposed to feed in the field in our model (25°C, surface layer, where cryptophyte showed the highest availability). In fact, a Q 10 specific for M. rubra and other mixotrophic ciliates has never been established and we preferred to not use a generalist one, like that defined by 47 for ciliates. On the other hand, by adopting a Max α of 0.131 d -1 , we kept a convergence between ingestion rate and the fraction of metabolic requirements gained via heterotrophy (i.e. 10%, see former sections) in M. rubra. Finally, since this organism performed photosynthesis intensively, the Min α was assumed to be close to zero, considering the possibility of M. rubra to induce division in captured plastids.
Specific ingestion rate for Strombidium rassoulzadeganii approached 3 d -1 at concentrations of food comparable to those found in our system in the surface layer (<0.5 g C m -3 ) but under a temperature of 19°C 44 . We set Max α for Strombidium spp. (FN 19) at 3 d -1 , despite temperature differences, for the same reason as above. Min α was set at 0.19 d -1 , since at saturating light (conditions at which the ingestion rate was at its minimum and photosynthesis at its maximum), the cell covered about the 81% of its metabolic need via photosynthesis (see Box 1). Concerning diet, the genus Strombidium performed kleptoplastidy at expenses of prasinophytes, cryptophytes, haptophytes and chlorophytes ( 12 and refs therein), these groups being included in FN 11 and 12 in our model. Moreover, a ciliate with an ESD of about 30 µm, alike Strombidium, ate different kinds of particles, from bacteria to the largest ciliates 63 . Nonetheless, since microalgae in the above-mentioned classes were crucial for the ciliate to pursue its phototrophic metabolism, we assigned a relatively higher weight to them in FN 19 diet-matrix in our FWM.
Laboea strobila (FN 18) ingested up to 5.9% of its biomass per hour, giving a Max α of 1.42 d -1 12 , assuming that ingestion rates were equivalent during day and night (but this is not always true, see 64 ). Min α was set at 0.36 d -1 by assumption, following the same rationale adopted for the other oligotrichous ciliates. Concerning diet, alike Strombidium, L. strobila, with an ESD of about 44 µm, ingested from bacteria to the largest ciliates. Yet, a higher ingestion of flagellate microalgae in FNs 11-12 was largely reasonable, for the reasons already discussed for Strombidium.
No information was available for the genus Tontonia (FN 17). The consumption rate was set at 3 d -1 by assumption -i.e. it was identical to that of Strombidium, for similarities of cellular biomass and metabolism (i.e. phototrophy/heterotrophy). Min α was set at 0.17 d -1 , according with the rationale above. Diet in FN 17 was identical to the other two oligotrichous plastidic ciliates.

α and diet of heterotrophic ciliates
Scanty information was available for Prostomatids (FN 22 and 36). The functional response for Tiarina fusus at 17°C was studied in the lab 46 . The highest specific ingestion rates were gathered when the species was fed with the dinoflagellate Lingulodinium polyedrum, a species with an ESD ~31 µm, while ingestion was considerably lower with smaller preys. T. fusus reached saturation for food concentrations above 1 gC m -3 (value never reached in the GoN during summer), while the weight-specific ingestion rate below this threshold was 0.21-2.1 d -1 . This range was taken as the Min-Max α range in the deep prostomatids in our model (i. e. FN 36). Considering a Q 10 of 2.8 specific for ciliates ( 47 and references therein), α for prostomatids in the surface layer (FN 22) was 0.77-4.7 d -1 .
Concerning diets of prostomatids, according to meta-analyses, prostomatids with an ESD of about 54 µm (as in FWM) could ingest all micro-plankters, from bacteria to ciliates 63 . We gave a relatively higher weight to trophic-links with food-items having an ESD > 12 µm (based on 46 ). In a laboratory experiment, diatoms were not selected for by a prostomatid ciliate and a Strobilidium species 65 , suggesting that these algae represent a low-quality food for ciliates. To be more neutral, due to the paucity of data, we let prostomatids eat also diatoms in our FWM, although we imposed a relatively lower predatory pressure over them.
