Functional redundancy in natural pico-phytoplankton communities depends on temperature and biogeography

Biodiversity affects ecosystem function, and how this relationship will change in a warming world is a major and well-examined question in ecology. Yet, it remains understudied for pico-phytoplankton communities, which contribute to carbon cycles and aquatic food webs year-round. Observational studies show a link between phytoplankton community diversity and ecosystem stability, but there is only scarce causal or empirical evidence. Here, we sampled phytoplankton communities from two geographically related regions with distinct thermal and biological properties in the Southern Baltic Sea and carried out a series of dilution/regrowth experiments across three assay temperatures. This allowed us to investigate the effects of loss of rare taxa and establish causal links in natural communities between species richness and several ecologically relevant traits (e.g. size, biomass production, and oxygen production), depending on sampling location and assay temperature. We found that the samples' biogeographical origin determined whether and how functional redundancy changed as a function of temperature for all traits under investigation. Samples obtained from the slightly warmer and more thermally variable regions showed overall high functional redundancy. Samples from the slightly cooler, less variable, stations showed little functional redundancy, i.e. function decreased when species were lost from the community. The differences between regions were more pronounced at elevated assay temperatures. Our results imply that the importance of rare species and the amount of species required to maintain ecosystem function even under short-term warming may differ drastically even within geographically closely related regions of the same ecosystem.


72
Sampling and on-board treatment and husbandry 73 74 We obtained pico-phytoplankton community samples during two RV ALKOR cruises 75 (AL505 and AL513 respectively) in 2018 (see Table S1 for a time line, Figure 1 and Table S2  76 for sampling dates and locations, Table S3 for decomposition analysis output regarding the 77 environmental fluctuations characterising the sampling regions) using a Niskin bottle at 5m. 78 The Niskin bottle was solitary, and dispatched via a controlled crane. As the CTD data 79 revealed that surface waters were fully mixed, one Niskin-sample of 10L was taken per 80 station. As the Baltic Sea is higher in biomass than for example oligotrophic ocean waters, we 81 found that of these, 2L sufficed for all experiments that followed. Water from each station 82 was immediately passed through a 35µm sieve to remove grazers and large debris, and then 83 further size fractioned via gentle filtration with a vacuum pump at the lowest setting. During 84 filtration, we first passed the water sample through a 2µm membrane filter (kept filtrate) to 85 remove organisms larger than the picoplankton fraction, and then an 0.2µm filter. On the 86 0.2µm filter, we concentrated the 2L-filtrate to an end volume of 200mL. Great care was 87 taken to not let the filter never fall dry to ensure that cells did not get stuck in or were 88 damaged by the pores on the membrane. The filter was rinsed gently with the remaining water 89 above the filter such that the organisms were continuously more concentrated. 30mL of the 90 0.2µm filtrate were frozen for nutrient analyses in technical duplicates. We used Whatman-91 Nuclepore polycarbonate track-etched membrane filters with a size of 47mm for all filtration 92 processes. 93 94 Throughout the cruise, acute thermal profiles of photosynthesis and respiration for the 95 communities were determined during on-board incubations in order to better be able to 96 estimate which temperatures to use as assay temperatures in the laboratory. Time taken for 97 sample preparation (filtration, incubation of samples in the dark prior to photosynthesis 98 measurements) is on the scale of hours. The measurement of a full photosynthesis-irradiance 99 curve on an oxygen electrode takes about 20 minutes, including a dark phase for respiration. 100 As such, we can be fairly certain that our measurements tracked responses to temperature 101 within the same generation. 102 On board, an aliquot of each community was immediately frozen in sorbitol for later (upon 103 return to Hamburg) analysis on the flow cytometer. 