Comparing the pigeon food compositions based on the number of reads and presence data indicated significant differences between Chichijima and Hahajima (Figs. 5, 6A). Although Morus australis and Ficus dominated both of the islands, the results may indicate different patterns of food selection by the pigeons on each of the islands. The native Lauraceae species was observed only in Chichijima at a high frequency. This finding may be explained by its significant fruiting abundance in the forest of Chichijima (Toyoda 2003), indicating its importance as a major native food resources on the island. In Hahajima, the read percentage of introduced Morus australis was much larger than that of Chichijima. These data may indicate the greater usage of this introduced species by the pigeons in Hahajima than those of Chichijima. Leucaena leucocephala, which exhibited the third highest frequency in terms of both read numbers and presence in Hahajima, is also an expanding introduced species in the Ogasawara Islands (Hata et al. 2010). This plant lives on the forest edge (Toyoda 2003), which is not the preferred forest habitat of the pigeon. Pigeons on Hahajima may eat Leucaena leucocephala to make up for a lack of food resources in the forest area. It is also interesting that the food composition varied among Hahajima samples (Fig. 6), despite the small sample size and restricted sampling site (five samples were collected in February around the same roost). The red-headed wood pigeons in Hahajima are more generalist than those of Chichijima because of a lack of food resources. Estimations of food resource availability and temporal diet variation on each island may reveal differences in pigeon feeding strategies between these locations.
We also compared data from the number of reads and the relative frequency of presence for each plant taxa. The reliability of quantitative diet analysis using the sequence count data has been discussed in several previous studies (e.g., Deagle et al. 2009, 2010; Soininen et al. 2009; Valentini et al. 2009a; Raye′ et al. 2011; Pompanon et al. 2012; Shehzad et al. 2012). The PCR efficiency and resulting number of sequences from each food item can be affected by several biological (e.g., number of DNA copies in a unit mass of tissues, digestion process) and technical factors (e.g., different sequence lengths among species, mismatches in primer-binding sites; Pompanon et al. 2012). Indeed, one of the shortest plant sequences dominates the dietary analysis of the alpine chamois Rupicapra rupicapra in Raye′ et al. (2011). Thus, Pompanon et al. (2012) recommended using the read count data for measuring spatial or temporal variations in the diet but not for absolute diet quantification. In the case of the present study, comparing the proportion of each plant read number between the two islands can be reliable (e.g., sequence proportion of Morus australis). The sequence lengths of the Gr. Lauraceae1 (87 bp, only found in Chichijima), Morus australis (89 bp), and Gr. Ficus1 (89 bp), which exhibited high-frequency read numbers, were not much shorter than those of the other detected food plants (ranging 72-121 bp, average 89.9 bp), and their frequencies of presence were also high. Thus, pigeons may have actually eaten these plants in large amounts, as reported by Deagle et al. (2010) regarding the accuracy of quantitative diet data from the number of reads. However, the pigeon consumption of some plants with low PCR efficiency may be underestimated in the analysis based on the number of reads. For example, the low sequence proportion of Gr. Poaceae2 in Chichijima (4%) may be explained by its low PCR efficiency, despite its high frequency (75%) (Fig. 5A and C). No significant difference between Chichijima and Hahajima in the NMDS analysis was based on the number of reads, which may reflect the dominance of specific plant sequences (e.g., Morus australis and Gr. Ficus1) throughout the samples and the underestimation of some plants with low PCR efficiency. Another possibility is that the small number of reads for some plants may indicate secondary predation (Sheppard et al. 2005), such as plants eaten by snails or other invertebrate herbivores eaten by pigeons, or that the number of reads reflects the biomass to some extent (Yoccoz et al. 2012). This is an important issue to be considered in estimating pigeon food preferences. Given that appropriate methods to correct various biases in quantitative analysis have never been presented, it is advisable to use not only the number of reads but also the frequency of presence to estimate dietary composition and diversity, as in Soininen et al. (2009), Raye′ et al. (2011), and Shehzad et al. (2012).