A detailed description of the mesocosm setup and the experimental manipulation is given in . Briefly, eight KOSMOS mesocosm units (cylindrical 18.7 m long polyurethane bags, 2 m diameter, 54.4 ± 1.3 m3 volume) were deployed on 23 February 2017, close to San Lorenzo Island about 6 km off Callao (12.0555 ∘ S, 77.2348 ∘ W, Fig. ). Nets (mesh size 3 mm) attached to both ends of the bags prevented larger plankton or nekton from entering the mesocosms during deployment. These meshes were removed as soon as the mesocosms were closed at the bottom with the sediment trap and the bags lifted above the surface, i.e., when the experiment started. On 25 February (defined as day 0), the mesocosms were closed at the bottom with a sediment trap. To simulate upwelling with differing inorganic N:P ratios (i.e., with an extreme and a moderate OMZ signature), water was collected on day 5 at station (hereafter St.) 1 from 30 m depth (12.03 ∘ S; 77.22 ∘ W) and on day 10 at St. 3 from 70 m depth (12.04 ∘ S; 77.38 ∘ W) of the Instituto del Mar del Perú (IMARPE) time series transect using the 100 m3 deep-water collectors described by . In each mesocosm ∼20 m3 of water was exchanged with deep water from St. 3 (moderate OMZ signature: mesocosms M2, M3, M6, M7) or St. 1 (extreme OMZ signature: M1, M4, M5, M8). Deep water was injected on day 11 and day 12 to similar depth ranges as water had been removed from each mesocosm before (14–17 and 1–9 m).

Oxygen minimum zones may reach very close to the surface (<10 m) in the near-coastal region off Peru . To conserve the low-O2 bottom layer in the mesocosms that had established after the addition of low-oxygen deep water over the entire experimental duration, water column stratification was artificially maintained by evenly injecting a concentrated NaCl brine solution into the bottom layers of each mesocosm on day 13 and day 33. Otherwise, convective mixing induced through heat exchange with the surrounding Pacific would have destroyed the oxygen minimum layer in the mesocosms. For more details on the exact procedure of water exchange and deep-water injection as well as addition of the brine solution see . According to these authors the mesocosm experiment was subdivided into three main phases based on the phytoplankton development: phase 1 lasted from day 1 until the OMZ water addition (day 11–12) and was characterized by a diatom-dominated phytoplankton community. Phase 2 started with the OMZ water addition until day 40 and was characterized by a bloom of the mixotrophic dinoflagellate Akashiwo sanguinea. Phase 3 continued from day 40 until the end of the experiment, when eutrophication by defecating seabirds (guanotrophication) triggered a late phytoplankton bloom in most mesocosms.

MeZP samples were always obtained with an Apstein net of 17 cm opening diameter equipped with a 100 µm net bag. Integrated vertical hauls were taken from 17 m depth of each mesocosm and the Pacific close to the mesocosm field. Due to technical and logistic constraints, regular sampling was not possible before day 18. Therefore, sampling intervals varied between 2 and 7 d until day 18. From then onwards, sampling was performed every 6 d. In total, two net hauls from each mesocosm and the Pacific were collected on 10 sampling days (day 0, 8, 10, 13, 18, 24, 30, 36, 42, 48). Contents from one net of each mesocosm were used for species abundance and biomass determination, and contents from the other were used for picking live organisms for fatty acid and elemental analyses (C, N, P, δ13C, δ15N). MeZP was sampled in the afternoon between 13:00 and 17:00 (local time). As soon as the abundance net haul was retrieved onboard, the zooplankton sample was emptied into sample bottles with filtered seawater (100 µm), and the net and cod end were subsequently rinsed to also wash zooplankton attached to the mesh into the sample bottle. The second net for live organisms was carefully poured into 5 L sampling containers prefilled with 4 L of filtered seawater at ambient temperature to accommodate zooplankton organisms and reduce stress. All net samples were stored in cooling boxes to prevent heating until returning to the laboratory and further processing.

Microzooplankton samples for elemental composition analysis (detailed in Sect. 2.8) were collected from integrated water samplers (see detailed description of water samplers in ). In total, samples were taken on 12 sampling days (day 0, 8, 10, 13, 16, 18, 20, 26, 30, 36, 42, 48) from each mesocosm and the Pacific. On the first sampling day 0, samples were collected from the depth intervals 5–0 and 17–5 m. From day 8 onwards, sampling depths were 10–0 and 17–10 m. Until day 20, samples were regularly taken from both sampling depth intervals; after day 20 only the upper interval (10–0 m) was sampled for microzooplankton because concentrations in the deeper layer were too low to reach detection limits.

In the laboratory, the net samples taken for abundance and biomass determination of each mesocosm were split in half (Motoda splitter), and one-half of each sample was preserved in 4 % formaldehyde–seawater solution for subsequent microscopic species identification and enumeration. The other half was used for ZooScan analyses. The formalin-preserved abundance subsample was split by applying the Huntsman Marine Laboratory (HML) beaker technique, producing count results with a coefficient of variation of 9 %–15 % from the effective mean . Subsequently, aliquots were counted under a stereomicroscope (Nikon SMZ 1270) until at least 50 individuals of the most abundant taxa were counted; rare species and taxa were counted from the whole sample. As usual, zooplankton abundances were calculated assuming 100 % filtering efficiency of the net. However, variation among samples is normally high due to plankton patchiness, and moreover variation is species- and taxon-specific .