104 105 All communities were transferred into full f/2 media [1] at the salinity of the sampling 106 location to rule out effects of parameters other than temperature and diversity during the 107 experiment. AL505 samples from an in-situ temperature of 1-2ºC were first stored at 4ºC for 108 24-48 hours, and then in a 10ºC cold storage room on board for the remainder of the cruise (2-109 12 days depending on cruise and day of sampling). We used LED light stripes for an 110 irradiance of approximately 100 µmol quanta m -2 s -1 , at a 12h/12h light/dark cycle. Irradiance 111 in Baltic Sea surface waters can fluctuate dramatically (between 30 µmol quanta m -2 s -1 and 112 3000 µmol quanta m -2 s -1 ) within even a day. 100 µmol quanta m -2 s -1 was found to be a light-113 intensity suitable for culture under the conditions on board and in our laboratory. We suggest 114 that for similar studies, each researcher carry out pilot experiments to establish the appropriate 115 light levels. 116 AL513 samples from an in-situ temperature of 21ºC to 23ºC were also stored in the cold room 117 (at 10ºC) using the irradiance and media conditions above. We have found that this does not 118 'shock' the samples, but puts them into a gentle stasis until further use, so as long as the 119 period at colder temperatures does not exceed 2-3 weeks. 120 121 Treatment and husbandry of communities in the laboratory 122 As during the time on board, to rule out effects of parameters other than temperature and 123 diversity during the experiment, all samples were grown in f/2 media [1] at the salinity of the 124 sampling location. Community samples grew in semi-continuous batch culture at 100 µmol 125 quanta m -2 s -1 (12:12 light/dark cycle) in 40mL of media using vented-cap bent-neck, fully 126 transparent Nunclon® flasks. Batch-transfers occurred fortnightly, and at least at these 127 (sometimes in between) detailed cytograms were taken to track community composition (see 128 below for details on flow cytometry). Communities from AL505 were kept at 15ºC for 11 129 months. Communities from AL513 were kept at 22ºC for 7 months. 130 131 Rationale for culturing temperatures in the laboratory until start of experiment 132 We had to walk a very fine line between multiple requirements (not all of them in our hands).: 133 i) biomass in the samples was, while not too low for metabolic measurements, too low for 134 MOTU analysis, so a growing period would have been necessary in any case (where time, 135 space, and logistics allow, filtration of much larger samples may also help) ii) this project was 136 carried out as part of an MSc/MRes thesis. To make sure that we stayed within a range of 137 parameters that allows for good growth and a time-frame that is manageable for such a 138 project, we had to choose a higher temperature than was found in situ at time of sampling 139 (growth of the community samples even at 5ºC-8ºC is extremely slow and the experiment 140 would have taken a year, and net photosynthesis rates are near the detection limit at these 141 temperatures), iii) 1-2ºC are not a common temperature for the Southern Baltic Sea in March. 142 Water temperatures between 4ºC and the low double digits are much more common (we 143 sampled during an unexpected cold snap). The photosynthesis measurements carried out on 144 board indicated that samples did best at temperatures exceeding 15ºC. This is not surprising 145 given the usual spring temperatures in the Baltic Sea, and a tendency for ectotherms to have 146 their thermal optima slightly above usual environmental levels [2]. 147 Nevertheless, we took great care that AL505 communities were gradually transferred to 148 warmer temperatures (see above) 149 For communities retrieved during AL513, we chose an incubator temperature of 22ºC based 150 on thermal performance curves. 151 152 Rationale for using a common garden approach 153 As is inherent to experiments carried out on samples obtained at different times of the year, 154 one faces the decision to either carry out experiments as the samples arrive and have a 155 confounding effect of time within the laboratory (e.g. effects of having to use different 156 batches of media, dealing with the shelve-life of lighting systems, which may change over the 157 course of weeks and months), or have one set of samples spend more time in the laboratory 158 than the other. We decided for the latter, and cultured samples in a common garden prior to 159 the beginning of the dilution experiment. 