More complex than you think: Taxonomic and temporal patterns of plant–pollinator networks of caraway (Carum carvi L.)

Caraway (Carum carvi L.) is a crop species that is gaining in importance in Europe, especially as a condiment and medicinal plant. Here, we present the plant–pollinator network of caraway in a central European agricultural landscape, focusing on two diverse potential pollinator taxa, Diptera: Brachycera (= true flies) and Hymenoptera (sawflies, bees, and wasps). We specifically studied qualitative differences in interactions between the two insect taxa as well as the intraday and intraseasonal variability of the network. Insect and pollen plant species determination was done via morphological identification and DNA (meta)barcoding. In total, 121 species representing 33 families of Hymenoptera and Brachycera were found to carry caraway pollen. These taxa included many nonhoneybee and nonhoverfly species, showing a wide taxonomic breadth of potential pollinators and a higher network complexity than previously anticipated. There are distinct qualitative differences between Brachycera and Hymenoptera networks, suggesting complementary roles of both taxa in the pollination of native and crop plants. Strong intraday differences in potential pollinator diversity make it necessary to collect insects and pollen at different times of the day to compile complete plant–pollinator networks. Intraseasonal analyses of the plant–pollinator network of caraway show the potential of caraway as an important food source for insect species with an activity peak in late summer.