For the scanning procedure, samples were split further (Motoda splitter) to ratios between 1:4 and 1:32 to avoid crowding in the scanning chamber. In the laboratory, samples were scanned on an Epson Perfection V750 Pro scanner in a modification of the ZooScan method and a scan chamber constructed of a 21 by 29.7 cm (DIN A4) glass plate with a plastic frame. Scans were 8-bit greyscale, 2400 DPI images (tagged image file format; *.tif). The scan area was partitioned into two halves (i.e., two images per scanned frame) to reduce the size of the individual images and facilitate the processing by ZooProcess/ImageJ. “Vignettes” and image characteristics of all objects were extracted with ZooProcess and sorted sequentially by first clustering and naming using MorphoCluster and then predicting the remaining unsorted images using deep-learning features in Ecotaxa (http://ecotaxa.obs-vlfr.fr/ (last access: 25 May 2021), ). Automated image sorting was manually validated by experts. Individual biomass was derived from the image area of each object using published taxon-specific relationships .

Grazing rates of the dominant copepods Paracalanus sp. were analyzed with the gut fluorescence (GF) method via experimental determination of gut clearance rates (clearance coefficient). Clearance rate determination assumes that maximum gut fullness of organisms equals initial gut pigment in the gut clearance experiment (time point 0 min, T0 min). The experimental identification of a clearance coefficient is accomplished through fluorescence determination of organisms' guts allowed for clearance over a series of time steps for a total period of ca. 1 h. At least eight time steps along the time axis are recommended. Hence, pre-determination of in situ daily maximum gut fullness is required if maximum ingestion rates are to be estimated .

Copepods often have diel feeding rhythms . Therefore, prior to clearance rate determination of mesocosm copepods, the feeding rhythmicity and time of maximum gut fullness of dominant copepods (Paracalanus sp.) were examined twice over a 24 h cycle for copepods collected in both the adjacent Pacific and in the mesocosms. Each time zooplankton net samples were taken, adult female copepods (the dominant copepodite stage) were picked and GF measured (see below for more details on the procedure). These investigations proved that copepods' feeding activities, i.e., times of maximum gut fullness, were highest at night between approximately 22:55 and 03:40 (Fig. ). Therefore, sampling of zooplankton in the mesocosms for clearance rate determinations was done at night on occasions: during the night of day 21–22 and during the night of day 34–35. MeZP samples were collected in all mesocosms with the same Apstein net of 17 cm opening diameter as mentioned above but equipped with a 200 µm net bag and a non-filtering cod end to collect organisms gently and prevent stress evacuation of their guts before retrieval of the net.

Determination of gut clearance rates and coefficients followed recommendations by . Immediately after retrieval of each net haul, the zooplankton sample was split into eight subsamples. One of the subsamples was immediately concentrated over a 200 µm mesh; the mesh was carefully folded not to squash organisms, wrapped in aluminum foil, and immediately shock-frozen in liquid nitrogen to preserve the in situ state of gut content upon catch (time point 0 min, T0 min). The other seven subsamples (T8 min–T56 min) were each incubated in 500 mL of filtered seawater (20 µm) to allow copepods to evacuate their guts. After each 8 min, one of the incubated subsamples was terminated (i.e., last sample after 56 min, T56 min) by concentrating the zooplankton on a 200 µm mesh; the mesh was wrapped in aluminum foil and immediately shock-frozen in liquid nitrogen. The shock-frozen subsamples (T0 min–T56 min) were stored at -80 ∘C until further processing in the home laboratories in Kiel.

Paracalanus (mostly Paracalanus parvus) was the dominant copepod taxon in all mesocosms and was therefore chosen for determination of gut clearance rate. In Kiel, depending on availability, between 8 and 52 individual adult female Paracalanus sp. were picked on crushed ice under a stereomicroscope (Wild Heerbrugg M3) from each of the frozen subsamples (T0 min–T56 min) into 2 mL cryovials placed in a labtop vial cooler at -15 ∘C. This procedure was done under dimmed room light, usually within half an hour to prevent destruction of fluorescent pigments . Loaded vials were stored at -80 ∘C until completion of all samples from all mesocosms. Subsequently, gut pigments were extracted in 1.2 mL 90 % acetone with the help of glass beads (0.5 mm diameter) in a Precellys Evolution HP homogenizer at 10 000 RPM (9168 g) for 15 s. Constant cooling of samples was ensured during the whole procedure to avoid pigment degradation. After 30 s of pause, the crushed samples were treated in a Sigma 3–18 K centrifuge for 10 min at 10 000 RPM (9168 g) at 4 ∘C to separate the liquid and solid phase. Finally, the liquid phase was filtered through 0.2 µm PTFE filters to remove any remaining solid body parts. GF was then measured with a Trilogy Laboratory Fluorometer (Turner Design, USA). The relative GF of each sample was measured three consecutive times and converted to absolute values (ng pigment µg DM-1) by means of a chlorophyll standard curve. To eliminate background fluorescence caused by astaxanthin carotenoids in copepod tissues, GF was corrected by GF measured for female Paracalanus collected in the surrounding Pacific and starved for 24 h to allow animals to empty their guts completely . An exemplary sample was analyzed for gut pigments (or their degradation products) also by means of reverse-phase high-performance liquid chromatography (HPLC, ) calibrated with commercial standards.