160 161 A common garden is an approach often found in ecology and evolutionary biology [3,4] (the 162 name originating from the plant sciences): here, species or communities from different native 163 environments are transplanted into a common environment that is different from either 164 species' or community's native environment. If the native environment did not matter, the 165 different organisms would rapidly display the same phenotypes in the common garden. This 166 means that any differences we measure despite the time in the common garden are robust and 167 indeed attributable to where the organisms came from. In the end, we agreed on a common 168 garden temperature of 18ºC. From another set of experiments that measured the thermal 169 tolerance curves (i.e. growth of the communities across a temperature gradient -currently in 170 prep for another publication), we know that at 18ºC, for samples from spring and summer, 171 community composition remains relatively stable, that samples can be grown to good biomass 172 concentration in a manageable time-frame, and that growth as well as photosynthesis rates can 173 be obtained easily. After culture in the common garden for two months (see illustrated time 174 line below), we carried out an in-depth pilot study using lower levels of replication. Including 175 all trouble-shooting and analysis, this took another 5 months, during which stocks were kept 176 in the common garden with regular tracking of growth rates and community composition, 177 until we were confident in the methods to start the experiment as described here. While we 178 did not find that our pilot study findings deviated from the results described here, we do not 179 report them due to the lower replication. 180 181 It is extremely likely that our communities as they entered the laboratory, and finally, the 182 common garden were not a perfect replicate of the communities in situ (especially on levels 183 not even measured here, e.g. the bacterial and viral component), and we would ideally have 184 kept the samples in the laboratory for much shorter time periods prior to the measurements. 185 However, even a sample taken from the body of water and used directly on board will not be 186 a perfect replicate (as e.g. some species might not be so amenable to the filtration process). To 187 find a compromise between investigating near-natural communities (rather than assembling 188 single species from culture collections) and still making use of the controllable nature of 189 laboratory experiments, we took great effort to continuously monitor the cytometric 190 characteristics of all samples. We provide estimates of phenotypic diversity (see below for 191 calculation) of samples at point of freezing on the ship, and at point of entering the common 192 garden in Figures S7 -S9. We note that this does not yield information on the identities of 193 species present (or changes in phenotype without underlying genetic change), but does tell us 194 how phenotypically diverse samples were throughout time, which is our main question here. 195 We found that while some phenotypic characteristics differed between samples at t0, the 196 initial incubator culture, and the common garden period (especially size -cells initially 197 became a bit larger in laboratory culture), phenotypic diversity declined slightly at first, but 198 then remained almost unchanged (see Figures S8 and S9 respectively), and were further found 199 to not change much during the growth cycle (Figures S10 and S11).. 200 8 Below, we provide a rough time-line of the experiment (note that back-ups of stocks were kept in the common garden throughout). Numbers refer to months and start in March 2018. Table S1: Time-line of experiment, detailing the time spent in incubation at 15ºC for spring samples (AL505), and 22ºC for summer samples (AL513), the common garden period, a pilot study, and the final experiment. Samples from the Kiel Basin grew faster than samples from the Bornholm Basin. Even though this resulted in the Kiel samples' spending more generations in the laboratory than the Bornholm samples, we can be positive that after an initial loss of species (see Figures S7-S9 Full study -growth curve 1 to µmax Full study -growth curve 2 to K Growth rate and composition checked at least fortnightly Table S2: Sampling station overview Below, we provide the coordinates (Long/Lat), time of sampling (spring or summer 2018 and official ALKOR identifier), as well as salinity, temperature, and nutrient content at time of sampling for each station. See also map in Figure 1 in the main text. Three technical replicates were established for each Station at each temperature and each level of dilution. StationID is as used throughout this manuscript, and not an official station identifier. The official WERUM ID is given in brackets. As each individual station was only sampled once for nutrient content, temperature, and salinity as is standard, we do not provide standard deviations as they would not carry any true meaning (technical replicates for nutrient analyses were established in the laboratory). Temperature and salinity data are as exported from the ship's CTD. Nutrient content was measured on a SEAL sequential analyser (AA3) following protocols of [12,13]  Setting up the dilution experiment 1 To set up the dilution experiment, we first counted cell numbers in the Kiel Bight community 2 samples (3 stations, liquid culture, non-frozen samples) and Bornholm Basin community 3 samples (3 stations, liquid culture, non-frozen samples) using a BD Accuri C6 flow 4 cytometer. More than 3 stations per basin had been obtained on board, but for the sake of 5 keeping the total number of experimental units within a manageable range, we focused on 3 6 stations per basin. The cell counts also yield flow cytometric fingerprints that allow for an 7 estimate of phenotypic diversity or trait-level diversity [5] which is largely based on photo-8 pigment composition and size [6] [7] (see below for more details). Samples were then diluted 9 in six 10-fold dilution steps at the appropriate salinity, down to the lowest point of dilution (in 10 theory containing no more than 1 species or pico-phytoplankton per mL). Six technical 11 replicates of each sample in the dilution series (i.e. region*station*dilution) were left to 12 regrow to 10 6 cells mL -1 at the assay temperatures of 15ºC, 18ºC, and 22ºC. These 13 temperatures are all within the ranges of temperatures commonly experienced during late 14 spring (15ºC), summer (18ºC), and the height of summer (22ºC). This resulted in a total of 15 648 unique experimental units. From the time of dilution, samples were cultured on 48 well 16 plates (1.5mL), which provide a space-(and plastic) saving alternative to larger culturing 17 vessels. We had tested beforehand that between-treatment differences did not change 18 significantly with the culture vessels used. 19 20 Then, we re-diluted all samples to 3000 cells mL -1 and tracked two consecutive growth 21 curves: One, where samples were harvested for net photosynthesis measurements at µmax, 22 followed by a full growth curve to carrying capacity (ca. 23 days, see below for details as not 23 all samples reached K the same day) in all experimental units at all temperatures, with 24 measurements taken on the flow cytometer every other day. We found that growth at µmax 25 did not differ between the first and second growth cycle. Yet, growth hinged on a 26 combination of dilution, region of origin, and assay temperature. As a result, the points of 27 µmax and carrying capacity were not reached on the same day for all samples in either growth 28 cycle. Supporting Figure S17 has the growth rates at µmax, and Table  29 "20200606_timetoK.csv" on data dryad details the times at which carrying capacity was 30 reached. We would like to point out that in order to keep the experiment manageable, 31 different growth rates indeed are an advantage rather than a disadvantage, as measurements 32 can be spread out throughout multiple days and importantly can be carried out at the same 33 time of day for each sample to account for effects of circadian rhythms on metabolic 34 processes. 35 36 Estimation of cell size 37 Cell size as diameter in µm was obtained from the flow cytometer's forward scatter after 38 calibration with size beads. Taking into account cell counts per mL and assuming on average 39 spherical shapes and using conversion factors after [8], we then calculated an estimate of pg 40 carbon per mL to obtain biomass produced. 41 42 Estimation of Net Photosynthesis 43 Net photosynthesis rates were obtained when samples were in exponential phase, on PreSens 44 ® SDR Sensor Dish optodes. Here, we aimed for a total of 10 4 -10 5 cells in 4mL 45 measurement vials (the optodes sit on the bottom of each vial). To achieve this cell density, 46 aliquots from the harvested experimental units had to be diluted in the appropriate media and 47 salinity. PreSens optodes are pre-calibrated by the manufacturer. A headspace of oxygen can 48 be eliminated by filling samples to the rim and sealing off with parafilm. We measured 49 oxygen production for 15 minutes in the light, and respiration for 15 minutes in the dark. 50 Whenever an experimental unit was run on the PreSens optode, it was also run (in its diluted 51 state) on the flow cytometer to allow for per cell estimates. Net photosynthesis was calculated 52 considering that phytoplankton in our set-up will only be able to photosynthesise during the 53 light phase (12 hours), but will respire throughout the day and night phase (24 hours). All 54 measurements were carried out at the same time of day ( ~ 9am to 11am) under the light-and 55 temperature conditions set in the incubator (i.e. all experimental units at their assay 56 temperatures). 