In addition, there is evidence that dipterans are more resilient to stressors such as land-use change in comparison to managed and wild bees (Rader et al., 2016) and are as efficient as honeybees (Rader et al., 2009).
Most studies of pollination by Diptera have focused on one family (Syrphidae) despite many other taxa frequently being listed as flower visitors (Orford et al., 2015;Ssymank et al., 2008). The understudied taxa are often excluded or deliberately ignored, which leads to the erroneous assumption that they do not play an important role in pollination (Larson et al., 2001;Rader et al., 2016Rader et al., , 2020 or miss the fact that some species can be pollinators as well as pest or weed control agents (Dunn et al., 2020;Moerkens et al., 2021;Rizza et al., 1988;Sheppard et al., 1995). Consequently, many governmental programmes to enhance pollinator diversity have been designed and developed for well-studied taxa, neglecting the importance of including and safeguarding the less-studied taxa (Orford et al., 2015;Rader et al., 2020). A deeper knowledge of plant-pollinator interactions is required to understand the underlying multifactorial processes causing the worldwide pollinator decline and the effects of landscape changes from agricultural activity (Arstingstall et al., 2021). At the same time methods that protect beneficial insects while also protecting the economic interests of farmers necessitate further investigation.
Providing flower strips or fallow land in the vicinity of crops has been shown to be successful at attracting pollinators (Batáry et al., 2015;Feltham et al., 2015;Garibaldi et al., 2016) as well as natural enemies of crop pests (Cahenzli et al., 2019;Tschumi et al., 2015). However, the cost of seed stock for flower strips that do not provide an income source as well as a delayed increase of crop production for several years often make them unattractive to farmers (Christmann et al., 2021). An approach that would devote small plots of ecologically beneficial mass-blooming crops or "magnet-species" (Thomson, 1978;Zych et al., 2007) in the vicinity of fields and crops could be a potential alternative or addition to flower strips to enhance the populations and diversity of agriculturally beneficial insects while providing an additional income source (Christmann et al., 2021).
Carum carvi L. (caraway) is an annual or biennial cultivated medical and spice plant with a worldwide increasing market (Stelter, 2014;Van Roon & Bleijenberg, 1964). It belongs to the family Apiaceae and has characteristic yellowish white protandrous flowers arranged in compound umbels (d'Albore, 1986;McGregor, 1976) which require insect pollination to transport the pollen to the stigma. Pollination exclusion can reduce seed yield up to 40% (Bouwmeester & Smid, 1995;Toivonen et al., 2022). In Central Europe, the main flowering period of caraway has a duration of 25-30 days (Németh et al., 1997) between late May and early July (Langenberger & Davis, 2002). After the main flowering period, there is sometimes a second flowering period in autumn, but to a lesser extent (Hegi, 1926). Nectar and pollen are easily available and they constitute a valuable source of protein and carbohydrates for many potential pollinators (d'Albore, 1986;Langenberger & Davis, 2002;McGregor, 1976;Toivonen et al., 2022). Syrphidae and other flower-visiting Brachycera that provide both economically important services of pollination and pest control (at their larval stage in the case of syrphids) are known to be important pollinators of caraway and related Apiaceae species (Colley & Luna, 2000;Lamborn & Ollerton, 2000;Pérez-Bañón et al., 2007;Toivonen et al., 2022;Wojciechowicz-Żytko, 2019;Zych, 2002Zych, , 2007Zych et al., 2014Zych et al., , 2019. These characteristics make caraway a possible crop to be added to an agricultural system to attract beneficial insects while offering an additional economic resource for growers. In the present study, we analysed the astounding complexity of the plant-pollinator network of caraway in a central European agricultural landscape, targeting taxon-specific roles and, so-far neglected, temporal patterns within the network. The two prevailing approaches to analyse plant-pollinator networks are observing the interaction between the flower and a potential pollinator (Classen et al., 2020;Toivonen et al., 2022) or morphological identification of pollen loads (Beattie, 1971;Erdtman, 2013). While both methodologies are widely used, they often underestimate the number of plant species visited, leading to an underestimation of the total number of interactions (Jędrzejewska-Szmek & Zych, 2013;Macgregor et al., 2019). Furthermore, many described plantpollinator networks are static records, not covering temporal shifts of species and interactions (Burkle & Alarcón, 2011;CaraDonna et al., 2017CaraDonna et al., , 2021CaraDonna & Waser, 2020). To gain a better estimate of plant species visited we implemented morphological identification as well as DNA metabarcoding for pollen determination. DNA metabarcoding is an emerging molecular tool that has the potential for more rapid pollen identifications with higher taxonomic resolution than possible with traditional morphology (Bell et al., 2017;Macgregor et al., 2019;Sickel et al., 2015). Previous studies utilizing DNA metabarcoding to investigate plant-pollinator networks have resulted in more complex networks when compared to those based on observational data alone (Michelot-Antalik et al., 2021;Pornon et al., 2017). In addition, the data contradicted the previous assumption that many pollinating species are specialists (Arstingstall et al., 2021). Still, many DNA metabarcoding studies target only those flower-visiting species with high pollen loads and do not cover the whole diversity of potential pollinators (Arstingstall et al., 2021;Bänsch et al., 2020;Bell et al., 2017;Cornman et al., 2015;Keller et al., 2015).
In this study, we aim to address the following questions: (i) Which Hymenoptera and Brachycera are involved in the plant-pollinator network of caraway? (ii) What are the qualitative differences in the plant-pollinator networks of Hymenoptera and Brachycera? (iii) What is the intraseasonal pattern of the plant-pollinator network? (iv) Are there any intraday differences in potential pollinator diversity of caraway?
The plot without flower strips had a length of 150 × 9 m (1350 m 2 ).
The plots were surrounded predominantly by winter barley and wheat and flowering orchards (cherry, apple). By sampling both areas we wanted to attract as many potential caraway pollinators as possible.
The caraway (bi-annual variety "Sprinter"; N. L. Chrestensen) and flower strip were sown on April 8, 2016, seeding was carried out on both fields at a row spacing of 50 cm and the sowing rate was set at 10 kg ha −1 . Field emergence on April 23, 2016 in the plot with flowering strips was 77% and the plot without flowering strips had a field emergence of 91%. For weed control, a preemergent Aclonifen herbicide was applied on both fields once at 3 L ha −1 on April 14. To keep the plots weed-free during the vegetation period, additional weed control was performed with a roller and hand hoe.

| Collection of potential pollinators
During the caraway flowering period (July 5-25, 2016), sampling took place on all rain-free days, resulting in 10 collection days in total. After the flowering period, we sampled an additional 5 days (August 13 to September 1, 2016) after a 15-day gap to establish a boundary between the interactions during and after the main flowering time (Table S2).
Both caraway plots were sampled for 30 min each at three different time intervals to sample variation of insect activity: 10-12 h (interval I), 12-14 h (interval II) and 14-16 h (interval III). To avoid time bias in our sampling of the two caraway fields, we alternated the order, starting with a different caraway plot each day.