To normalize GF, dry mass (DM) of female Paracalanus individuals was determined for each time-point subsample. Single female copepods were picked for DM determination of each sample in triplicate (i.e., three samples with one female from each mesocosm if enough organisms were available). Organisms picked for DM determination were rinsed in Milli-Q and transferred into pre-weighed tin cups before drying at 60 ∘C for 24 h and mass determinations on a Mettler Toledo XP2U Ultra Micro Balance (accuracy: 0.000058–0.0034 µg). Additionally, females from day 21–22 samples were picked for C:N determinations, usually from T0 min and T56 min (13–38 individuals per sample depending on availability). Carbon and nitrogen elemental analyses were done on a Euro EA – CHNSO element analyzer according to .

Adult females (the dominant developmental stage) of the dominant copepods Paracalanus sp. and Hemicyclops sp. were picked as soon as possible after collection for lipid analysis from the second net (for live organisms). Due to time constraints, female copepods were only picked from six mesocosms (M2, M3, M6 of the moderate treatment; M1, M4, M5 of the extreme treatment). Lipid samples of female copepods from the adjacent Pacific could only be picked during phase 3 because of limited abundance during phases 1 and 2. If available, up to 80 individuals (minimal 16) of each species were sorted to ensure sufficient lipid mass above detection limit. Some samples were directly transferred to dichloromethane : methanol (2:1, v/v) in 8 mL analytical vials equipped with Teflon-sealed screw caps, and some were transferred to Eppendorf caps without solvent. The latter allowed for determination of DM after lyophilization. DM of specimens directly stored in dichloromethane : methanol (2:1, v/v) could not be determined. Samples were stored at -80 ∘C until further analysis in the home laboratory at Bremen University.

DM of the copepods was determined after lyophilization for 48 h (CHRIST Alpha). Total lipids were extracted from the samples after and with dichloromethane : methanol (2:1 v/v). For quantification of total fatty acids (TFA), tricosanoic acid (23:0) was added as an internal standard prior to extraction and TFAs were expressed as the percentage of fatty acids in relation to the DM of the sample (TFA % DM). For Paracalanus, TFAs were related to the mesocosm-specific mean DM of females used for gut fluorescence determination because DM determined after lyophilization of lipid samples was inaccurate. Lipid extraction of samples stored in solvent followed the same procedure as for the lyophilized samples.

Fatty acids were converted to their methylester derivatives (FAME) by transesterification for 4 h at 80 ∘C in hexane and methanol containing 3 % concentrated sulfuric acid . FAMEs were extracted with aqua bidest and hexane and analyzed by gas chromatography (Agilent Technologies, GC model 7890A). The device was equipped with a DB-FFAP column (30 m length, 0.25 mm inner diameter) and a programmable temperature vaporizer injector, operating with helium as a carrier gas . Fatty acid and alcohol components were detected using a flame ionization detector (FID) and identified by their retention times in comparison to known fatty acid and alcohol standard compositions (FAMEs and free alcohols of the copepod Calanus hyperboreus and Supelco 37 Component FAME Mix). In some cases, the lipid mass was very low with a stronger impact of impurities so that gas chromatography did not result in reliable data. Therefore, data presented here are those that have at least 70 % purity of fatty acids relative to impurities. The fatty acid compositions were evaluated according to the fatty acid trophic marker (FATM) concept of .

Bulk samples of copepods or polychaetes were used for the analysis of stable isotope signatures (δ13C and δ15N) and carbon (C), nitrogen (N), and particulate organic phosphorus (POP) content. Copepod samples mainly contained the taxa Paracalanus sp. and Hemicyclops sp., whereas polychaetes were separated into the species Paraprionospio sp. and Pelagobia longicirrata. Copepods and polychaetes were pipetted through a 55 µm nylon mesh attached to a polypropylene tube of ∼2 cm length so that the organisms remained on the mesh. Organisms were shortly rinsed with Milli-Q water. Bulk samples were transferred to pre-weighed tin capsules (5×9 mm, Hekatech) and dried at 60 ∘C for at least 24 h. After drying, tin capsules were closed at the top and stored in a 96-well plate covered with Parafilm. In the home laboratories, samples were dried again and weighed on a microbalance. Subsamples for phosphorus were taken whenever enough biomass of the sample was available. Target sample masses were 1–4 mg for C:N and 1–10 mg for POP. Stable isotopes (13C and 15N) as well as C and N contents were analyzed at the UC Davis Stable Isotope Facility using a PDZ Europa ANCA-GSL elemental analyzer interfaced to a PDZ Europa 20–20 isotope ratio mass spectrometer (Sercon Ltd., Cheshire, UK) with helium as a carrier gas. Vienna Pee Dee Belemnite and atmospheric air were used as standards to determine the carbon and nitrogen ratio, respectively. Stable isotope ratios are reported with reference to a standard and expressed in parts per thousand (‰) according to the formula δHX=[(RSAMPLE/RSTANDARD)], where X is the respective element, H gives the heavy isotope mass of that element, and R is the ratio of the heavy to the light isotope.

POP was determined photometrically on a Hitachi U-2900 spectrophotometer as orthophosphate after oxidative decomposition by adding spatula-tip Oxisolv^®, 5 mL of Milli-Q, and the sample to Pyrex tubes and autoclaving. After cooling to room temperature, a subsample (dilution 1:5 or 1:10) was measured according to .

