Poison frogs bioaccumulate alkaloids for chemical defense from their arthropod diet. Although many alkaloids are accumulated without modification, some poison frog species can metabolize pumiliotoxin (PTX 251D) into the more potent allopumiliotoxin (aPTX 267A). Despite extensive research characterizing the chemical arsenal of poison frogs, the physiological mechanisms involved in the sequestration and metabolism of individual alkaloids remain unclear. We first performed a feeding experiment with the Dyeing poison frog (Dendrobates tinctorius) to ask if this species can metabolize PTX 251D into aPTX 267A and what gene expression changes are associated with PTX 251D exposure in the intestines, liver, and skin. We found that D. tinctorius can metabolize PTX 251D into aPTX 267A, and that PTX 251D exposure changed the expression level of genes involved in immune system function and small molecule metabolism and transport. To better understand the functional significance of these changes in gene expression, we then conducted a series of high-throughput screens to determine the molecular targets of PTX 251D and identify potential proteins responsible for metabolism of PTX 251D into aPTX 267A. Although screens of PTX 251D binding human voltage-gated ion channels and G-protein coupled receptors were inconclusive, we identified human CYP2D6 as a rapid metabolizer of PTX 251D in a cytochrome P450 screen. Furthermore, a CYP2D6-like gene had increased expression in the intestines of animals fed PTX, suggesting this protein may be involved in PTX metabolism. These results show that individual alkaloids can modify gene expression across tissues, including genes involved in alkaloid metabolism. More broadly, this work suggests that specific alkaloid classes in wild diets may induce physiological changes for targeted accumulation and metabolism.

2.1 Alkaloid feeding

Lab-reared (non-toxic) Dendrobates tinctorius were housed in terraria with live plants, a water pool, and a shelter. Ten adult females were size-matched, randomly assigned to control or experimental groups (N=5 per group), and then housed individually. To measure the specific effects of PTX 251D compared to a background toxicity, the control group was fed 0.01% DHQ (Sigma-Aldrich, St. Louis, USA) in a solution of 1% EtOH and the experimental group was fed a solution of 0.01% DHQ and 0.01% PTX 251D (PepTech, Burlington, MA, USA) in a solution of 1% EtOH in water. Each frog was fed 15 µL each day for five days by pipetting the solution directly into the mouth between 10am-12pm. On the afternoon of the fifth day, frogs were euthanized by cervical transection and the dorsal skin, liver, intestines, and oocytes were dissected into Trizol (Thermo Fisher Scientific, Waltham, USA). All procedures were approved by the Institutional Animal Care and Use Committee at Stanford University (protocol number #32870).

2.2 RNA extraction and library preparation

RNA extraction followed the Trizol (Thermo Fisher Scientific, Waltham, MA, USA) protocol outlined in Caty et al. 2019 [33] and according to the manufacturer’s instructions. After the first spin, the organic layer was saved for alkaloid extraction (see below). Poly-adenylated RNA was isolated using the NEXTflex PolyA Bead kit (Bioo Scientific, Austin, USA) following manufacturer’s instructions. RNA quality and lack of ribosomal RNA was confirmed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, USA). Each RNA sequencing library was prepared using the NEXTflex Rapid RNAseq kit (Bioo Scientific). Libraries were quantified with quantitative PCR (NEBnext Library quantification kit, New England Biolabs, Ipswich, USA) and a Agilent Bioanalyzer High Sensitivity DNA chip, both according to manufacturer’s instructions. All libraries were pooled at equimolar amounts and were sequenced on four lanes of an Illumina HiSeq 4000 machine to obtain 150 bp paired-end reads.

2.3 Transcriptome assembly and differential expression analysis

We created a reference transcriptome using Trinity [34] and filtered the raw assembly by removing contigs with BLAST hits belonging to microorganisms and invertebrates in the Swiss-Prot database [35], as these represent likely parasites, prey items, or other contaminants. Overlapping contigs were clustered using cd-hit-est [36,37] and contigs that were less than 250bp long were removed from the assembly. We mapped the paired quality-trimmed Illumina reads to the reference transcriptome using kallisto [38]. Samples were compared across treatment groups (DHQ vs DHQ+PTX) for the skin, liver, and intestines, as these tissues contained higher levels of PTX. Differences in gene expression levels were calculated using DESeq2 [39] [P<0.05 false discovery rate (Benjamini–Hochberg FDR), 4-fold change]. Contigs with significant expression differences were compared to the non-redundant (nr) database using BLAST with an E-value cutoff of 1e-5. Many contigs did not have a BLAST hit, or aligned to hypothetical or non-vertebrate proteins. Contigs with annotations of interest were chosen based on candidates from existing literature. Boxplots were made with R package ggplot2 (R version 3.6.3) using TMM (trimmed mean of M-values) normalized expression. All scripts are detailed in supplementary materials.

2.4 Alkaloid extraction and detection

To isolate alkaloids, 0.3 mL of 100% EtOH was added to 1mL of organic layer from the Trizol RNA extraction, inverted 10 times, and stored at room temperature for 2-3 minutes to precipitate genomic DNA, which was pelleted by centrifugation at 2000g for 5 minutes at 4°C. Then, 300 µL of supernatant was transferred to a new microfuge tube. Proteins were precipitated by adding 900 µL of acetone, mixing by inversion for 10-15 seconds, incubating at room temperature for 10 min, and centrifuging at max speed for 10 min at 4°C. Then, 1 mL of the supernatant containing alkaloids was moved into a glass vial and stored at -20°C until dried down completely under a gentle nitrogen gas flow.

Samples were resuspended in 200 µl of methanol:chloroform 1:1 and 1 µM Nicotine-d3 (used as an internal standard). A 10-point standard curve was prepared in the same solution with DHQ and PTX. A QE+ mass spectrometer coupled to an Ultimate3000 LC (ThermoFisher) was used for analysis. Five µl of each sample were injected on a Gemini C18 column (100x2mm, Phenomenex). The mobile phases were A: water and B: acetonitrile, both with 0.1% formic acid. The gradient was 0% B for 1 min, then increased to 100% B in 14 min, followed by 5 min at 100% B and 3.5 min at 0% B. Data were quantified using accurate mass, using the standard curve for DHQ and PTX for absolute quantification. aPTX was identified by accurate mass and MS/MS fragmentation similarity to PTX.

2.5 Alkaloid statistical analyses

R version 3.6.3 was used for all statistical analyses, and all plotting and statistics code is provided in a supplementary file. There were instances in the LC-MS/MS data where the molecules of interest (DHQ, PTX 251D, or aPTX 267A) were not detected, and these were converted to zeros prior to statistical analyses and visualization. A generalized linear mixed model was used (glmmTMB package in R [40]) to test for differences in alkaloid abundance across tissues and treatment type with the frog as a random effect, using a negative binomial error distribution and a single zero-inflation parameter applied to all observations. PTX 251D and DHQ were analyzed separately. The abundance of aPTX 267A was approximated using the area-under-the-curve divided by the internal nicotine standard, as there is no standard for aPTX 267A, and therefore exact pmol values could not be calculated. A Wilcoxon rank-sum test (wilcox.test) was used to compare the aPTX values in the skin between treatment groups and the Kruskal-Wallis test (kruskal.test) with a post-hoc Dunn test (dunnTest from the FSA package [61]) was used to compare the aPTX values across tissues. Boxplots used to visualize alkaloid abundance values were created in R using ggplot.