Philippa Kaur

All experimental procedures were permitted by the Institutional Animal Care and Usage Committees of the Grassland Science College of Gansu Agricultural University (GSC-IACUC-2015-0011). Our experiments were conducted according to their guidelines, which are in accordance with the Guide for the Care and Use of Laboratory Animals (the Constitution of Experimental Animal Ethics Committee of Gansu Agricultural University). All experiments were performed in accordance with ARRIVE guidelines.Animals and laboratory conditionsAdults of both sexes (three males and three females) were captured in April 2015. Specifically, the animals were captured in Mayin Tan (37°12′N, 102°46′E; Tianzhu Tibetan Autonomous County, China) using live traps25 set at fresh surface mounds. The individuals were then transported to the laboratory and housed in an acrylic box with a pipeline covered with soil. The box and pipeline were covered by black cloth to simulate the dark environment of plateau zokors. The temperature in the room was maintained between 20 and 25 °C. Food was supplied daily and consisted of potatoes, lettuce and carrot. After three days of acclimatization to the laboratory the animals were used in the different experiments. Our laboratory is located 2 km away from the field site. At the end of the experiments all animals were returned to the capture site in good health.Laboratory testing arenaThe experimental setup for the laboratory experiments was as follows (Fig. 1): A transparent Perspex tube (8 cm × 8 cm × 80 cm) was joined to the side of the dark acrylic box (40 cm × 40 cm × 40 cm). A rubber stopper was inserted into one end of the tube to avoid effects from the external environment. Treatment apparatus was placed into the rubber stopper (see “Laboratory treatment apparatus” section, below), and, to avoid the apparatus being damaged by the animals, wire mesh (8 cm × 8 cm × 0.5 cm) was placed about 15 cm from one end of the tube. A mercury thermometer was inserted into the tube in the middle to monitor the tube’s temperature. Between experiments with different animals, the box and tube were wiped with 95% alcohol and then with distilled water.Figure 1Schematic drawing of the setup used to test burrow-sealing behavior in plateau zokors in the laboratory. (1) acrylic box covered with soil 30 cm in depth; (2) experimental animal; (3) mercury thermometer; (4) transparent Perspex tube; (5) the pipe’s support clip; (6) wire mesh (8 cm × 8 cm × 0.5 cm); (7) rubber stopper (8 cm × 8 cm × 5 cm).Full size imageLaboratory treatment apparatusA rubber stopper with seven holes was used for plugging one end of the tube (Fig. 2). The oxygen concentration, light, temperature, sound and gas flow were considered in this design.Figure 2Schematic diagram of the rubber stopper used to simulate the entrance plug of the burrow. (1) power supply; (2) light bulb switch; (3) electric wire switch; (4) oxygen cylinder; (5) in situ three-parameter soil gas analyzer; (6) voice recorder; (7) negative pressure drainage device; (8) rubber plug; (9) LED bulb; (10) the iron rod; (11) heating cord; (12) AVOXIVY speaker with 5 cm diameter.Full size imageOxygen treatmentTo avoid the oxygen that was delivered into the tube causing the gas to flow too strongly, become drier, and create a sound, a steel oxygen cylinder and thin hose (0.3 cm in diameter) were selected, and one end of the hose was connected directly to the oxygen cylinder with a humidifier bottle, while the opposite end was inserted into the rubber stopper (Fig. 2). Before beginning the experiment, we allowed the oxygen cylinder to sit for two hours at laboratory temperature to remove any temperature effects. A three-parameter soil gas analyzer (13.05.03Pro, Shanghai SAFE Biotech Co., Ltd, China) was used to monitor the oxygen concentration in the tube (Fig. 2).Light treatmentThe average light intensity—that is, 360 Lux from 8:00 am to 8:00 pm—was measured in the field. One end of a wire was connected to an LED light (1 Watt), and the other end to the power supply (Fig. 2).Temperature treatmentThe temperature in the burrow entrance in the field was about 3 °C warmer than that at a tunnel depth of 10 cm. As such, one end of a wire was connected to a heater strip and the other end to the electrical power supply (Fig. 2). A thermometer was inserted into the tube to monitor the temperature inside the tube (Fig.1). During the experiment period in the laboratory, we switched on or off to make sure the relatively constant temperature inside the tube. The temperature range inside the tube was 3.2 ± 0.27 °C .Sound treatmentWhen a burrow is opened, wind whistle can be produced around the burrow entrance. Accordingly, a voice recorder (PCM-D50, frequency response 50 Hz–40 kHz, Sony, Japan) was placed at the burrow entrance in the field to record the burrow-entrance sound, the duration of which was 30 min. In the laboratory, the two ends of a wire were connected to an AVOXIVY loudspeaker (diameter: 5 cm; impedance: 4 Ω; 50 Hz–20 kHz) and a voice recorder, respectively (Fig. 2). The recorded sound was played back with a 60 dB sound pressure level, as measured at the burrow entrance in the field (XL2 sound level meter, Nti Audion, Switzerland). The sound was repeatedly played within one hour.Gas flow treatmentTo avoid ambient atmosphere entering the tube, a negative pressure drainage ball with plastic tube (12 cm long, 2 cm in diameter) connected the tube through a rubber stopper (Fig. 2). The tunnel gas was inhaled by the ball, then we pinched the ball to blow the gas into the tunnel as gas flow treatment.Field treatment apparatusFor the field experiment, the apparatus consisted of a tube (40 cm long, 8 cm in diameter) and an alarm device. The alarm device was made up of a loudspeaker, two slide rails (15 cm long), two metal plates (approximately 7 cm in length and 3 cm in width), and three coiled metal springs (5 cm long, 2 cm in diameter). The three springs were joined to one of the metal plates, while the other metal plate was fixed on the slide rails. The two metal plates were touched by the plateau zokor when it was plugging, which triggered the alarm device, thus enabling us to know whether or not burrow-sealing behavior was occurring (Fig. 3). The aluminum tube with an oxygen device was embedded into the burrow. The soil covering the tube served as an excellent insulator, buffering the tube from the aboveground temperature (Fig. 4A). A steel oxygen cylinder and thin hose (0.3 cm in diameter) were applied by connecting one end of the hose directly to the oxygen cylinder with a humidifier bottle, and then the opposite end of the hose was inserted into the tube (Fig. 4A). A three-parameter soil gas analyzer (13.05.03Pro, Shanghai SAFE Biotech Co., Ltd, China) was used to monitor the oxygen concentration in the tube (Fig. 4A). Allowing sunlight to enter the burrow, a glass bottle, open at one end but closed at the other, was embedded into the burrow. We also used soil to cover the bottle, and there was a 5 cm gap at the surface (Fig. 4B). The aluminum tube with high thermal conductivity was embedded into the burrow. Again, we used soil to cover the bottle and retained a 20 cm gap (Fig. 4B).Figure 3Schematic drawing of the apparatus used to test the burrow-sealing behavior of plateau zokor in the field. (1) tube; (2) loudspeaker; (3) slide rail; (4) metal plate; (5) coiled metal springs.Full size imageFigure 4(A) Schematic drawing of the apparatus used in the oxygen treatment placed in the tunnel of the plateau zokor. (B) Schematic drawing of the apparatus used for the temperature and light treatments placed in the tunnel of the plateau zokor. (1) tunnel of the plateau zokor; (2) oxygen cylinder; (3) three-parameter soil gas analyzer; (4) plateau zokor.Full size imageProcedureIn the laboratory experiment, we tested three males and three females for their responses to each treatment. To avoid generating stress and habituation to treatments, zokors were tested for 12 h each day and there was one hour interval between treatments, and five days interval between round of testing for the same individual (Table 1). We performed a control experiment in which a rod was inserted into the burrow but no further treatment was applied, which allowed us to evaluate whether it was the treatment that was causing the burrow-sealing behavior. Before beginning treatment experiment, each zokor was tested 24 times (12 h × 2 days) under the control experiment. We determined the rod movement as occurrence of burrow-sealing behavior.Table 1 Times of the experiments for each treatment in the laboratory simulation.Full size tableIn the field experiment, we tested three zokors (one male, two females), and six zokors were caught in the cold season and warm season (three males and three females, respectively). We then fastened radio collars (Ag357, Biotrack, Ltd., UK) to each captured individual to allow us to track the position in foraging tunnels of each zokor. Each zokor was used three times in the experiments under each treatment, and, after finishing each experiment, we changed the position of the foraging tunnel to ensure the test tunnel was not an abandoned tunnel. According to radio-tracking data, the straight-line distance between the test tunnel and the nest for each treatment was about 5 m. We conducted a control experiment that whether plateau zokor move to the test tunnel or not during the time between treatments. In the cold season, from 4 October 2015 to 2 November 2015, the burrow-sealing behavior of each zokor was tested under different treatments during their active time (12:00–18:00) and inactive time (09:00–11:00) for a total of 27 days (Table 2). The same was done in the warm season but for a total of 18 days from 15 May 2016 to 5 June 2016, in which the active time was 14:00–20:00 and the inactive time was 08:00–13:00 (Table 2).Table 2 Times of the experiments for each treatment in the active and inactive periods of plateau zokors during the warm and cold season.Full size tableData analysisThe occurrence of burrow-sealing was recorded as “1”, and non-sealing was recorded as “0”. The frequency of burrow-sealing was the number of times the burrow was sealed divided by the total number of experiments for each treatment26, and we considered the frequency for each individual as a replicate. The latency to reseal the burrow was the period from the start of the treatment to the sealing of the burrow, and we considered each instance of latency to reseal the burrow as a repeat. The latency to reseal the burrow for non-sealing under each treatment was unavailable data and was therefore removed. The presence of a normal distribution in the initial data was determined using the Kolmogorov–Smirnov test. All data followed a normal distribution. A comparison of males and females in their frequency of sealing the burrow and in their latency to reseal the burrow under each treatment was performed with an independent-samples T-test. Multiple comparisons were made for the frequency of burrow-sealing and the latency to reseal the burrow under different treatments by using the least significant difference method at the significance level of P = 0.05. In the field experiment, the number of replicates was fewer than three for frequency and the latency to reseal the burrow, we did not conduct multiple comparisons.Preliminary statistical analysis of the data was performed using Excel 2013 and SPSS 19.0. All the figures and tables were produced in GraphPad Prism 8.0 and Excel 2013. More

Site, meadow, and fertilisationThe grassland utilized in this study was located in Sedico (BL: 420 m a.s.l., eastern Italian Pre-Alps), where the annual mean temperature is 10.6 °C and the annual rainfall is approximately 1366 mm (389, 326, 401, and 250 mm in spring, summer, autumn, and winter, respectively). The site was level and had an alluvial calcareous substratum. The soil was sandy-loam textured with 12.2% gravel content, 14.6% total carbonate content, and a pH of 7.5. Since 1977, a section of the meadow has been used for a fertilisation trial organised as three completely randomized blocks with 24 m2 plots and twenty-seven treatments obtained by combining three levels of yearly N, P, and K applications per ha: 0, 96, and 192 kg N as ammonium nitrate; 0, 54, and 108 kg P2O5 as triple superphosphate; and 0, 108, and 216 kg K2O as K sulphate. Since 2010, the grassland has been cut twice per year and surveyed for seed production in three treatments: no fertilisation (000), fertilization with no N and intermediate levels of P and K (011), and fertilisation with the highest nutrient rates (222).The vegetation of the three treatments (Annex 1) corresponded to the following meadow types: type 000, vegetation intermediate between a poor-soil form of the Arrhenatherum elatius meadow (Ar0) and a Bromus erectus meadow (Br), with high species richness and low legume abundance; type 011, an Ar0 meadow with high species richness and legume abundance; and type 222, a grass-rich form of the Arrhenatherum elatius meadow with low species richness and legume abundance.Plant sampling and laboratory analysisDuring the two growth periods within each of the years from 2012 to 2017, fertile shoots were sampled from the three fertilization treatments. 15–30 shoots (5–10 in each plot) from each flowering species were collected at the optimal seed maturation stage (most fruits/inflorescences still intact, i.e., no seed shedding). At the sub-plot level, all fertile shoots were collected at the time of meadow mowing on one 1-m2 sub-plot per plot. Collected shoots were put separately per species into porous paper bags, dried, and preserved in a refrigerator until laboratory analysis.During the autumn and winter after collection, the 15–30 shoots of each species were analysed for the number of compound and/or simple inflorescences and the flowers per simple inflorescence or shoot. For species with flowers or inflorescences too numerous to be rapidly counted, an inflorescence length or diameter which could be related to the number of flowers was also measured (e.g., the panicle length in grasses). In sample flowers, intact fruits, or simple inflorescences, the number of ovules per flower and the number of ovules transformed to seed were observed under a binocular microscope. Mature seeds were weighed and tested for germinability and viability according to ISTA19. Germination trials were performed with three seed samples per species, which were placed on filter paper in petri-dishes and moved to a germinator for 4 weeks (8 h light/25 °C and 16 h darkness/15 °C) with weekly observation and extraction of germinated seeds. At the end of the germination test, seeds that had not germinated were checked for viability with the tetrazolium test. Total viability was calculated as the sum of germinability and viability of non-germinated seeds.All shoots collected on the sub-plots were counted and measured for the number of inflorescences and flowers. When inflorescences and flowers were too numerous to be counted rapidly (e.g., in all grasses), only the same length/diameter measured on the 15–30 shoot samples was recorded.A more detailed description of the laboratory analyses is available in Scotton20.Data analysisThe value of each reproductive trait was calculated for each year and growth period at the plot level for each species. The values of the traits describing the size of the reproductive system were obtained from the shoots collected on the sub-plots. However, for species with too many flowers per shoot, a relation was calculated between the flowers per shoot and the length/diameter of the inflorescences measured on the 15–30 shoot samples. This relationship was then used to calculate the flower number for each shoot. The number of ovules per flower, the portion of ovules transformed to seed (ovule site utilisation, i.e. the filled seed/ovule ratio), the 1000-seed weight, germinability, and viability were calculated from the results of lab analyses of the 15–30 shoot samples.Because all the species collected were not always present in the six study years, only the thirty-two species (fifteen grasses and seventeen forbs: Table 1) found in at least three of the study years were considered in this paper to obtain enough reliable results. For all grasses, data were available only for the first growth period. For four forbs (see Annex 3) data were available for the first and second growth periods: in these cases, the average values of the two periods were used in the analyses.Table 1 Species studied for reproductive traits in a grassland fertilisation trial in the Italian eastern Alps.Full size tableThe statistical analyses were performed at the levels of individual species and the two grassland functional groups (grasses and forbs). Nine main reproductive traits describing the whole process of gamic reproduction were considered: number of simple inflorescences per shoot, flowers per simple inflorescence, ovules per flower, ovules and viable seeds per shoot, OSU (ovule site utilization), percent viability, germinability, and seed weight. Percent dormancy (the difference between percent viability and germinability) and the shoot density recorded in the subplots were also considered in some analyses.Only sixteen species were present in all of the fertilisation treatments, presenting a challenge in the tests that included all of the species together because a balanced among-treatments comparison was only possible by discarding the data from species not present in all of the treatments. To overcome this issue, we assumed that due to symbiotic N-fixation, the high presence of legumes in the 011 treatment (fertilization with P and K) was equivalent to a yearly N fertilisation of about 3.5 kg/ha per percent point of legume abundance in the species composition21,22. Therefore, treatment 011 (30% more legume abundance than in treatment 222: Annex 1) was regarded as an intermediate N addition of 105 kg per ha per year (from 3.5 kg N × 30% legume abundance). The values of the reproductive traits were then calculated for two fertilisation levels, low (LowFert) and high (HighFert). For species present in 000 and 011, LowFert was 000 and HighFert was 011. For species present in 011 and 222, LowFert was 011 and HighFert was 222. For species present in three fertilisation treatments, LowFert was 000 and HighFert was the average between 011 and 222. Statistical analysis considering only the species present in all fertilisation treatments yielded a similar pattern of fertilisation effects to those found in analysis of the two separate fertilisation levels. The analysis of the two fertilisation levels was therefore utilized because it was representative of a larger number of species.Statistical analyses (see summary in Annex 2) were conducted with three main aims: 1. studying the fertilisation effect on the reproductive behaviour of individual species and the two species groups of grasses and forbs; 2. finding species biological and ecological traits explaining their response to fertilisation; and 3. identifying multispecies correlations among reproductive traits and the possible effects of fertilisation on their patterns.For the first aim, the fertilisation effect was tested for the reproductive traits of each individual species through application of a mixed linear model under a repeated measure approach. In the model, fertilisation treatment, year, and block were input as class factors, and a plot identifier was used as a random factor. In case of significant fertilisation effects, the among-treatment differences were tested using the Tukey multiple comparison adjustment. Prior to performing the mixed model, data were checked for homoscedasticity and normality and, if necessary, log-transformed.From the individual species mixed models, a table was calculated containing the frequency of cases with fertilisation effects (three levels: no, positive, or negative) for each reproductive trait and species group. To check if grasses and forbs differed for the obtained frequencies, for each trait a chi-square test on the frequency table “fertilisation effect x species group” was performed.In a following set of analyses, the effect of the grassland functional group (grasses or forbs) on the multi-year means of each reproductive trait was tested with general linear models (GLM). Prior to the analysis, the data were sometimes log-transformed to mitigate homoscedasticity and normality problems. In these analyses, species were considered as replicates within the species group (therefore not included as a class factor) and the fertilisation level was input as a class factor. The effect of the fertilisation level on each reproductive trait was tested separately for the two species groups. In this case, the GLM included both fertilisation level and species as class factors.For the second aim, possible biological and ecological traits explaining the species response to fertilisation were investigated by relating the percent variation due to fertilisation in two important size traits (ovule and viable seed number per shoot: variables Y) to the following explanatory (X) variables: average values of the nine reproductive traits, the seven Ellenberg bioindicator values23, and the percent variation of shoot density. The relationships were fitted according to a linear regression approach for grasses and forbs together or separately and checked for the parametric assumptions of residual normality and homoscedasticity. For the percent variation due to fertilisation of individual species of OSU, seed germinability, viability and weight, one-way analyses of variance were performed where three traits of the species reproductive biology (type of reproduction, breeding system and pollen vector4: Annex 1) were used as categorical factors. A GLM approach was also used in this case.For the third aim, multispecies correlations were analysed by in-pairs relating the reproductive trait values of individual species averaged across fertilisation treatments and years. Fertile shoot density recorded in the subplots. was used as a supplementary characteristic. Nonlinear relationships were made linear with a log-transformation. Because the purpose of the analysis was not to predict one trait from the other but to efficiently summarise the relationships between traits, the standardised major axis (SMA) approach was used instead of the linear regression method24. The analyses were performed for grasses and forbs both together and separately. In order to verify if fertilisation could affect the characteristics of the evaluated relationships, a second set of SMA analyses were performed by separating the two fertilisation levels and the lines obtained were tested for common slope and elevation according to Warton et al.24.The year effect will be reported in a forthcoming paper and is therefore not discussed here, despite its inclusion in the statistical analyses.We used SAS/STAT software 12.325 with procedures MIXED, GLM, REG, and UNIVARIATE, and R 3.0.026 with package SMATR.Additional statementsThe experimental research and the collection of plant material was done according to relevant institutional, national, and international guidelines and legislation. No grassland species considered in the research is included in the list of endangered species according to the IUCN, European Union, Italian national and regional classifications. The collection of plant specimen was done with permission of the grassland owner during the hay-making operations for forage production which do not need a special permission from the concerned local authorities. The plant species were identified by the first author, Michele Scotton. A voucher specimen of each plant species considered in the research was stored in the laboratory of the authors’ Department (DAFNAE) and the authors have provided an ID number for each voucher specimen. More