57 58 Molecular analysis of diversity (as species richness) 59 We obtained two measures of biodiversity in our samples. One, following CTAB DNA 60 extractions [9], a subset of representative samples was sent for DNA-meta-barcoding at 61 biome-id (16S primers: forward CCTACGGGNGGCWGCAG, and reverse 62 GACTACHVGGGTATCTAATCC, 18S primers: forward CCGCGGTAATTCCAGCTC and 63 reverse CCTTGGTCCGTGTTTCTAGAC), resulting in a MOTU (meta-barcoding 64 operational taxonomic units) estimate for those samples. A MOTU is grouped by DNA 65 sequence similarity of a specific taxonomic marker gene, here 16S and 18S. 66 In total, we sent off three DNA pellets for each region for each dilution step. As we found that 67 MOTU scales well with phenotypic diversity, we forewent further MOTU analyses in favour 68 of cheaper and faster phenotypic diversity measurements. 69 70 Flow cytometric analysis of diversity 71 As molecular analyses are infamously costly, we chose phenotypic diversity [7] as our 72 second measure of diversity. This was assessed using the parameters returned by the flow 73 cytometer (abbreviations in Table S4). 74 75 On slow sampling rates of 14µL/minute, we used an aliquot of 50µL of each unique 76 experimental unit to obtain detailed cytograms (for tracing growth curves, 10-20µL often 77 suffice and flow rates can be chosen at faster settings). Larger aliquots do not yield better 78 cytograms, and only serve to clog up the flow cytometer. The aliquot taken from each 79 experimental unit was replaced by nutrient-free medium of the correct salinity and the 80 resulting (small) dilution was incorporated in the growth rate measurements. 81

82
We first stained aliquots of the sample with SYBR Gold, alongside a 0.2µm filtered MiliQ 83 sample. This allows us to distinguish debris and cytometer noise from living matter (see 84 below) on the FL1 channel (FITC in cytogram display). Below, we show an example for a 85 thresholded fraction containing debris, as well as bacterial (E4), viral (R3), and assumedly 86 pico-eukaryotic matter (E3 along -side beads of known size (R2, 1µm microspheres from 87 invitrogen), with FITC on the y and SSC on the x axis. Our gating strategies are in line with 88 [10]. Knowing where the DNA positive clouds lie, and which parts to exclude as debris/cytometer 97 noise, we then further gated for FL3 (Chl-a proxy) positive organisms (DNA positive but FL3 98 negative were also tracked to get an idea of the heterotrophic fraction, but not used for this 99 study) comparing known bacteria, known single species phytoplankton, and community 100 samples. Depending on whether one is interested in tracking the bacterial compound, one 101 either choses to only count organisms within the FL3 positive gate or quadrant, or 102 alternatively, one can set the thresholds so that very small, low-FL3 events are automatically 103 excluded from the display. 104 105 Below, we provide an example of FL3 (PerCP) against FSC with a fairly high threshold of 106 2500 on FSC and FL3. This is the fingerprint of a fairly high diversity Baltic Sea community 107 sample. We would consider events in Q1-UR for further analysis.  resulting from a breakdown of our flow cytometer half-way through the experiment, which 119 necessitated repeated re-calibrations (e.g. of the rental flow cytometer, and the original flow 120 cytometer after repair) associated with a shift in the 'absolute' location of the cloud. 121 122 Importantly, the chosen gates, quadrants, or thresholds, still allow us to pick up organisms 123 that might not have their main photosynthetic pigments detected by FL3, as even organisms 124 higher in FL2 and FL4, but low in FL3 will fall within this gate, but not the debris. An 125 example can be found below for one of our high and low diversity experimental units: 126 We also compared the chosen gating region to beads of known size and a known 138 Ostreococcus (a picoplankton of about 1.5µm diameter) sample, to make sure we were 139 capturing the full picoplankton community. Beads of known size run without an organism 140 will result in cytograms akin to this (beads close to the size of the fraction under examination 141 can also be run alongside the sample for direct comparison. On data dryad 142 The matrices containing the raw flow cytometry data can also be used to compare samples to 163 each other (also throughout time), akin to beta diversity, via any code that creates similarity or 164 dissimilarity matrices, i.e. a simple PCA or NMDS plot for graphic representation or a 165 PERMANOVA for statistical analysis (we use the R package vegan for this purpose). We find 166 that for comparing how samples change through time or how samples from different regions 167 differ from each other, a similarity/dissimilarity matrix based on means rather than individual 168 measures, yields the same results as calculations based on individual cell measurements, but 169 at much faster computing speeds (a few minutes compared to more than an hour). We make a 170 point that where time or computer power is a limiting factor, using mean data frames is a 171 valid option. 172