Samples for X-ray microanalysis (XRMA) were prepared following . 500 mL water samples were taken from each mesocosm and the Pacific Ocean and were pre-filtered through a 200 µm nylon mesh to remove larger MeZP. The 200 µm filtered water samples were further concentrated via inverse filtration (vacuum pump, 15 µm nylon mesh) to a volume of approximately 2–4 mL. These 2–4 mL remaining water samples were pipetted through a 15 µm nylon mesh so that the organisms remained on the mesh, where they were rinsed several times with ice-cooled Milli-Q brought to pH of 8.3–8.5 with NaOH. The last ∼50 µL was pipetted to a slide (in several drops) and checked under a stereomicroscope for microzooplankton organisms. 5 µL was transferred to a transmission electron microscopy (TEM) grid (coated with a formvar carbon film on copper on 75 mesh, Agar Scientific) in triplicate and again shortly checked under the stereomicroscope to ensure that there was a sufficient number of organisms from the same species to be analyzed. TEM grids were air-dried for about 1 h and afterwards transferred to TEM grid storage boxes and stored in a desiccator container (Fisherbrand™ Circular Bottom Desi-Vac™ container, Fisher Scientific) until arriving in the home laboratories in Germany. Samples were then kept in vacuum bags until final analysis.

XRMA was done at the Joseph Banks Laboratories in Lincoln, UK. Single cells of microzooplankton (200–20 µm) were imaged and analyzed for C, N, O, and P elemental content in an FEI Inspect scanning electron microscope (SEM) equipped with an energy-dispersive spectrometer Oxford X-Act silicon drift detector (SDD). X-ray spectra were acquired from an area that circumscribes the cell at 20 kV of accelerating voltage, accumulation time of 120 s (lifetime), spot size 6.0, and 10 mm working distance. Following the analysis of each cell, a blank spectrum (i.e., formvar only) close to the cell was acquired for 30 s lifetime. Spectra from cells that drifted, changed shape, or charged during analysis were discarded. An image before and after the analysis was taken using the INCA Oxford EDS software, and the area of analysis was calculated with a MATLAB routine. X-ray bremsstrahlung from each spectrum was removed using NIST DTSA-II Jupiter 6 November 2017, and the intensity of each peak was also calculated with a MATLAB routine. The standards used for quantification were latex beads of 3 and 5 µm for C and ADP for N, O, and P. Latex beads of 3 µm were analyzed as internal standards several times during each analysis session to correct for drift and for changes in day-to-day beam intensity. A detailed description of the measuring approach is given in .

We applied repeated measures ANOVA (rmANOVA) to test for possible treatment effects. Response variables (listed in Table ) were modeled as a function of deep-water treatment (two levels: “moderate” and “extreme”) and experiment day with mesocosm as a random factor to test for significant differences between the added deep waters with different OMZ signature. Normality and variance homogeneity of residuals were tested graphically. In the case of significant results, if assumptions were not met, data were log10-transformed and re-analyzed. Unfortunately, ∼50 % of the fatty acid samples showed a substantial level of impurities due to the generally very low lipid content of the copepods and were therefore omitted from the data set; i.e., only samples with at least 70 % purity are included in the present study. This led to a restricted data set with unbalanced sampling design and an unequal and low number of replicates that did not allow for more comprehensive statistics (rmANOVA). Instead, we calculated mean percentages and their 95 % confidence interval (CI) of the dominant fatty acids per mesocosm treatment and experimental phase to allow for basic statistical inference. Comparing any two means, evidence of significance can be inferred if their CIs do not overlap . Similarly, initial GF values of copepods of each mesocosm (with standard deviation, SD) are shown together with treatment means and their 95 % CIs because measured GF in the copepods' guts of both OMZ treatments was too low to allow for grazing rate estimations.

Pearson correlations between abundance of the dominant zooplankton taxa, the copepods Paracalanus sp. and Hemicyclops sp., and counts of dominant phytoplankton and microzooplankton groups (diatoms, phytoflagellates, coccolithophores, dinoflagellates, silicoflagellates, ciliates, all in cells L-1) as well as with concentrations of extracted phytoplankton pigments (Chloro-, Dino-, Crypto-, Prymnesio-, and Pelagophyceae, diatoms, Synechococcus, all in µg L-1) and with total chlorophyll a (µg L-1) were performed to gain insight into food web relations.

All statistics were performed using R 4.2.2 (libraries emmeans, lme4, lmerTest, and HighstatLib.V6; ).

Average total abundance of MeZP was very similar at the start of the experiment, and no significant differences developed between the two OMZ deep-water treatments over the course of the experiment (Table ). However, the established treatment differences with respect to N:P signatures were significant but only small. Thus, any potential small-scale treatment differences in MeZP abundance, if existent at all, would have most likely been hidden by methodical uncertainties associated with net sampling, patchy distribution, and counting bias .

