Sexual size dimorphism (SSD) arises when natural selection and sexual selection act differently on males and females. Male-biased SSD is rare in insects and usually indicates strong selection pressure on a population or species. Patterns of SSD can also vary between populations of species that are exposed to different environmental conditions, such as differing resource availability and diversity. Here, we investigate intra-specific variation in SSD as well as relative investment in precopulatory (horn length) and postcopulatory traits (sperm length and testes weight) in a tropical rainforest dung beetle Onthophagus babirussa across Singapore and Peninsular Malaysia. Overall, three out of four populations displayed significant male-biased SSD and SSD was greater in populations with smaller overall body size. Average male body size was similar across all populations while female body size was significantly smaller in Singapore, suggesting that the pronounced SSD may also be due to stronger sexual selection on male body size in Singapore populations. All populations showed significant investment in horns as a weapon used in precopulatory competition, while postcopulatory traits showed no clear scaling relationship with body size, suggesting a higher priority on precopulatory sexual traits in the mating system of this species.

2.1.1 Study Sites

Dung beetles examined in this study were sampled in SG and MY. Figure 1 depicts a map of all sampling sites, and a full list of coordinates are appended in a supplement to the main manuscript (Appendix 1: Table S1). Sampling in SG spanned over fifteen months from August 2018 to February 2018 and May 2019 to December 2019. Specimens from MY were sampled across four sites, Perak, Gombak, Kenyir and Langkawi, from August to November 2016, as well as in July 2019. Beetles from sites A, B, C and D were pooled because these sites were part of a continuous stretch of forest in the central region of SG. Pulau Ubin is an island separate from mainland SG and was treated as a population on its own. Sites G, H and I were pooled as they were all mainland MY sites with connected forests. Langkawi is an island separate from mainland MY and was also treated as its own population. Thus, for all analyses, specimens were separated into four populations – Central Catchment Nature Reserve on mainland SG (Central Catchment SG), Pulau Ubin (Pulau Ubin Island SG), central MY (Central Peninsular MY), and Langkawi (Langkawi Island MY). Literature search was conducted to compile a checklist of non-volant mammal species present in each of the four study sites with species body size and consumer type, and these are presented and summarised in the appendix (Appendix 1: Tables S2 and S3).

 2.1.2 Sampling and sorting protocol

Dung beetle sampling was conducted using baited pitfall traps and baited funnel pitfall traps with human dung as the bait because it is widely accepted to be the best bait to attract a wide variety dung beetles (Howden & Nealis, 1975; Kudavidanage et al., 2012; Larsen & Forsyth, 2005). Exact details of trap materials and construction are appended (Appendix 1, Figure S1). Traps were retrieved after 24 to 48 hours, and captured beetles were brought back to the laboratory for morphological identification and sorting using an Olympus SZX10 microscope.

Onthophagus babirussa were separated from other species via sorting by morphological characters (see Appendix 2, Figures S2 & S3) and DNA barcoding. Specimens used for DNA barcoding were killed and preserved in 70% molecular grade ethanol. DNA was extracted from 739 specimens from Singapore populations (CCNR=129 and Pulau Ubin=167) and Malaysian populations (Central Peninsular MY=109 and Langkawi=334). For these specimens, the right mid femur was dissected into 7 μl of QuickExtract solution and the DNA was extracted by following the manufacturer’s protocol (Lucigen, 2018). Then, 313bp fragments of the COI gene were amplified via PCR (see Appendix 2 for detailed protocol), sent for Next Generation sequencing (NGS) and used for DNA barcoding. Sequence analysis was then conducted with reference to the analysis pipeline detailed by (Meier et al., 2016), and a well-established 3% threshold for uncorrected pairwise distances was used to delimit different species (Hebert et al., 2003; Meiklejohn, Wallman, & Dowton, 2011; Srivathsan & Meier, 2012). All specimens examined in this study fell within the same molecular cluster under this 3% threshold, and a cluster fusion diagram with representatives from each population is appended in Appendix 2 (Figure S4), along with the full protocol for morphological and molecular sorting. The molecular barcodes were congruent with our morphological sorting and general consensus with the geographical sampling.

2.3. Documenting reproductive trait variation

2.3.1 Precopulatory trait measurements

To investigate the sexual size dimorphism in the four populations of O. babirussa, maximum pronotum width (Figure 2) of males and females were measured as a proxy for body size with the eyepiece reticle on the Olympus SZX10 microscope. This is widely used proxy for body size because the pronotum width does not change in adulthood and has been found to be the most appropriate measure for body size in dung beetles (Emlen, 1997; Knapp & Knappová, 2013). 