| Identification of potential pollinators
In the laboratory, all insect specimens were identified based on external morphology (Table S3 for keys used). For those insect specimens which could not be morphologically identified to the species level, we followed a DNA barcoding reverse-taxonomy approach (Morinière et al., 2019) using molecular sequences of the mitochondrial cytochrome c oxidase subunit I gene (COI; Hebert et al., 2003) ( Figure 1).

| DNA barcoding of Brachycera and Hymenoptera
DNA was extracted from one to three legs per specimen in larger taxa (>5 mm long) or using a nondestructive protocol for lysis extraction of the whole specimen in small taxa (<5 mm long) (Gilbert et al., 2007) with the BioSprint96 magnetic bead extractor (Qiagen). Subsequently, specimens were pinned, mounted, or kept in ethanol, and labelled. Voucher specimens are deposited at the Zoologisches Forschungsmuseum Alexander Koenig (ZFMK, Leibniz Institute for the Analysis of Biodiversity Change). PCR (polymerase chain reaction)_ amplification followed the Canadian Centre for DNA Barcoding (CCBC) protocol for COI amplification with the primer pair HCO2198-JJ (AWACT TCV GGR TGV CCA AAR AATCA) and LCO1490-JJ (CHACW AAY CAT AAA GAT ATYGG) (Astrin & Stüben, 2008). PCR products were sequenced at the Beijing Genomics Institute (BGI; https://en.genom ics.cn/). Sanger sequences were imported into Geneious version 7.1.9 (Kearse et al., 2012) and prepared with the Laboratory Information Management System plug-in (lims; Biomatters). Reverse and forward sequences were trimmed, filtered and de novo assembled.
Assembled COI sequences were inspected and manually corrected if necessary. Sequences were submitted to GenBank (OQ611071 F I G U R E 1 Visual summary of the methods applied to detect potential pollinators of caraway. -OQ611458) and the GBOL reference database (GBOL DNA Barcode Reference Library 2022; Geiger et al., 2016). Sequences were then searched in BOLD (https://www.bolds ystems.org/; Ratnasingham & Hebert, 2007). If all matches in BOLD with >99% similarity resulted in the same species name, this species name was used.

| Preparation and morphological identification of pollen loads
Pollen was collected by swabbing the insect specimens with a lentilsized piece of Kaisers phenol-free glycerol gelatin (Carl Roth), with a focus on the areas where the pollen was present in the greatest concentration. Then, the glycerol gelatin fragment was mounted on a slide over a 55°C heating plate and covered with a cover slip ( Figure 1) (Beattie, 1971). Morphological identification of their pollen load was done (i) for all specimens per species when the number of specimens per species was five or fewer per plot (data from both plots were pooled later, see below), (ii) from five specimens when the number of specimens per species was 6-50, ensuring that all intraday intervals were covered, and (iii) 10% of specimens when the number of specimens per species was >50 individuals, optimizing for intraseasonal collection dates. Slide-mounted pollen samples were then photographed to serve as a voucher, as the downstream method of DNA metabarcoding is destructive, and then identified with a microscope (Olympus, 400-1000× magnification) using the keys of von der Ohe and von der Ohe (2007) and the pollen reference collection of the Specialized Centre for Bees and Beekeeping in Mayen, Germany (www.biene nkunde.rlp.de) ( Figure 1). "Types" represent taxonomic units that cannot be identified further to lower taxonomic levels and may contain several species or even genera.