1.Parmesan, C. & Yohe, G. A globally coherent fingerprint of climate change impacts across natural systems. Nature 421, 37–42 (2003).CAS 
Article 

Google Scholar 
2.Root, T. L. et al. Fingerprints of global warming on wild animals and plants. Nature 421, 57–60 (2003).CAS 

Google Scholar 
3.Scheffers, B. R. et al. The broad footprint of climate change from genes to biomes to people. Science 354, aaf7671 (2016).
Google Scholar 
4.Cohen, J. M., Lajeunesse, M. J. & Rohr, J. R. A global synthesis of animal phenological responses to climate change. Nat. Clim. Change 8, 224–228 (2018).
Google Scholar 
5.Piao, S. et al. Plant phenology and global climate change: current progresses and challenges. Glob. Change Biol. 25, 1922–1940 (2019).
Google Scholar 
6.Merilä, J. & Hendry, A. P. Climate change, adaptation, and phenotypic plasticity: the problem and the evidence. Evol. Appl. 7, 1–14 (2014).
Google Scholar 
7.Chevin, L. M. & Hoffmann, A. A. Evolution of phenotypic plasticity in extreme environments. Philos. Trans. R. Soc. Lond. B Biol. Sci. 372, 20160138 (2017).
Google Scholar 
8.Fox, R. J., Donelson, J. M., Schunter, C., Ravasi, T. & Gaitán-Espitia, J. D. Beyond buying time: the role of plasticity in phenotypic adaptation to rapid environmental change. Philos. Trans. R. Soc. Lond. B Biol. Sci. 374, 20180174 (2019).
Google Scholar 
9.Thackeray, S. J. et al. Trophic level asynchrony in rates of phenological change for marine, freshwater and terrestrial environments. Glob. Change Biol. 16, 3304–3313 (2010).
Google Scholar 
10.Kharouba, H. M. et al. Global shifts in the phenological synchrony of species interactions over recent decades. Proc. Natl Acad. Sci. USA 115, 5211–5216 (2018).CAS 

Google Scholar 
11.Radchuk, V. et al. Adaptive responses of animals to climate change are most likely insufficient. Nat. Commun. 10, 3109 (2019).
Google Scholar 
12.Visser, M. E. & Gienapp, P. Evolutionary and demographic consequences of phenological mismatches. Nat. Ecol. Evol. 3, 879–885 (2019).
Google Scholar 
13.Kharouba, H. M. & Wolkovich, E. M. Disconnects between ecological theory and data in phenological mismatch research. Nat. Clim. Change 10, 406–415 (2020).
Google Scholar 
14.Samplonius, J. M. et al. Strengthening the evidence base for temperature-mediated phenological asynchrony and its impacts. Nat. Ecol. Evol. 5, 155–164 (2021).
Google Scholar 
15.Charmantier, A. et al. Adaptive phenotypic plasticity in response to climate change in a wild bird population. Science 320, 800–803 (2008).CAS 

Google Scholar 
16.Tomotani, B. M. et al. Climate change leads to differential shifts in the timing of annual cycle stages in a migratory bird. Glob. Change Biol. 24, 823–835 (2018).
Google Scholar 
17.Moyes, K. et al. Advancing breeding phenology in response to environmental change in a wild red deer population. Glob. Change Biol. 17, 2455–2469 (2011).
Google Scholar 
18.Lane, J. E., Kruuk, L. E. B., Charmantier, A., Murie, J. O. & Dobson, F. S. Delayed phenology and reduced fitness associated with climate change in a wild hibernator. Nature 489, 554–557 (2012).CAS 

Google Scholar 
19.Todd, B. D., Scott, D. E., Pechmann, J. H. K. & Whitfield Gibbons, J. Climate change correlates with rapid delays and advancements in reproductive timing in an amphibian community. Proc. R. Soc. Lond. B Biol. Sci. 278, 2191–2197 (2011).
Google Scholar 
20.Taylor, S. G. Climate warming causes phenological shift in pink salmon, Oncorhynchus gorbuscha, behavior at Auke Creek, Alaska. Glob. Change Biol. 14, 229–235 (2008).
Google Scholar 
21.Mills, L. S. et al. Camouflage mismatch in seasonal coat color due to decreased snow duration. Proc. Natl Acad. Sci. USA 110, 7360–7365 (2013).CAS 

Google Scholar 
22.Lameris, T. K. et al. Arctic geese tune migration to a warming climate but still suffer from a phenological mismatch. Curr. Biol. 28, 2467–2473.e4 (2018).CAS 

Google Scholar 
23.Singer, M. C. & Parmesan, C. Phenological asynchrony between herbivorous insects and their hosts: signal of climate change or pre-existing adaptive strategy? Philos. Trans. R. Soc. Lond. B Biol. Sci. 365, 3161–3176 (2010).
Google Scholar 
24.Charmantier, A. & Gienapp, P. Climate change and timing of avian breeding and migration: evolutionary versus plastic changes. Evol. Appl. 7, 15–28 (2014).
Google Scholar 
25.Keogan, K. et al. Global phenological insensitivity to shifting ocean temperatures among seabirds. Nat. Clim. Change 8, 313–317 (2018).
Google Scholar 
26.Both, C. & Visser, M. E. Adjustment to climate change is constrained by arrival date in a long-distance migrant bird. Nature 411, 296–298 (2001).CAS 

Google Scholar 
27.Both, C., van Asch, M., Bijlsma, R. G., van den Burg, A. B. & Visser, M. E. Climate change and unequal phenological changes across four trophic levels: constraints or adaptations? J. Anim. Ecol. 78, 73–83 (2009).
Google Scholar 
28.Cresswell, W. & McCleery, R. How great tits maintain synchronization of their hatch date with food supply in response to long-term variability in temperature. J. Anim. Ecol. 72, 356–366 (2003).
Google Scholar 
29.Visser, M. E., Van Noordwijk, A. J., Tinbergen, J. M. & Lessells, C. M. Warmer springs lead to mistimed reproduction in great tits (Parus major). Proc. R. Soc. Lond. B Biol. Sci. 265, 1867–1870 (1998).
Google Scholar 
30.Sanz, J. J., Potti, J., Moreno, J., Merino, S. & Frías, O. Climate change and fitness components of a migratory bird breeding in the Mediterranean region. Glob. Change Biol. 9, 461–472 (2003).
Google Scholar 
31.Marrot, P., Charmantier, A., Blondel, J. & Garant, D. Current spring warming as a driver of selection on reproductive timing in a wild passerine. J. Anim. Ecol. 87, 754–764 (2018).
Google Scholar 
32.Burgess, M. D. et al. Tritrophic phenological match–mismatch in space and time. Nat. Ecol. Evol. 2, 970–975 (2018).
Google Scholar 
33.Visser, M. E., Holleman, L. J. M. & Gienapp, P. Shifts in caterpillar biomass phenology due to climate change and its impact on the breeding biology of an insectivorous bird. Oecologia 147, 164–172 (2006).
Google Scholar 
34.Pearce-Higgins, J. W., Yalden, D. W. & Whittingham, M. J. Warmer springs advance the breeding phenology of golden plovers Pluvialis apricaria and their prey (Tipulidae). Oecologia 143, 470–476 (2005).CAS 

Google Scholar 
35.Nussey, D. H., Clutton-Brock, T. H., Elston, D. A., Albon, S. D. & Kruuk, L. E. B. Phenotypic plasticity in a maternal trait in red deer. J. Anim. Ecol. 74, 387–396 (2005).
Google Scholar 
36.Husby, A. et al. Contrasting patterns of phenotypic plasticity in reproductive traits in two great tit (Parus major) populations. Evolution 64, 2221–2237 (2010).
Google Scholar 
37.Matthysen, E., Adriaensen, F. & Dhondt, A. A. Multiple responses to increasing spring temperatures in the breeding cycle of blue and great tits (Cyanistes caeruleus, Parus major). Glob. Change Biol. 17, 1–16 (2011).
Google Scholar 
38.Fisher, J. I., Mustard, J. F. & Vadeboncoeur, M. A. Green leaf phenology at Landsat resolution: scaling from the field to the satellite. Remote Sens. Environ. 100, 265–279 (2006).
Google Scholar 
39.Duparc, A. et al. Co-variation between plant above-ground biomass and phenology in sub-alpine grasslands. Appl. Veg. Sci. 16, 305–316 (2013).
Google Scholar 
40.Hinks, A. E. et al. Scale-dependent phenological synchrony between songbirds and their caterpillar food source. Am. Nat. 186, 84–97 (2015).
Google Scholar 
41.Lambrechts, M. M., Blondel, J., Maistre, M. & Perret, P. A single response mechanism is responsible for evolutionary adaptive variation in a bird’s laying date. Proc. Natl Acad. Sci. USA 94, 5153–5155 (1997).CAS 

Google Scholar 
42.Dawson, A. Control of the annual cycle in birds: endocrine constraints and plasticity in response to ecological variability. Philos. Trans. R. Soc. Lond. B Biol. Sci. 363, 1621–1633 (2008).
Google Scholar 
43.Visser, M. E. et al. Phenology, seasonal timing and circannual rhythms: towards a unified framework. Philos. Trans. R. Soc. Lond. B Biol. Sci. 365, 3113–3127 (2010).CAS 

Google Scholar 
44.Caro, S. P., Schaper, S. V., Hut, R. A., Ball, G. F. & Visser, M. E. The case of the missing mechanism: how does temperature influence seasonal timing in endotherms? PLoS Biol. 11, e1001517 (2013).CAS 

Google Scholar 
45.Bourgault, P., Thomas, D., Perret, P. & Blondel, J. Spring vegetation phenology is a robust predictor of breeding date across broad landscapes: a multi-site approach using the Corsican blue tit (Cyanistes caeruleus). Oecologia 162, 885–892 (2010).
Google Scholar 
46.Bison, M. et al. Best environmental predictors of breeding phenology differ with elevation in a common woodland bird species. Ecol. Evol. 10, 10219–10229 (2020).
Google Scholar 
47.Bernhardt, J. R., O’Connor, M. I., Sunday, J. M. & Gonzalez, A. Life in fluctuating environments. Philos. Trans. R. Soc. Lond. B Biol. Sci. 375, 20190454 (2020).
Google Scholar 
48.Gienapp, P., Reed, T. E. & Visser, M. E. Why climate change will invariably alter selection pressures on phenology. Proc. R. Soc. Lond. B Biol. Sci. 281, 20141611 (2014).
Google Scholar 
49.Lönnstedt, O. M., McCormick, M. I., Chivers, D. P. & Ferrari, M. C. O. Habitat degradation is threatening reef replenishment by making fish fearless. J. Anim. Ecol. 83, 1178–1185 (2014).
Google Scholar 
50.Pellerin, F., Cote, J., Bestion, E. & Aguilée, R. Matching habitat choice promotes species persistence under climate change. Oikos 128, 221–234 (2019).
Google Scholar 
51.Firth, J. A., Verhelst, B. L., Crates, R. A., Garroway, C. J. & Sheldon, B. C. Spatial, temporal and individual-based differences in nest-site visits and subsequent reproductive success in wild great tits. J. Avian Biol. 49, e01740 (2018).
Google Scholar 
52.Naef-Daenzer, B. & Keller, L. F. The foraging performance of great and blue tits (Parus major and P. caeruleus) in relation to caterpillar development, and its consequences for nestling growth and fledging weight. J. Anim. Ecol. 68, 708–718 (1999).
Google Scholar 
53.Naef-Daenzer, B. Patch time allocation and patch sampling by foraging great and blue tits. Anim. Behav. 59, 989–999 (2000).CAS 

Google Scholar 
54.Bouwhuis, S., Sheldon, B. C., Verhulst, S. & Charmantier, A. Great tits growing old: selective disappearance and the partitioning of senescence to stages within the breeding cycle. Proc. R. Soc. Lond. B Biol. Sci. 276, 2769–2777 (2009).CAS 

Google Scholar 
55.Cole, E. F. & Sheldon, B. C. The shifting phenological landscape: within- and between-species variation in leaf emergence in a mixed-deciduous woodland. Ecol. Evol. 7, 1135–1147 (2017).
Google Scholar 
56.Wint, W. The role of alternative host-plant species in the life of a polyphagous moth, Operophtera brumata (Lepidoptera: Geometridae). J. Anim. Ecol. 52, 439–450 (1983).
Google Scholar 
57.Keller, L. F. & van Noordwijk, A. J. Effects of local environmental conditions on nestling growth in the great tit Parus major L. Ardea 82, 349–362 (1994).
Google Scholar 
58.Wilkin, T. A., Garant, D., Gosler, A. G. & Sheldon, B. C. Density effects on life-history traits in a wild population of the great tit Parus major: analyses of long-term data with GIS techniques. J. Anim. Ecol. 75, 604–615 (2006).
Google Scholar 
59.Wilkin, T. A. & Sheldon, B. C. Sex differences in the persistence of natal environmental effects on life histories. Curr. Biol. 19, 1998–2002 (2009).CAS 

