Experimental aphid species were either sourced from the wild (Uroleucon spp. and cabbage aphids, Brevicoryne brassicae; Suppl. material 3) or from laboratory-reared colonies (corn leaf aphids, Rhopalosiphum maidis; Russian wheat aphids, Diuraphis noxia; and bird cherry-oat aphids, Rhopalosiphum padi; Suppl. material 3). Colonies of R. maidis and D. noxia were maintained on caged barley (Hordeum vulgare) plants grown from seeds in the laboratory at 28 °C under a 12:12 light:dark cycle for four weeks prior to feeding experiments. Uroleucon spp. and B. brassicae were collected from sow thistle (Sonchus sp.) and field mustard (Brassica rapa) plants, respectively, on the La Trobe University Bundoora campus, Victoria, Australia, an hour prior to feeding experiments. As positive controls for the metabarcoding experiments, we used a subset of aphids from each of the sources listed above, as well as California maple aphids (Periphyllus californiensis) collected from a Japanese maple tree (Acer palmatum) in the Horsham Botanic Gardens, Victoria, Australia. We also collected wild psyllids from an Acacia tree on the same campus to be used as non-target ‘feeder’ insects. For feeding experiments, we used spotted amber ladybeetles (Hippodamia variegata), a known predator in Australian grain ecosystems (Martoni et al. 2023a), that can be commercially sourced as eggs and adults (Biological Services, Loxton, Australia). We also collected large spotted ladybeetle (Harmonia conformis) larvae and adults from garden plants on the campus of La Trobe University, Bundoora, Victoria, Australia (37°43'18.8"S, 145°03'03.3"E) for additional comparisons.

The first feeding experiment simulated ladybeetles gorging during a period of high pest abundance, followed by a short starvation period (17 hours) and then exposure to low abundances of additional prey (Fig. 1A). Here, we used ladybeetles (H. variegata) that were housed in mesh cages containing barley plants and were provided with ad libitum R. maidis (chosen because it was our most abundant aphid colony) as prey for nine days, and then starved for 17 hours prior to the experiment. Ladybeetles were stored individually in sterile petri dishes, and five aphids (either all Uroleucon spp. or D. noxia – we used different species to allow measurements of detectability after a specific feeding event, as R. maidis was likely still present in the ladybeetles guts) were introduced to each dish. Ladybeetles were given 1.5 hours to feed, and we recorded the number of aphids consumed, after which the remaining aphids were removed from the petri dish. We then collected ladybeetles into sterile tubes containing 100% ethanol at the following time points after feeding: 0 hr (i.e. immediately after the feeding period concluded), 1 hr, 2 hr, 4 hr, and 72 hr. We were unable to include timepoints between 4 and 72 hours due to logistical constraints. To provide controls for sequencing, a subset of ladybeetles (which had fed on R. maidis in the mesh cage, but were not fed on Uroleucon spp. or D. noxia) and samples from each aphid species (not exposed to ladybeetles) were selected at random, at the same time that feeding trials began, into 100% ethanol (in separate sterile vials).

The second feeding experiment simulated the start or end of the season, with pest prey present in low abundance, to test the effects of a longer starvation period (96 hours) and to reduce the effect of dominant prey on DNA detectability (Fig. 1B). Here, we transferred all ~70 adult ladybeetles (H. variegata) from our colony evenly across two aphid-free barley cages (30 cm × 30 cm × 70 cm), where they were left for 96 hours without food prior to the feeding experiment. As in the Short starvation experiment, each ladybeetle was transferred to sterile petri dishes and provided with five aphids (either R. maidis or B. brassicae – we used these species as they were the most readily available at the time of the experiment). The ladybeetles were left for 1.5 hours to feed. After this feeding period, we recorded the number of aphids consumed, removed remaining aphids, and then transferred ladybeetles to 100% ethanol (in separate sterile vials) at one the following time points: 0 hr, 1 hr, 2 hr, 4 hr, 24 hr, 48 hr, and 96 hr. Different time points were used compared with the Short starvation experiment again due to logistical constraints. To maintain biologically realistic metabolic rates, ladybeetles in the 48 hr and 96 hr collection groups were fed two psyllids after 24 hr, and, for the 96 hr treatment, an additional two psyllids were provided after 72 hr (Weber and Lundgren 2009). As above, a subset of ladybeetles, aphids, and psyllids were transferred, from colonies or wild collections as applicable, into 100% ethanol (in separate sterile vials) at the beginning of the feeding period as controls for sequencing.

