Flower visits of cockroaches (Insecta: Blattodea) in the Iberian Peninsula: Are they neglected pollinators?

Although several insect orders have been deeply studied in plant–animal interactions (e.g. pollination) cockroaches have traditionally been ignored in taxonomic and ecological studies. However, they could be playing a role in the reproduction of several plants. To date, 8 plant species use cockroaches as a pollination agent. In our study, we have reviewed 2865 records from citizen science platforms and our own data from the Iberian Peninsula to find flower visits made by cockroaches. We have detected 51 interaction records involving at least 8 different cockroach species and 35 plant species. Furthermore, nearly half of the detected cockroaches carried pollen on various parts of their bodies. These insects were found to mainly visit white, yellow and pink flowers from Apiaceae, Asteraceae and Cistaceae plant families (among others) mainly in late spring and early summer. However, for the confirmation of effective pollination, new studies must be carried out. Additionally, although the existence of pollination syndrome is far from being understood, we provide new insights that could help shed some light on this unknown relationship. Here we provide the first approximation of cockroach floral perception and we have detected that white flowers show the best fit and higher conspicousness to cockroach colour vision, as suggested for other neglected pollinator insects.


| INTRODUC TI ON
The co-evolution over millions of years between angiosperms and arthropods has been key to the great diversification of both groups (Grimaldi, 1999;Pellmyr, 1992;Van der Kooi & Ollerton, 2020). Insects are the most abundant and important pollination agents as they perform an essential service in plant reproduction (Ashman et al., 2004;Gallai et al., 2009;Klein et al., 2007), acting as a vector for nearly 90% of angiosperms (Ollerton et al., 2011). However, not all flower visitors are pollinators. Approximately 30% of arthropod species benefit in some way from flowers, either to search for food, mate or other resources (Wardhaugh, 2015). To be considered a pollinator, the agent needs to transport pollen from the male floral organs (stamens) to the receptive area of the female ones (stigma) (Faegri & Van der Pijl, 1979). There is a wide variety of pollinator insects. Their efficiency as pollen vectors is linked to the quantity and quality of the transport (Herrera, 1987). Beetles (Coleoptera), flies (Diptera), butterflies and moths (Lepidoptera) and bees and wasps (Hymenoptera) have traditionally been the four best-studied insect pollination orders due to their efficiency and abundance. However, there are other insects that have not received as much attention as the previous ones and whose importance as pollinators has been overlooked (Wardhaugh, 2015).
In order to improve the knowledge of this «neglected relation-

| Photographs survey from citizen science platforms and own data
An exhaustive review of all photographs uploaded on citizen science platforms was carried out. We have tried to locate all records of cockroaches visiting wildflowers in the Iberian Peninsula. All available photographic material from Biodiversidad Virtual (BV; https://www.biodi versi dadvi rtual.org/) and iNaturalist (iNat; https://www.inatu ralist.org/) were reviewed in depth. All records of cockroaches touching a flower were first included. Only photographs uploaded until 31 September 2022 were considered.
These data were complemented with our own records and photographs provided by other authors specifically for this study. All cockroaches were identified attending to diagnostic characters of the Iberian fauna following the keys of Bohn, 1989, 1991, 1993and Knebelsberger & Miller, 2007 In all the flowervisit events, we have tried to include the cockroach and the plant species (or, at least, the lowest taxonomic level possible; no identifiable species were classified besides into distinguishable morphotypes), the month of the observation, the location and whether the cockroaches are carrying pollen grains.

| Spatial and temporal occurrence
We analysed the temporal occurrence of observations in order to understand the phenology of plant-crockroach interactions and elucidate whether it is a markedly seasonal phenomenon. To test for seasonality in the activity peak of plant-cockroach interactions, we used circular statistics following Morellato et al., 2010. We performed the Rayleigh test (Z) to evaluate the significance of the mean date. The degree of seasonality is then measured as the length of the mean vector (r), which ranges from 0 (no seasonality) to 1 (highest degree of seasonality). For the spatial distribution, we used the exact location, or in case the coordinates were not available, an approximation of the nearest locality. Maps were generated with ArcGis ArcMap 10.8.2.