Concerning Strobilidium (FN 23 and 37, in surface and deeper layers, respectively), rates of weightspecific ingestion ranging 0.03-0.16 h -1 were reported in the literature for Strobilidium cf. spiralis fed different kinds of preys 66 : the higher value in the range was gained at saturation, corresponding to a concentration of food of 250 mgC m -3 , a value in the range of our system of study during summer. Assuming that ingestion rates at light and dark periods were comparable for Strobilidium (according to 64 ), Max-Min α for Strobilidium in our model (FN 23) were set at, respectively, 3.84 and 0.72 d -1 (based on 66 ). This range referred to a single temperature (i.e. 20°C, intermediate between our environmental extremes) but we considered it as representative for both surface and deeper layers populations (FNs 23 and 37). Concerning diet-matrix, being Strobilidium size comparable with that of prostomatids, the same prey-range was applied to it.
Concerning Tintinnids, ingestion rates at 25°C for Tintinnopsis acuminata ranged 0.73-7.3 d -1 (recalculated data from 67 ), assuming that ingestion rates during day and night were comparable (no info available). These were our Max-Min α for Tintinnids in the surface layer (FN 24). Applying a Q 10 of 2.8, α ranged 0.26-2.61 d -1 for Tintinnids in the deeper layer (FN 38). Diet of tintinnids was established based on the dimension of the tintinnid assemblage present in our system. The maximum particle size ingested by tintinnids was around 43% of the diameter of lorica at oral opening 68 . The average diameter for FN 24, weighted to each species relative abundance, was about 20 µm. The probable maximum size edible by tintinnids in our model was thus 9 µm. Moreover, in the laboratory, tintinnids eating phytoflagellates (FN 12 in our FWM) grew at rates similar to those eating diatoms lacking significant external processes (FN 7 in our FWM) 69 . Moderate growth was observed when tintinnids ate dinoflagellates of edible size (FN 14) while extensive mortality was reported in tintinnids fed with Synechococcus (FN 1) 69 . The diet of FN 24 and 38 was set based on considerations above.
Nanociliates (FNs 25 and 39, in surface and deeper layers, respectively) in our model included all ciliates with a cell-size < 20 µm (average ESD = 16 µm). The marine nanociliate Uronema (ESD about 12 µm, a size comparable to those in our model) ingested up to 31 Synechococcus cells ciliate -1 hour -1 at light conditions 70 . Since the ingestion rate by the nanociliates was 120% higher in dark than in light conditions (data for Balanus comatum, 64 ), we estimated that the whole amount of cells ingested per day by one nanociliate was 68.2, giving a specific rate of ingestion of 1.89 d -1 (recalculation of original data). Since the experimental data considered were gained at 20°C (intermediate between 15 and 25°C) 70 , we set 1.89 d -1 as the Max α for both the surface and deeper populations (FN 25 and 39). Since the functional response was unknown for nanociliates, the Min α at surface and deeper layers were assumed to be proportional to the Max α and the ratio between Min and Max α of other heterotrophic ciliates (i.e. Prostomatids and Strobilidium). The assumed Min α for FN 25 and 39 was 0.33 d -1 . Concerning diet of nanociliates, published meta-analyses indicated that they had been always tested in feeding experiments with cell-preys smaller than 9 micron ESD 63 . We were confident that that value represented an upper threshold-size for nanociliate feeding. Furthermore, the prostomatid Uronema spp. ate preferentially more Synechococcus than Prochlorococcus cells and heterotrophic bacteria 70 .