Google Scholar 
60.Gagen, M. et al. The tree ring growth histories of UK native oaks as a tool for investigating chronic oak decline: an example from the Forest of Dean. Dendrochronologia 55, 50–59 (2019).
Google Scholar 
61.Sturrock, R. N. et al. Climate change and forest diseases. Plant Pathol. 60, 133–149 (2011).
Google Scholar 
62.MacColl, A. D. C. The ecological causes of evolution. Trends Ecol. Evol. 26, 514–522 (2011).
Google Scholar 
63.Grant, P. R. & Price, T. D. Population variation in continuously varying traits as an ecological genetics problem. Integr. Comp. Biol. 21, 795–811 (1981).
Google Scholar 
64.Hereford, J. A quantitative survey of local adaptation and fitness trade-offs. Am. Nat. 173, 579–588 (2009).
Google Scholar 
65.Hadfield, J. D. The spatial scale of local adaptation in a stochastic environment. Ecol. Lett. 19, 780–788 (2016).
Google Scholar 
66.Porlier, M. et al. Variation in phenotypic plasticity and selection patterns in blue tit breeding time: between- and within-population comparisons. J. Anim. Ecol. 81, 1041–1051 (2012).
Google Scholar 
67.Hidalgo Aranzamendi, N., Hall, M. L., Kingma, S. A., van de Pol, M. & Peters, A. Rapid plastic breeding response to rain matches peak prey abundance in a tropical savanna bird. J. Anim. Ecol. 88, 1799–1811 (2019).
Google Scholar 
68.Caro, S. P., Lambrechts, M. M., Balthazart, J. & Perret, P. Non-photoperiodic factors and timing of breeding in blue tits: impact of environmental and social influences in semi-natural conditions. Behav. Process. 75, 1–7 (2007).CAS 

Google Scholar 
69.Bourret, A., Bélisle, M., Pelletier, F. & Garant, D. Multidimensional environmental influences on timing of breeding in a tree swallow population facing climate change. Evol. Appl. 8, 933–944 (2015).
Google Scholar 
70.Nussey, D. H., Wilson, A. J. & Brommer, J. E. The evolutionary ecology of individual phenotypic plasticity in wild populations. J. Evol. Biol. 20, 831–844 (2007).CAS 

Google Scholar 
71.Morris, D. W. Toward an ecological synthesis: a case for habitat selection. Oecologia 136, 1–13 (2003).
Google Scholar 
72.Long, R. A. et al. Linking habitat selection to fitness-related traits in herbivores: the role of the energy landscape. Oecologia 181, 709–720 (2016).
Google Scholar 
73.Morris, D. W. Spatial scale and the cost of density-dependent habitat selection. Evol. Ecol. 1, 379–388 (1987).
Google Scholar 
74.Patten, M. A. & Kelly, J. F. Habitat selection and the perceptual trap. Ecol. Appl. 20, 2148–2156 (2010).
Google Scholar 
75.Ponchon, A., Garnier, R., Grémillet, D. & Boulinier, T. Predicting population responses to environmental change: the importance of considering informed dispersal strategies in spatially structured population models. Divers. Distrib. 21, 88–100 (2015).
Google Scholar 
76.Nilsson, A. L. K. et al. Hydrology influences breeding time in the white-throated dipper. BMC Ecol. 20, 70 (2020).
Google Scholar 
77.Nilsson, A. L. K. et al. Location is everything, but climate gets a share: analyzing small-scale environmental influences on breeding success in the white-throated dipper. Front. Ecol. Evol. 8, 542846 (2020).
Google Scholar 
78.Martin, R. O., Cunningham, S. J. & Hockey, P. A. R. Elevated temperatures drive fine-scale patterns of habitat use in a savanna bird community. Ostrich 86, 127–135 (2015).
Google Scholar 
79.Bailey, L. D. et al. Habitat selection can reduce effects of extreme climatic events in a long-lived shorebird. J. Anim. Ecol. 88, 1474–1485 (2019).
Google Scholar 
80.Kirby, K. J. et al. Changes in the tree and shrub layer of Wytham Woods (southern England) 1974–2012: local and national trends compared. Forestry 87, 663–673 (2014).
Google Scholar 
81.Perrins, C. & McCleery, R. Laying dates and clutch size in the great tit. Wilson Bull. 101, 236–253 (1989).
Google Scholar 
82.Wilkin, T. A., Perrins, C. M. & Sheldon, B. C. The use of GIS in estimating spatial variation in habitat quality: a case study of lay-date in the great tit Parus major. Ibis 149, 110–118 (2007).
Google Scholar 
83.Perrins, C. M. Population fluctuations and clutch-size in the great tit, Parus major L. J. Anim. Ecol. 34, 601–647 (1965).
Google Scholar 
84.Wesołowski, T. & Rowiński, P. Timing of bud burst and tree-leaf development in a multispecies temperate forest. For. Ecol. Manage. 237, 387–393 (2006).
Google Scholar 
85.Gibson, C. W. D. in Woodland Conservation and Research in the Clay Vale of Oxfordshire and Buckinghamshire (eds Kirby, K. J. & Write, F. J.) 32–40 (JNCC, 1988).86.Dawkin, H. C. & Field, D. R. B. A Long-Term Surveillance System for British Woodland Vegetation. Commonwealth Forestry Institute, Oxford, Occasional Paper No. 1. (1978).87.Horsfall, A. S. & Kirby, K. J. The Use of Permanent Quadrats to Record Changes in the Structure and Composition of Wytham Woods, Oxfordshire Research and Survey in Nature Conservation No. 1 (JNCC, 1992).88.Wilkin, T. A., King, L. E. & Sheldon, B. C. Habitat quality, nestling diet, and provisioning behaviour in great tits Parus major. J. Avian Biol. 40, 135–145 (2009).
Google Scholar 
89.Van Noordwijk, M. & Purnomosidhi, P. Root architecture in relation to tree–soil–crop interactions and shoot pruning in agroforestry. Agrofor. Syst. 30, 161–173 (1995).
Google Scholar 
90.Bailey, L. D. & van de Pol, M. climwin: an R toolbox for climate window analysis. PLoS ONE 11, e0167980 (2016).
Google Scholar 
91.van de Pol, M. et al. Identifying the best climatic predictors in ecology and evolution. Methods Ecol. Evol. 7, 1246–1257 (2016).
Google Scholar 
92.Simmonds, E. G., Cole, E. F. & Sheldon, B. C. Cue identification in phenology: a case study of the predictive performance of current statistical tools. J. Anim. Ecol. 88, 1428–1440 (2019).
Google Scholar 
93.Oksanen, J. et al. vegan: Community Ecology Package: R Package v.2.5-6 (2019); https://CRAN.R-project.org/package=vegan94.Sturges, H. A. The choice of a class interval. J. Am. Stat. Assoc. 21, 65–66 (1926).
Google Scholar 
95.Hadfield, J. D. MCMC methods for multi-response generalized linear mixed models: the MCMCglmm R package. J. Stat. Softw. https://doi.org/10.18637/jss.v033.i02 (2010).96.R Core Team R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2020); http://www.R-project.org/ More

Summary of sequencing dataWe obtained 168 million and 180 million 250 bp reads from Asian Badger and Northern Hog Badger, respectively. After removing transcripts and unigenes below 200 bp, we obtained 335,772 transcripts and 285,159 unigenes belonging to Asian Badger and 413,917 transcripts and 362,075 unigenes belonging to Northern Hog Badger (Table 1). Next, we analysed the length distribution of the unigenes and transcripts in these two species (Fig. 1). Their N50 of transcript length is longer than 1000 bp, and their N50 of unigene length is longer than 600 bp. The average GC content of the transcriptome data of Asian Badge was 52.71%, a value slightly higher than that of the Northern Hog Badger, which was 52.12% (Table 1).Table 1 Summary of the transcriptome of Asian Badgers and Northern Hog Badgers.Full size tableFigure 1Length and quantity distribution of transcripts and unigenes.Full size imageFunctional annotation and classification of the assembled unigenesThe success rate of annotation of these research data in the seven databases is shown in Table 2. In total, 34,150 (ZH) and 31,632 (GH) unigenes had GO terms (Table 2). Among them, there were three GO items related to digestion: positive regulation of the digestive system process (GH and ZH both have one gene), digestive tract development (GH and ZH both have four genes), and digestion (GH has five genes, ZH has three). Next, we compared the GO terms of Asian Badger and Northern Hog Badger transcriptomes and found that the distributions pattern of gene functions from these two species were particularly similar (Fig. 2). This predictable result indicates that there is no bias in the construction of the libraries from the Asian Badger and Northern Hog Badger. For both species, in the three main partitions (cellular component, molecular function, and biological process) of the GO classification, ‘Cellular process’, ‘Binding’ and ‘Metabolic process’, terms were principal individually (Fig. 2). In total, 8915 (ZH) and 10,203 (GH) unigenes had KOG terms (Table 2). In addition, 15,667 (ZH) and 17,823 (GH) were mapped to the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways (Table 2) and grouped into 32 subclasses. Interestingly, the digestive system subcategory contains 695 and 611 unigenes in Asian Badger and Northern Hog Badger, respectively, involving 9 pathways, namely, bile, gastric acid, pancreatic, salivary secretion, carbohydrate, protein, vitamin, fat digestion and absorption, and mineral absorption.Table 2 Gene annotation success rate statistics.Full size tableFigure 2GO term Top20 for GH and ZH.Full size imageAnalysis of orthologous genesThe transcriptome evolution of different species can be understood by comparing transcriptome data. We analysed the possible orthologous genes between the transcriptome of Asian Badger and Northern Hog Badger obtained in this study. We selected a total of 5227 homologous gene pairs from these four species. After 5227 pairs of homologous genes were optimized and screened, 943 orthologous gene pairs were obtained (Supplementary Table S1).To explore whether the genes related to small intestinal digestion in Asian Badger and Northern Hog Badger have undergone adaptive evolution. We can predict the genes that affect the evolution of the two species through selection pressure on orthologous genes12. We selected 473 orthologous gene pairs with Ka/Ks  > 1 called divergent orthologous genes from the Ka/ks analysis results. We obtained 1263 orthologous gene pairs with Ka/Ks  More

1.Eitzel, M. V. et al. Citizen science terminology matters: Exploring key terms. Citiz. Sci. 2(1), 1 (2017).Article 

Google Scholar 
2.European Citizen Science Association (ECSA). ECSA’s Characteristics of Citizen Science. Apr 2020 [cited 3 July 2020]. In ECSA Our Documents [Internet]. https://ecsa.citizen-science.net/wp-content/uploads/2020/05/ecsa_characteristics_of_citizen_science_-_v1_final.pdf.3.Shirk, J. L. et al. Public participation in scientific research: A framework for deliberate design. Ecol. Soc. 17(2), 29 (2012).Article 

Google Scholar 
4.Follett, R. & Strezov, V. An analysis of citizen science based research: Usage and publication patterns. PLoS One 10(11), e0143687 (2015).Article 

Google Scholar 
5.Tauginienė, L. et al. Citizen science in the social sciences and humanities: The power of interdisciplinarity. Palgrave Commun. 6, 89 (2020).Article 

Google Scholar 
6.European Citizen Science Association (ECSA). Ten Principles of Citizen Science. Sept 2015 [cited 1 Dec 2020]. In ECSA Our Documents [Internet]. https://ecsa.citizen-science.net/wp-content/uploads/2020/02/ecsa_ten_principles_of_citizen_science.pdf.7.Vayena, E. & Tasioulas, J. Adapting standards: Ethical oversight of participant-led health research. PLoS Med. 10(3), e1001402 (2013).Article 

Google Scholar 
8.Resnik, D. B., Elliott, K. C. & Miller, A. K. A framework for addressing ethical issues in citizen science. Environ. Sci. Policy 54, 475–481 (2015).Article 

Google Scholar 
9.Riesch, H. & Potter, C. Citizen science as seen by scientists: Methodological, epistemological and ethical dimensions. Public Underst. Sci. 23(1), 107–120 (2014).Article 

Google Scholar 
10.Rothstein, M. A., Wilbanks, J. T. & Brothers, K. B. Citizen science on your smartphone: An ELSI research agenda. J. Law Med. Ethics 43(4), 897–903 (2015).Article 

Google Scholar 
11.Guerrini, C. J., Majumder, M. A., Lewellyn, M. J. & McGuire, A. L. Citizen science, public policy. Science 361(6398), 134–136 (2018).ADS 
CAS 
Article 

Google Scholar 
12.Resnik, D. B. Citizen scientists as human subjects: Ethical issues. Citiz. Sci. 4(1), 11 (2019).
Google Scholar 
13.Woolley, J. P. et al. Citizen science or scientific citizenship? Disentangling the uses of public engagement rhetoric in national research initiatives. BMC Med. Ethics 17, 33 (2016).Article 

Google Scholar 
14.Rasmussen, L. M. Research ethics in citizen science. In The Oxford Handbook of Research Ethics (eds Iltis, A. S. & MacKay, D.) (Oxford University Press, 2021).
Google Scholar 
15.Chesser, S., Poster, M. M. & Tuckett, A. G. Cultivating citizen science for all: Ethical considerations for research projects involving diverse and marginalized populations. Int. J. Soc. Res. Methodol. 23(5), 497–508 (2020).Article 

Google Scholar 
16.Vayena, E. & Tasioulas, J. “We the scientists”: A human right to citizen science. Philos. Technol. 28, 479–485 (2015).Article 

Google Scholar 
17.Rasmussen, L. M. “Filling the ‘Ethics Gap’ in Citizen Science Research”: A Workshop Report. 2017 [cited 15 July 2020]. In NIEHS Partnerships for Environmental Public Health [Internet]. https://www.niehs.nih.gov/research/supported/translational/peph/webinars/ethics/rasmussen_508.pdf.18.Louviere, J. J., Flynn, T. N. & Marley, A. A. J. Best-Worst Scaling: Theory, Methods and Applications (Cambridge University Press, 2015).Book 

Google Scholar 
19.Mühlbacher, A. C., Kaczynski, A., Zweifel, P. & Johnson, F. R. Experimental measurement of preferences in health and healthcare using best-worst scaling: An overview. Health Econ. Rev. 6, 2 (2016).Article 

Google Scholar 
20.Marti, J. A best-worst scaling survey of adolescents’ level of concern for health and non-health consequences of smoking. Soc. Sci. Med. 75, 87–97 (2012).Article 

Google Scholar 
21.Erdem, S. & Rigby, D. Investigating heterogeneity in the characterization of risks using best worst scaling. Risk Anal. 33(9), 1728–1748 (2013).Article 

Google Scholar 
22.Peay, H. L., Hollin, I. L. & Bridges, J. F. P. Prioritizing parental worry associated with Duchenne muscular dystrophy using best-worst scaling. J. Genet. Couns. 25, 305–313 (2016).Article 

Google Scholar 
23.Flynn, T. N., Louviere, J. J., Peters, T. J. & Coast, J. Best-worst scaling: What it can do for health care research and how to do it. J. Health Econ. 26, 171–189 (2007).Article 

Google Scholar 
24.Finn, A. & Louviere, J. J. Determining the appropriate response to evidence of public concern: The case of food safety. J. Public Policy Mark. 11(2), 12–25 (1992).Article 

Google Scholar 
25.Louviere, J. J. & Flynn, T. N. Using best-worst scaling choice experiments to measure public perceptions and preferences for healthcare reform in Australia. Patient 3(4), 275–283 (2010).Article 

Google Scholar 
26.Janssen, E. M., Benz, H. L., Tsai, J.-H. & Bridges, J. F. P. Identifying and prioritizing concerns associated with prosthetic devices for use in a benefit-risk assessment: A mixed methods approach. Expert Rev. Med. Devices 15(5), 385–398 (2018).CAS 
Article 

Google Scholar 
27.Auger, P., Devinney, T. M. & Louvier, J. J. Using best-worst scaling methodology to investigate consumer ethical beliefs across countries. J. Bus. Ethics 70, 299–326 (2007).Article 

Google Scholar 
28.Bridges, J. F. P. et al. Conjoint analysis applications in health—A checklist: A report of the ISPOR Good Research Practices for Conjoint Analysis Task Force. Value Health 14, 403–413 (2011).Article 

Google Scholar 
29.Coast, J. et al. Using qualitative methods for attribute development for discrete choice experiments: Issues and recommendations. Health Econ. 21, 730–741 (2012).Article 

Google Scholar 
30.Johnson, F. R. et al. Constructing experimental designs for discrete-choice experiments: Report of the ISPOR Conjoint Analysis Experimental Design Good Research Practices Task Force. Value Health 16, 3–13 (2013).Article 

Google Scholar 
31.Hauber, A. B. et al. Statistical methods for the analysis of discrete choice experiments: A report of the ISPOR Conjoint Analysis Good Research Practices Task Force. Value Health 19, 300–315 (2016).Article 