Variability of the manual counts of total MeZP abundance (individuals per liter: Ind L-1) was high between mesocosms and sampling days. Average total MeZP abundance varied between 6.3 Ind L-1 (CI 3.5) on day 30 and 52.7 Ind L-1 (CI 14.0) on day 42 in the moderate-treatment mesocosms and between 7.2 Ind L-1 (CI 3.3) on day 30 and 33.9 Ind L-1 (CI 30.1) on day 42 in the mesocosms with the extreme treatment. Total abundance of MeZP sampled in the adjacent Pacific station was in a similar range as numbers estimated in the mesocosm treatments (9.7–129.3 Ind L-1) and peaked on day 48 (Fig. ). Individual mesocosms developed some more distinct abundance peaks at the beginning and towards the end of the study with maximum numbers of 112.3 Ind L-1 in M1 on day 10, 127.9 Ind L-1 in M4 on day 42, and 129.0 Ind L-1 in M3 on day 48. Deep-water additions on day 11 and day 12 after removal of approximately 20 m3 of water from each of the mesocosms, respectively , resulted in distinct deviation from average MeZP abundance between day 13 and day 18 with temporarily higher numbers in the moderate mesocosms. This deviation was only short-term, not significant, and due to some higher abundance in M7 (moderate treatment), causing the relatively large confidence intervals (Table , Fig. S1 in the Supplement). It had equalized by day 24 but was indicative of different MeZP densities of the two injected deep-water masses.

In terms of abundance, copepods dominated the zooplankton communities at all times both in the mesocosms and the adjacent Pacific. Other taxa (in descending order of contribution: Polychaeta, Echinodermata, Euphausiacea, Mollusca, Hemichordata, Cnidaria, Chordata, Cirripedia, Rotifera) contributed only between 0.2 % and 18 %. Higher shares of other taxa were found only until day 18, and afterwards they were negligible. This pattern essentially reflected the occurrence of copepods and other taxa in the surrounding Pacific (0.3 %–24 %), although in the Pacific other taxa contributed higher shares to the zooplankton community compared to the values found in the mesocosms also after day 18 (Fig. a).

Among other taxa, in the mesocosms, Polychaeta, Echinodermata, Euphausiacea, and Mollusca (taxa ordered in decreasing trend) were most abundant, sometimes constituting more than 5 % of total abundance. Remaining taxa consistently contributed less than 5 % to total MeZP abundance. Towards the end of the study, the numbers of Chordata (ichthyoplankton) and Cnidaria increased in M1, M2, M4, M5, and M6 to values >1 %. In the Pacific in the early phase of the experiment, Polychaeta (up to 21 %) and Echinodermata (up to 11 %) dominated, followed by Hemichordata (2 %) and Rotifera (3 %). In the second half of the experiment, Polychaeta, Cirripedia, Mollusca, and Chordata (each <3 %) dominated the generally low share of other taxa in total abundance. Chordata (ichthyoplankton) increased because fish eggs were added to the mesocosms on day 31; however, they did not stay for long in the mesocosms (Fig. b).

For the dominant taxa, the ZooScan method reduced the systematic variability and added the size information for each object, allowing biomass estimation. For rare taxa, manual enumeration is to be preferred because of the low split ratios in the scans.

Zooplankton biomass (as estimated based upon scanned area of individuals) was dominated by copepods (Fig. a), which in the mesocosms ranged between 11.8 and 75.6 µg DM L-1 in the first phase, increased to between 8.58 and 137.9 µg DM L-1 in the second phase, and declined again to between 4.8 and 40.8 µg DM L-1 in the third phase. The second most important contributor to zooplankton biomass were polychaetes (Fig. b), which in the mesocosms were hardly detected on day 1 and ranged between 0 and 39.0 µg DM L-1 in the first phase, between 0 and 74.7 µg DM L-1 in the second phase, and between 0 and 0.65 µg DM L-1 in the third phase. The highest biomass values were observed for both total copepods and polychaetes in mesocosms M2 and M3 (both in the moderate mesocosm treatment). The main contributors to copepod biomass were Paracalanus sp. and Hemicyclops sp. Due to the high variability between mesocosms, however, this treatment effect was insignificant (Table ). In the adjacent Pacific, total copepod and polychaete biomass tended to be lower than in the experimental mesocosms but followed a similar temporal pattern (peaking in phase 2). Among the copepods, the numerically dominant genera Paracalanus and Hemicyclops were also the main biomass contributors (Fig. a and b), with roughly equal share and peaking in phase 2, in particular in the moderate OMZ signature treatment. Acartia, which was only present in lower abundance and biomass (Fig. c), tended to reach higher biomass in the extreme OMZ signature treatment (in particular in M4). Oncaea and other or unidentified cyclopoids (Fig. d, f) were mainly present in phase 1, with the lowest values in phase 2 and a slight increase in phase 3.

In contrast to MeZP abundance, biomass (copepods) peaked between day 10 and day 20, but the highest abundance values were found towards the end of the experiment when biomass was at a minimum. This discrepancy is most likely due to a shift in the abundance–biomass relationship of dominating zooplankton organisms towards smaller-sized or lower-mass individuals, respectively.

In general, initial (T0 min) GF of Paracalanus sp. females was very low, but averages for the moderate and the extreme treatment, respectively, on day 34–35 (0.14 and 0.29 ng pigment µg DM-1) were higher than on day 21–22 (0.03 and 0.08 ng pigment µg DM-1), indicating somewhat higher feeding activities of Paracalanus females on autotrophic food particles later in the experiment (Table ). Due to such low GF, no clearance rate measurements could be inferred because GF of incubated copepods showed no clear decreasing trend in GF over time (data published at PANGAEA, https://doi.org/10.1594/PANGAEA.947833, .) Female Paracalanus of the adjacent Pacific had even lower fluorescence in their guts (0.005 and -0.004 ng pigment µg DM-1) than measured in mesocosm organisms.