α and diet of cladocerans
The feeding behaviour of Penilia avirostris (FN 42) from the Mediterranean Sea was studied in the laboratory and weight-specific ingestion rates ranged 0.26-1.57 d -1 at Summer conditions (natural photoperiod and temperature 22-27°C) 17 . A wider range for the weight-specific ingestion rate, i.e. 0-7 d -1 , was reported at about 22°C 18 , with Min and Max values corresponding to food concentrations of about 0 and above 250 μg C L -1 (saturating condition). A value of 3.50 d -1 (recalculated data) was reported at about 203 μg C L -1 , i.e. the maximum potential food concentration in our model system (including the bacterial fraction too). Since mesozooplankton in our model was able to go up and down the seasonal thermoclyne, Max and Min α values for P. avirostris were chosen at the extremes of the natural food range in the two food-webs. Based on regression derived from the literature 18  Concerning diet, P. avirostris showed a lower prey-size of about 1 µm, corresponding to the ESD of the cyanobacterium Synechococcus ( 17 , grazing experiment using natural diet). The highest grazing pressure was exerted on flagellates, diatoms and dinoflagellates, while ciliates were grazed at lower rates due to their motility and larger size 17 . Most diatoms would be excluded from the preferred food of cladocerans, since long cell-colonies are not easily managed by these animals. Yet, we chose to consider all diatoms as a source of food to cladocerans based on the observation that these microalgae produced short or no chain in the GoN during summer (Sarno & Zingone personal communication). The diet-matrix for Penilia was set according to grazing-selectivity gained in experimental observations 18 . These reported: i) the highest grazing rates (selectivity index W' = 1) for food particles between 15 and 70 µm in size, corresponding to 6-28 micron µm and ii) no selection for bacterioplankton (neither auto-nor heterotrophic), although passive interception of these feeding particles was also possible. The grazing pressure on particles between 28 and 40 µm ESD and below 6 µm, excluding bacteria, was lower (W' =0.6-0.74). Conversely, in the range 40-60 µm ESD, selection was the lowest possible (W' = 0.25). Above 60 µm ESD, no feeding was detected.
The feeding performances for the other cladocerans in our model, including the genera Evadne and Pseudoevadne (FN 43), had been less investigated in comparison to P. avirostris and we set the ingestion rates for FN 43 equal to that set for FN 42. However, food-selectivity data for Evadne normanni suggested that selection by this species was operated on food particles different from those selected by Penilia 71 : the highest grazing pressure was exerted on particles with an ESD between 40 and 60 micron (W' = 1) and between 3 and 6 micron (W' = 0.76) (recalculated data).

α and diet of calanoid copepods
Feeding performances of copepods have been extensively studied at both in situ and laboratory conditions. By compiling data from several published papers, some authors conducted meta-analyses to sort for global relations between feeding rates and i) body-biomass, ii) amount of food available and iii) temperature 72,73 . It was suggested that, in this order, food concentration and body weight were the main drivers of changing of feeding rates of copepods in nature. Food concentration alone explained 52% of the variance in the (log) ingestion rates. Temperature had a limited influence of ingestion rates. Based on experimental and field data, multiple regression analysis were described to derive mathematically the ingestion rate from these driving factors 72,73 . We used these mathematical relations to find the most probable values of α for the main calanoid copepods in our model, i.e. P. parvus, A. clausi, T. stylifera, C. typicus, other calanoids and juvenile calanoids (respectively, . Each of these FNs were relatively small copepods, i.e. having a body-weight < 10 μg C ind -1 . Known variables in our calculations were i) FNs' body-biomass (see Table 1) and ii) the amount of carbon allocated into microbial FN with ESD ≥ 3 μm -i.e. the lowest threshold size for food particles available to copepods -at Green and Blue states and in the surface and deeper layers (i.e. 203-21 and 37-17 mgC m -3 , respectively). Data of feeding rates for copepods are shown in Dataset 1.
Calanoid copepods in our FWM ate wide range of prey, i.e. all unicellular organisms with the exception of prokaryotes. Nonetheless, selectivity was modulated differently for each copepod species according to the feeding preferences listed in the next paragraphs (Diet matrices in Dataset 2).
Paracalanus spp. (FN 44) collected small particles (< 5 µm, ESD 2 µm) only passively 74 ; yet, in the laboratory, animals appeared to select for diatoms < 30 μm in size in a mixed-diatoms diet 20 . Information for Paracalanus feeding was anyway scanty and we assigned relatively higher preferences towards small microalgae and lower towards protozooplankton.
Prey of Acartia clausi (FN 45) ranged 7.5-250 µm in size 18 . Yet, selectivity was operated in the range of 70-100 µm in size, thus over mainly large diatoms, which were the most abundant food typology in the experimental observation considered herein 18 . It was thus possible that A. clausi was a peak tracker (i.e. it selected for the most abundant food). To overcome these uncertainties, we assigned low feeding selectivity to A. clausi in our model with the only exception of a extremely low pressure exerted over smaller organisms (ESD ~ 3 µm, i.e. nanoflagellates).