Google Scholar 
32.Flynn, T. N. Valuing citizen and patient preferences in health: Recent developments in three types of best-worst scaling. Expert Rev. Pharmacoecon. Outcomes Res. 10(3), 259–267 (2010).Article 

Google Scholar 
33.Louviere, J., Lings, I., Islam, T., Gudergan, S. & Flynn, T. An introduction to the application of (case 1) best-worst scaling in marketing research. Int. J. Res. Mark. 30, 292–303 (2013).Article 

Google Scholar 
34.Jannsen, E. M., Marshall, D. A., Hauber, A. B. & Bridges, J. F. P. Improving the quality of discrete-choice experiments in health: How can we assess validity and reliability?. Expert Rev. Pharmacoecon. Outcomes Res. 17(6), 531–542 (2017).Article 

Google Scholar 
35.Gallego, G., Bridges, J. F. P. & Flynn, T. Using best-worst scaling in horizon scanning for hepatocellular carcinoma technologies. Int. J. Technol. Assess. Health Care 28(3), 339–346 (2012).Article 

Google Scholar 
36.Bowser, A., Shilton, K., Preece, J. & Warrick, E. Accounting for privacy in citizen science: Ethical research in a context of openness. In CSCW ’17: Proceedings of the 2017 ACM Conference on Computer Supported Cooperative Work and Social Computing 2124–2136 (Association for Computing Machinery, 2017).37.Pandya, R. E. A framework for engaging diverse communities in citizen science in the US. Front Ecol. Environ. 10(6), 314–317 (2012).Article 

Google Scholar 
38.Citizen Science Association (CSA). Who We Are; Who We Serve. n.d. [cited 1 Sept 2021]. In CSA About [Internet]. https://www.citizenscience.org/about/.39.Sawtooth Software. The MaxDiff System Technical Paper v. 9. Oct. 2020 [cited 25 Jan 2021] In Technical Papers [Internet]. https://sawtoothsoftware.com/resources/technical-papers/maxdiff-technical-paper.40.English, P. B., Richardson, M. J. & Garzon-Galvis, C. From crowdsourcing to extreme citizen science: Participatory research for environmental health. Annu. Rev. Public Health 39, 335–350 (2018).CAS 
Article 

Google Scholar 
41.Conrad, C. C. & Hilchey, K. G. A review of citizen science and community-based environmental monitoring: Issues and opportunities. Environ. Monit. Assess. 176, 273–291 (2011).Article 

Google Scholar 
42.Cashman, S. B. et al. The power and the promise: Working with communities to analyze data, interpret findings, and get to outcomes. Am. J. Public Health 98(8), 1407–1417 (2008).Article 

Google Scholar 
43.Citizen Science Association (CSA). Trustworthy Data Practices. n.d. [cited 5 Sept 2020]. In CSA Events [Internet]. https://www.citizenscience.org/data-ethics-study/.44.Cheung, K. L. et al. Using best-worst scaling to investigate preferences in health care. Pharmacoecon. 34, 1195–1209 (2016).Article 

Google Scholar 
45.Hastings, J. J. A. When citizens do science. Narrat. Inq. Bioeth. 9(1), 33–34 (2019).Article 

Google Scholar 
46.Guerrini, C. J., Trejo, M., Canfield, I. & McGuire, A. L. Core values of genomic citizen science: Results from a qualitative interview study. BioSocieties https://doi.org/10.1057/s41292-020-00208-2 (2020).Article 

Google Scholar 
47.Sharon, T. Self-tracking for health and the Quantified Self: Re-articulating autonomy, solidarity, and authenticity in the age of personalized medicine. Philos. Technol. 30, 93–121 (2017).Article 