173
We would like to add that for this manuscript, the gating on the Accuri software is merely to 174 aid the researcher as they observe the samples being counted (e.g. to immediately spot 175 contaminants or issues with the cytometer). We exported the full raw fcs data files for gating 176 and de-noising to be carried out in R (following the same gating steps) within the PhenFlow() 177 package. We provide higher quality versions of the cytograms shown here on datadryad 178 (https://doi.org/10.5061/dryad.0p2ngf1xw). The authors are happy to provide raw fcs files 179 upon reasonable request. 180 181

Statistical analysis 182
All data were analysed in the R programming environment (version 3.5.3.). To analyse the 183 shape of the growth curves, non-linear curve fitting of a baranyi growth model [11] was 184 carried out using the 'nlsLM' function in the R package, 'minpack.lm'(version 1.2-1). 185 Parameter estimation was achieved by running 1000 different random combinations of 186 starting parameters for cell count at carrying capacity, duration of lag phase, and maximum 187 growth rate picked from a uniform distribution. The script then retains the parameter set that 188 returned the lowest Akaike information criterion (AICc) score. Parameters (biomass and cell 189 size at carrying capacity, net photosynthesis during exponential growth) were then compared 190 through a mixed effects model (within the nlme package, version 3.1-137). There, the 191 respective parameters were explained by a global model that included sampling location (Kiel 192 Bight or Bornholm Basin), assay temperature (15ºC,18ºC, or 22ºC), and dilution step (from 193 lowest to highest) and sampling season (spring or summer) as fixed factors in full interaction. 194 Sampling station was computed as a nested random effect within region. In all cases, 195 seasonality was found to not explain the data better and was subsequently dropped from the 196 fixed factors to avoid over-parameterisation of the model. For multi-model selection, we 197 computed small sample-size corrected AIC scores (AICc) and then compared the models by 198 calculating delta AICc values and AICc weights using the "MuMIn" package (version 1.42-199 1). We picked the model where delta AICc was > 2 for refitting with REML. PERMANOVAs 200 were carried out using the "ecodist" (2.01) packages. Distance matrices using the Bray-Curtis 201 index were created from these, on which we ran PERMANOVAs to test for separation of 202 samples by treatment. Pairwise contrasts between treatments were examined via the function 203 permdisp() followed by TukeyHSD post-hoc tests. 204

205
For graphical presentation of data, we used the ggplot2 (version 3.2) and vegan (version2 .4)  206 packages. While NMDS plots are common for the comparison of ecological sampling sites, 207 we found that the distance matrices did not differ significantly from PCA plots, and are 208 presenting the latter throughout for their more direct compatibility with PERMANOVA 209 results. The relationship between the logarithms of dilution and MOTU count reveals that the 220 dilutions successfully reduced species richness in Kiel Bight (orange) and Bornholm Basin 221 (blue) samples. Kiel samples had slightly higher original MOTU counts, which was driven 222 largely by a slightly lower species count and higher predominance of cyanobacteria in the 223 Bornholm Basin samples during the summer (see also below). The boxplots are displayed as 224 is standard, with the girdle band indicating the median, and the whiskers extending to the 25th 225 and 75th percentile. For each unique treatment combination (dilution*region*temperature). 226 Due to the high costs involved in meta barcoding we sent off 3 samples for each region and 227 dilution (after common garden culture at 18ºC). (This is a larger version of subpanel C in 228 Figure 1 in the main manuscript). returned from meta-barcoding showed that any differences were driven solely by the summer 247 months seeing a higher abundance of cyanobacteria in the Bornholm region. 