Referenced Transcriptomics Identifies a Core Set of Cytochrome P450 Genes Driving Broad-Spectrum Insecticide Detoxification in Phthonandria atrilineata

Abstract

Phthonandria atrilineata, also known as the mulberry looper, is a major defoliator of mulberry trees. This feeding behavior directly affects the growth of the trees and reduces the quality and yield of mulberry leaves for its use in sericulture. Despite its importance the molecular basis of its resistance to insecticides remains poorly understood. Therefore, this study aimed to comprehensively characterize the cytochrome P450 monooxygenases (P450s) gene family in P. atrilineata and identify key effectors responsible for responses to diverse chemical stressors. We integrated genome-wide re-annotation, phylogenetic analysis, and comparative transcriptomics following exposure to five chemically distinct insecticides. We identified a high-confidence set of 70 P450 genes, dominated by the CYP6 and CYP4 families, whose expansion was driven by tandem gene duplication. Transcriptomic analysis revealed a powerful yet highly selective “elite-driven” response, wherein a small subset of P450s was strongly induced by multiple insecticides. Random Forest and Support Vector Machine (SVM) models converged with differential expression data to pinpoint a core trio of P450s as primary drivers of detoxification: two generalists, CYP6(09521) and CYP6(04876), responsive to all compounds, and one potent specialist, CYP4(04803), exhibiting massive induction to a specific subset of insecticides. Our findings uncover a complex, energy-efficient metabolic strategy in P. atrilineata and identify pivotal P450 genes for broad-spectrum detoxification. These genes represent high-priority targets for developing molecular diagnostic tools for resistance monitoring and informing scientifically guided insecticide rotation strategies.

1. Introduction

Sustainable management of agricultural pests demands a fundamental understanding of their adaptive mechanisms at the molecular level, particularly those governing insecticide metabolism and resistance development [1,2]. Among diverse enzyme systems mediating xenobiotic detoxification, cytochrome P450 monooxygenases (P450s) serve as a critical interface between pest physiology and chemical control efficacy [3,4]. These versatile enzymes catalyze the oxidative biotransformation of structurally diverse insecticides, often acting as the primary determinant of field-level control failures [4,5].

The mulberry looper, Phthonandria atrilineata (Lepidoptera: Geometridae), is an ideal model for investigating P450-mediated adaptations, as this economically important pest encounters multiple insecticide classes across its agricultural range in East Asia [6,7]. However, current molecular studies on this serious mulberry pest are so limited that we cannot find any literature focusing on such subjects. Molecular characterization of detoxification responses, building on prior genomic data, has potential applications in pest management. For instance, identifying P450 genes induced by specific insecticide classes could serve as markers for monitoring resistance development in real time, allowing adjustments to control strategies prior to field failures [7]. Additionally, analyzing the range and specificity of P450 induction patterns may guide the development of insecticide rotation schemes that reduce selection pressure on shared detoxification pathways. This approach supports a shift from reactive to predictive resistance management [8,9].

Recent advances in genome sequencing and transcriptional profiling have enabled comprehensive cataloging of detoxification gene families in non-model organisms [4,10,11]. For P. atrilineata, available genomic resources now permit systematic investigation of its P450 complement and responses to chemical stressors [6]. This knowledge is particularly valuable given the pest’s exposure to insecticides with diverse modes of action, from acetylcholinesterase inhibitors like organophosphates to mitochondrial complex I disruptors [7]. Investigating P450 responses to these compounds can elucidate baseline detoxification mechanisms and inducible defenses that contribute to cross-resistance in field populations.

This study presents a comprehensive analysis of the P450 gene family in P. atrilineata, combining computational identification from genomic sequences with functional characterization via controlled insecticide exposures. By examining transcriptional responses to representative compounds from organophosphates and mitochondrial inhibitors, we delineated conserved and compound-specific P450 expression signatures. Our findings establish a molecular framework for metabolic adaptation in this pest, providing actionable insights for resistance monitoring tools and optimized chemical control within integrated pest management programs. This work bridges fundamental molecular biology with practical agricultural applications, addressing the need for science-based approaches to sustain effective pest control.