To complement our experimental feeding trials, we sampled two ladybeetle species (H. conformis and H. variegata) collected directly from the field, and we analysed frass collected from ladybeetles during the Short starvation experiment (Fig. 1C). As a pilot trial, we collected frass (transferred into ATL buffer; see below) from three H. variegata larvae (from the barley cage colony) after feeding, and from two H. conformis (field-collected) and one H. variegata (lab colony) adults before and after feeding them field-caught aphids overnight. To minimise the amount of frass derived from food consumed in the pre-feeding period (Davidson and Evans 2011), larvae were moved to a new petri dish after feeding, where they remained for 24 hours, at which point the larvae and any frass were transferred to 100% ethanol in separate vials.

We performed destructive DNA extractions on most specimens, and non-destructive DNA extractions on some specimens, using the DNeasy Blood and Tissue kit (Qiagen, Germany) with an overnight incubation period (~17 hours) at 56 °C as previously described (Martoni et al. 2021). Initial pilots of non-destructive extractions from beetles produced inconsistent PCR bands, so destructive extractions were primarily used for our experiments. For destructive extractions, we dissected the ladybeetle into three parts (head and thorax together, and abdomen in two halves) and immediately transferred all three together into ATL buffer. These dissections took place in a biosafety cabinet with a new scalpel blade and new pair of gloves for each dissection and also included DNA extraction blank controls for each batch of samples processed. For comparison, we performed additional non-destructive extractions, without the dissection step (i.e. extracted from whole specimens), for some ladybeetles, as well as all frass, aphid, and psyllid specimens. For both extraction methods, we submerged specimens in 180 μL of ATL buffer and 20 μL of proteinase K at 56 °C overnight.

We purified the 200 μL lysate per sample on DNeasy spin columns, following the manufacturer’s instructions. After DNA extraction, specimens processed non-destructively were preserved in high grade ethanol (100%) for further morphological examination, if required. We used a two-step PCR protocol, based on previous work (Martoni et al. 2023a, 2023b), to amplify target DNA and index sequences for Illumina sequencing. First, polymerase chain reactions (PCRs) were performed, targeting a ~434 bp fragment of the standard barcode region (Hebert et al. 2003) within the 5’ portion of subunit I of the cytochrome c oxidase (COI) mitochondrial gene. Custom primers were designed to match sequences in aphids (target species: R. maidis, R. padi, D. noxia, Myzus persicae) but not ladybeetles (non-target species: H. variegata, H. conformis, Coccinella transversalis), to ensure target specificity. Partial Illumina adapter sequences (in bold italic) were added as primer tails at the 5’ end of the aphid-specific primers, Aphid_F: 5’-ACACTCTTTCCCTACACGACGCTCTTCCGATCTTAAGACAAATTAATTCAATTATYAA-3’; Aphid_R: 5’ - GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTGGGAAAAGTGGAATTTGRTT-3’. The PCR was conducted in 25 μL reactions consisting of 14.7 μL Bovine Serum Albumin (New England Biolabs, USA), 5 μL of 5× Bioline MyFi reaction buffer (Meridian Bioscience, USA), 1 μL of each primer (10 μM), 0.8 μL MyFi DNA polymerase (Meridian Bioscience, USA) and 2.5 μL of template DNA. The PCR cycling conditions were as follows: 2 minutes at 94 °C; 30 cycles of denaturation at 94 °C for 30 seconds, annealing at 49 °C for 45 seconds, and extension at 72 °C for 45 seconds; and then a final extension phase of 1 minute at 72 °C. Amplification was verified by gel electrophoresis (1% w/v agarose). After the initial PCR amplification, the amplicons were diluted 1/10 and run through a second 6 cycles of real-time PCR to attach unique dual indexes and the remainder of the Illumina adapter sequence. Each indexing real-time PCR reaction (50 μL volume) contained 32.5 μL of ddH₂O, 10 μL of 5× Phusion HF Buffer (Thermo Fisher Scientific, USA), 1 μL of dNTP mix (10 mM), 1 μL of SYBR Green I Mix (Thermo Fisher Scientific, USA) diluted 1/1000 in ddH₂O, 0.5 μL Phusion DNA polymerase (Thermo Fisher Scientific, USA), 4 μL of sample-specific indexing primers (2.5 μM) and 1 μL of the diluted PCR product. The real-time PCR cycling conditions were: 98 °C for 30 seconds, followed by 6 cycles of 98 °C for 10 seconds, 65 °C for 30 s, and 72 °C for 30 seconds, with fluorescence measurement conducted in the 72 °C phase. The amplification curve was visually inspected in real time and stopped while still in the exponential phase, during the extension phase of the 6^th cycle, to prevent over-amplification artefacts. Amplicons were purified and normalised using the SequalPrep Normalization Plate Kit (Thermo Fisher Scientific, USA) following the manufacturer’s protocol, but eluting the final product in 12 μL instead of 20 μL. Normalised and cleaned real-time PCR amplicons were then pooled together and the resulting library was quality checked, sized and quantified using a High Sensitivity D1000 ScreenTape assay performed on a 2200 TapeStation (Agilent Technologies, USA). The final pooled library was diluted to a concentration of 7 pM, spiked with 15% PhiX, and sequenced using V3 chemistry (2 × 250 bp reads) across two flow cells on an Illumina MiSeq system (Illumina, USA). Raw sequence reads were demultiplexed using bcl2fastq, allowing for a single mismatch in the indexes, and the FASTQ files were archived at NCBI SRA BioProject PRJNA1267727.