248 Here, each "Kiel" or "Bornholm" identity on the plot contains the mean information on 249 phenotypic characteristics per unique experimental unit (a plot with this information per cell 250 per sample would be beyond readable). (This is a larger version of subpanel C in Figure 1 in 251 the main manuscript). In Figures S8 and S9 we show that while this phenotypic composition 252 differed slightly to samples at t0 (i.e. frozen directly after filtration), phenotypic composition 253 then remained largely stable between the time spent in the incubators at 15º/22ºC and the 254 common garden at 18ºC. Further, phenotypic diversity also remained largely unchanged, e.g. 255 while cells on average increased in size after being brought to the laboratory, diversity 256 eventually stabilised. 257 24 258 259 Figure S8: Phenotypic characteristics for Kiel Bight and Bornholm Basin remained largely stable between incubation at 15ºC or 22ºC and 260 the common garden period, but differed slightly from t0 261 Phenotypic characteristics as detailed in the methods and also shown in Figure S2 remained stable between culture in the incubators set to 15ºC for 262 March 2018 samples/22ºC for July/August 2018 samples and the common garden at 18ºC. As is to be expected, there were some differences to the 263 original samples ("t0") frozen in sorbitol immediately after on-board filtration. Phenotypic diversity decreased slightly during laboratory culturing, 264  but eventually stabilised (see Figure S9). Each "t0" or "temp incubator" or "common garden" identity on the plot contains the mean information on 265 phenotypic characteristics per unique experimental unit, with several measurements carried out for each time point per station per replicate. 266 In the cytometric output for communities from the Kiel Bight (upper row, red tones) and the Bornholm Basin (lower row, blue tones), we can see 282 that there was no overall significant change in community characteristics throughout the growth curve (PERMANOVA F2,13= 2.35, p =0.08). 283 However, at 22ºC Kiel Bight communities at carrying capacity seemed to develop a lower chlorophyll phenotype, and during the lag phase, 284 Bornholm Basin communities at 22ºC showed on average higher cell size. We show in Figure S11 that this did not affect phenotypic diversity 285 significantly throughout the growth cycle. Lag is for lag phase, exp for exponential phase, and K for carrying capacity 286 Biomass at carrying capacity (here in pg C per mL, displayed as LOG10 for clarity) in 297 samples from the Kiel sampling stations (orange, upper) and the Bornholm sampling stations 298 (blue, lower) was influenced by assay temperature (individual panels) and dilution (labelled 299 as 'species richness'. Here, displayed as the LOG10 of phytoplankton cells after dilution, 300 which is a good indicator for species richness (see Figure S5 and main manuscript Figure 1). 301 A slope that does not deviate significantly from 0 (see also Log phytoplankton biomass mL -1 at carrying capacity Species richness dilution). The boxplots are displayed as is standard, with the girdle band indicating the 308 median, and the whiskers extending to the 25th and 75th percentile. For each unique treatment 309 combination (dilution*region*temperature), n=6. Standard deviations for the slopes can also 310 be found in Table S5. Shaded areas in the plot are confidence intervals generated in R and 311 mainly for graphical representation. 312 313 314 Figure S13: Cell count at carrying capacity 315 Cell count mL -1 at carrying capacity (here displayed as LOG10 for clarity) in samples from 316 the Kiel sampling stations (orange, upper) and the Bornholm sampling stations (blue, lower) 317 was influenced by assay temperature (individual panels) and dilution (labelled as 'species 318 richness'. Here, displayed as the LOG10 of phytoplankton cells after dilution, which is a good 319 indicator for species richness (see Figure S5 and main manuscript Figure 1). A slope that does 320 not deviate significantly from 0 (see also displayed as is standard, with the girdle band indicating the median, and the whiskers 327 extending to the 25th and 75th percentile. For each unique treatment combination 328 (dilution*region*temperature), n=6. Standard deviations for the slopes are in Table S5. 329 Shaded areas in the plot are confidence intervals generated in R and mainly for graphical 330 representation. 