Google Scholar  More

Wet and dry weight of aerial partsDry weight of aerial parts was significantly affected by drought stress and genotype treatments and their interactions (Table 2). With increasing drought stress the amount of dry weight of aerial parts in all genotypes was decreased. Dry functions in I2, I3, and I4 levels in H genotype (Lavandula gngustifolia cv. Hidecot) were 15.68%, 40.35% and 48.15%, respectively. In S genotype (Lavandula stricta) these amounts were 0.78%, 48.58% and 51.72%, respectively; and in M genotype (Lavandula angustifolia cv. Muneasted) they were 22.29%, 49.38% and 52.63%, respectively. Compared to the control group, the most reduction in dry weight of aerial parts was in M genotype. The highest amount of dry weight (11.40 g in plant) was observed in H genotype in drought stress of 90–100% of field capacity. The lowest amount of dry weight of aerial parts (3.07 g) was seen in S genotype in drought stress of 30–40% field capacity (Fig. 2).Table 2 Variance analysis of the effect of drought stress on enzymatic activity of antioxidant enzymes, and quantity of essential oil from different lavender genotypes.Full size tableFigure 2The effect of drought stress on dry weight of aerial parts in different lavender genotypes.Full size imageIn this study drought stress had a negative effect on biomass of lavender plants. This effect can be due to water shortage. Because drought stress cause reduction in swelling, total water potential in cell and withering, it also results in closing stomata, reduction in cell division, and cell enlargement47,48. Reduction in cell division and cell enlargement as a result of drought, reduce the leaf surface, photosynthesis and growth function of the plant. In other words, reduction in photosynthesis products, cause reduction in leaf’s surface; and reduction in transfer of assimilated materials to aerial part, as a result of drought, cause decrease in aerial yield of the plant49. In this regard, Abbaszadeh et al. (2020) reported that due to drought stress of 30% and 60% of field capacity, dry weight of aerial parts in Rosmarinus officinalis L. has decreased. While contrary to our results Rhizopoulou and Diamantoglou (1991) observed that dry weight of leaves from Marjoram plant (Origanum majorana) was increased with increased soil moisture deficiency; which can be due to differences in plant species and ecological conditions50,51.Proline content of leavesThe results of variance analysis showed that drought stress, genotype and their interactions have significantly affected proline content of leaves (Table 2). With increasing drought stress the proline content was increased. The highest amount of proline content (4.96 mg per g) was observed in H genotype in I4 drought level (30–40% of field capacity). While the lowest amount (1.08 mg per g) was observed in S genotype in irrigation of 90–100% of field capacity (Fig. 3). In each genotype separately, in I2 to I3 drought levels the amount of proline was equal, but in H and M genotypes with increasing drought stress, the amount of proline was increased, While in S genotype with increasing water deficit proline did not show a significant increase. This indicates that two genotypes (H and M) have a similar function for using these types of osmolyte to deal with this level of drought. Which this result may be exist another osmolite production as a resistance mechanism in S genotype52.Figure 3The effect of drought stress on proline content in different genotypes of lavender.Full size imageOne change that happens in biological and non-biological stresses is increasing the amount of osmolytes in plant. To prevent negative effects of drought stress, the plant increases the amount of its osmolytes including proline53. Proline is an amino acid which in addition to act as an osmolyte, plays an important role in maintaining and stabilizing membranes by adding membrane phospholipids and changing the hydrated layer around macromolecules. Proline is also recognized as a stabilizer for cellular homeostasis under stressful conditions. This is due to high ability of proline to stabilize sub-cellular structures such as proteins and cell membranes and its ability to eliminate free radicals54. In present study, increasing proline content in different genotypes of lavender as a result of drought, can be for the same reason. It is proved that in some plants, changes in amount of proline is related to their ability to tolerate and adapt with drought stress; so, the proline content can be used as an indicator to select drought-resistant plants. Hosseinpour et al. (2020), reported that in response to drought stress, accumulation of compatible metabolites such as proline can participate in water absorption. In accordance with our results, an increase in proline content in different genotypes of Calendula officinalis plant due to drought stress has been reported as well55,56. However, in some plant species, other osmolites are produced under biological stress, the most important of them is glycine betaine. So that it is probable that the relationship between glycine betaine accumulation and stress tolerance, such as drought stress, is species- or even genotype specific57. As a results, the S genotype likely produced glycine betaine under drought stress, obviously, completed studies are needed to confirm this hypothesis.Relative water content of leavesThe relative water content (RWC) of leaves was significantly affected by drought stress, genotype and their interaction (Table 2). The highest amount of RWC (87.43%) was observed in H genotype in no drought stress condition. The lowest RWC (19.60%) was observed in S genotype in 30–40% of field capacity (Fig. 4). The results of comparing average data showed that in highest level of drought stress RWC in H, S and M genotypes is 57.25%, 65.19% and 58.88%, respectively; which compared to the control group, it is decreased in all genotypes. This suggests higher resistance of H genotype to maintain RWC of leaves (Fig. 4). In all evaluated genotypes, with increasing drought, RWC was decreased.Figure 4The effect of drought stress on RWC of leaves in different lavender genotypes.Full size imageRWC is a suitable indicator for water stress in plants. Drought stress by reducing RWC and total water potential of cell, result in reduction in growth of plants. The osmoregulation mechanisms in drought-resistant plants, maintains high RWC in them. Reduction in RWC of leaves as a result of water deficiency stress, is due to reduction in amount of water in tissue, reduction in amount of water in soil, and the negative soil water potential58. Alinejad et al. (2020), reported that RWC of leaves in Datura stramonium L. plant was decreased due to drought, in a way that the highest amount of RWC (80.22%) was seen in 55% of field capacity, compared to 35% and 15% of field capacity59. Also Mohammadi et al. (2018) suggested that RWC of leaves in Thymu vulgari L. was decreased to 18.41%, after being exposed to drought60.Total phenolic and flavonoids contents in leavesDrought, genotype and their interaction had a significant effect on total phenolic content of leaves (Table 2). The results suggest that in different levels of drought, total phenolic content was different in lavender genotypes. In the highest level of drought, total phenolic content in H, S, and M genotypes was respectively increased 18.64%, 28.57% and 98.07% in comparison with the control group. The highest difference in total phenolic content compared to control group was observed in M genotype (Fig. 5).Figure 5The effect of total phenolic content of leaves in different genotypes of lavender.Full size imageTotal flavonoids content of leaves was significantly (p ≤ 0.01) affected by drought and genotype (Table 2). The results of comparing averages showed that the highest amount of total flavonoids (1.12 mg quercetin per g of fresh weight) was in H genotype, and the lowest amount (0.95 mg quercetin per g of fresh weight) was in M genotype (Table 3). Moreover, our results showed that drought level from I2 to I4 caused an increase of 12.74%, 14.61% and 15.38% in total flavonoid content of leaves, respectively. Which indicates an increase in flavonoid amount with increasing drought level (Table 3). Table 3 Comparing simple effects of genotype and drought stress on traits of lavender plant.Full size tableTotal phenolic content is related to stress-resistance, indirectly by helping cell protection, and directly as an antioxidant61. Phenolic compounds due to their reductive properties, act as a free radical remover62. Our findings are similar to those of a study on growth of Mentha piperita in drought stress54.Total antioxidant activityTotal antioxidant activity was significantly affected by drought stress and genotype (Table 2). With increasing drought, antioxidant activity in H and S genotypes was increased. The results of comparing average data showed that compared to the control group, in drought levels of I2, I3 and I4, antioxidant activity in H genotype was increased by 98.43%, 98.36% and 118.78%, respectively; and in S genotype this amounts were increased by 89.85%, 111.78%, and 131.90% respectively (Table 5). In M genotype the antioxidant activity has reached its highest amount (49.38 mg/g) in I3 level of drought, and then with increasing drought stress the antioxidant activity was decreased, in a way that in highest drought level it had the lowest antioxidant activity (23.18 mg/g). M genotype was used as control (Fig. 6). Our results indicate that in highest drought level, antioxidant activity of S genotype was more than others. Figure 6The effect of drought stress on antioxidant activity in different lavender genotypes.Full size imageAntioxidant enzymesEnzymatic activity of antioxidant enzymes in lavender leaves was significantly affected by genotype and drought stress (Table 2). Our results showed that the highest activity of SOD (304.75 μmol min−1 mg−1 protein) was observed in interaction of H genotype and I4 drought level, and the lowest activity of SOD (144.52 μmol min−1 mg−1 protein) was observed in S genotype with no drought (Fig. 7). Moreover, our observations showed that in I2 and I3 drought levels, the highest amount of SOD enzymatic activity was related to M genotype (Fig. 7). In the highest drought level, enzymatic activity of SOD was increased in H and S genotypes, and it decreased in M genotype.Figure 7The effect of drought stress on enzymatic activity of SOD, POX and CAT in different lavender genotypes.Full size imageEnzymatic activity of peroxidase (POX) enzyme was increased in all three genotypes, with increasing drought. In all drought levels, H genotype had the highest amount of POX activity, compared to other genotypes. There was no significant difference in POX activity in S and H genotypes. The results showed that the highest amount of POX activity (274.48 μmol min−1 mg−1 protein) was observed in interaction of H genotype and 30–40% field capacity, and the lowest amount (117.66 μmol min−1 mg−1 protein) was observed in interaction of S genotype and no drought condition (control) (Fig. 7).Catalase (CAT) enzyme was affected by drought, genotype and their interaction (Table 2). The results of catalase enzyme activity assessment showed that with increasing drought, catalase activity is different in H, M and S genotypes. The most different reaction in production of CAT was related to H genotype, which with increasing drought stress up to I3 level, the enzyme activity was increased. But regarding M and S genotypes, with increasing drought level, CAT activity was increased in both genotypes. In this study the highest amount of CAT (460.51 μmol min−1 mg−1 protein) was observed in interaction of S genotype with 30–40% of field capacity; and the lowest amount (157.06 μmol min−1 mg−1 protein) was observed in interaction of H genotype with 90–100% of field capacity (Fig. 7).No significant effect was observed for APX enzyme in interaction of genotype and drought (Table 2). The results of comparing average data, suggest that the highest amount of APX activity (284.96 μmol min−1 mg−1 protein) was observed in H genotype (Table 3). Also the results showed that I2, I3 and I4 drought level resulted in an increase in APX enzyme activity by 32.38%, 49.16%, and 65.53% respectively. This indicates that APX enzymatic activity increases with increasing drought level (Table 3).Using physiological and biochemical mechanisms to reduce effects of stress shows that to overcome drought, oxidative stress and to eliminate ROS, plants will increase the amount of antioxidant content55. One of major mechanisms to cope with oxidative stress in plants, is activation of antioxidant enzymes61. Findings of the present study indicates that different lavender genotypes showed partial resistance against drought. In this research, increased activity of antioxidant enzymes in lavender genotypes under drought condition, was considered as an important drought-resistance factor. Among all antioxidant enzymes, SOD can have a good response against drought stress. In a way that H, S, and M genotypes of lavender in the highest level of drought stress (I4), showed an increased amount of SOD, by 57.42%, 35.85% and 60.69% compared to normal conditions (Fig. 7).In this study, the minimum enzymatic changes were related to the POX enzyme and the highest enzymatic changes were related to the CAT enzyme. Moreover, it was observed that the highest amount of catalase enzymatic activity was in H genotype. In a way that in plants under drought stress CAT activity was increased up to I3 drought level; but, after this level with increasing drought (I4 drought level) CAT enzymatic activity was decreased. CAT and POX are among important plants enzymes which can protect plant cells against free radicals63. In this study, in drought period, enzymatic activity of CAT and POX was increased, this means that lavender genotypes, in the face of stress produce antioxidant enzymes to protect themselves. While in H genotype compared to other genotypes, in high drought stress, CAT activity was decreased which this response indicates the different function of this genotype in dealing with ROS. Enzymatic response to drought condition was different in various lavender genotypes. Generally, the negative effect of drought is shown by production of reactive oxygen species (ROS). Increased enzymatic activity of antioxidant enzymes, particularly CAT and POX can reduce the negative effects of drought64, 65. In this regard, increased activity of antioxidant enzymes in different genotypes of Calendula officinalis plant was reported to56.Malondialdehyde (MDA) contentReaction of different lavender genotypes under drought stress was different in terms of malondialdehyde (MDA) production and accumulation (Table 2). With increasing drought, MDA content was significantly increased in M and H genotypes. The highest amount of MDA in these genotypes was 14.34 and 9.50 nmolg − 1 FM respectively, which was observed in drought level of 30–40% of field capacity. This indicates a significant increase in MDA content with increasing drought (Fig. 8). While the process of production and accumulation of MDA in S genotype was different at various drought levels. For S genotype, in first level of drought (I2), MDA content was increased which showed the vulnerability of the cell membrane at this drought level. But with increasing drought, gradually, the S genotype plants adapted to the dry environment, which in this level cell membrane damage was not obvious. Then, increasing in drought stress resulted in increased MDA content. Generally, in I2 and I3 drought levels, lavender genotypes underwent varying degrees of damage, which in M and H genotypes followed by increasing enzymatic activity, and in S genotype it resulted in decreased enzymatic activity. But in the highest level of drought (I4), the cell membrane was seriously damaged and in all three genotypes and MDA content was significantly increased (Fig. 8).Figure 8The effect of drought stress on MDA content in different lavender genotypes.Full size imageMembrane lipid peroxidation due to the accumulation of active oxygen species leads to cell damage and death. In plants this lipid peroxidation happens under drought stress66. MDA is the final product of membrane peroxidation and membrane processes. Simultaneously with peroxidation, the MDA content increases significantly67. So the MDA content can be considered as an indicator of drought-resistance in plants. Among lavender genotypes, in the highest level of drought, MDA content in M genotype was significantly increased compared to others genotypes; whish suggests that M genotype is more vulnerable in comparison with the two other genotypes. An increase in MDA content under drought stress, was reported in Thymus species as well66.Quantity and quality of essential oilMutual interaction between drought stress and genotype had a significant effect on percentage and yield of essential oil in lavender plants (Table 2). Our findings suggested a different essential oil percentage for each genotype in various levels of drought stress. With increasing drought to I3 level, the essential oil percentage was increased in M and H genotypes, but after that with increasing drought to a higher level (I4), essential oil percentage in these genotypes was decreased. While in S genotype, increasing essential oil percentage totally had an upward trend (Fig. 9).Figure 9The effect of drought stress on essential oil percent in different lavender genotypes.Full size imageEvaluation of essential oil percentage in different levels of drought, showed that in I2 drought level, the highest amount of essential oil (0.81%) was observed in H genotype; and in I3 and I4 drought levels, the highest amounts of essential oil were 1.29% and 1.68% respectively, which were observed in S genotype. Moreover, our results suggest that the highest difference in essential oil percentage in the studied genotypes compared to the control, was related to S genotype (Fig. 9). Totally, the highest percentage of essential oil was observed in S genotype in I4 drought level. This shows the high capacity of this genotype to produce essential oil under drought stress.Essential oil yield was significantly affected by genotype and drought. The results showed that the essential oil yield in S genotype was different from the others. So that the highest yield of essential oil (0.055 g per plant) was observed in this genotype in I3 drought level. While in H and M genotypes the highest amounts were 0.068 g and 0.065 g respectively, which were gained in I2 drought level (Fig. 10). Results of comparing average data showed that the highest yield of essential oil at I2 and I3 levels was obtained with 151/85% and 122.22% difference compared to the control, respectively, and they gained from H genotype. This indicates the high potential of H genotype to maintain biomass and produce essential oil in drought stress. Also our results suggest that in the highest drought level (I4), the highest essential oil yield (0.046 g per plant) was observed in M genotype (Fig. 10).Figure 10The effect of drought stress on essential oil yield in different lavender genotypes.Full size imagePrincipal component analysis (PCA)PCA analysis was performed to identify susceptibility of genotypes in irrigation regimes. According to physiological traits in the PCA analysis (Fig. 11a, b), the first factor (PC1) explains about 90% of the total variance of variables, and the second factor (PC2) about 8%.Figure 11Principal component analysis (PCA) for genotypes (a) and physiological traits (b) based on water status calculated for physiological traits. (R 4.0.4 packages, https://rstudio.com/products/rstudio/).Full size imageThe results of PCA analysis of different irrigation regimes showed that in the first component, which shows 89.91% of changes, the best traits are antioxidant enzymes CAT, SOD, APX, while in the second component, with 8.10% changes, only the trait Catalase is the best trait. Also, in total, the first and second components, which show 98.01% of the changes, show CAT as the most effective trait (Fig. 11a).The results of PCA analysis in lavender genotypes showed that the first and second main components could explain 98.91% of the existing changes. So that the first main component with 91.13% and the second component with 7.78% had a share in the total variation. Therefore, using these two components and ignoring other components will only cause the loss of a small part of about 1.09% of the data changes (Fig. 11b). These two principal components include peroxidase, ascorbate peroxidase, and superoxide. Physiological responses of Lavandula genotypes (L. angustifolia cv. Hidcote, L. angustifolia cv. Munstead, and L. stricta) submitted to drought stress were evaluated through principal component analysis (PCA), and the results are illustrated in Fig. 11a. Lavandula stricta presents higher levels of CAT activity than L. angustifolia cv. Hidcote and L. angustifolia cv. Munstead. In addition, APX and CAT increase in stress-treated in 30–40% FC. This result shows that L. stricta exhibits the most affected physiological changes while trying to adjust to changes in the water status of the environment, under the imposed conditions and shows the highest resistance.The results of analysis of essentials oils from H, S and M genotypes is shown is Tables 4, 5 and 6. The trend of changes in essential oils composition is described in all three genotypes. By studying the mass spectra and the Kovats retention index, 23 compounds were identified in the H genotype’s essential oil (Table 4). The yield of H genotype essential oil from I1 to I4 drought levels was 99.89%, 82.78%, 81.09% and 82.85%, respectively. The main components of H genotype essential oil in I1 to I4 drought levels, include 1.8-Cineol compounds (5.94%, 7.73%, 4.24% and 3.50%), Linalool (23.20%, 16.30%, 11.90% and 10.57%), Camphor (3.41%, 4.65%, 2.32% and 2.87%), Borneol (4.89%, 3.34%, 3.65% and 3.01%), Bornyl formate (27.32%, 16.04%, 19.45% and 20.03%), Lavandulyl acetate (1.40%, 4.21%, 6 and 8.35%), Caryophyllene oxide (10.92%, 11.77%, 12.16% and 19.91%), α-Muurolene (4.38%, 3.20%, 1.20% and 0%) (Table 4). The results of grouping the essential oil compounds showed that the amount of hydrocarbon monoterpenes from I1 to I4 drought level were 12.88%, 8.86%, 8.53% and 6.06%, respectively. The amount of oxygen monoterpenes was 64.76%, 50.70%, 43.32% and 42.45%; and hydrocarbon sesquiterpene compounds were 13.12%, 11.45%, 13.03% and 13.96%. The amount of oxygen sesquiterpene compounds were 10.92%, 11.77%, 16.21%, and 19.91%; which shows that increasing drought level, result in decreasing monoterpene compounds, and increasing sesquiterpene compounds.Table 4 Chemical composition of essential oils extracted from Lavandula angustifolia cv. Hidcote plants under different irrigation regime.Full size tableTable 5 Chemical composition of essential oils extracted from Lavandula stricta plants under different irrigation regime.Full size tableTable 6 Chemical composition of essential oils extracted from Lavandula angustifolia cv. Munstead plants different irrigation regime.Full size tableHeat map for the essential oil profile in Lavandula angustifolia cv. Hidcote corresponding to the different irrigation regime The similar discrimination was also supported by the heatmap constructed for essential compounds. Accordingly, 22 rows and 4 columns were achieved. α- pinene, β-Pinene, δ-3-Carene, type of Cymene, 1,8-Cineol, Camphor and Linalool from the main compounds, peaked at control. Moreover, lavandulyl acetate, Myrtenyl acetate, caryophyllene oxide, camphene and γ-Cadinene revealed highest percentage at 30–40% FC, Some compounds, such as Camphor and Linalyl acetate, are at the levels of the intermediate irrigation regime (Fig. 12). It is remarkable that as the water limit increases, the amount of monoterpene compounds decreases and the amount of sesquiterpene compounds increases.Figure 12Heatmap for the essential oil profile in aerial parts of Lavandula angustifolia cv. Hidcote corresponding to irrigation regimes (CIMminer, https://discover.nci.nih.gov/cimminer/oneMatrix.do).Full size imageWith evaluation of the essential oil from S genotype, 18 compounds were identified (Table 5). The amount of essential oil in I1 to I4 drought levels was 99.41%, 98.48%, 99.53% and 99.93% respectively (Table 5). Among identified compounds in S genotype the followings were accounted for the highest amount of components in the essential oil in I1 to I4 levels respectively; Linalool (32.60%, 28.45%, 20.12% and 19.12%), decanal (10.26%, 15.21%, 18.56% and 19.27%), 1-Decanol (8.01%, 10.31%, 17.88% and 21.34%), Kessane (2.44%, 4.43%, 9.99% and 11.50%), Hexadecane (1.26%, 5.77%, 6.10% and 11.9%), 2-methyl-1-hexadecanol (11.1%, 9.32%, 8.15% and 2.37%) and Hexahydrofarnesyl acetone (6.8%, 6.34%, 3.78% and 1.26%) (Table 5). The most obvious point was the high percentage of Linalool, decanal and 1-Decanol in the S genotype. With increasing drought, Linalool compounds were decreased and decanal and 1-Decanol compounds were increased. The grouping of essential oil components also showed that among the 18 compounds identified, the following were the highest in I1 to I4 drought levels, respectively; 3 hydrocarbon monoterpenes with total of (5.34%, 5.44%, 4.57% and 4.34%), 6 oxygen monoterpenes with total of (60.49%, 61.03%, 59.57% and 60.45%), 3 hydrogen sesquiterpenes with total of (5.69%, 10.27%, 11.85% and 15.24%) and 6 oxygen sesquiterpenes with total of (27.89%, 28.09%, 29.44% and 18.32%). With increasing drought, the amounts of hydrocarbon monoterpenes and oxygen sesquiterpenes were decreased; while the amount of hydrocarbon sesquiterpenes was increased. Also the highest amount of oxygen monoterpenes, by 61.03%, was seen in I2 drought level.Heat map for the essential oil profile in Lavandula stricta corresponding to the different irrigation regime The parallel discrimination was also supported by the heatmap constructed for essential compounds. Accordingly, 18 rows and 4 columns were achieved. α- pinene, Amyl isovalerate, Citronellol, β-Ionone and Linalool from the main compounds, peaked at control. Moreover, α-Thujene, decanal, 1-Decanol, Sesquiphellandrene, Kessane and Hexadecane revealed highest percentage at 30–40% FC (Fig. 13). These results confirm the results obtained from the Lavandula angustifolia cv. Hidcote so that as the water limit increases, the amount of monoterpene compounds decreases and the amount of Sesquiterpene compounds increases.Figure 13Heatmap for the essential oil profile in aerial parts of Lavandula stricta corresponding to irrigation regimes (CIMminer, https://discover.nci.nih.gov/cimminer/oneMatrix.do).Full size imageEssential oil yield in M genotype from I1 to I4 drought levels was obtained 99.90%, 98.38%, 93.08% and 87.04% (Table 6). As it is shown in Table 6, analysis of the essential oil from M genotype included 27 compounds which its major part was consisted of Camphor (16.82%, 16.32%, 17.11% and 18.30%), Borneol (44.96%, 42.80%, 37.54% and 30.99%) and Caryophyllene oxide (14.68%, 15.21%, 15.90% and 17.21%) from I1 to I4 drought levels, respectively. comparison of essential oil components (Table 6) showed that from 27 identified compounds in M genotype, the followings were the most prevalent from I1 to I4 levels respectively, including hydrocarbon monoterpene with total of (17.82%, 17.45%, 13.91% and 9.96%), 12 total oxygen monoterpene compounds with total of (65.95%, 62.05%, 56.96% and 50.42%), 4 hydrocarbon sesquiterpenes with total of (1.58%, 23.23%, 5.42% and 8.09%) and 2 oxygen sesquiterpenes with total of (14.91%, 15.65%, 16.79% and 18.37%). The highest drought level resulted in 31.76% and 17.23% increase in Camphor and Caryophyllene oxide. It also caused 31.07% decrease in Borneol compared to the control (Table 6). Totally, with increasing drought level, monoterpene compounds were decreased and sesquiterpene compounds were increased in lavender genotypes.The major components of essential oil were different in various lavender genotypes in the highest level of drought (I4). In this study in H genotype, the compounds Linalool, Bornyl formate and Caryophyllene oxide; in S genotype the compounds Linalool, decanal, 1-Decanol, Kessane and Hexadecane; and in M genotype the compounds Camphor, Borneol and Caryophyllene oxide, were the most prevalent components of essential oil. In this study, Borneol compound was not observed in S genotype. regarding the fact that essential oil extraction was performed on flowering branches in all three genotypes, and they were studied under similar drought conditions; and also comparing the results of this study with finding of other studies shows that the difference in types and percentage of essential oil’s components can be due to the effect of genetic differences; and to some extent, environmental factors on essential oil in different genotypes.A total comparison of essential oil analysis results for different lavender genotypes under drought stress showed that oxygen monoterpenes are the most prevalent components of the essential oil, which will decrease with increasing drought level. Sarker et al. (2012) reported that the essential oil of lavender (Lavandula angustifolia) contains high amounts of linalool and linalool acetate, along with scares amount of other monoterpenes68. A study by Hassan et al. (2014) showed that the compounds carvacrol, phenol-2-amino-4, 6-bis, trans-2-caren-4-ol, and n-hexadecanoic acid are the main constituents of Lavandula stricta plants which were collected from the Shaza Mountains in southern Saudi Arabia69. Total results from essential oil analysis in this study showed that Linalool was the main ingredient of essential oils in H and S genotypes. This compound is an oxygen monoterpene with a density of 0.85 and a pleasant smell, and is the main component of the essential oil from lavender plant. While in M genotype, Borneol was the main component of the essential oil, which is a circular monoterpene compound with density of Mohammadnejad ganji et al. (2017) suggested that the difference in natural quality of the essential oil from lavender plants is related to intrinsic factors (genetic or heredity capabilities and maturity), and external factors including sunlight, water, heat, pressure, latitude, and soil which affect plant growth and essential oil production70.Heat map for the essential oil profile in Lavandula angustifolia cv. Munstead corresponding to the different irrigation regime The parallel discrimination was also supported by the heatmap constructed for essential compounds. Accordingly, 18 rows and 4 columns were achieved. α- pinene, Tricycle, Camphene, Thuja-2,4(10)-diene, δ-3-Carene, ρ-Cymene, Borneol and limonene from the main compounds, peaked at control. Moreover, Camphor, α-Santalene, γ-Cadinene, δ-Cadinene, Caryophyllene oxide, α-Muurolene and Ledene oxide-(II) revealed highest percentage at 30–40% FC (Fig. 14). The results showed that the composition of the compounds was similar to the previous two genotypes and the water limit increases, the amount of monoterpene compounds decreases and the amount of Sesquiterpene compounds increases.Figure 14Heatmap for the essential oil profile in aerial parts of Lavandula angustifolia cv. Munstead corresponding to irrigation regimes (CIMminer, https://discover.nci.nih.gov/cimminer/oneMatrix.do).Full size imageEssential oils are generally in the group of terpenoids and The structure of terpenoids consists of two main precursors, isopentenyl pyrophosphate (IPP) and its isomer, dimethylallyl pyrophosphate (DMAPP). These compounds are synthesized via the cytosolic pathway of mevalonic acid (MVA) or plasticity of methylerythritol phosphate (MEP)71. The MVA pathway is primarily responsible for the synthesis of Sesquiterpenoids and triterpenoids, while the MEP pathway is used for the biosynthesis of monoterpenoids, diterpenoids and tetraterpenoids72. Monoterpenes and Sesquiterpenes are the main constituents of essential oils that play a role in aroma, flavor, photosynthetic pigments and antioxidant activities73.In drought conditions, the amount of these isoprenes does not decrease in relation to the mediators of the MEP pathway and in contrast sometimes increases. Therefore, sesquiterpene compounds increase in drought conditions because most of these compounds are synthesized through the MVA pathway74. Another reason for the decrease in MEP path flux is the location of this path, which has a significant impact in drought conditions. In this case, plastids are not able to provide the required IPP of this path, so most monoterpene compounds are reduced75.Also, since the quality of the essential oil is due to the presence of linalool and linalyl acetate76. According to the results obtained from heatmaps related to essential oils, three genotypes are identified, the highest amount of linalool amount in S genotype was remained under mind- (I2) till severe-drought (I4) condition. This indicates more compatibility with maintaining the desired quality of drought conditions in this plant than the other two commercial genotypes. And then the H genotype is in the second stage due to the presence of important compounds.Comparing the grouping created in the heat maps related to the essential oil of 3 genotypes, it is clear that the two genotypes S and H were divided into two groups I1, I2 and I3, I4 in the genotype. But in the genotype M, the results were divided into I4 and I3 groups I2 were divided into genotypes. This can be due to differences in the resistance mechanism of plants in different genotypes, so in genotypes S and H of the plant through increasing sesquiterpene compounds showed resistance to drought stress, while in genotype M increased resistance to drought levels through higher monoterpene compounds. Another conclusion that can be drawn from these heat maps is that in genotypes S and H, the rate of drought resistance in the first and second levels of drought with the third and fourth levels has shown more changes in the type of essential oil compounds, while in the third genotype (M) these changes in the last level drought has been most evident.At a glance, it seems Genotype S has a different mechanism in reducing the negative effects of drought compared to genotypes M and H, So that, among the enzymatic and non-enzymatic mechanisms, it tends to use the enzymatic pathway more. In association with the production of “proline”, drought stress index osmolyte, genotype S has a different trend from genotypes H and M and this osmolyte in this plant has a lower production flux compared to other genotypes. Also, due to the fact that the production of soluble sugars in this plant has been moderate compared to other genotypes, it is expected this genotype replace proline with another osmolyte or uses an enzymatic mechanism to deal with drought, as the results of antioxidant enzyme “catalase” related to genotype S had the highest value with a significant difference under drought stress, while, in the H and M genotypes, the SOD enzyme was responsive to drought.On the other hand, the high resistance of genotype S can be attributed to the greater activation of the pathway of essential oil compounds. Because by examining the constituents of the essential oil (monoterpene and sesquiterpene), it can be concluded that genotype H and then M at high drought levels still retain the ability to produce monoterpene compounds, while in genotype S with increasing drought, the amount of semi-heavy compounds (sesquiterpene) has increased significantly (Fig. 15), this can confirm the existence of a different resistance mechanism in the S genotype. Because some structural compounds of the membrane, such as sterols, are made from the mevalonic acid (MVA) pathway of acetyl coenzyme A origin. For this reason it seems that S genotype by setting up terpenoid pathways involved in the production of steroids another solution to drought is by preserving its plasma membrane. Steroids are derivatives of triterpenes that, along with phospholipids, are major components of plasma membranes70. Also, the study of MDA content as the final product of membrane lipid peroxidation in genotypes at the fourth level of drought (the most severe drought) showed the M genotype is most sensitive to drought. In this way, the two genotypes S and H have almost equal MDA content, so that it can be said that with a small difference from genotype S, genotype H has less composition.Figure 15The amount of monoterpenes and sesquiterpene compounds in different genotypes under irrigation regimes.Full size imageContinuous production of isoprene under drought conditions shows that despite the reduction in the synthesis of osmolyte and relative increasing of MDA (with very little difference from genotype H) that occurs under these conditions, the function of this pathway is essential for the S genotype. Isoprene has long been used to protect plants from drought, high temperatures and oxidative stress are recommended77. Of course, it was showed which is possible with increasing drought, sufficient isoprene is not produced to counteract and launch defense pathways and instead used as a general signal to increase drought tolerance78,79.Reasons such as further activation of terpenoid skeletal pathways towards the production of semi-heavy (sesquiterpene) compounds, production of steroids via the MVA pathway could be a reason for lower susceptibility of S genotype and high resistance of this genotype through these mechanisms compared to other genotypes. In contrast, on the one hand, H genotype using proline production, soluble sugar levels and decreased MDA in response to stress caused by drought and on the other hand, the ability to produce substances important monoterpenes, such as Linalool and Linalyl acetate, with the aim of using medicine and aromatherapy76, It (H genotype) can be considered as a cultivar with high commercial value and significant resistance to M genotype. More