T. stylifera (FN 46) was reported to select for cells larger than those selected by A. clausi -i.e. particle retention efficiency was accounting for 100% for cell-ESD > 10 µm, and only ~25% for cell of 5 µm 75 . T. stylifera individuals from the GoN selected mainly cells and colonies ranging 30-200 µm 20 . In our model, the grazing pressure was established based on observations above: the highest pressure was on particles exceeding 10 µm in ESD, moderate pressure between 7.5 and 10 µm, lower between 7.5 and 5 µm, and the lowest for ESD smaller than 5 but larger than 2 µm.
Laboratory studies have reported C. typicus (FN 47) feeding upon a wide size-range of organisms including small algae 76 , Appendicularia juveniles and eggs 77 and juvenile copepods 78 . The optimal preysize for C. typicus was represented by particles >10 µm, but regarding the upper size limit, this copepod ingested prey larger than its own size, like yolk-sac fish larvae ( 79 and references therein). Differently from T. stylifera, which was reported as a cruiser/suspension feeder, C. typicus was capable also of ambush predation and selected for large motile preys, such as ciliates or dinoflagellates, both in nature and laboratory 79 . In our model, we gave i) the highest selectivity rank to motile preys exceeding 10 µm in ESD, ii) an intermediate value to non-motile preys above 10 µm and iii) the lowest rank to those below this threshold.
The diet of other calanoids (FN 48), dominated by the small copepods Clausocalanus spp., was assumed similar to that of P. parvus for taxonomical affinity between the latter and the copepods species included in FN 48 (i.e. small copepods in the genera Clausocalanus and Paracalanus).
Trophic behavior of copepodites of calanoid copepods (FN 49, including mainly Paracalanus and Clausocalanus) was set similar to that of adults of Paracalanus. Concerning diet of Oithona spp., according to 82 , faecal pellets provided the 20-30% of daily ratio of these animals, but Oithona spp. ate also copepod juveniles, so that carnivorous feeding accounted for at least 25% of the daily ratio 83 . It was also reported that ciliates accounted for at least the 25% of the daily up-taken carbon and that these protists were anyway preferred within the microplankton 84 . Moreover, ciliates and dinoflagellates represented up to 80% of the diet of O. similis 85 . Concerning phytoplankton, a lower food-size limit of 10 µm was reported 86 . To this respect, we assumed that the diet of Oithona included 40% of ciliates and dinoflagellates, 25% of detritus (as faecal pellets), 25% of juvenile copepods and 10% of phytoplankton (size > 10 µm).

α and diet of Appendicularia
The bioenergetics of the appendicularian Oikopleura dioica was described in detail in laboratory conditions (T=15°C) 28,29 . The Max-Min α in our model for Appendicularia (FN 50) were defined based on published regressions 28 ( Table 2 in that paper), by relating α to the body-weight of these animals in our model and the four key food-concentrations at the two system states (see above). Values for surface layers have been increased according to a Q 10 of 1.78 specific for grazing rates of O. dioica 57 .
Being filter feeders, Appendicularia ingested all particulate material, from bacteria to microzooplankton, and even small detritus particles. In our model, we excluded detritus from the spectrum of food available to these organisms for some reasons: firstly, in surface waters, during the physically stable summer season and in the course of phytoplankton matter production and retention in the mixed layer, aggregates exceeding the size of appendicularians considered in the model (~ 1 mm) are more frequently developed; secondly, in the deeper layer, a large fraction of detritus would be composed by large and heavy faecal pellets sinking to the bottom. The weight-specific clearance of Oikopleura spp. was put in relation with the size of food particles: this rate showed the highest values for particles with an ESD between 20 and 60 µm 87 . Above this range, the clearance was 10 times lower; below it, the rate was 100-10 times lower than in the range mentioned above. In line with these results, weight-specific ingestion of eukaryotes with ESD <13 µm was reported almost three times higher than that of prokaryotes 88 . Diet of FN 50 was arranged accordingly.