| DNA metabarcoding of pollen loads
Following morphological identification, pollen slides were used in DNA metabarcoding. To maximize pollen content (and therefore higher DNA content) and reduce cost, all slides from a particular insect species (one to six) were combined in a single 2-mL SafeSeal microcentrifuge tube (Sarstedt) to create a species-and plot-specific, but not specimen-or time-specific sample with a total number of 206 samples. To minimize cross-contamination, the slides were wiped externally with Molecular BioProducts DNA AWAY before gently removing the coverslips with a sterile scalpel blade. Then, the glycerol gelatin sample was removed from the slide with the same sterile scalpel. In most cases removal of the coverslip and obtaining the gelatin sample required little effort, but in several instances, the slides were heated at 50°C for 5 s, and in a few cases, specimen samples were discarded due to glass fragmentation.
Following the sample creation, 1 g of 1.4-mm ceramic beads was added to the 2-mL tube and DNA was extracted with a Nucleomag 96 Plant Kit (Macherey Nagel). All reagents were used at 25% of the factory protocol, except for elution buffer MC6. Prior to lysis incubation lysis buffer MC1 and 5 μL Proteinase K (10 mg mL −1 ) were added and the sample was homogenized for 2.5 min on a Mixer Mill MM 400 (Retsch) at 30 Hz, then incubated at 65°C for 60 min after which, 5 μL RnaseA (10 mg mL −1 ) was added and incubated at room temperature (20 ± 2°C) for 30 min. Following all other protocol steps, 35 μL of elution buffer MC6 was added and incubated at 55°C for 5 min to remove residual ethanol, then 25 μL was removed for further processing and 2 μL for DNA quantification with a Qubit 4 fluorometer (Thermo Fisher Scientific).
Polymerase chain reaction was performed with three replicates per sample, with the addition of two DNA extraction-negative controls and two PCR-negative controls to evaluate contamination, and two positive controls. Amplification was performed with an adaptation of the Canadian Centre for Barcoding Platinum Taq Protocol (Ivanova et al., 2007) with the addition of 0.25 μL BSA (bovine serum albumin; 0.01 mg mL −1 ) and 1.25 μL of 50% DMSO (dimethyl sulphoxide) in a total reaction volume of 12.5 μL. Universal plantspecific ITS2 primers were used: forward: ITS-3p62plF1, ACBTR GTG TGA ATT GCA GRATC and reverse: ITS-4unR1, TCCTC CGC TTA TTK ATATGC (Kolter & Gemeinholzer, 2021b). PCR cycling conditions were 95°C for 3 min, followed by 35 cycles of 95°C for 30 s, 50°C for 30 s, 72°C for 45 s and a final extension of 72°C for 10 min.
Following PCR cycling, the three replicates were combined by the addition of 5 μL of each replicate for a total volume of 15 μL and purified with Thermo Scientific Exonuclease 1. The pooled replicates of nonindexed PCR products were sent to LGC Genomics for sequenc- Sequencing data were processed with usearch (Edgar, 2010) and dada2 (Callahan et al., 2016) in R (R Core Team, 2021). Sequencing primers were trimmed and quality filtered with a maximum expected error of 1.0 in usearch. Dada2 was then used for error learning, denoising by the error profile (pseudo pooling) and merging of reads.
Chimeras were removed with uchime3. The resulting amplicon sequence variants (ASVs) were identified by implementation of the SINTAX algorithm (Edgar, 2016) using the PLANiTS database (Banchi et al., 2020) and submitted to NCBI SRA (Accession nos.: PRJNA935259 and PRJNA935270). The resulting ASVs with fewer than five reads per sample, as well as fungal contaminants, were discarded. Taxa that do not occur in Germany and were probably present due to laboratory contamination were removed from further analysis. Taxa with ambiguous species-level identifications, due to lack of coverage in the identification reference database, were given genus-level identifications.

| Curation of the data set
The data initially kept separately by plot (i.e., plot with and without flower strip) were pooled into two data sets: one based on morphological identification of pollen only (data set 1) and one based on the morphological identification combined with the DNA metabarcoding identifications of pollen (data set 2). Both data sets therefore differ in terms of identified plant/pollen species but include the same data on potential pollinators (i.e., insect species). Data set 1 is semiquantitative (i.e., includes the number of samples containing the respective interaction). Data set 2 was converted into a single presence/absence data set (qualitative data set). Since presence/absence data sets can over-accentuate interactions with rare plant taxa, we excluded plant species involved in less than 1% of the total number of interactions from the analysis, following Lucas et al. (2018).