2. Materials and Methods

2.1. Data Source and Experimental Design

The raw RNA-sequencing data used in this study were generated from a previously described experiment and have been deposited in the NCBI Sequence Read Archive (SRA) under accession number PRJNA1288676. Briefly, third-instar larvae of P. atrilineata were collected from mulberry fields in Hechi, Guangxi, China, and maintained as a laboratory colony under controlled conditions (25 °C, 70% relative humidity, 14:10 light: dark cycle). The strain was susceptible to the tested insecticides, with no prior exposure history. Treatments included phoxim, trichlorfon, trichlorfon-malathion mixture (organophosphates), tolfenpyrad (METI complex I inhibitor), and chlorfenapyr (mitochondrial uncoupler). Controls used solvent-treated leaves. Three independent biological replicates per group (six groups total) yielded 18 libraries, sequenced on Illumina NovaSeq 6000 (150 bp paired-end reads). For each biological replicate, 22–25 larvae were exposed to sublethal concentrations (based on preliminary LC50 assays: phoxim (32%), trichlorfon (21%), trichlorfon malathion (27%), tolfenpyrad (30%), and chlorfenapyr (0% mortality but with visible symptoms of poisoning including green regurgitation and body contortion [7].

2.2. Genome Re-Annotation and P450 Gene Identification

The reference genome assembly of P. atrilineata (GenBank accession: GCA_014332785.1) was retrieved from the National Center for Biotechnology Information (NCBI) database. To improve gene model accuracy for this study, we performed a comprehensive, transcriptome-guided re-annotation. First, clean RNA-seq reads from all 18 samples were mapped to the genome using the splice-aware aligner Hisat2 (v2.2.1) [12]. Second, transcripts were assembled for each sample using StringTie (v2.2.1) [13], and the resulting assemblies were merged using stringtie to create a unified, non-redundant set of transcripts. This evidence, plus prior protein homology, was integrated into GETA v2.7.1 (https://github.com/chenlianfu/geta, accessed on 17 May 2025) for high-confidence protein-coding gene prediction. Putative P450 genes were identified from the re-annotated proteome via InterProScan (v5.60-92.0), selecting proteins with Pfam domain PF00067 and GO term GO:0004497 (“monooxygenase activity”). This yielded 70 high-confidence P450 genes for subsequent analyses.

2.3. Phylogenetic, Genomic, and Structural Analysis

To classify the 70 high-confidence P450 proteins, a UPGMA phylogenetic tree was constructed. The full-length protein sequences were aligned using MAFFT (v7.505) [14] with the accurate L-INS-i algorithm. The UPGMA tree was subsequently inferred using MEGA11 v11.0.13 [15] with 1000 bootstrap replicates to assess nodal support. The final tree was visualized and annotated using the Chiplot web server [16]. The physical coordinates of the P450 genes were extracted from the final GFF3 annotation file. Their distribution and clustering on the chromosomes were visualized using the “show gene location” function in TBtools-II (v2.360) [17]. Tertiary structures of the three top candidate P450 proteins—CYP6(09521), CYP4(04803), and CYP6(04876)—were predicted by homology modeling using the SWISS-MODEL server automated pipeline [18]. Models underwent quality assessment: GMQE scores (0.85–0.92) and QMEAN scores (>−1.5) indicated reliable predictions.

2.4. Transcriptomic Analysis and Differential Gene Expression

Reads were pre-processed with fastp (v0.23.2) to remove adapters, low-quality bases (Phred < 20), and short reads (<50 bp). Clean reads mapped to the re-annotated genome via Hisat2. Expression was quantified as FPKM using RSEM (v1.3.3) [19]. Differential expression used edgeR (v3.42.4) in R (v4.3.1) [20].