α and diet of doliolids and salps
The functional response (ingestion vs. food concentrations) was described in detail for Doliolum denticulatum from the North Western Mediterranean in individuals cultivated in laboratory conditions (T ~22°C) and fed with natural microplankton community 18 . These animals had a size comparable to that of D. nationalis, the doliolid species most frequently recorded in the GoN. The α rates of Doliolids in our FWM (FN 51) were estimated based on regressions derived from the study above at the four key concentrations in our model.
Doliolids' diet included food-particle within a wide range of size, from bacteria to microplankton. A higher selectivity was reported for particles between 2.5-7.5 µm in size (Selectivity index W' = 1) 18 . Nonetheless, relatively high indexes were also reported for slightly smaller and larger particles (W' = 0.99 and 0.94 for particles around 2.5 µm and ranging 7.5-15 µm in size, respectively), while particles larger than 100 µm were not grazed. Selectivity indexes for picoplankton (including bacteria) and for microplankton between 15 and 100 µm were around half the maximum mentioned above, thus relevant. Since in our model only a few tintinnid species exceeded 100 µm in the longest linear dimension, we included all pico-, nano-and microplankton in Doliolids' diet in our FWM, giving different weights to the different trophic links based on the considerations above.
Ingestion rates of Salps (FN 52) have been less easily investigated than those of Doliolids, since, in comparison with the latter, salps are bigger organisms hardly kept alive at laboratory conditions. Salps with a size < 10 mm (the dimensional category more easily detected in mesozooplankton samples at the coastal station MC) showed weight-specific ingestion rates ranging 2-8.3 % of the body carbon per hour (α ~ 0.24-1 d -1 , assuming that animal ate 12 hours a day) ( 36 and reference therein). These values were in the range of those gained in nature for salps at extreme trophic conditions -bloom vs. non bloom conditions in the Southern Ocean -(i.e. 0.1-1 d -1 , respectively, 89 ), thus at temperature <15°C. As for salps living at higher temperature (like in the Gulf of Naples), assuming a Q 10 of 1.78, the Max α for both Green and Blue conditions was 1.78 d -1 . In our model, we assumed that α ranged 0.24-1.78 d -1 for salps, thus enough to sustain the potentially high µ rates assumed for these organisms (see section 2.6). These values were also in the range of those gained in other studies on salps (e.g. 90,91 ). Moreover, based on data regressions 91 , saturation of salps' filter-feeding was gained at about 10 µgC L -1 . Since in our system of study the concentration of food available to salps always exceeded that threshold, we adopted the same value of Max α for both Green and Blue conditions. Salps are non-selective feeders, eating from bacteria to the largest microplankton and small detritus-aggregates. In our model, we did not included detritus in the diet of salps for the same reason we did for doliolids. In the course of a feeding experiment with microspheres, a high affinity for particles with size < 3 µm, i.e. bacteria, virus and colloids, was reported for the gut of salps, suggesting that these could be an important source of food at low microplankton concentration 92 . In the experiment mentioned above, about the 60 % of particles in the gut of the tested salps were within 1 µm in size at concentrations in the order of 10 3 cell ml -1 , thus comparable to those found in our system at any state and depth of the water column. According to the authors of that study, even though flagellates and small diatoms (particles > 1 μm) represented a larger carbon-pool in nature due to their abundance, prokaryotes and smaller particles (in the range 0.1-1 μm) were more quickly digestible because they presented a relatively higher surface area and they could fully satisfy salp energetic needs. Although these observations were not easy to be generalized for our modelling purposes -since they did not considered particles larger than 3 µm (the majority in our model) -they supported the view on possible selective feeding in salps. Thus, we included in our model the constraint that the 60% of the daily ration of Salps (FN 52) came from bacteria, cyanobacteria and prochlorophytes.