| Terminology, statistical analysis, plantpollinator networks and indices
We constructed bipartite networks composed of two node divisions (insect and pollen species) connected by a link defined as an interaction between the plant and a potential pollinator (Dormann et al., 2009).
All bipartite plant-pollinator networks analyses were carried out in R (version 1.4.) (R Core Team, 2021), using the function plotweb of the bipartite package for the network analyses and using the function networklevel for network indices (Dormann et al., 2008). We created five different plant-pollinator networks: (i) a network with all potential hymenopteran and brachyceran pollinating species of caraway, (ii) a plant-pollinator network with only brachyceran species, and (iii) with only hymenopteran species ((i-iii) based on the qualitative data set 2), and two intraseasonal networks, (iv) during and (v) after the main flowering period (based on the semiquantitative data set 1).
For the description of the main differences between the plant-pollinator networks, we calculated the Connectance (C), Nestedness (N) and the mean number of links per species for each network. C is defined as the number of links in proportion to all possible links (Dormann et al., 2009;Dunne et al., 2002).
The overall N of a network (in this case 0 being highly nested) describes the specialization asymmetry, that is the proportion between specialists and generalists in the network (Bascompte & Jordano, 2007;Dormann et al., 2008Dormann et al., , 2009). Since the morphological identification of pollen grains was only possible at genus, type or family levels in most cases, some plant species in data set 2 may be represented in several nodes.

To analyse the differences in interactions between
Hymenoptera and Brachycera, we used a permutational multi- The differences were also plotted as a nonmetric multidimensional scaling (nMDS) with the function metaMDS and ellipses were generated with the function VeganCovEllipse (package vegn : Oksanen et al., 2016). We also generated a UpsetR-plot (function upset in the R package UpsetR; Conway et al., 2017) to study the key intervals to sample the highest number of potential pollinators of caraway.

| RE SULTS
We collected 1021 insect specimens (844 brachycerans and 177 hymenopterans). These specimens represent 121 species from 33 families (87 Brachycera taxa from 20 families and 34 Hymenoptera taxa from 12 families) ( Figure S1). In total, 707 specimens were identified morphologically (559 Brachycera and 148 Hymenoptera) and 516 specimens (331 Brachycera and 185 Hymenoptera) were identified via DNA barcoding following the reverse-taxonomy approach. Of the 1021 specimens collected, 457 were selected for analyses of their pollen load, representing all collected insect species.

| Plant-pollinator network of caraway
The plant-pollinator network of caraway included 121 potential pollinator species and 139 plant taxa of different taxonomic levels.
Overall, we found 859 links, from which 199 links were identified only by pollen morphology, 617 links were recorded only by DNA metabarcoding of pollen samples and 43 links were recorded by both methodologies.
We found a significant difference in pollen load composition

| Intraseasonal pattern of the plant-pollination network of caraway
We  Table 1).

| Intraday differences of potential pollinator diversity
In total, 75 out of the 121 potential pollinators species of caraway