An attempt to analyze gut fluorescence with HPLC to identify degradation products (pheopigments) of chlorophyll a and draw conclusions on potential food sources revealed no identifiable peak(s) besides astaxanthin, which is indicative of the copepods' carapax, supporting very low GF and thus very low feeding on autotrophic food particles.

Mean C:N values for copepods in mesocosms treated with St. 3 (extreme) and St. 1 (moderate) OMZ water were 5.40 and 4.99, respectively, and had overlapping CIs; thus, no significant differences are suggested. Availability of female Paracalanus during day 34–35 was not sufficient to reach the detection limit for C:N analysis.

Fatty acid (FA) analyses revealed a relatively high degree of impurities of the samples. This is a common problem in small copepods with rather low total lipid mass close to the detection limit . For this reason, identification of FA peaks in the gas chromatogram was sometimes difficult or even impossible. Therefore, data presented here only include results of samples for which at least 70 % of all peaks of a chromatogram could be identified and assigned to a specific FA (i.e., 70 % purity).

Pearson correlations (data not shown; please see the Supplement) revealed no particularly strong relationships between protist groups (diatoms, phytoflagellates, coccolithophores, dinoflagellates, silicoflagellates, Chloro-, Dino-, Crypto-, Prymnesio-, and Pelagophyceae, Synechococcus) and adult and copepodite stages of the copepods Paracalanus sp. and Hemicyclops sp. The majority of correlations with a higher correlation coefficient were influenced by a single value; thus, we do not consider these correlations meaningful in spite of their significance. Apart from that, no consistent associations were identified across the single mesocosms or the two OMZ treatments. Some more meaningful positive correlations were determined for both genera with Pelagophyceae, silicoflagellates, total Chl a, Cryptophyceae, and Chlorophyceae (p<0.05, Table S1 in the Supplement). More conclusive negative correlations existed with dinoflagellates and Prymnesiophyceae. However, none of these correlations were consistent in all mesocosms; i.e., we could not identify a consistent food web relation pattern for the dominant Paracalanus and Hemicyclops species in the different mesocosms. Apparently, these copepods fed very opportunistically with no major preference for any of the occurring protist groups. In the adjacent Pacific, positive correlations for Paracalanus sp. existed with silicoflagellates, Dinophyceae, Chlorophyceae, and Chl a (p<0.05, Table S1).

δ15N and δ13C values of copepods (bulk samples) were not impacted by treatment (except a marked difference in M4) (Table ) but varied over time with a general increase in δ15N and a decrease in δ13C during the experimental period (Fig. ).

During phase 1, δ15N of copepods increased by ∼1 ‰ in all mesocosms; a further increase of 1 ‰–2 ‰ in δ15N occurred during phase 2, while copepod δ15N slightly decreased during phase 3. δ13C was highest during phase 1 and fluctuated between phases 2 and 3. In mesocosm M4, the decrease in δ13C was particularly pronounced. Generally, δ15N values were substantially higher and δ13C values slightly lower in the mesocosms than in the copepods from the adjacent Pacific (Fig. ). Copepod pooled C:N was unaffected by treatment, with an average value of 4.8 (±0.2), and ranged between 5.1 and 5.8 for Hemicyclops sp. and Paracalanus sp., respectively, based on species- and genus-specific determinations. For the two polychaete species, temporal patterns of δ15N and δ13C values were difficult to resolve due to incomplete sampling and did not reveal a consistent pattern (data available at PANGAEA, https://doi.org/10.1594/PANGAEA.947833, ). C:N ratios in the two polychaetes were slightly lower than those of the copepods, with 4.3 (± 0.44) and 4.2 (±0.44) in Paraprionospio sp. and P. longicirrata, respectively. DM-specific organic C, N, and P contents were also lower in the two polychaetes than in the copepods. In all MeZP groups, C:P and N:P ratios were higher than 200 and 40, respectively (Table ). While the analysis of parallel samples for C:N and P may explain some of the variability in the MeZP, the comparatively low P content was also determined in the microzooplankton, for which all elemental contents were analyzed per cell. We analyzed microzooplankton cells of the genera Dinophysis cf. punctata (n=127), Tripos (e.g., T. fusus, T. furca, n=25), and Protoperidinium sp. (n=3). In the analyzed dinoflagellates, cellular elemental ratios deviated even more from the canonical Redfield ratio (Table ). Because the N:P additions in the different mesocosms were not very different, we pooled the analysis of microzooplankton per sampling day. Comparing single-cell analysis of Dinophysis cf. punctata from this study (Table ) with the same taxa from the Mediterranean Sea, all species analyzed here had more C, N, and O but similar P contents .

Our experiment coincided with a concurrent strong coastal El Niño, a situation of generally warmer waters and weaker upwelling conditions that favor small-sized zooplankton occurrence in the coastal upwelling areas that consequently dominated our MeZP community in the mesocosm bags at the start of the experiment. The MeZP community comprised various taxonomic groups with copepods dominating throughout the experiment in both the mesocosms and the surrounding Pacific; therefore, we focus our discussion on copepods. The four main genera were Paracalanus, Hemicyclops, Acartia, and Oncaea. Species of the genus Paracalanus, Acartia, and Oncaea are common in Peruvian coastal and neritic waters e.g.,, whereas the occurrence of the cyclopoid copepod Hemicyclops sp. off Callao has only been reported by . During our study, Hemicyclops sp. regularly occurred in the surrounding Pacific with different developmental stages including older copepodids.