| Plant-cockroach species interactions
We have developed a bipartite graph between cockroach genus and the family of the flowers or their colours to improve the visualization of the data. Given the difficulties of identifying the plant species in many photos, interaction networks at the species level have not been considered. We have chosen not to compute network parameters, as the sampling effort for each individual record was unknown.
These graphs were created with RawGraphs 2.0.

| Floral spectral reflectance and colour perception
The flowers were categorized into different colours based on the human perception: white, yellow and pink. We used floral reflectance data already measured by the authors for other projects, matching with some of the visited plant species (N = 6; Cistus albidus L., Cistus salviifolius L., Crepis vesicaria L., Daucus carota L., Erica australis L. and Halimium halimifolium (L.) Willk.). We measured UV-visible flower spectral reflectance using a FLAME spectro-photometer (Ocean Insight Inc., Orlando, USA), and processed them before performing further analysis using PAVO R-package (Maia et al., 2019) following methodology applied in León-Osper and Narbona (2022). While measuring floral reflectance, we recorded data for apex and base of the flower or inflorescence. This let us determine the presence of UV reflectance changes along the flower. We embraced those UV reflectance differences to test their impact on cockroach colour perception. We provided a description of these reflectance curves which contains hue and maximum brightness for the main reflectance peak and whether there was a secondary peak of reflectance, the wavelength where it occurs and its intensity as a relative measure of its brightness compared to the maximum (Table 1).
As a trait overlooked in flower-cockroach interactions studies, we have performed an analysis of floral perception by cockroaches using focal visited flowers which colour data were available. For this purpose, we need cockroach colour visual sensitivity and floral colour. Vision systems of cockroaches have not been broadly studied, lacking data for all of our sampled species. We solved this by modelling the visual system of Blattella germanica (L.) which peaks at 365 nm and 490 nm (Koehler et al., 1987;Van der Kooi et al., 2021). This is the closest species to those recorded by us (same family Blatellidae) (Van der Kooi et al., 2021). However, the three known cockroach species' visual systems are very similar, showing two peaks: one in the UV band of light spectrum (around 360 nm) and another in the blue band (around 500 nm). Thus, we expect no big differences between the Blattella germanica visual system and our sampled species. We used the visual model for dichromats developed by Kelber et al. (2003), setting an average of green leaves as background (provided in Chittka et al., 1994), 'D65' as a standard illuminant and von Kries' chromatic adaptation transformation (Kemp et al., 2015). We chose to model the visual system of cockroaches in daylight in congruence with our observations in which all registered data occurred in daytime. Some authors suggested that nocturnal invertebrates have developed neural processes to compensate for varying light conditions such as the dim light (i.e. colour constancy; Hurlbert, 2007;Warrant & Somanathan, 2022) and likely rely on achromatic cues (i.e. brightness cues) to detect flowers in these conditions (Van der Kooi & Kelber, 2022). However, we were not able to calculate achromatic contrast as the visual system of Blattodea is far from well known and there is no known method to calculate it. We calculated the 'chromatic contrast' as the Euclidean distance from a given stimulus to the achromatic centre (Menzel & Backhaus, 1991) using the 'coldist' function from PAVO R-package (Maia et al., 2019). This parameter is important because most pollinators rely on chromatic signals to detect a target when foraging Ilse, 1949;Koshitaka et al., 2008;Martínez-Harms et al., 2012). For the honey bee and common drone fly there are known chromatic contrast values that let us determine whether a stimulus is conspicuous (Dyer et al., 2012;García et al., 2021). For cockroaches, discriminability thresholds are not available. To check whether the colour of visited species is adapted to cockroaches' colour perception, we calculated the overlapping of flower reflectance spectra marker points with their wavelength of best colour perception (Chittka & Menzel, 1992;Dyer et al., 2012). Following Chittka and Menzel (1992), visual systems best discriminate colour in the wavelengths where two photoreceptors' sensitivity overlap. For dichromat cockroaches, we can calculate this wavelength based on the visual sensitivity modelled with R-package PAVO. We have located the sensitivity peak of the overlapping region at 408 nm. Thus, we used that region of the light spectrum for marker points analyses. We used the online tool 'Spectral-MP' (Dorin et al., 2020) to calculate marker points as any change of >10% reflectance in <50 nm range (Shrestha et al., 2016). Then we calculated the mean absolute deviation (MAD) and the minimal absolute deviation (minAD) (Shrestha et al., 2013).
MAD measures the average proximity of flower marker points to each wavelength of maximum discrimination of a given vision system, while minAD measures the minimal distance between a marker point and a given wavelength of maximum discrimination (Shrestha et al., 2013); in both measurements, lower values indicate better colour discrimination of the insect visual system. Because cockroaches are dichromats, they only present one wavelength of best colour perception. Thus, MAD and minAD values are equal when the flower reflectance spectrum shows just one marker point. Chromatic contrast, MAD and minAD values that did not meet normality and/or homogeneity were corrected using boxcox transformation with the 'box-cox' function in R (Crawley, 2007). To test for differences between visited flowers' colour categories, we performed ANOVA analyses followed by Tukey post hoc test. For analyses between colour categories, we merged flowers that look the same in the human visible spectrum, meaning we did not account for UV reflectance differences mainly due to the low number of samples. TA B L E 1 Mean absolute deviance (MAD), minimal absolute deviance (minAD 408 nm ) and descripting variables of the reflectance curves of sampled species.