| DISCUSS ION
To our knowledge, the present survey is the first study character- pollinators of caraway (Bouwmeester & Smid, 1995;d'Albore, 1986), we found a much higher diversity of potential pollinators, in particular featuring many species of Brachycera. This aligns with the results by Toivonen et al. (2022), which provided evidence of higher flower visiting rates by Brachycera on caraway than of bee species combined.
The overall plant-pollinator network with all species had an overall low nestedness and connectance, which usually makes networks more prone to external disturbances (Bascompte et al., 2003).
However, when calculating these network indices, we encountered These results align with previous studies (Phillips et al., 2018;Rader et al., 2011). Based on these numbers, hymenopterans might appear as more effective pollinators, but the higher abundance of brachycerans as flower visitors compared to hymenopterans (this study; Garibaldi et al., 2013;Garratt et al., 2014;Innouye et al., 2015;Rader et al., 2016), their higher resilience to land-use changes in comparison with bees (Rader et al., 2016;Ricketts et al., 2008), their ability to carry pollen to greater distances, and their potential additional ecosystem service as biocontrol agents (Dunn et al., 2020) make them equally important ecosystem service providers. Moreover, it has been pointed out that pollen transport and diversity in a species do not correlate with pollination effectiveness (King et al., 2013;Popic et al., 2013). Further studies are required to assess the difference in the effectiveness of hymenopteran and brachyceran potential pollinators, despite some studies on other Apiaceae species already noting a high effectiveness of Brachycera (Niemirski & Zych, 2011;Pérez-Bañón et al., 2007;Zych, 2007;Zych et al., 2014).
Brachyceran species with a high number of interactions were mainly anthropophilic syrphid species with a general preference for white or yellow umbels (Innouye et al., 2015;Speight, 2018).
Nonetheless, the high number of generalized syrphid species may be attributed to the occurrence of short-term specialized feeding bouts between individuals of the same species (Lucas et al., 2018). Eleven syrphid species found in our study are aphidophagous in their larval stage and therefore are highly suitable as biocontrol agents (Dunn et al., 2020;Moerkens et al., 2021;Nelson et al., 2012;Tenhumberg & Poehling, 1995), which implies that caraway is not only an attracting resource for pollinators but also for the adult stages of natural enemies of crop pests.
Within Hymenoptera, as expected, Apidae species and other wild bees presented the highest number of interactions, except for two sawflies species, Athalia rosae (Linnaeus, 1758) (Tenthredinidae, turnip sawfly) and Tenthredo notha (Klug, 1817), indicating the importance of sawflies as potential generalist pollinators of caraway and other Apiaceae (Lamborn & Ollerton, 2000). Athalia rosae had the highest number of links of all Hymenoptera species in the network, but the larva of this species is known as a common pest of Brassicaceae crop species (Oishi et al., 1993), particularly of oilseed rape, Brassica napus subsp. napus, one of the most common oil crops throughout Europe (Woźniak et al., 2019). In addition to detecting non-Apidae species with high numbers of interactions, we also found two species, Lasioglossum pauperatum (Halicitidae: Brullé, 1832) and Anthidium strigatum (Megachilidae; Panzer, 1805), as potential caraway pollinators that are listed in the German Red List as severely endangered and prewarning list, respectively (Haupt et al., 2009).
This shows the potential of caraway as a relevant food source for some endangered species.
F I G U R E 2 Bipartite network of potential Brachycera pollinators (right), based on data set 2 (qualitative data set based on the morphological identification and DNA metabarcoding of pollen loads). We corroborate previous studies showing significant differences in flower affinity between Hymenoptera and Brachycera (Lowe et al., 2022). Hymenopterans showed a preference for Fabaceae and Boraginaceae flowers, which are highly adapted to Hymenoptera pollination (Faegri & van der Pijl, 1979;Sedivy et al., 2013;Westerkamp, 1996;Wood et al., 2021) and have restricted access to nectar, making long-tongued bees more suited for retrieving nectar rewards (Jeiter et al., 2020). Brachycerans prefer flowers with mainly white, sometimes yellow, umbrella-like inflorescences that provide pollen and exposed nectar throughout the flowering period, in addition to a resting place (Woodcock et al., 2013). Apiaceae have generally been considered nonspecialized in pollination biology, but studies have shown an increased visitation by dipterans (Niemirski & Zych, 2011;Wojciechowicz-Żytko, 2019;Zych, 2007;Zych et al., 2014Zych et al., , 2019. The composite flowers of Asteraceae also provide a large surface area, and sometimes shelter, as well as easy access to floral resources shown to be preferentially exploited by dipterans (Branquart & Hemptinne, 2000;Morales & Köhler, 2008). Stinging nettle, Urtica dioica L., is a primarily wind-pollinated species with highly reduced and inconspicuous flowers and is generally not considered of great interest to pollinating insects. However, pollination of U. dioica by insects is known to occur (Taylor, 2009) and it is known to provide important habitat for beneficial insects, including predatory and parasitoid flies and parasitoid wasps that can also serve as pollinators (Alhmedi et al., 2007;James et al., 2015). These two factors, combined with its prevalence in the study area, could in part explain the high abundance. In addition, U. dioica was in peak or moderate pollen flight through the duration of the study (2016 Archive; Stiftung Deutscher Polleninformationsdienst; https:// www.polle nstif tung.de) and was probably ubiquitous in the air and on surfaces through the study area where insects could pick up pollen grains during contact with these surfaces. We therefore believe that the presence of U. dioica found from metabarcoding is not a false positive caused by sample or laboratory contamination, but a real occurrence in the environment.
The presence of three other wind-pollinated taxa both in the morphological and metabarcoding data (Betula spp., Picea spp. and Pinus spp.) and one insect-pollinated taxon (Salix spp.) is surprising, as pollen production occurs in early spring, well before our collection dates. The occurrence of wind-pollinated taxa is probably a result of persistence on surfaces in the environment, contact with nest provisions by nesting bees or by active collection of species which depend on anemophilous pollen (Ssymank & Gilbert, 1993   temperature, wind or cloudiness) and therefore it was anticipated that some potential pollinators of caraway will not be present over the whole day (Innouye et al., 2015;Koul et al., 1993;Willmer, 1983).
Therefore, the strong intraday differences in potential pollinator diversity make it necessary to collect insects and pollen at different times of the day to compile complete plant-pollinator networks.
By combining the results of both pollen identification methods, we were able to distinguish over four times more interactions than with the morphological identification of the slide-mounted pollen alone, and 1.3 times more interactions than with the pollen identification via DNA metabarcoding alone. While DNA metabarcoding has an overall higher species identification resolution than morphological identification, its accuracy is limited by the quality of the database (Meiklejohn et al., 2019;Michelot-Antalik et al., 2021), and by the barcode marker used (Kolter & Gemeinholzer, 2021a). Also, possible cross-contamination can produce false positives, and PCR bias has the potential to produce both false positives and negatives. In our study, F I G U R E 5 Semiquantitative bipartite network during and after the main flowering period (FP) of caraway, based on data set 1 (semiquantitative data set based on the morphological identification of pollen loads). Plant taxa on the left side, potential pollinators on the right side. The width of the link indicates the total number of samples analysed containing the respective link. Timeline on the top illustrates the number of sampling days during and after the flowering period (FP).
we benefited from a well-curated database for our region and incorporated best practice protocols of sterile techniques and a dense system of negative controls to account for contamination as well as positive controls to confirm methods. Unfortunately, there is no definitive way to account for false negatives. DNA metabarcoding is a developing method and several factors could have influenced our results. Our samples for the most part comprised very low pollen loads (<5000 pollen grains) that could be highly prone to false positives (Alberdi et al., 2018). In addition, cross-contamination in the field and laboratory processes can be mitigated but are impossible to eliminate, and some scrutiny needs to be applied to the results.
We also observed a great difference in the abundances of the species, ranging from a single specimen in 83 species up to 108 specimens in Athalia rosae. Further sampling over multiple years, with a higher number of specimens per species, would be necessary to get the full picture of the plant-pollinator network of caraway and other possible hidden links. This multiyear sampling would also account for possible interannual and spatial variations of caraway, which could influence the patterns of the plant-pollinator network of caraway.