1.Gupta A, Sharma VK. Using the taxon-specific genes for the taxonomic classification of bacterial genomes. BMC Genom. 2015;16:1–15.Article 
CAS 

Google Scholar 
2.Gil R, Silva FJ, Pereto J, Moya A. Determination of the Core of a Minimal Bacterial Gene Set. Microbiol Mol Biol Rev. 2004;68:518–37.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
3.Mira A, Martín-Cuadrado AB, D’Auria G, Rodríguez-Valera F. The bacterial pan-genome: a new paradigm in microbiology. Int Microbiol. 2010;13:45–57.CAS 
PubMed 

Google Scholar 
4.Escalas A, Troussellier M, Yuan T, Bouvier T, Bouvier C, Mouchet MA, et al. Functional diversity and redundancy across fish gut, sediment and water bacterial communities. Environ Microbiol. 2017;19:3268–82.PubMed 
Article 

Google Scholar 
5.Jurburg SD, Salles JF. Functional Redundancy and Ecosystem Function — The Soil Microbiota as a Case Study. In: Lo Y-H, Blanco JA, Shovonlal R, editors. Biodiversity in Ecosystems—Linking Structure and Function. BoD–Books on Demand; 2015. p. 29–49.6.Louca S, Polz MF, Mazel F, Albright MBN, Huber JA, O’Connor MI, et al. Function and functional redundancy in microbial systems. Nat Ecol Evol. 2018;2:936–43.PubMed 
Article 

Google Scholar 
7.Polz MF, Hunt DE, Preheim SP, Weinreich DM. Patterns and mechanisms of genetic and phenotypic differentiation in marine microbes. Philos Trans R Soc Lond B Biol Sci. 2006;361:2009–21.PubMed 
PubMed Central 
Article 

Google Scholar 
8.Young JPW. Bacteria Are Smartphones and Mobile Genes Are Apps. Trends Microbiol. 2016;24:931–2.CAS 
PubMed 
Article 

Google Scholar 
9.Boon E, Meehan CJ, Whidden C, Wong DHJ, Langille MGI, Beiko RG. Interactions in the microbiome: communities of organisms and communities of genes. FEMS Microbiol Rev. 2014;38:90–118.CAS 
PubMed 
Article 

Google Scholar 
10.Escalas A, Hale L, Voordeckers JW, Yang Y, Firestone MK, Alvarez-Cohen L, et al. Microbial Functional Diversity: from Concepts to Applications. Ecol Evol. 2019;5:12000–16.Article 

Google Scholar 
11.Barberán A, Casamayor EO, Fierer N. The microbial contribution to macroecology. Front Microbiol. 2014;5:1–8.Article 

Google Scholar 
12.Shade A, Dunn RR, Blowes SA, Keil P, Bohannan BJM, Herrmann M, et al. Macroecology to Unite All Life, Large and Small. Trends Ecol Evol. 2018;33:731–44.PubMed 
Article 

Google Scholar 
13.Chase AB, Martiny JB. The importance of resolving biogeographic patterns of microbial microdiversity. Microbiol Aust. 2018;1:5–8.Article 

Google Scholar 
14.Shoemaker WR, Locey KJ, Lennon JT. A macroecological theory of microbial biodiversity. Nat Ecol Evol. 2017;1:e1450v4.Article 

Google Scholar 
15.Bachy C, Worden AZ. Microbial ecology: finding structure in the rare biosphere. Curr Biol. 2014;24:R315–R317.16.Lynch MDJ, Neufeld JD. Ecology and exploration of the rare biosphere. Nat Rev Microbiol. 2015;13:217–29.CAS 
PubMed 
Article 

Google Scholar 
17.Pedrós-Alió C. The Rare Bacterial Biosphere. Ann Rev Mar Sci. 2012;4:449–66.PubMed 
Article 

Google Scholar 
18.Rabinowitz D. Seven forms of rarity and their frequency in the flora of the British Isles. In: Soulé ME, editors. Conservation biology: the science of scarcity and diversity. Sinauer Associates; Massachusetts; 1986.19.McGeoch MA, Gaston KJ. Occupancy frequency distributions: patterns, artefacts and mechanisms. Biol Rev Camb Philos Soc. 2002;77:311–31.PubMed 
Article 

Google Scholar 
20.Blackburn TM, Cassey P, Gaston KJ. Variations on a theme: Sources of heterogeneity in the form of the interspecific relationship between abundance and distribution. J Anim Ecol. 2006;75:1426–39.PubMed 
Article 

Google Scholar 
21.Buckley HL, Freckleton RP. Understanding the role of species dynamics in abundance-occupancy relationships. J Ecol. 2010;98:645–58.Article 

Google Scholar 
22.Gaston KJ, Blackburn TM, Greenwood JJD, Gregory RD, Quinn RM, Lawton JH. Abundance-occupancy relationships. J Appl Ecol. 2000;37:39–59.Article 

Google Scholar 
23.Miranda LE, Killgore KJ. Abundance–occupancy patterns in a riverine fish assemblage. Freshw Biol. 2019;64:2221–33.Article 

Google Scholar 
24.Suhonen J, Jokimäki J. Temporally stable species occupancy frequency distribution and abundance-occupancy relationship patterns in urban wintering bird assemblages. Front Ecol Evol. 2019;7:129.Article 

Google Scholar 
25.Webb TJ, Barry JP, McClain CR. Abundance–occupancy relationships in deep sea wood fall communities. Ecography. 2017;40:1339–47.Article 

Google Scholar 
26.Amend AS, Oliver TA, Amaral-Zettler LA, Boetius A, Fuhrman JA, Horner-Devine MC, et al. Macroecological patterns of marine bacteria on a global scale. J Biogeogr. 2013;40:800–11.Article 

Google Scholar 
27.Barberán A, Bates ST, Casamayor EO, Fierer N. Using network analysis to explore co-occurrence patterns in soil microbial communities. ISME J. 2012;6:343–51.PubMed 
Article 
CAS 

Google Scholar 
28.Barnes CJ, Burns CA, van der Gast CJ, McNamara NP, Bending GD. Spatio-temporal variation of core and satellite arbuscular mycorrhizal fungus communities in Miscanthus giganteus. Front Microbiol. 2016;7:1–12.
Google Scholar 
29.Fillol M, Auguet JC, Casamayor EO, Borrego CM. Insights in the ecology and evolutionary history of the Miscellaneous Crenarchaeotic Group lineage. ISME J. 2016;10:665–77.PubMed 
Article 

Google Scholar 
30.Jeanbille M, Gury J, Duran R, Tronczynski J, Agogué H, Saïd OBen, et al. Response of core microbial consortia to chronic hydrocarbon contaminations in coastal sediment habitats. Front Microbiol. 2016;7:1–13.
Google Scholar 
31.Lindh MV, Sjöstedt J, Ekstam B, Casini M, Lundin D, Hugerth LW, et al. Metapopulation theory identifies biogeographical patterns among core and satellite marine bacteria scaling from tens to thousands of kilometers. Environ Microbiol. 2017;19:1222–36.CAS 
PubMed 
Article 

Google Scholar 
32.Logares R, Audic SS, Bass D, Bittner L, Boutte C, Christen R, et al. Patterns of Rare and Abundant Marine Microbial Eukaryotes. Curr Biol. 2014;24:813–21.CAS 
PubMed 
Article 

Google Scholar 
33.Michelland R, Thioulouse J, Kyselková M, Grundmann GL. Bacterial Community Structure at the Microscale in Two Different Soils. Micro Ecol. 2016;72:717–24.CAS 
Article 

Google Scholar 
34.Unterseher M, Jumpponen A, Öpik M, Tedersoo L, Moora M, Dormann CF, et al. Species abundance distributions and richness estimations in fungal metagenomics – Lessons learned from community ecology. Mol Ecol. 2011;20:275–85.PubMed 
Article 

Google Scholar 
35.Grime JP. Benefits of plant diversity to ecosystems: Immediate, filter and founder effects. J Ecol. 1998;86:902–10.Article 

Google Scholar 
36.Grime JP. Dominant and subordinate components of plant communities: implications for succession, sta- bility and diversity. In: Gray AJ, Crawley MJ, editors. Colonization, Succession and Stability. Oxford:Blackwell Scientific Publications; 1984. p. 413–28.37.Hanski I. Dynamics of Regional Distribution: the Core and Satellite Species Hypothesis. Oikos. 1982;38:210.Article 

Google Scholar 
38.Magurran AE, Henderson PA. Explaining the excess of rare species in natural species abundance distributions. Nature. 2003;422:714–6.CAS 
PubMed 
Article 

Google Scholar 
39.Newton R, Shade A. Lifestyles of rarity: understanding heterotrophic strategies to inform the ecology of the microbial rare biosphere. Aquat Micro Ecol. 2016;78:51–63.Article 

Google Scholar 
40.Shade A, Jones SE, Caporaso JG, Handelsman J, Knight R, Fierer N, et al. Conditionally rare taxa disproportionately contribute to temporal changes in microbial diversity. MBio. 2014;5:e01371–14.PubMed 
PubMed Central 
Article 
CAS 

Google Scholar 
41.Shade A, Gilbert JA. Temporal patterns of rarity provide a more complete view of microbial diversity. Trends Microbiol. 2015;23:335–40.CAS 
PubMed 
Article 

Google Scholar 
42.Koch AL. Oligotrophs versus copiotrophs. BioEssays. 2001;23:657–61.CAS 
PubMed 
Article 

Google Scholar 
43.Cobo-Simón M, Tamames J. Relating genomic characteristics to environmental preferences and ubiquity in different microbial taxa. BMC Genom. 2017;18:1–11.Article 
CAS 

Google Scholar 
44.Tu Q, Yu H, He Z, Deng Y, Wu L, Van Nostrand JD, et al. GeoChip 4: a functional gene-array-based high-throughput environmental technology for microbial community analysis. Mol Ecol Resour. 2014;14:914–28.CAS 
PubMed 

Google Scholar 
45.Xu X, Wang N, Lipson D, Sinsabaugh R, Schimel J, He L, et al. Microbial macroecology: in search of mechanisms governing microbial biogeographic patterns. Glob Ecol Biogeogr. 2020;29:1870–86.Article 

Google Scholar 
46.Reich PB, Knops J, Tilman D, Craine J, Ellsworth D, Tjoelker M, et al. Plant diversity enhances ecosystem responses to elevated CO2 and nitrogen deposition. Nature. 2001;410:809–12.CAS 
PubMed 
Article 

Google Scholar 
47.Field CB, Chapin FS, Chiariello NK, Holland EA, Mooney HA. The Jasper Ridge CO2 Experiment: Design and Motivation. In: Mooney HA, Koch GW, (Editors). Carbon Dioxide and Terrestrial Ecosystems. San Diego, California: Academic Press; 1996. p. 121–45.Chapter 

Google Scholar 
48.Luo C, Rodriguez-R LM, Johnston ER, Wu L, Cheng L, Xue K, et al. Soil microbial community responses to a decade of warming as revealed by comparative metagenomics. Appl Environ Microbiol. 2014;80:1777–86.PubMed 
PubMed Central 
Article 
CAS 

Google Scholar 
49.Mauritz M, Bracho R, Celis G, Hutchings J, Natali SM, Pegoraro E, et al. Nonlinear CO2 flux response to 7 years of experimentally induced permafrost thaw. Glob Chang. Biol. 2017;23:3646–66.
Google Scholar 
50.Natali SM, Schuur EAG, Mauritz M, Schade JD, Celis G, Crummer KG, et al. Permafrost thaw and soil moisture driving CO2 and CH4 release from upland tundra. J Geophys Res Biogeosci. 2015;120:525–37.CAS 
Article 

Google Scholar 
51.Yang Y, Gao Y, Wang S, Xu D, Yu H, Wu L, et al. The microbial gene diversity along an elevation gradient of the Tibetan grassland. ISME J. 2014;8:430–40.CAS 
PubMed 
Article 

Google Scholar 
52.Yang Y, Wu L, Lin Q, Yuan M, Xu D, Yu H, et al. Responses of the functional structure of soil microbial community to livestock grazing in the Tibetan alpine grassland. Glob Chang Biol. 2013;19:637–48.PubMed 
Article 

Google Scholar 
53.Zhang Y, Cong J, Lu H, Li G, Xue Y, Deng Y, et al. Soil bacterial diversity patterns and drivers along an elevational gradient on Shennongjia Mountain, China. Micro Biotechnol. 2015;8:739–46.Article 

Google Scholar 
54.Zhang Y, Cong J, Lu H, Deng Y, Liu X, Zhou J, et al. Soil bacterial endemism and potential functional redundancy in natural broadleaf forest along a latitudinal gradient. Sci Rep. 2016;6:1–8.Article 
CAS 

Google Scholar 
55.Paula FS, Rodrigues JLM, Zhou J, Wu L, Mueller RC, Mirza BS, et al. Land use change alters functional gene diversity, composition and abundance in Amazon forest soil microbial communities. Mol Ecol. 2014;23:2988–99.PubMed 
Article 

Google Scholar 
56.Rodrigues JLM, Pellizari VH, Mueller R, Baek K, Jesus EDC, Paula FS, et al. Conversion of the Amazon rainforest to agriculture results in biotic homogenization of soil bacterial communities. Proc Natl Acad Sci USA. 2013;110:988–93.CAS 
PubMed 
Article 