Analysis of FASTQ files was performed using the Freyr metabarcoding analysis pipeline (commit 9d4f34a) available here: https://github.com/AVR-biosecurity-bioinformatics/freyr. Freyr was run with the parameters file “loci_params.csv” (Suppl. material 4), along with the command-line parameters “--primer_n_trim --miseq_internal”. Briefly, reads per sample were grouped by their primer sequence (allowing a single mismatch) and then trimmed of primers, using Cutadapt (Martin 2011). Reads were then filtered using DADA2 (Callahan et al. 2016), retaining reads with <=1 expected error, a minimum length of 20 bp, and no ambiguous bases. Reads were then truncated to 225 bp and denoised using DADA2 in a two-step process, using sequences from the first round as priors for the second round. Paired-end reads were merged and the resulting ASVs were filtered with a minimum length of 220 bp and a maximum length of 390 bp. Taxonomy was assigned to ASVs using IDTAXA (Murali et al. 2018) with the aphid COI reference database described below used as reference, with a confidence threshold of 60%.

The reference database of aphid COI sequences (Suppl. material 5) was generated using the Mimir DNA barcoding reference database curation pipeline (commit e2ad316) available here: https://github.com/AVR-biosecurity-bioinformatics/mimir. Mimir was run with the command-line parameters “--target_taxon 33385 --target_rank superfamily --marker COI --min_length 200 --genbank_fetch_size 5000 --add_root --remove_unclassified any_ranks --placeholder_as_unclassified --max_group_size 3 --train_idtaxa --idtaxa_max_iterations 1”, and used the April 4^th 2025 data package from the BOLD database (Ratnasingham et al. 2024). Briefly, all AphidoideaCOI sequences and whole mitochondrial genomes were downloaded from GenBank and BOLD and subset to the COI coding region using a profile hidden Markov model from InterPro (Blum et al. 2025). These sequences were filtered to remove those with ambiguous bases and any unclassified taxonomic ranks, then clustered at a threshold of 97% and pruned of sequences that taxonomically disagreed with >=80% of their cluster at the genus level. Remaining sequences were grouped by species, identical sequences removed (leaving one per species), remaining sequences were aligned and those with genetic distances > 5% to the median sequence within each species were removed. Up to 3 sequences were retained per species, and an IDTAXA model was trained using a single iteration.

To further analyse the metabarcoding data in R, an index switching “noise floor” per flow cell was calculated from the estimated rate of index switching multiplied by the total number of reads retained in the Freyr output; ASVs with fewer reads than the noise floor had their read count set to 0 for subsequent analyses, on a per-sample basis. ASVs were removed if they were not assigned to the family Aphididae, with the exception of a number of 226 bp ASVs that had high sequence identity to COI from Uroleucon spp. but had stop codons and frameshifts. As these appeared in samples where Uroleucon spp. were expected to appear, we interpreted these sequences as nuclear mitochondrial (NUMT) COI pseudogenes from Uroleucon spp. and merged their read counts with all other Uroleucon sequences. ASVs with the same species assignment were merged into single species, with their read counts summed. Raw output and “cleaned” data files are available as Suppl. materials 6, 7, respectively.