| Photographs survey from citizen science platforms and own data
We

| Spatial and temporal occurrence
In general, most of the flower-visiting observations occurred in June (37.7%) followed by May (17%) and July (13.2%) (Figure 2a).

| Plant-cockroach species interactions
We found at least 8 different cockroach species (  and 450 nm with intensity ranging from 0.54 to 0.78 ( Table 1). Out of the identified species (N = 6), three of them showed a secondary colour within their corolla, whether it is a visible colour (e.g. Cistus salviifolius and
There were no significant differences between the three main colour categories of visited flowers (F = 0.786; p = 0.505). Mean chromatic contrast for white flowers was 0.54 ± 0.08 EU, for yellow ones was 0.37 ± 0.13 EU and 0.25 ± 0.17 EU for pink ones (Figure 4). depend more on the presence of floral resources for feeding in dry Mediterranean ecosystems than tropical ones (Schapheer et al., 2017). In relation to phenology, it is important to note that this study is based mainly on citizen science photographs that could be biased to the season when observers and cockroaches are more active. However, our results show that cockroach-flower interactions peak in June. The majority of plant communities in the Iberian peninsula present a flowering peak in May, although it is common that such a peak extends into June (i.e. Arroyo, 1988;Rodríguez-Gallego & Navarro, 2015).  -McKey et al., 2010;Raguso, 2008), energy resources such as nectar (Pyke, 2016) or the development of important floral visual signs to be easily detected (Dyer et al., 2012;García et al., 2021;León-Osper & Narbona, 2022).
In this context, flower perception by cockroaches has not been deeply studied before. Despite not finding significant differences in chromatic contrast between the colours of visited flowers, the values range of white flowers is more consistent than the remaining colours. White flowers commonly lack UV reflectance compared to other colours but reflect light throughout the whole visible light spectrum (Dyer, 1996;Kevan et al., 1996). Thus, these flowers are easily detectable during nighttime when the illumination is longwavelength-shifted (Johnsen et al., 2006). In fact, white and pale colours are often related to bat and moth pollination, which are nocturnal pollinators like cockroaches (Domingos-Melo et al., 2021;Faegri & Van der Pijl, 1979;Fenster et al., 2004;Johnsen et al., 2006;Kelber et al., 2003). It is important to note that colour perception usually works as a 'colour opponent' mechanism (Song & Lee, 2018) meaning that equally excited photoreceptors would counteract We also want to highlight the importance of the citizen science platforms. Despite the fact that they must be carefully analysed due to its biases (e.g. phenology or distribution), they offer very valuable information on the ecology of many organisms, some of them little studied by scientists. For this reason, we encourage society to continue using these platforms, so in the future, we can continue extracting valuable information from them.

| CON CLUS IONS
We