We monitored flower-visiting insects in six experimental fields in Fukuoka (33.50° N, 130.57° E), Gifu (35.44° N, 136.70° E), Hiroshima (34.33° N, 132.82° E), Kagawa (34.30° N, 133.93° E), Kumamoto (32.64° N, 130.72° E), and Wakayama (34.28° N, 135.41° E) prefectures and at four farms in Fukushima (37.82° N, 140.44° E), Ibaraki (36.27° N, 140.25° E), Shimane (34.69° N, 131.87° E), and Shizuoka (34.86° N, 137.74° E) prefectures. We monitored cultivars ‘Aizu-mishirazu’ (Fukushima), ‘Fuyu’ (Hiroshima), ‘Fuyu’ and ‘Matsumotowase-Fuyu’ (Ibaraki), ‘Fuyu’ and ‘Taishuu’ (Fukuoka and Kagawa), ‘Saijo’ (Shimane), ‘Soshu’ (Gifu), ‘Taishuu’ (Kumamoto and Wakayama), and ‘Tateishi’ (Shizuoka). These cultivars were selected for monitoring of flower-visitors because enough flowers were available in the study sites. No domesticated pollinators were placed in any of the fields except for the farm in Ibaraki, where two hives of A. mellifera were placed. For detailed information about the study sites, see Table S1.
We performed pollination experiments in the experimental field in Hiroshima (Site A) and the farm in Ibaraki (Site B), which were also used for the monitoring of flower-visiting insects, and at a farm in Shimane (Site C; 35.46° N, 132.80° E). ‘Fuyu’ was used for the pollination experiments at Sites A and B and ‘Matsumotowase-Fuyu’ at Site C. Detailed information about Sites A–C is given in Table S1.
Identification of flower-visiting species
We captured 453 flower-visiting insects in the 10 orchards; each was captured in a 5-mL vial placed on the flower that it had entered. Insects were captured on different trees from those used for monitoring if the tasks were conducted simultaneously. We recorded the sex of the flowers on which insects were captured.
The captured insects were morphologically and genetically analyzed. We first identified collected insects to the species level on the basis of morphological traits, and then confirmed if necessary by COI DNA barcoding. The primer combination used for PCR amplification and sequencing of COI of the bees was BarbeeF (forward) and MtD9 (reverse) (Françoso and Arias 2013; Simon et al. 1994). That used for the others was LCO1490 (forward) and HCO2198 (reverse) (Folmer et al. 1994). The PCR products were treated with ExoSAP-IT (USB Corp., Cleveland, OH, USA) and directly sequenced in both directions. Similar sequences were searched using “All Barcode Records on BOLD” (http://www.boldsystems.org/index.php/IDS_OpenIdEngine) in BOLD (the Barcode of Life Data Systems) or the Basic Local Alignment Search Tool (BLAST; https://blast.ncbi.nlm.nih.gov/Blast.cgi) in GenBank (NCBI, National Centre for Biotechnology Information). Species were identified on the basis of highly similar best matches (> 98% identity). The sequences were deposited in the DNA Data Bank of Japan (DDBJ accession numbers LC682290–LC682345; see Table S2). For details of the procedure, see Nikkeshi et al. (2019).
Monitoring flower visitors
Visits by insects to D. kaki flowers were recorded in 2019, with the exception of the farm in Ibaraki that was monitored in 2021. The ten surveyed prefectures all ranked in the upper half of persimmon-producing prefectures in Japan (Statistics Bureau, Ministry of Internal Affairs and Communications, Japan 2021). Monitoring was performed at the peak blooming date or a few days earlier. We set several monitoring plots in a study site. Each monitoring plot measured approximately 2 m width × 2 m height, which we found to be the maximum area in which any researcher could adequately track all visitors in a plot. Tree height was approximately 3 to 4 m. We set each monitoring plot at eye level, because it was difficult to monitor the flowers on the upper branches. After the numbers of male and female flowers in a plot had been recorded, visits by insects to the flowers in the plot were recorded for 30 min. Every flower-visiting insect was observed and classified into five species that can be identified by appearance: Apis cerana japonica Radoszkowski, A. mellifera, B. ardens ardens, Bombus diversus diversus Smith, and Xylocopa appendiculata circumvolans Smith (all Hymenoptera: Apidae); the others were classified by functional group: medium-sized bees, small bees, dipterans including syrphid flies, and beetles. Consecutive visitations to a flower by a single individual were counted independently. Each monitoring plot was surveyed several times.
Annual fluctuations of flower visitation frequency and fruit set
To examine the range of annual fluctuations of fruit set, the frequency of flower visitors to D. kaki and fruit set were monitored over 4 years (2018–2021) in Hiroshima, where pollination experiments were also carried out (Site A; see Table S1). The flower visitors were monitored as above. To assess fruit set, we conducted pollination experiments on flowers, with a no pollination (NP) treatment in which female flower buds were covered with paper bags before and during anthesis and an open-pollination (OP) treatment in which flower buds were tagged and kept uncovered. Each year, the female flower buds were thinned beforehand to one per shoot. In 2021, for example, we selected five ‘Fuyu’ trees adjacent to at least one tree onto which ‘Zenjimaru’, a pollinizer cultivar, had been top-grafted, and set up six replications of each treatment on a single tree (n = 30 for each treatment). Different trees were selected for assessing fruit set each year. Treatment dates also varied among years owing to the variation in peak blooming dates (when 80% of flowers had bloomed) of ‘Fuyu’ in this field: 21 May 2018, 23 May 2019, 27 May 2020, and 22 May 2021. In 2021, for example, the NP female flower buds were covered with paper bags on 18 May, and visitor insects were monitored on 18 and 19 May. Fruit set in each pollination treatment was assessed on 29 July, when physiological fruit drop in the early development stage had finished. The numbers of female flowers used for the NP and OP treatments were 25 and 25, respectively, in 2018, 50 and 19 in 2019, 30 and 30 in 2020, and 30 and 29 (one tug was lost) in 2021. The fruit-set probability was compared between treatments and among years.