Horn lengths of male O. babirussa (Figure 2) were measured to document variation in this precopulatory trait. Images were taken of the anterior habitus. Heads of the beetles were separated and suspended with Durex KY Jelly, with horns aligned parallel to the lens of the camera. Images were captured using the EOS 800D and 6D camera body with the Canon MP-E 65mm f/2.8 1-5x lens at 5x optical zoom. The camera was suspended on the Dun, Inc. P-51 and the Camlift controller V2.9.3.0 software was used to take multiple images at different heights for focus stacking.  EOS Utility Launcher software was used to access the images and stack them using the Zerene Stacker V. 1.04. software. Stacked images were imported to Adobe Photoshop CS5 V. 12.0 x64 and a 1mm scale bar was added to each image. Next, processed images were imported to ImageJ V. 1.51 and the horns were measured from the tip to the bottom of the outer edge of each horn, following previous studies (Moczek & Emlen, 1999).

2.3.2 Postcopulatory trait measurements of male specimens

Abdomens of male O. babirussa specimens were dissected into 1x phosphate-buffered solution (PBS) to measure the following postcopulatory traits: testes weight and sperm length (Figure 2). Testes were isolated and transferred onto pre-weighed aluminium sheets and dried in a Memmert Gravity Basic Digital Oven D overnight. Then, total weight was measured on the Mettler Toledo ML104 Newclassic ml Analytical Balance. Weight of the testes were calculated by subtracting the weight of the aluminium sheet from the total weight. 

To measure the sperm length, seminal vesicles containing the mature sperm were first isolated and transferred onto a drop of PBS on a frosted slide. Then, sperms were teased out from the vesicles using an insect pin. Slides were dried in the oven and sperms were fixed onto the slides with a solution of three parts methanol and one part acetic acid for two minutes. Next, the slides were washed in 1x PBS for one minute and the sperms were stained for five minutes in the dark with 4′,6-diamidino-2-phenylindolev (DAPI), which binds to DNA to form a fluorescent complex to allow for visualisation of sperm heads under a fluorescent microscope. Following that, the slides were washed in 1x PBS and placed in the dark to dry. When the slides were dried completely, one to two drops of glycerol were added on the stained regions, coverslips were placed, and the edges were sealed with clear nail polish and left to dry in the dark. The sperms were visualized using an Olympus BX50 fluorescence microscope and measured using μManager and ImageJ V. 1.51 software. Based on previous studies, five to ten sperms were measured per specimen (García-González & Simmons, 2007; Simmons & Kotiaho, 2002; Werner & Simmons, 2011).

2.4. Statistical analyses

Box plots of average pronotum width were constructed with confidence intervals using the R packages ggplot2 (Wickham, 2016), dplyr (Wickham et al., 2020) and plotrix (Lemon, 2006) and tested for significance in body size difference between the sexes within each population using ANOVA, checking the residuals for normality after. To test if SSD varied between populations, we ran linear models testing for significant sex by location interaction. Post-hoc analyses using Dunn test were also conducted to determine which populations differ from the other for male and female body size.  In addition, the sexual dimorphism index (SDI) was calculated for each population following the formulation by Gibbons and Lovich (1990), where the mean size of the larger sex is divided by the mean size of the smaller sex. A negative sign is arbitrarily added to the SDI as the males are larger (Gibbons and Lovich, 1990). 

To determine whether populations differed with respect to relative investments in precopulatory and postcopulatory traits, the static allometries were calculated by first constructing log-log scatterplots of trait size against pronotum width. As the log-log scatter plot of horn length against pronotum width displayed a clear non-linear relationship, we followed the recommendations by Knell (2009) and Parrett et al. (2021) and fitted (1) linear model, (2) quadratic model, (3) cubic model and (4) breakpoint model using the R package segmented (Muggeo, 2008) to the pooled data with all four populations to characterise the trait size-body size relationship Figure 3). Model selection was then conducted with the Akaike information criterion (AIC). The breakpoint model had the lowest AIC score for horn length (Table 1), indicating that this model is the best model for explaining the relationship between the variables (Knell, 2009). Following this, allometries were also calculated for the overall data separated by 1) population, and 2) minor or major morphs as determined by the breakpoint models applied to each population (see Appendix 3, Figure S5).

Microsoft Excel, R or RStudio