α and diets of meroplankton, detritivores and carnivores
Feeding rates of meroplanktonic larvae (FN 53) typically found during summer in the GoN and belonging to Polychaeta, Echinoidea, Cirripeda, Bivalvia and Gastropoda were measured over natural microplankton communities and a specific ingestion rate of 2.5 d -1 was reported for each of these groups 31 . These rates referred to a natural phytoplankton-bloom in the Vancouver bay during July with temperature ranging 13-17°C and at relatively high food concentration -estimated via chlorophyll concentration, ranging 2-6 μg L -1 (thus assimilated to Green conditions in the Gulf of Naples in surface waters). Assuming Q 10 =2.8 in general for marine planktonic animals 47 , the Max α during summer in the GoN accounted for 7 d -1 , which was a very high rate. Yet, meroplanktonic larvae have high energetic requirements to pursue an intense somatic growth and such high rates were not improbable. Being information on functional response of meroplankton virtually absent in the literature, the minimum rate was set as proportional to the lowest Chl a concentration in our model-system: i.e. Min α = Max α * 0.1 = 0.7 d -1 , where 0.1 is the ratio between Min and Max Chl a concentrations in our model system (see previous sections). An identical α range was assumed for both Green and Blue conditions.
Concerning meroplankton diet, meroplankton ate different kinds of particles, from nanoflagellates (both photo-and heterotrophic), diatoms, dinoflagellates (of different sizes) and ciliates 31 . According to that study, the highest pressure was exerted on the most abundant species, the harmful photosynthetic flagellate Heterosigma akashiwo (Raphidophycea, in our model, assimilated to algae in FN 12), followed by dinoflagellates, nanoflagellates and ciliates, while diatoms were weakly impacted by meroplankton grazing. We must stress here the fact that the above-mentioned study referred to a so-called 'harmful algal bloom condition', with H. agashiwo abundance largely exceeding that of diatoms (H. agashiwo was ten times more abundant than all diatoms taken collectively). Thus, it was not improbable that meroplanktonic larvae behave as peak-trackers at specific trophic conditions. In our model, due to the apparently controversial information of the trophic behavior of meroplanktonic larvae, these animal were enabled to eat all unicellular plankton, with the exception of prokaryotes.
Detritivores (FN 55) in our model includes only cyclopoid and harpacticoid copepods feeding specifically on Appendicularia houses. Among them, Oncea mediterranea ate up to 100% its body-biomass, at 20°C 93 . According to the regression described by 72 , Max α for FN 55, was 1.23 d -1 , considering its bodysize. The Min α was ~0.1 d -1 , a value similar to the minimum ingestion rate of other copepods. Carnivores in our model (FN 56) included chaetognaths, pteropods, siphonophors and some strictly carnivore copepods (Candacia spp. and Pleuromamma spp.). Chaetognats were the most abundant group, with pteropods, the second in rank, being about one fifth of the former. In the model, we assumed that the ingestion rate of the carnivore FN was mainly driven by chaetognats' behaviour. The feeding performances (by analyzing the gut content) for different Sagitta species were studied in detail at different maturity stages and across a coast-offshore trophic gradient in the Western Mediterranean during summer: ingestion rates ranged 0.13-3.76 preys ingested per chaetognath per day and no relation with food concentration was detected 94 . Relating to our model, considering an average weight of 18.35 µg C per animal for the FNs which were potentially eaten by chaetognaths, α ranged 0.013-0.37 d -1 , at both Green and Blue states. Among the identified preys in the animal gut, copepods were the most abundant ones, while cladocerans were one tenth less 94 . Among copepods, relatively larger species (i.e. C. typicus, T. stylifera, Candacia spp., Pleuromamma spp.) were more intensively caught than smaller ones (Paracalanus spp., Acartia spp., Clausocalanus spp.). Other crustaceans and chaetognats were also present in the diet, but at lower abundance. The diet of FN 56 in our model was set according to the food preferences listed above.

Derivation of Not-Assimilated fraction of Consumed Biomass (ε)
The parameter ε is the complement to 1 of the assimilation efficiency (AE), i.e. the amount of the food consumed that is destined to respiration and growth (i.e. assimilated). AE was here considered the better proxy for defining the Ecopath parameter ε, i.e. the not-assimilated fraction of consumed biomass.
The assimilation efficiency of mixo-and heterotrophic unicellular organisms in our model was assumed ~1, according to 12 , and ε was set at 0.01.