2.5. Machine Learning-Based Biomarker Identification

Two supervised models identified key P450s in R (v4.3.1). First, a Random Forest (RF) model was implemented using the randomForest package (v4.7-1.1) [21]. The model was trained on the FPKM expression matrix of selected P450 genes from all 18 samples to classify samples into their respective treatment groups. Default parameters (ntree = 500) were used, and the importance of each P450 gene as a predictor was quantified using the Mean Decrease Accuracy metric. Second, a Support Vector Machine (SVM) model with a linear kernel was applied to identify P450s involved in a generalist detoxification response. This model was also trained on the FPKM expression data of P450 genes found to be upregulated in all five insecticide treatments. Using the e1071 package (v1.7-13) [22], a recursive feature elimination approach was employed to rank the importance of each gene in distinguishing the pooled treated samples from the control samples.

2.6. Statistical Analyses and Data Visualization

All statistical plots, including the PCA plot, volcano plots, and bar charts, were generated using the ggplot2 package (v3.4.2) in R [23]. The genomic landscape circular plot was created using the OmicStudio cloud platform at https://www.omicstudio.cn/tool (accessed on 30 September 2025) [24].

3. Results

3.1. Genome Re-Annotation and Transcriptome-Wide Expression Landscape

Using RNA-seq from six conditions, we re-annotated the P. atrilineata genome, yielding 16,841 protein-coding genes. Comprehensive functional annotations (NR, InterProScan) covered 15,878 genes (94.3%), integrating data from NR, Pfam, CDD, SMART, SUPERFAMILY, NCBIfam, PANTHER, InterPro, and GO terms. The 336.55 Mb genome spans 31 chromosomes (Figure 1). An overview of the transcriptional activity revealed that a substantial portion of the genome is actively expressed under both control and insecticide-stressed conditions.

In the control group, a total of 11,870 genes (70.5% of the annotated genome) were expressed. Following insecticide exposure, the number of expressed genes varied depending on the compound. The highest number of expressed genes was observed in the phoxim (12,649 genes, 75.1%) and trichlorfon (12,633 genes, 75.0%) treatments. The tolfenpyrad (12,447 genes, 73.9%) and trichlorfon-malathion mixture (12,394 genes, 73.6%) treatments also showed a broad transcriptional response. The chlorfenapyr treatment resulted in the expression of 11,730 genes (69.6%), a number similar to that of the control group. This comprehensive expression atlas, depicted across the 31 chromosomes in Figure 1, establishes the baseline transcriptional landscape for investigating specific P450 gene family responses to diverse chemical pressures.

3.2. Identification and Annotation of the P450 Gene Complement

A multi-evidence annotation strategy, querying seven functional and protein domain databases (Table S1), delineated 112 putative P450 genes in the re-annotated Phthonandria atrilineata proteome. Notably, 70 genes (62.5%) were unanimously supported by all sources, establishing a high-confidence core set for downstream analyses (Figure 2). An additional 31 genes received majority endorsement from domain databases but lacked the “monooxygenase activity” GO term, while a minority were single-source detections (e.g., eight via GO alone, three via Pfam or CDD). These peripheral candidates, potentially representing pseudogenes or fragments, were excluded to prioritize functional integrity.

Mapping these 70 genes revealed an uneven chromosomal distribution across 22 chromosomes, characterized by 11 prominent clusters (Figure 3). A striking feature of their distribution was the extensive formation of gene clusters, suggesting that gene duplication events have played a significant role in the family’s evolution. We identified a total of 11 distinct P450 gene clusters. The largest clusters belonged to the CYP6 family, including a remarkable tandem array of eight genes on Chr13 (from CYP6(07978) to CYP6(08024)) and another cluster of four genes on Chr16 (CYP6(09800) to CYP6(09810)). Significant clusters were also observed for the CYP4 family (e.g., three genes on Chr05 and three on Chr07) and the CYP24A1 family (e.g., three genes on Chr05). This clustering pattern strongly indicates that tandem gene duplication has been a primary mechanism driving the expansion of the P450 gene family in P. atrilineata, particularly within the CYP6 and CYP4 clans, which are frequently associated with xenobiotic metabolism.