Evaluating pollination efficiency by a single visit
To evaluate pollination efficiency of major visitor species, we performed five sets of field experiments to limit the number of visits at Sites A–C from 2019 to 2021. The number of pollen grains on a stigma and the fruit-set probability were compared among the following five pollination treatments: (1) NP; (2) single visitation by A. mellifera (SV-Am); (3) single visitation by B. ardens ardens (SV-Baa); (4) OP; and (5) hand-pollination (HP). At Site C in 2021, for example, on 21 May we covered flowers with paper bags to prevent insect pollination; after confirming that the bagged buds had bloomed, we removed the bags and used these flowers for the SV-Am and SV-Baa treatments on 23, 24, and 26 May. We re-bagged the flowers immediately after we confirmed the visitation to prevent additional pollination. For the HP treatment, we used pollen collected from the male flower buds of ‘Zenjimaru’. On 24 May, male flower buds from which the sepals and petals had been removed were dried at 25 °C overnight. On the following day, we dipped a fingertip in the dried pollen in a Petri dish and attached the pollen to the stigma of a female flower. After hand-pollination, the flowers were left open. The procedures in the NP and OP treatments were as described in the previous subsection. In this set of pollination experiments, we used 30 flowers in the NP treatment, 26 in SV-Am, 40 in SV-Baa, 30 in OP, and 30 in HP. The other four sets of a sequence of pollination experiments (at Site A in 2019 and 2021, at Site B in 2021, and at Site C in 2020) were conducted as described above, but the number of treatments (3–5) depended on the set. At Site A, we omitted SV-Am owing to the extremely rare flower visitation by A. mellifera. At Site B, we omitted SV-Am, and also omitted HP owing to the limited number of female flowers available for our experiment.
The stigmas of the female flowers used for the pollination treatments were collected to count the number of pollen grains deposited; they were collected 24 h after pollination or later in the SV-Am, SV-Baa, and HP treatments, or 3 days after anthesis in the NP and OP treatments. This procedure is harmless to ovary growth, because it has been confirmed that removal of stigmas 10 h after hand-pollination does not reduce persimmon fruit set and seed formation (Kawagoe and Inoue 1958). The collected stigmas were taken to our laboratory and stored individually in tubes at − 30 °C in the freezer until measurement. Under a microscope (Eclipse E200; Nikon, Tokyo, Japan), we counted the numbers of pollen grains of D. kaki deposited on each stigma without dyeing by continuously shifting the focal distance. We compared the number of pollen grains deposited by a single visit between A. mellifera and B. ardens ardens using the data obtained at Site C in 2020 and 2021.
We then compared the fruit-set probabilities between pollination treatments using the data from the pollination experiment conducted at Site B in 2021, where the NP treatment had the lowest fruit-set probability among our five datasets. The effect of a single visit by a pollinator species on the fruit-set probability was analyzed as the difference between the NP and SV-Baa treatments.
Relationship between the number of pollen grains on a stigma and yields
To test the relationship between the number of pollen grains deposited on a stigma and the resulting fruit set and seed numbers, we used fruits that had developed from the flowers from which the stigmas were collected for counting pollen grains. Fruit set was assessed as in the previous subsection. The number of seeds was counted in each immature fruit harvested in late July every year. To analyze the relationship between the number of pollen grains on a stigma and fruit sets or the number of seeds, we used the data set of the pollination experiment conducted at Site B in 2021, where the NP treatment had the lowest fruit-set probability among our five datasets. Forty-five female flowers were used for this experiment.
To compare the fruit-set probability between the OP and NP treatments, we used a generalized linear model (GLM) in which the response variable was the fruit-set success, assumed to follow a binomial distribution (with a logit link function), and the explanatory variables were treatment and year. In addition, we estimated the fruit-set probability distribution using a Bayesian model in which we assumed that the fruit-set probability followed a Bernoulli probability distribution. To compare the number of pollen grains deposited on a stigma by a single visit between A. mellifera and B. ardens ardens at Site C in 2020 and 2021, we used a GLM in which the response variable was the number of pollen grains on a stigma and the explanatory variables were year, visitor species, and their interaction. In this model, the response variable was assumed to follow a quasi-Poisson distribution because the residual deviance was far greater than the degree of freedom when a Poisson distribution was assumed in a preliminary analysis. To compare the fruit-set probability between the OP, NP, and SV-Baa treatments, we used a Bayesian model similar to that above. Then we used a generalized linear mixed model (GLMM) to examine the relationship between the number of pollen grains on a stigma and the fruit-set success. In the model, the response variable was the fruit-set success, assumed to follow a binomial distribution (with a logit link function), and the explanatory variable was the number of pollen grains deposited on a stigma and a random effect was the trees. In the GLMM assessing the relationship between the number of pollen grains on a stigma and the number of seeds in a fruit, the response variable was the number of seeds, of which the maximum is eight, assumed to follow a binomial distribution, and the explanatory variable was the number of pollen grains on a stigma and a random effect was the trees. Data from both developed and dropped fruits (all assumed to have no seed) were used to assess the relationship between the number of pollen grains on a stigma and the “potential” number of seeds. All statistical analyses were performed in R v. 4.1.1 software (R Development Core Team 2021). For Bayesian inferences, we used Stan v. 2.21.2 software (Stan Development Team 2020). To fit a GLMM, we used “lme4” package (Bates et al. 2015).