Binomial regression modelling of aphid DNA detection predicted by sampling time in hours was performed using the glm() function in R (“logit” link function), with the predictor log₂(x+1) transformed, and the number of aphids consumed by each beetle (0–5) used as a sample weight. For each sample, “detection” was defined as a non-zero number of reads assigned to the fed aphid species. The 50% effective time (ET₅₀) -- time taken after feeding to reduce detection probability to 50% -- for each fed aphid experiment was determined from the regression model using the “dose.p()” function in the MASS package and back-transforming to the unit of hours.

Our negative controls (beetles that were not provided with aphids during the feeding experiments) showed no reads assigned to any aphid species, except for R. maidis in the Short starvation experiment, which is the prey species the beetles were maintained on prior to the feeding experiment. Our positive controls (each aphid species alone) showed only reads for their own species, while psyllids (fed to some beetles in feeding experiments) produced no reads assigned to aphids. Finally, our technical controls (extraction blanks), showed no reads for aphids, apart from slight contamination in one extraction blank (Suppl. materials 1, 2).

In the Short starvation experiment, R. maidis (consumed ad libitum prior to the feeding period) and D. noxia (consumed during the feeding period) aphids remained detectable at all measured time points up to and including 72 hours post-feeding (except for a sample at 4 hr where no D. noxia aphids were consumed). Overall, the number and proportion of reads attributable to Uroleucon spp. (consumed during the feeding period) was generally lower compared to either R. maidis or D. noxia. Specifically, Uroleucon spp. DNA was broadly detectable up to 4 hr post-feeding, but detectability was reduced at 72 hr post-feeding, with only one of the six replicates producing Uroleucon spp. reads (Fig. 2).

In the Long starvation experiment, R. maidis and B. brassicae (consumed during feeding period) were consistently detected at 0 hr and 1 hr post-feeding, but detectability decreased in the 2 hr and 4 hr samples for both species. After 24 hr, aphid DNA was largely undetected, apart from a single replicate at 48 hr for R. maidis (Fig. 3). We detected R. maidis between 0 hr and 4 hr in beetles that were given B. brassicae, although at lower reads compared to the beetles fed on R. maidis. Although initial pilots for non-destructive extractions gave inconsistent PCR results (above), the non-destructively extracted samples in the Long starvation experiment showed qualitatively similar metabarcoding results to the destructively extracted samples (Suppl. material 2).

The presence/absence of DNA from aphids fed during the feeding periods in the Short starvation experiment and Long starvation experiment, predicted by time since feeding (transformed as log₂(hr + 1)), was modelled using logistic regression to produce “50% effective time” (ET₅₀) estimates, or the time after which the detection probability was 50% (Fig. 4). For D. noxia in the Short starvation experiment, no model could be fitted as detection occurred in nearly all samples (Fig. 4A), but significant models were produced for Uroleucon spp. in the Short starvation experiment (p = 5.7 × 10^-5; Fig. 4B), B. brassicae in the Long starvation experiment (p = 0.029; Fig. 4C), and R. maidis in the Long starvation experiment (p = 5.6 × 10^-9; Fig. 4D). Back-transformed ET₅₀ values were 14.0 hours for D. noxia in the Short starvation experiment and 4.36 and 4.45 hours for B. brassicae and R. maidis in the Long starvation experiment, respectively, indicating a consistent decrease in the detectability window for aphid DNA from beetles starved for 96 hr compared with those starved for 17 hr.

Our preliminary data show that target aphids are also detectable in ladybeetle frass collected after, but not before, a feeding event (Suppl. material 1). For example, ladybeetle Hc1 was fed a Uroleucon spp. aphid, and this aphid was detected in frass produced after (Hc1P2), but not before (Hc1P1), the feeding event. Similarly, ladybeetle Hc5 showed a reading for the aphid Macrosiphum corydalis after, but not before, feeding. Ladybeetle Hv1, which was provided Uroleucon spp. during the feeding event, showed the same pattern, but also showed readings for other aphid species – B. brassicae, M. corydalis, Macrosiphum euphorbiae, and Macrosiphum melampyri (none of which were provided during the feeding event) – in frass from before feeding, and these other aphid species were not detected in frass collected on the following day after the feeding event.

Specimens of H. conformis collected from wild populations produced reads for multiple aphid species. All field specimens showed reads assigned to R. maidis, and field-collected larvae also showed reads assigned to R. padi, Hyperomyzus carduellinus, and Uroleucon spp. (Suppl. material 1).