The observed developmental delay of copepodids and adults and especially the very low abundance of nauplii may be a consequence of hypoxic conditions in the mesocosms below ∼10 m. Copepod nauplii were rare in the mesocosms, particularly after deep-water addition between day 18 and day 30 (<1 % or absent), although females of the sac-spawning Hemicyclops and also the less abundant Oncaea were frequently observed carrying egg sacs (personal observations). In contrast, except for the first days of the experiment, nauplii almost consistently comprised around 20 % of all copepod stages in the surrounding Pacific until day 30 and reached almost 80 % on day 36. This could be explained by advection of copepod populations carrying higher numbers of nauplii or turbulence and upwelling keeping eggs in the oxygenated layer and promoting hatching. A lack of nauplii is also reflected in the development of copepodids and adults over the experimental duration and is particularly evident for Acartia sp. with adult copepods dominating the population almost over the entire study period. Despite species-specific tolerance levels, copepods generally respond to hypoxic conditions with avoidance (adults) and decreasing rates of survival, egg production, and population growth resulting in significant effects on population dynamics e.g.,.

Eggs of Paracalanus sp., the dominant broadcast spawning (i.e., eggs released freely into the water column) copepod in the mesocosms, were probably particularly affected by hypoxia when sinking into low-oxygen waters. Field studies showed that maximum abundances of Paracalanus cf. indicus eggs in Mejillones Bay, Chile, were found in the oxygenated surface waters, whereas egg abundance decreased significantly from the oxygenated layer and the oxycline to the OMZ . Sublethal and lethal hypoxia levels (<67 and <31 µmol L-1 O2, respectively; ) occurred in all mesocosms and the surrounding Pacific consistently throughout the study, where the oxycline was between 5 and 15 m and dissolved oxygen concentrations decreased at depth to <50 µmol L-1 . Sinking velocities of Paracalanus parvus eggs determined in the lab and of P. cf. indicus determined in the field (Mejillones Bay, Chile) ranged 2.4–16.9 m d-1 . At a temperature range of 18–20 ∘C, P. parvus nauplii hatch within 0.40–0.46 d, and the generation time of Paracalanus is about 18 d at 18 ∘C . During our study, due to the coastal El Niño, surface temperatures usually varied between 20 and 22 ∘C, exceptionally reaching 24 ∘C between day 14 and day 22. Averaged over the entire mesocosm water column and all mesocosms, temperatures ranged between 18.4 and 20.2 ∘C from days 1 to 38 and between 17.9 and 18.6 ∘C thereafter . Thus, the experimental duration was long enough to expect Paracalanus eggs in the mesocosms. Eggs released at the mesocosms' surface would sink to 0.96–6.76 m depth within 0.4 d (9.6 h). Eggs released at greater depth would reach the oxycline and hypoxic layers faster, where development was likely halted or eggs may have suffered mortality. The accumulation of adults and a minimum contribution of copepodids around day 24, after oxygen concentrations in the mesocosms had decreased below 50 µmol O2 L-1 below 10 m depth , further support our notion of hindered egg development due to hypoxic conditions.

A late increase in copepod nauplii discernible in the mesocosm treatments on day 36 (extreme) and day 42 (moderate) might have resulted from a co-occurring deepening of hypoxic layers (<55 µmol L-1) from ∼10 m to 14–15 m and slightly higher oxygen concentrations resulting from a phytoplankton bloom event facilitated through guanotrophication . Similarly, a short intrusion of oxygen-richer waters up to ∼10 m occurred in the adjacent Pacific, parallel to the small nauplii increase on day 30. Together with the concomitant increase in food availability this may have supported an increase in eggs and nauplii in both mesocosm treatments.

Upwelling is an important factor shaping copepod population dynamics operating through an optimal window of upwelling intensity that fosters copepod abundance and biomass, whereas increasing upwelling may be unfavorable for copepod populations . This is in close connection with the fact that especially smaller copepods such as Paracalanus and Acartia require well-oxygenated waters for egg development and survival of early-stage copepods for population success . Hence, our findings of failed egg and nauplii development strongly support these earlier notions and underline the importance of mesoscale variability and its implications for copepod reproduction in oxygenated waters and upwelling events, respectively.

Exhausted inorganic nitrogen sources and the dominance of the mixotrophic and facultative osmotrophic dinoflagellate Akashiwo sanguinea in the phytoplankton community in most mesocosms (absent in M4) seem the most likely reasons for the extremely low GF of Paracalanus females during phase 2 of the experiment. Pigment contents determined for Paracalanus in our study were mostly an order of magnitude lower than those of Acartia, Pseudocalanus, and Temora from Danish waters and of Acartia off southern California . Paracalanus parvus prefers phytoplankton >5 µm over particulate matter as a food source but may also feed on dinoflagellates when these dominate phytoplankton biomass . Furthermore, visual inspection of guts of Paracalanus females revealed that guts were often empty and that they were not feeding directly on phytoplankton.