| CON CLUS IONS
Our results highlight the unexpected complexity of the studied network and the high diversity of nonhoneybee Hymenoptera and Brachycera species involved as potential caraway pollinators.
Furthermore, we observe significant network differences over the course of the year, as well as strong qualitative differences between main potential pollinator taxa. We emphasize the importance of caraway as a food source outside the peak flowering periods of other crops and natural plants for pollinating insect communities.
Moreover, we highlight the potential of caraway as a complement to flowering strips and other biodiversity-fostering methods in agricultural areas, providing farmers with an additional source of income.
We also argue that upscaling this type of study to cover intraseasonal and intraday variation as well as all main pollinator species is crucial to obtain complete data. Implementing beneficial and evaluating beneficial as well as detrimental measures will rely on this comprehensive understanding of the plant-pollinator networks in agro-ecosystems.

AUTH O R CO NTR I B UTI O N S
ICK, SJS, XM, AH and RSP designed the study. ICK collected the samples. SJS conducted the laboratory part for pollen analysis. ICK processed and analysed the data. SJS, XM, BG, AH, JWW and RSP prepared, contributed to and approved the final manuscript.

ACK N O WLE D G E M ENTS
We Access funding enabled and organized by Projekt DEAL.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declare there are no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Sequencing data generated and analysed in this study are publicly available through the NCBI SRA (Accession nos. PRJNA935259 and PRJNA935270) for the pollen analyses via metabarcoding and in  Total number of species