Phylogenetic reconstruction via UPGMA classified the 70 genes into four robust clades, aligning with major P450 families (Figure 4). The largest clade belongs to the CYP6 family, comprising a remarkable 38 members. The second-largest clade represents the CYP4 family, containing 18 members. The remaining two clades were identified as the CYP24A1 family with 11 genes and the CYP15A1 family with 3 genes. The segregation of these four major families was strongly supported by high bootstrap values, many at 100%, indicating a reliable classification that aligns perfectly with their CDD annotations. The vast number observed in the CYP6 and CYP4 families, which together account for 80% (56 of 70) of the high-confidence P450s, strongly suggests their critical and diversified roles in the detoxification of xenobiotics and metabolism of endogenous compounds in P. atrilineata.

3.3. Transcriptional Response of P450 Genes to Insecticide Exposure

We began by exploring the overall expression patterns of the 70 high-confidence P450 genes across our 18 samples using principal component analysis (PCA). This approach uncovered a stark division between control and insecticide-treated groups, driven largely by the first principal component (PC1), which captured 96.8% of the variance (Figure 5). The control replicates formed a compact cluster, well separated from the treated ones, highlighting a consistent transcriptional shift triggered by chemical exposure. Interestingly, while the various insecticide treatments showed some spread along PC2 (accounting for 2.1% of variance), their general clustering points to shared response mechanisms, despite differences in chemical structures and modes of action.

Then, we generated a heatmap based on their log₂ fold change (log₂FC) values following exposure to the five insecticide treatments to dissect the expression patterns of individual P450 genes (Figure 6). The analysis revealed highly heterogeneous responses. A majority of the P450 genes exhibited either downregulation (blue) or minor changes in expression (green), suggesting that a broad, non-specific induction of the entire P450 family does not occur. Instead, a distinct and small subset of P450s was powerfully upregulated (yellow to red) across multiple treatments. Among the most potently induced genes were CYP4(04803), which showed exceptionally high upregulation (>10-fold) in response to four treatments, and CYP6(09807), which was consistently upregulated across all five insecticides. Other notably induced genes included CYP4(03422), CYP4(10594), and CYP6(04876). Intriguingly, some genes displayed compound-specific regulation; for example, CYP4(03422) was strongly induced by four compounds but was significantly suppressed by Phoxim, indicating complex regulatory mechanisms. The hierarchical clustering further grouped these co-regulated genes, highlighting specific modules of P450s that are likely involved in core detoxification pathways.

Building on this, we applied a Random Forest model to pinpoint genes most predictive of treatment status, using mean decrease in accuracy as the ranking metric (Figure 7). This machine learning step highlighted a handful of standout performers, such as CYP6(09521) (MDA = 6.05), CYP4(10594) (5.9), and CYP6(04876) (5.63), which aligned closely with the most induced genes from the heatmap. Notably, these top predictors, mainly from the CYP6 and CYP4 families, not only flagged exposure reliably but also hinted at their frontline roles in metabolic defense.

Focusing further on genes upregulated universally across all treatments, we used a Support Vector Machine (SVM) with recursive feature elimination to assess their discriminatory value against controls (Figure 8). The heatmap, displaying the Z-score normalized expression values, visualizes the relative importance of these genes across conditions. Genes such as CYP6(04876), CYP6(09807), CYP4(10594), and CYP6(03564) consistently exhibited high importance values (indicated by red shading) in multiple insecticide treatments, reinforcing their role in a generalized stress response. Hierarchical clustering of these importance scores grouped genes with similar response profiles, suggesting potential co-regulation or functional redundancy in metabolizing a wide array of substrates. For instance, the cluster containing CYP6(09807) and CYP4(10593) shows high importance in response to Tolfenpyrad and Trichlorfon. The False Discovery Rate (FDR) values associated with each gene’s importance score provide statistical confidence in their role as biomarkers. Notably, several genes, including CYP6(04876), CYP6(09807), CYP6(02923), CYP6(09521), and CYP4(04804), showed highly significant FDR values (FDR ≤ 0.05), establishing them as statistically robust indicators of a general xenobiotic detoxification response in P. atrilineata.