Google Scholar 
57.He Z, Deng Y, Van Nostrand JD, Tu QC, Xu MY, Hemme CL, et al. GeoChip 3.0 as a high-throughput tool for analyzing microbial community composition, structure and functional activity. Isme J. 2010;4:1167–79.CAS 
PubMed 
Article 

Google Scholar 
58.He Z, Gentry TJ, Schadt CW, Wu L, Liebich J, Chong SC, et al. GeoChip: a comprehensive microarray for investigating biogeochemical, ecological and environmental processes. ISME J. 2007;1:67–77.CAS 
PubMed 
Article 

Google Scholar 
59.Li X, He Z, Zhou J. Selection of optimal oligonucleotide probes for microarrays using multiple criteria, global alignment and parameter estimation. Nucleic Acids Res. 2005;33:6114–23.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
60.Tu Q, He Z, Deng Y, Zhou J. Strain/species-specific probe design for microbial identification microarrays. Appl Environ Microbiol. 2013;79:5085–8.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
61.Wu L, Liu X, Schadt CW, Zhou J. Microarray-based analysis of subnanogram quantities of microbial community DNAs by using whole-community genome amplification. Appl Environ Microbiol. 2006;72:4931–41.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
62.Wu L, Liu X, Schadt CW, Zhou J. Microarray-based analysis of subnanogram quantities of microbial community DNAs by using whole-community genome amplification. Applied and Environmental Microbiology. 2006;72:4931–41.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
63.Oksanen J, Blanchet FG, Kindt R, Legendre P, Minchin PR, O’hara RB, et al. Package ‘vegan’. Community ecology package, version. 2013;2:1–295.
Google Scholar 
64.Anderson MJ, Bueno AS. A new method for non-parametric multivariate analysis of variance. Austral Ecol. 2001;26:32–46.
Google Scholar 
65.Crow EL, Patil GP. Applications in Ecology. In: Cros E, Shimizu K, editors. Lognormal Distributions. New York and Basel:Marcel Dekker; 1988. p. 303–30.66.Ser-Giacomi E, Zinger L, Malviya S, De Vargas C, Karsenti E, Bowler C, et al. Ubiquitous abundance distribution of non-dominant plankton across the global ocean. Nat Ecol Evol. 2018;2:1243–9.PubMed 
Article 

Google Scholar 
67.Wu L, Ning D, Zhang B, Li Y, Zhang P, Shan X, et al. Global diversity and biogeography of bacterial communities in wastewater treatment plants. Nat Microbiol. 2019;4:1183–95.CAS 
PubMed 
Article 

Google Scholar 
68.Locey KJ, Lennon JT. Scaling laws predict global microbial diversity. Proc Natl Acad Sci. 2016;113:5970–5.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
69.Louca S, Mazel F, Doebeli M, Parfrey LW. A census-based estimate of Earth’s bacterial and archaeal diversity. PLoS Biol. 2019;2:1–30.
Google Scholar 
70.Tokeshi M. Dynamics of distribution in animal communities: theory and analysis. Res Popul Ecol (Kyoto). 1992;34:249–73.Article 

Google Scholar 
71.Logares R, Deutschmann IM, Junger PC, Giner CR, Krabberød AK, Schmidt TSB, et al. Disentangling the mechanisms shaping the surface ocean microbiota. Microbiome. 2020;8:55.PubMed 
PubMed Central 
Article 

Google Scholar 
72.Azovsky A, Mazei Y. Do microbes have macroecology? Large-scale patterns in the diversity and distribution of marine benthic ciliates. Glob Ecol Biogeogr. 2013;22:163–72.Article 

Google Scholar 
73.Noguez AM, Arita HT, Escalante AE, Forney LJ, García-Oliva F, Souza V. Microbial macroecology: highly structured prokaryotic soil assemblages in a tropical deciduous forest. Glob Ecol Biogeogr. 2005;14:241–8.Article 

Google Scholar 
74.Thompson LR, Sanders JG, McDonald D, Amir A, Ladau J, Locey KJ, et al. A communal catalogue reveals Earth’s multiscale microbial diversity. Nature. 2017;551:457–63.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
75.Papp L, Izsák J, Papp L, Izsak J. Bimodality in Occurrence Classes: a Direct Consequence of Lognormal or Logarithmic Series Distribution of Abundances- A Numerical Experimentation. Oikos. 1997;79:191.Article 

Google Scholar 
76.Verberk WCEP, van der Velde G, Esselink H. Explaining abundance-occupancy relationships in specialists and generalists: A case study on aquatic macroinvertebrates in standing waters. J Anim Ecol. 2010;79:589–601.PubMed 
Article 

Google Scholar 
77.Liao J, Cao X, Zhao L, Wang J, Gao Z, Wang MC, et al. The importance of neutral and niche processes for bacterial community assembly differs between habitat generalists and specialists. FEMS Microbiol Ecol. 2016;92:fiw174.PubMed 
Article 
CAS 

Google Scholar 
78.Slatyer RA, Hirst M, Sexton JP. Niche breadth predicts geographical range size: a general ecological pattern. Ecol Lett. 2013;16:1104–14.PubMed 
Article 

Google Scholar 
79.Fierer N, Barberán A, Laughlin DC. Seeing the forest for the genes: using metagenomics to infer the aggregated traits of microbial communities. Front Microbiol. 2014;5:1–6.Article 

Google Scholar 
80.Rivett DW, Bell T. Abundance determines the functional role of bacterial phylotypes in complex communities. Nat Microbiol. 2018;3:767–72.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
81.Wertz S, Degrange V, Prosser JI, Poly F, Commeaux C, Guillaumaud N, et al. Decline of soil microbial diversity does not influence the resistance and resilience of key soil microbial functional groups following a model disturbance. Environ Microbiol. 2007;9:2211–9.PubMed 
Article 

Google Scholar 
82.Wertz S, Degrange V, Prosser JI, Poly F, Commeaux C, Freitag T, et al. Maintenance of soil functioning following erosion of microbial diversity. Environ Microbiol. 2006;8:2162–9.CAS 
PubMed 
Article 

Google Scholar 
83.Mendes LW, Tsai SM, Navarrete AA, de Hollander M, van Veen JA, Kuramae EE. Soil-Borne microbiome: linking diversity to function. Micro Ecol. 2015;70:255–65.CAS 
Article 

Google Scholar 
84.Sunagawa S, Coelho LP, Chaffron S, Kultima JR, Labadie K, Salazar G, et al. Structure and function of the global ocean microbiome – SM. Science. 2015;348:1261359–1261359.PubMed 
Article 
CAS 

Google Scholar 
85.Wohl DL, Arora S, Gladstone JR. Functional redundancy supports biodiversity and ecosystem function in a cloased and constant environment. Ecology. 2008;85:1534–40.Article 

Google Scholar 
86.Kurm V, Geisen S, Gera Hol WH. A low proportion of rare bacterial taxa responds to abiotic changes compared with dominant taxa. Environ Microbiol. 2019;21:750–8.PubMed 
Article 

Google Scholar 
87.Bergkessel M, Basta DW, Newman DK. The physiology of growth arrest: Uniting molecular and environmental microbiology. Nat Rev Microbiol. 2016;14:549–62.CAS 
PubMed 
Article 

Google Scholar 
88.Hofer U. Life in the slow lane. Nat Rev Microbiol. 2019;26:266–7.Article 
CAS 

Google Scholar 
89.Baho DL, Peter H, Tranvik LJ. Resistance and resilience of microbial communities – Temporal and spatial insurance against perturbations. Environ Microbiol. 2012;9:2283–92.Article 

Google Scholar 
90.Jousset A, Bienhold C, Chatzinotas A, Gallien L, Gobet A, Kurm V, et al. Where less may be more: how the rare biosphere pulls ecosystems strings. ISME J. 2017;11:853–62.PubMed 
PubMed Central 
Article 

Google Scholar 
91.Aanderud ZT, Jones SE, Fierer N, Lennon JT. Resuscitation of the rare biosphere contributes to pulses of ecosystem activity. Front Microbiol. 2015;6:1–11.Article 

Google Scholar 
92.Lawson CE, Strachan BJ, Hanson NW, Hahn AS, Hall ER, Rabinowitz B, et al. Rare taxa have potential to make metabolic contributions in enhanced biological phosphorus removal ecosystems. Environ Microbiol. 2015;17:4979–93.CAS 
PubMed 
Article 

Google Scholar 
93.Zhou J, He Z, Yang Y, Deng Y, Tringe SG, Alvarez-Cohen L. High-throughput metagenomic technologies for complex microbial community analysis: open and closed formats. MBio. 2015;6:e02288–14.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
94.Zhou J, Wu L, Deng Y, Zhi X, Jiang YH, Tu Q, et al. Reproducibility and quantitation of amplicon sequencing-based detection. ISME J. 2011;5:1303–13.CAS 
PubMed 
PubMed Central 
Article 

Google Scholar 
95.Shi Z, Yin H, Van Nostrand JD, Voordeckers JW, Tu Q, Deng Y, et al. Functional Gene Array-Based Ultrasensitive and Quantitative Detection of Microbial Populations in Complex Communities. mSystems. 2019;4:99–117.
Google Scholar  More