The AE of cladocerans (FN [42][43] was in the range of 0.70-0.45, respectively, from the lowest to higher food concentrations, within the range of variability of our study system 18,71 , giving ε ranging 0.55-0.30 for both Green and the Blue states. AE of calanoid copepods (FN 44-48) was set based on information available for A. clausi 18 : under different food concentrations, that copepod showed AE ranging 0.78-0.45, within a wide range of food concentrations, i.e. from ~0 to over 800 μg C L -1 . Nevertheless, the regression of AE vs. food concentration was not statistically significant 18 , since some values were comprised between 0.6 and 0.8 at supersaturating food concentrations, opening to the possibility that assimilation could be more efficient at some conditions. In the range of food variability of our study system, and considering a linear regression fitting data-points, AE ranged ~0.68-0.40, with a consequent ε ranging 0.32-0.60, for both Green and Blue states. This ε range was assumed to be extensible to all calanoid copepods in our model , excepting juvenile stages (FN 49), whose efficiency was set at a fixed value of 0.73, assuming that it was equal to that of Oithona davisae juveniles at 20°C 95 . Such a high value was related to higher somatic growth shown by juvenile copepod (absent in adults). ε of juvenile calanoids was thus fixed at 0.27.
AE of Appendicularia (FN 50) was inversely related to food concentration 28,29 . The latter authors described a regression linking AE to food concentration, used herein to infer AE values for our concentration and temperature extremes (recalculated data, considering also a Q 10 of 1.78 as reported by 57 . In our model system, ε ranged 0.091-0.604 in the Blue state, and 0.091-0.945 in the Green state. The wide range of AE for Appendicularia (and also other gelatinous mesozooplankton) stemmed from the higher influence exerted by external conditions in metabolic processes (as feeding) in these organisms, if compared with copepods.
AE of Doliolids (FN 51) ranged 0.72-0.32 in the range of food of our study system, with a slight negative relation to food concentration for values of the latter variable exceeding 400 mg C m -3 18 . The calculated ε for our system using the regression described by the above-mentioned study ranged 0.25-0.58 and 0.28-0.68 in Blue and Green conditions, respectively.
AE of Salps (FN 52) within 10 mm in length (as in our model) would account for 67 % 96 , but this could be only a value in a range and homologous data are very scanty in the literature. However, comparing to the other filter feeders, salps' AE seems to be relatively less dependent from food concentration. Salps of 10 mm showed AE comparable to smaller ones, ranging 75-87% of the body carbon-biomass at two different states of carbon concentration and microplankton composition during an iron fertilization experiment 97 . We can thus suppose that the assimilation efficiency is quite stable under different food concentrations, since salps can also eliminate water and food in excess by means of backwashing, i.e. expelling water in excess and regulate feeding 36 . ε of Salps in our model (FN 52) was set as ranging 0.13-0.25 at both Green and Blue states.
AE ranging 65-86% for different life stages was estimated for the cyclopoid copepod Oithona (FN 54) 95 . No other data were available for this copepod and we assumed that ε ranged 0.24-0.35 at both Green and Blue states. Considering that Oithona spp. were ambush feeders, it was probable that saturation emerged at concentrations higher than those critical for suspension feeders, thus giving a narrow range of variability for AE ε.
Information about AE for meroplankton (FN 53) were not available. We assumed that ε was 0.2 in this group at both Green and Blue states. This was a relatively high value, but it was plausible since larvae should convert in somatic growth a large part of the matter ingested.
Assimilation efficiency of detritivora (FN 55) -a minor group of mesozooplankton composed mainly by cyclopod and harpacticoid copepods -was assumed equal to that of Oithona. Being detritivors opportunistic feeders, similarly to Oithona, this assumption is highly plausible.
Carnivorous mesozooplankton (FN 56), mainly chaetognats in our model, were reported to have a high AE, i.e. ranging 80-82% 98,99 . These rates were similar to other strictly carnivorous plankton (e.g. siphonophors, a minor sub-group of animals in our model ( 35 ). ε in FN 56 ranged 0.18-0.2, at both Green and Blue states.