331 332 At carrying capacity, cell diameter in samples from the Kiel sampling stations (orange, upper) 335 and the Bornholm sampling stations (blue, lower) was influenced by assay temperature 336 (individual panels) and dilution (here, displayed as the LOG10 of phytoplankton cells after 337 dilution, which is a good indicator for species richness (see Figure S5 and main manuscript 338 Figure 1). At 15ºC, dilution did not significantly affect cell size. At 18ºC, dilution affected 339 cell size only in samples from the Bornholm region. At the highest temperature (22ºC), cell 340 size strongly decreased when communities were more diverse in samples from both regions. 341 A slope that does not deviate significantly from 0 (see also richness has a strong impact on the trait under investigation, although the implications of a 344 slope deviating from 0 are less clear for size than for biomass and photosynthetic activity. For 345 each unique treatment combination (dilution*region*temperature), n=6. The boxplots are 346 displayed as is standard, with the girdle band indicating the median, and the whiskers 347 extending to the 25th and 75th percentile Standard deviations for the slopes can be found in 348 Table S5. Shaded areas in the plot are confidence intervals generated in R and mainly for 349 graphical representation. (blue, lower) was influenced by assay temperature (individual panels) and dilution (labelled 359 as 'species richness'. Here, displayed as the LOG10 of phytoplankton cells after dilution, 360 which is a good indicator for species richness (see Figure S5 and main manuscript Figure 1). 361 We fitted a slope through LOG10 transformed NP data as this transforms the otherwise 362 exponential relationship into a linear one (see Figure S16). Here, we show the non-363 transformed data for easier visualisation, as LOG10 transformed data of very small values 364 will be negative. A slope that does not deviate significantly from 0 (see also This is a LOG10 transformed version of Figure S15 for better visualisation of the slopes. All 385 details are as in Figure S15). 386

387
Figure S17: Growth rates (at µmax) 388 During exponential growth, growth rates in samples from the Kiel sampling stations (orange, 389 upper) and the Bornholm sampling stations (blue, lower) was influenced strongly by 390 geographical origin and assay temperature (individual panels) but only to a smaller degree by 391 dilution (labelled as 'species richness'. Here, displayed as the LOG10 of phytoplankton cells 392 after dilution, which is a good indicator for species richness (see Figure S5 and main 393 manuscript Figure 1). The growth rate values mainly serve to show that experimental units 394 reached the time-points of µmax, and hence carrying capacity, at different points in time and 395 therefore had to be harvested/measured across several days. For each unique treatment 396 combination (dilution*region*temperature), n=6. The boxplots are displayed as is standard, 397 with the girdle band indicating the median, and the whiskers extending to the 25th and 75th 398 percent. We provide the time points at which K was reached at in the data dryad files.     This is a statistics summary (degrees of freedom and F statistics). We recommend that readers look at Tables S7 to S9 in this document for details. In all tables ":" denotes an interaction between factors. Temperature refers to the assay temperature. The F statistics are reported for the model found to be the best model based on AICc scores, not the global model. The denominator DF is lower than the total number of samples minus the number of treatment groups because of the nested nature of the model.

selection (A) output (B) for investigating the effect of dilution (abbreviated to D), assay temperature (abbreviated T), region (abbreviated R), and season (abbreviated S) on biomass produced at carrying capacity.
In the mixed model, D (from 1 -highest richness to 1e-05 -lowest richness), T (15ºC,18ºC, 22ºC), R (Kiel Area, Bornholm Basin), and S (spring, summer) selection regimes, i.e. nutrient (low nutrient and replete), were fitted as fixed effects. Stations were treated as a random factor. Technical replicates were not fitted. Here and in all other model selection tables, the header indicates the factors considered by a model. When a factor is part of the model, this is shown by a +. When a factor is not considered by a model, this is shown by NA. The best model is highlighted in bold, and is the model with the smallest AICc, where delta AICc to the next best model is >2. By tracing the "+" and "NA" we can see which factors in which combination are or are not part of the model. df for degrees of freedom; logLik for log likelihood ratio. : indicates an interaction term. We display only the first 10 models for clarity. The global model formula was lme.formula(K~D*R*S*T, random=~1|bio.stat.id, data=dataframe.K, method="ML) . The model used for the model output