The low GF is in accordance with the very low occurrence or absence of phytoplankton biomarker fatty acids determined in Paracalanus and Hemicyclops females in this study. Together, these point to the fact that development of these copepods depends on immediate food supply. Due to their very low lipid (TFA) levels, the fatty acid signatures of the copepods analyzed were largely dominated by biomembrane fatty acids (phospholipids), e.g., 16:0,20:5(n-3), and 22:6(n-3). However, dietary signals of marker fatty acids are conserved in the storage lipids, i.e., in wax esters or triacylglycerols . The marker fatty acid 16:1(n-7) typical of diatoms decreased in Paracalanus females from initially ca. 5 % to 1% TFA–2 % TFA in phase 3, suggesting diatom ingestion only before or at the beginning of the experiment. However, Paracalanus females collected in the adjacent Pacific during phase 3 had higher 16:1(n-7) concentrations of 10 %, indicative of diatom feeding in the wild. In contrast, the fatty acid 18:4(n-3) typical of flagellates was not detected in significant amounts in any of the copepods. The near absence of wax esters (storage lipid) agrees with previous studies that egg production in the studied species is related to immediate food supply rather than a period prior to actual egg production, when energy reserves may have been accumulated .

Moreover, of the detected gut fluorescence no clear pheopigment peaks were identifiable with HPLC, which points to a high degree of degradation of fluorescent pigments found in the guts. Possibly, this “aged” gut fluorescence originated from ingestion of particulate matter, which could also explain why no particularly strong correlations existed between any of the phytoplankton groups and female abundance. The A. sanguinea bloom was associated with a marked increase in particulate organic matter and in the C:N ratio (from about 5–6 to 8–12) in the water column of most mesocosms . However, phytoplankton availability or dietary nitrogen (i.e., protein) is at times the direct limiting nutrient for egg production of Paracalanus . Hence, nutritional conditions for Paracalanus females were apparently insufficient in the mesocosms during phase 2, leading to low egg production and explaining the low numbers of nauplii found in the mesocosms.

Like Paracalanus, Hemicyclops may have lived at suboptimal conditions. Gut fluorescence was not detected in Hemicylops sp. The fatty acid composition of Hemicyclops females was analyzed only from day 30 onwards and was very similar to that determined for Paracalanus. Except for very low concentrations (2 % TFA–3 % TFA) of the diatom marker 16:1(n-7), no trophic marker fatty acids were identified and no fatty alcohols were determined; hence, no wax esters (storage lipid) were present. Correlation analysis suggested no particular food preference of Hemicyclops for any of the prevailing phytoplankton groups. The diet of Hemicyclops species with a coupled pelagic–benthic life cycle consists of microbenthos and meiobenthos, while detrital particles and also bacteria might be additional food sources and references therein. Unfortunately, no studies are available on food preferences of H. thalassius. Although Hemicyclops is an egg-sac spawner, virtually no nauplii were found over the experimental runtime. Using a 100 µm mesh net, we may have missed the younger nauplii but should have captured the older naupliar stages . From the absence of nauplii despite the frequent presence of egg-sac-carrying females, we conclude that the Hemicyclops populations in the mesocosms, like Paracalanus, also lived at suboptimal conditions in terms of both nutrition and dissolved oxygen concentrations.

Isotopic signatures of zooplankton are determined by source and trophic fractionation. The latter is particularly the case for δ15N, for which an increase of 3 ‰–4 ‰ per trophic level is usually assumed for marine food webs . Zooplankton δ15N values ranging approximately between 5 and 13 ‰ occur in (and close to) upwelling areas e.g.,, whereas lower zooplankton δ15N values (∼2 ‰–4 ‰) were found in regions with a high diazotroph contribution to the pelagic food web . δ15N values of zooplankton (copepods) in the mesocosms following a simulated upwelling pulse were rather high (∼13–17 ‰) compared to the δ15N (mean 7.6 ‰) measured by for bulk copepod samples of the HCS. The δ13C ratio of copepods from M4 differed profoundly from the other mesocosms in phase 2 (day 13–36). The decrease in δ13C seems to correlate with an increase in Chlorophyceae, which made up >90 % of the phytoplankton community on day 36 . Turnover time of δ13C in copepods varies from days to weeks . An increase in Chlorophyceae and following constant supply should be seen in the C signal of the copepods after several days. Also, in contrast to the other mesocosms, δ15N was generally lower in M4 in the first two phases. M4 was the only mesocosm lacking the dinoflagellate Akashiwo sanguinea . Mixotrophy by this flagellate could thus have led to increased nitrogen fractionation in the other mesocosms.

A comparison of δ15N among different studies without clearly defined baseline values (phytoplankton) is difficult, as the δ15N baseline is strongly affected by the variability in the intensity of the OMZ along the Peruvian coast . A recent study by revealed an enrichment in δ15N of ∼5 ‰ for several copepod species of the HCS from north (8.5∘ S) to south (16∘ S). The δ15N ratios of seston also increased with depth, i.e., between seston from surface oxygenated waters and the OMZ. Dissolved inorganic nitrogen taken up by phytoplankton often creates the basis of the isotopic composition of the whole food web . Under anoxic conditions in OMZs, a preferred uptake of the lighter isotope by denitrifiers leads to elevated δ15N in the remaining nitrate . OMZs such as the strong and shallow one in the northern HCS are thus sites of intense nitrogen loss , leading to an increase in δ15N baseline values . The higher δ15N of copepods within the mesocosms can therefore be a result of or may be explained by the source water injection to the mesocosms, which was reduced in oxygen and most likely enriched in δ15N. Starvation would lead to loss of body mass and preferential metabolism of the lighter isotope, with a resulting increase in delta values for both tracers .