Overview of the FluxNet methodThe FluxNet approach is based on a mechanistic model, which includes multiple species/types of phytoplankton, bacteria, dissolved and particulate organic matter (DOM, POM), inorganic nutrients, micronutrients and inhibitors (see Table 1). For phytoplankton—bacteria carbon flux, which is the focus here, phytoplankton produce organic carbon by exudation and death. For exudation, living phytoplankton produce total DOM at constant and photosynthesis-proportional rates (ke, ef), with a composition defined by an exudation fraction (Fe) for each DOM species. These parameters vary by phytoplankton type. For example, for green algae (gre), the constant exudation rate is kegre and the fraction of glucose-containing HMW DOM (gl2) is Fegre,gl2. For one phytoplankton type the total DOM production varies in time with the photosynthesis rate, but the composition is constant. Phytoplankton die by a general death function and inhibition. The death function is time-variable (a bell-shaped function with a maximum at a specific time of year) and does not differentiate between various death mechanisms like zooplankton grazing or viral lysis, but presumably it represents mostly grazing in this case. Upon death, the phytoplankton biomass is converted to POM and DOM, where e.g., the content of chrysolaminarin (chr) for the diatom Rhizosolenia styliformis (rst) is defined by a composition fraction (Fxrst,chr). POM dissolves to DOM at a first-order rate. Bacteria consume DOM using Monod-level kinetics, where e.g. the affinity for Polaribacter (pol) for chrysolaminarin is defined by a half-saturation constant (Kshpol,chr).Table 1 Model components.Full size tableThe novel aspect is the upscaling to hundreds of state variables and thousands of parameters, which is accompanied by several conceptual and practical modeling challenges. To balance mass and account for the action of unobserved components, cryptic or hypothetical species are included [17], like DOM types d01-d15, which may represent e.g., threonine [18]. To simulate a diverse community with a smaller number of drivers (“paradox of the plankton”) and control chaos, interaction via micronutrients and inhibitors, as well as dormancy is included [19,20,21,22]. Parameters are optimized/calibrated to minimize the discrepancy between the model and observations. Which parameters are optimized and the corresponding ranges is based on available information (complete model equations and parameters are in Table S1–S25). For example, the constant DOM production rate (ke) is optimized for all phytoplankton, with a range adopted from a previous modeling study [23]. For rst (Rhizosolenia styliformis), the exudation fractions for most DOM components, like the cryptic species d01 (Ferst,d01), are optimized. Others, like glucose-containing HMW DOM (Ferst,gl2), are fixed based on literature (Table S14). The optimization is challenging because of the many components, nonlinear interactions, and resulting local optima in the objective function. We developed an optimization routine customized for microbial ecosystems with a number of key features.First, the method mimics natural speciation, where a coarse-grained model is gradually de-lumped to a finer resolution, a strategy also used in manual model development [13, 24, 25]. This is illustrated in Fig. 1, which shows how the model starts with just one component in each ecological compartment (Fig. 1E). This model is optimized until a threshold is reached, and then all species are de-lumped/split into two, followed by another round of optimization and so on. During the course of the optimization, with time or model runs, the number of components and parameters increase, and the total error generally decreases, although there can be a transient increase when new species are introduced (Fig. 1A, B). This way the optimization routine works with a smaller model on average and computational effort can be directed to a smaller set of parameters corresponding to newly introduced species, and the performance increases (Fig. 1C).Fig. 1: FluxNet inference method illustration.A Number components and optimized parameters. B Error for entire model (Total) and selected individual observations (rst = R. styliformis, pol = Polaribacter, lam = particulate chrysolaminarin). Best of 128 replicate runs. C Diversification of chrysolaminarin uptake affinity (max. heterotrophy rate/half-saturation constant). D Method performance with and without de-lumping. E Network corresponding to different de-lump levels. See Table 1 for component names and abbreviations.Full size imageAt each de-lumping level, the new species generally inherits the parameter values (i.e., the genome [26]) from the old species. Subsequent optimization then diversifies the population. This is illustrated in Fig. 1C, which shows the uptake affinity of all bacteria species for chr. However, different parameter values can also be specified for the new species, and then they are adopted and overwrite those inherited from the old species. This is used, for example, to assign species-specific cell sizes or prevent species from taking up a substrate. In Fig. 1C, those species that are not capable of assimilating chr, like rei (Reinekea), have an affinity equal to 0. The method thus allows for natural and automated expansion of the model to very large scale, yet provides a way to constrain/curate it based on available information.Second, the routine includes multi-parameter optimization (Nelder-Mead simplex method) on selected subsets of dependent parameters, like those involved in the production and consumption of chrysolaminarin (chr) or directly affecting the photosynthesis of the diatom R. styliformis (rst). Dependence between parameters, like max. photosynthesis rate and nutrient half-saturation constant, are explicitly considered. Also, Monte Carlo scans are performed on selected parameter sets at various points in the process.Application to Helgoland time seriesThe FluxNet method is applied to a four-year time series at Helgoland [27], including near-daily observations of 15 phytoplankton and 38 heterotrophic bacteria types (e.g., species, strains) and various bulk and auxiliary parameters (e.g., Chlorophyll a, DAPI, temperature, nitrate+nitrite, ammonium, phosphate, light extinction) (Tables S19 and S20). Data from more focused studies characterizing DOM and POM are also included [28, 29] (Table S21).In addition to the time-series data, the model is informed by literature information. Model parameters, incl. general properties like phytoplankton exudation fraction or bacteria growth efficiency, are constrained based on past models and data. Also, constraints are implemented for parameters controlling composition, exudation and utilization for the specific components included in the model. Those were based on a literature meta-analysis, where we searched primarily for studies with strains from Helgoland, but included strains from other locations if necessary. These constraints include, for example, for the phytoplankton storage polysaccharide chrysolaminarin, the typical content (~30% for diatoms, none for dinoflagellates) and ability of bacteria to assimilate it (yes for Polaribacter, no for Roseobacters and Reinekea) (Tables S4 and S11). Imposing constraints from the literature generally results in a worse agreement with the observations, but also increased realism of the model. Removing the constraints of phytoplankton composition (Table S4) significantly improves the agreement with observations, but also predicts substantial glycogen content of diatoms (e.g., Fxmhe,gly+ply = 0.19). Removing uptake constraints by bacteria (Table S11) reduces the error, but not significantly, suggesting that there is enough flexibility of the model to reproduce the observations even with this constraint. However, that model also includes features that disagree with literature, like substantial uptake of chr by s11 (Kshs11,chr = 25 L/mmolC/d).Carbon fluxes through and within in the ecosystemThe final model includes 210 components and their behavior and interaction are described by a total of 8200 calibrated parameters of 50 different parameter types (e.g., the composition of each of the 53 microbes is described by 76 fractions Fx, or 4000 total parameters) (Fig. 1), and it constitutes a mass-balancing, mechanistically-constrained, quantitative representation of the ecosystem. It reproduces many of the observed patterns of summary parameters like Chlorophyll a (chl), total bacteria (dap), particulate chrysolaminarin (lam), various high-molecular weight (HMW) DOM compounds, as well as absolute concentrations of individual phytoplankton and bacteria species (Fig. 2A–C). Only subset of the hundreds of model components is shown in Fig. 2B, C, which were selected based on (a) importance (e.g., rst is the dominant OM producer in 2009), (b) availability of data (e.g., chrysolaminarin, [29]) and (c) illustration of co-blooming (panel B) and succession (panel C). All model-data comparisons are presented in the SI (Fig. S1). The model under-predicts total DOM (doc), probably because a large fraction of observed DOM is more refractory allochthonous material, which is not considered in the model.Fig. 2: FluxNet model results and comparison to data.A All model types lumped. Phytoplankton (chl, μgChla/L), POM (poc, incl. microbes, μmolC/L ×0.1*), DOM (doc, μmolC/L ×0.1*), bacteria (dap, 1e6/mL ×3*). Gray shading are spring blooms, defined as the first time of the year the phytoplankton exceeds 3 µgChla/L plus 28 days. B Selected types for 2009 spring bloom. Rhizosolenia styliformis (rst, centric diatom, 1e6/L ×1.2*), Thalassiosira nordenskioeldii (tno, centric diatom, 1e6/L ×0.05*), particulate chrysolaminarin (lam = phr + phytoplankton content, μmolC/L ×0.002*), dissolved chrysolaminarin (chr, μmolC/L ×0.002*, no data available), Polaribacter (pol, DAPI × CARD-FISH, 1e6/mL ×0.1*), NS3a marine group (ns3, 1e6/mL ×0.2*). C Selected types for 2010 spring bloom. Mediopyxis helysia (mhe, centric diatom, 1e6/L), Thalassiosira nordenskioeldii (tno, centric diatom, 1e6/L), glucose-containing HMW DOM (glc, μmolC/L ×0.01*), arabinose-containing HMW DOM (ara, μmolC/L ×3*), Reinekea (rei, 1e6/mL ×5*), Alteromonas (alt, 1e6/mL ×1.5*). Lines are model and symbols are data [27,28,29]. *Individual concentration series scaled to illustrate dynamics. See Fig. S1 for all model-data comparisons. Upside-down triangles mark various bloom stages for networks in (D) and Fig. 4A. D Inferred carbon flux network. Nodes are components. Size indicates in/outflux (μmolC/L/d), color varied randomly within each ecological compartment. Lines are fluxes. Thickness is proportional to log flux (μmolC/L/d), colored based on the source node, lines below a threshold distance are colored gray to highlight most important fluxes. Italic numbers are total fluxes (μmolC/L/d). Flux cut off is 0.01%. See Table 1 for component names and abbreviations. See Movie S1.Full size imageIt is important to understand that the model was calibrated to these observations, so this is not a prediction per se. The main information produced by this analysis (emergent property) are the mass fluxes. Predicted ecosystem-level fluxes can be compared to independent estimates, which were not used as input here. For the period 2009–2012, the gross primary production rate in the model is 28 (±1.2 standard deviation) mmolC/m2/d. Uncertainty of fluxes and parameters are based on top 5% of 128 replicate runs, as in [23]. This flux compares well to a regional estimate of 29 (26–33) mmolC/m2/d for the Transitional East Region of the North Sea for the same period [30]. At the end of March, the bacterial production rate in the model is 0.32 (±0.041), 0.14 (±0.017), 0.20 (±0.025) and 0.45 (±0.057) μmolC/L/d for the 4 years, respectively. This is consistent with measurements of 0.20 μmolC/L/d in 1992 ~30 km from Helgoland [31].These comparisons provide confidence in other aggregate fluxes predicted by the model. The C, N and P fluxes to the sediment bed, via settling of phytoplankton and POM, are 5.8 (±0.91) mmolC/m2/d, 0.87 (±0.14) mmolN/m2/d and 0.054 (±0.0085) mmolP/m2/d, which constitute 20%, 16% and 18% of the input via photosynthesis (C) or external input (N, P) (see Fig. S2). External “new” input of N is 0.66 μmolN/L/d, which is 6.0 time higher than the 0.11 (±0.023) μmolN/L/d released or “recycled” by bacteria.The resulting flux network includes quantitative carbon fluxes between all components at each time point, like 28 days into the 2009 spring bloom (Fig. 2D, Dataset S1 list all fluxes). The dominant source of organic matter is rst at 0.36 (±0.19) μmolC/L/d, 30% of which is dissolved and particulate chrysolaminarin (chr + phr). These instantaneous fluxes exhibit a higher uncertainty than the integrated fluxes discussed in the previous paragraph, which can be explained by small timing differences (Table S26). The DOM is consumed by a diverse consortium of bacteria, mostly Polaribacter (pol) at 0.46 (±0.22) μmolC/L/d, 35% of which is chr. chr has a through-flux of 0.25 (±0.049) μmolC/L/d and a turnover time of 8.8 (±2.0) days. In the model, phytoplankton and bacteria interact via DOM, but the carbon flux can be traced and used to quantify phytoplankton – bacteria associations. Here, the carbon flux via all DOM types from rst to pol is 0.27 (±0.20) μmolC/L/d, 58% of carbon to pol, making this the second-strongest (after ns3) microbial linkage in the system at this time. This who produces/consumes how much of what when information is the main output of the FluxNet method, and it is critical for moving our understanding of microbial ecosystem functioning beyond bulk parameters like respiration and photosynthesis rates towards a higher resolution.Whereas the 2009 spring bloom illustrates co-blooming of phytoplankton and bacteria, the 2010 bloom shows succession of phytoplankton, DOM and bacteria. Several factors control this pattern in the model. Reinekea (rei) is negative for chrysolaminarin (chr) based on literature (Table S11), but is predicted to have a relatively high affinity for other glucose-containing DOM (gl2) (khrei / Kshrei,gl2 = 63 (±22) L/mmolC/d). A substantial fraction of gl2 is produced relatively early by phytoplankton exudation, and it is the primary substrate for rei at bloom stage 14 days. Alteromonas (alt) is predicted to have a low affinity for gl2 (khalt / Kshalt,gl2 = 0.015 (±0.0097) L/mmolC/d), but it is positive for chr based on literature and predicted to have a high affinity (khalt / Kshalt,chr = 52 (±4.7) L/mmolC/d). Chr is a death (i.e. grazing) product of phytoplankton and produced relatively later in the bloom, and it is the primary substrate for alt at this time. The substrate spectra of bacteria emerge in the analysis, within literature constraints, and can be considered a prediction testable with modern experimental techniques [6].Oligotrophic and copiotrophic carbon processingThe network includes concentrations and fluxes for each bacteria type, and a natural question is to what extend they are correlated. There is increasing awareness that high abundance may not necessarily mean high importance and vice versa, including the over-proportional role of rare species in biogeochemical cycles [32]. In the model, there is a strong correlation between concentration and carbon flux of bacteria, but for the same concentration there is also about an order of magnitude variation in flux (Fig. 3). The spread reflects differences in growth rates during the bloom periods. Some species, like the oligotroph SAR11 (s11), have consistently lower flux and others, like the copiotroph Polaribacter (pol), have consistently higher flux. There are also some, like the cryptic alphaproteobacteria (alx), that go in different directions in different years.Fig. 3: Correlation between spring bloom abundance and importance.Concentration and carbon flux for all model bacteria types during spring bloom periods (see Fig. 2 caption for definition). Lines: All(dashed)/Olig.(thick)/Copi.(thin), log Flux = –0.93/–1.03/–0.81 + 0.93/1.00/0.94 × log Conc., R2 = 0.88/0.92/0.92.Full size imageIt is important to realize that, in dynamic systems, microbial interactions and the corresponding networks are not static [3, 33]. The dynamics of the entire Helgoland flux network over the four-year period is illustrated in an animation, which shows the production of DOM and POM during and after phytoplankton blooms and later blooming of bacteria (Movie S1). These features are also evident in the phytoplankton – DOM – bacteria interactions at two selected time points during the 2009 spring bloom (Fig. 4A, B). At the onset of the bloom, the oligotroph SAR11 (s11) consumes the most DOM, primarily the cryptic species d08, which comes mostly from grazing death of green algae (gre) and exudation by rst. After 28 days the copiotroph pol dominates, which consumes primarily chr, a death product of mostly rst. SAR11 continues to be a major carbon processor in the early parts of the bloom, which was unexpected, because it is an inferior competitor at this time (growth rate s11 = 0.051 vs. pol = 0.15 1/d, bloom average), but can be explained by the higher biomass concentration (s11 = 0.68 vs. pol = 0.13 μmolC/L, bloom start). The flux is proportional to concentration and growth rate, and neither measure alone is a good proxy for the importance of a species [4]. Across all four years, oligotrophic bacteria, defined based on below-average growth rates (literature classifications are often ambiguous), dominate carbon processing for the first 18 days, generally past the phytoplankton peak (Fig. 4C).Fig. 4: Carbon processing during the course of blooms.A Phytoplankton—DOM—bacteria carbon flux network for the start and +28 days of 2009 spring bloom. See Fig. 2 legend. Flux cut off is 0.3%. B Cell concentrations, growth rate and relative carbon processing for s11 and pol for 2009 spring bloom. C Fraction of DOM processed by oligotrophic bacteria and exudate fraction in DOM pool for all blooms. Oligotrophs are defined based on literature as shown in Table S23 or based on below-average growth rates (kg). For the later, the oligotrophic fraction or weight given for type i, is based on fOLIi = kgAVEn / (kgAVEn + kgin), n = 5. kg is the net growth rate calculated from biomass change, plus dilution rate.Full size imageThe use of d08 by s11 and chr by pol in 2009 suggests are more general pattern, i.e., use of exudation products earlier by oligotrophs and death (i.e., grazing) products later by copiotrophs. Across all years, the fraction of DOM produced by exudation decreases during the course of the bloom (Fig. 4C), a common feature of phytoplankton blooms [33]. This is reflected in the diet of these bacterial groups, i.e., for oligotrophs (vs. copiotrophs), exudates make up a higher portion of the diet (27 vs. 18%), and they have a higher affinity for exudates (39 vs. 35 L/mmolC/d), which is also consistent with experimental evidence from another system [7].After the model was developed, while this paper was in peer review, metaproteomic data for the Helgoland Island spring bloom in 2016 were published that suggest that algal storage compounds (e.g., chrysolaminarin) are used throughout the bloom, whereas cell wall-related compounds (e.g., fucose-containing) are used at later bloom stages [34]. Our model also predicts an increase in the consumption of cell-wall vs. storage compounds at later bloom stages (Fig. 5), which validates our outcomes, although a direct comparison is not possible because of the different time.Fig. 5: Consumption of cell wall vs. storage compounds during the course of blooms.Total consumption (all bacteria) in µmolC/L/d of cell wall compounds divided by storage compounds. Cell wall compounds = man (mannan) + glo (glucoromannan) + fcs (FCSP). Storage compounds = chr (chrysolaminarin) + gly (glycogen) + sta (starch). Averages for all four years.Full size imagePhytoplankton functional similarity decouples them from bacteriaAn important question is to what extent the patterns recur from year to year [27]. We compare networks of phytoplankton producers, DOM exchanged and bacteria consumers, as well as phytoplankton – bacteria interactions quantified in absolute (μmolC/L/d moving between phytoplankton X to bacteria Y) and relative (% of carbon for bacteria Y supplied by phytoplankton X) terms (Fig. 6A). All networks show significant similarity so there is recurrence from year to year. The recurrence is higher for DOM than phytoplankton, suggesting that different phytoplankton produce similar DOM, which is expected considering similar composition (e.g., chr in diatoms). There are no phytoplankton producers that recur in the top quartile every year, but chr and others are in the top quartile of DOM exchanged (produced and consumed) every year. The recurrence is lower for bacteria consumers suggesting factors beyond DOM shape the bacteria community.Fig. 6: Recurring patterns and comparison of FluxNet and LSA methods.A Similarity of networks for spring blooms. Error bars are 95% confidence limits. Bray-Curtis similarity was calculated as 1 – Bray-Curtis dissimilarity. Text on top of symbols lists components that recur in the top quartile every year, listed in order of average rank. B Carbon flux networks for top recurring bacterial consumer, top four DOM sources and top coupled phytoplankton. (C&D) LSA network (showing top 15% of significant local similarity scores) and sample time series.Full size imageAn important question is how specific interactions are and how tightly networks are interconnected [35, 36], which depends on the mechanisms of interaction and will affect the recurrence. Consistent with the relatively low recurrence of phytoplankton producers, phytoplankton—bacteria coupling shows relatively low recurrence, i.e. low specificity. The primary substrate for the consumer pol is mostly chr and gl2, although it does change from year to year with varying DOM, consistent with the known assimilation capabilities of pol (Polaribacter) [37] (Fig. 6B). However, the primary associated phytoplankton for pol is different each year, although it is always a diatom. The de-coupling of phytoplankton production and bacteria consumption was also concluded from the lower recurrence of phytoplankton and higher recurrence of bacteria abundance in the same dataset [27]. It suggests that carbon processing is resilient to changes in phytoplankton, which may arise from factors like species invasion or climate change.The above discussion focused on one-way/commensal (phytoplankton  > DOM  > bacteria) interactions, but the network also includes specific two-way/mutualistic phytoplankton-bacteria interactions. Phaeocystis (pha) has the highest exudation fraction and Bacteroidetes nvi the highest affinity for DOM d04, whereas nvi has the highest exudation fraction and pha the highest requirement for micronutrient m15. Such mutualism is observed in other systems and the interactions predicted here can be tested experimentally [20]. Alternatively, experimentally-observed interactions could be used as input to the method, as constraints.Robustness of the analysisTo understand the effect of some of the choices made in the model structure we repeated the analysis with added or removed components or processes. Models without micronutrients or inhibitors produce significantly worse fit to the data (Fig. 7A), highlighting the need for a two-way interaction between phytoplankton and bacteria to maintain diversity. Models with more micronutrients or inhibitors are similar to the basecase. Together, these results provide some justification for the complexity (i.e., number of parameters) in the basecase model. The analysis including osmotrophy (aka absorbotrophy, i.e., phytoplankton can perform heterotrophy) produces a better fit to the observations, but that model was not adopted as basecase, because the osmotrophy process is poorly constrained and includes some probably unrealistic features/fluxes, like significant exudation and uptake of the same substance by one phytoplankton species. Importantly, excluding the runs with worse fit to the observations, the main conclusions (as shown in Figs. 4C and 5A) are the same, confirming that the results are reproducible and robust to some of the choices made in model structure.Fig. 7: Reproducibility of main results.A Total error for runs with different models. “w” or “b” indicates performance is significantly worse (open bars, think lines) or better than basecase, p  ns9 interaction ranks in the top 1% for LSA and FluxNet (relative interaction). However, the lack of mechanistic constraints is evident. One of the strongest links for the 2009 spring bloom (rank 13%) is between the diatom Chaetoceros debilis (cde) and Roseobacters (ros) (Fig. 6C, D). The shifted peaks line up nicely, but the bacteria biomass is higher than that of the phytoplankton and genome analysis suggests ros do not assimilate chrysolaminarin [37], which is a major death product of diatoms. Considering this, growth yield and other competing consumers, it is unlikely that cde is a major source of carbon to ros.Summary and outlookModern observational tools are generating high-resolution descriptions of the components of microbial ecosystems, and an ongoing grand challenge is to use these data to understand how systems function. Our method predicts dynamic mass fluxes between marine phytoplankton and bacteria, which provides insights into the functioning of the ecosystem. Specifically, it showed that there is a strong correlation between concentration and flux of bacteria during blooms, but oligotrophs are relatively less important than copiotrophs. However, due to their higher biomass, they are major carbon processors during early phases of blooms, well past the peak. Oligotrophs grow preferentially on exudation products, which are more abundant earlier in the bloom. Also, our results suggest that phytoplankton are functionally similar in terms of what organic carbon species they produce, and that this decouples them from bacteria.FluxNet is an inference method for microbial time series data that serves the same general purpose as existing empirical inference methods, like LSA [38]. In general, both approaches have strengths and weaknesses (see Introduction) and may complement each other. A main advantage of FluxNet is that it produces quantitative concentrations and fluxes, and associated conclusions (e.g., preferential use of exudates by oligotrophs). Also, it is constrained by mass balance and additional information from the literature (i.e., beyond the time series data), which make the results more realistic.The existing FluxNet code can readily be applied at a higher resolution (microdiversity), explicit representation of other ecosystem components, like viruses and zooplankton, and more processes, like photoheterotrophy and mixotrophy. It may also be applied to understand other microbial ecosystems, like the human gut or wastewater treatment plants. For an inference method it is important to be applicable to various types of observations, including modern environmental -omics observations, like transcript, protein and metabolite levels, and the present model will have to evolve in this direction [39]. The present model includes a relatively simple representation of the various processes, and the current biological understanding supports increasing the mechanistic realism (and complexity). For example, the present version assumes constant composition of DOM produced by phytoplankton, but observations show that it changes with physiology and interaction with bacteria [18, 40]. Also, the model assumes simple first-order dissolution of POM to DOM and direct utilization by bacteria, whereas break-down of especially polysaccharides is often mediated by extracellular enzymes [41]. More