3.4. Convergent Evidence Pinpoints a Trio of P450s as Key Effectors of Broad-Spectrum Resistance

Synthesizing our differential expression, Random Forest, and SVM results yielded a focused set of P450 genes central to P. atrilineata’s insecticide tolerance. Three emerged as frontrunners—CYP6(09521), CYP4(04803), and CYP6(04876)—each with compelling evidence of key involvement, though their profiles suggested varied strategic roles. CYP6(09521) and CYP6(04876) exemplified versatile responders, ranking highly in both models and showing uniform upregulation across all insecticides tested. This broad reactivity positions them as likely generalists, adept at processing diverse chemical threats and forming the backbone of the pest’s adaptive arsenal. CYP4(04803), by contrast, exhibited a more specialized pattern: it surged over 10-fold in four treatments but was notably suppressed by phoxim, indicating substrate-specific prowess and regulatory finesse.

To probe their molecular foundations, we modeled their 3D structures via homology (Figure 9), revealing the classic P450 architecture in each. All three models adopted the canonical, well-conserved globular fold characteristic of cytochrome P450 enzymes, lending structural plausibility to their predicted function. The models for CYP6(09521) and CYP6(04876) were based on templates from other lepidopteran species (Phalera bucephala and Ostrinia furnacalis) with high sequence identity (73.28% and 70.25%, respectively), indicating a reliable structural prediction. Similarly, the CYP4(04803) model was built using a template from Heortia vitessoides with 70.09% identity. The surface-rendered models highlight regions (shown in orange) that correspond to areas of high sequence variability, which typically encompass the substrate recognition sites (SRSs) that define the enzyme’s substrate specificity. The distinct topographies of these regions across the three models may underlie their differential capacities to bind and metabolize a diverse array of insecticide molecules. Collectively, the convergence of robust transcriptional induction, high predictive importance in machine-learning models, and conserved structural architecture provides compelling, multi-faceted evidence that these three P450s are high-priority targets for functional validation and are likely pivotal players in the evolution of insecticide resistance in P. atrilineata.

4. Discussion

This study examined the cytochrome P450 monooxygenase gene family in the mulberry looper, P. atrilineata, an important agricultural pest. Insecticide resistance threatens global food security, making it essential to understand the molecular mechanisms pests employ to detoxify chemicals [25,26]. We combined genomic, transcriptomic, and machine-learning methods to identify the P450 genes in P. atrilineata and determine those involved in broad-spectrum detoxification. These results contribute to understanding metabolic adaptation in this pest and support the development of evidence-based resistance management strategies.

Our systematic re-annotation of the P. atrilineata genome identified a high-confidence set of 70 P450 genes, a number comparable to that found in other lepidopteran pests, such as the rice borer Chilo suppressalis [4,11,27]. The phylogenetic analysis revealed that this gene family is dominated by the CYP6 and CYP4 clans, which together constitute 80% of the entire repertoire. This clan distribution is a well-established evolutionary signature in insects, where the expansion and diversification of the CYP3 and CYP4 clades (to which the insect CYP6 family belongs) are strongly correlated with adaptation to chemical challenges, including both plant allelochemicals and synthetic insecticides [28,29]. A particularly telling finding was the pronounced clustering of these genes on the chromosomes, with a notable tandem array of eight CYP6 genes on chromosome 13. This genomic architecture provides compelling evidence that tandem gene duplication has been the primary evolutionary engine driving the expansion and subsequent functional specialization of the P450 family in P. atrilineata, like in the brown planthopper Nilaparvata lugens [30], equipping it with a versatile enzymatic toolkit to confront diverse xenobiotic threats [31,32].

A key aspect of this study was analyzing the transcriptional responses of P450 genes to insecticide exposure. Principal component analysis (PCA) indicated a consistent response, with the first principal component (PC1) explaining 96.8% of the variance between control and treated samples. This suggests activation of a general detoxification pathway in response to diverse insecticides, including acetylcholinesterase inhibitors and mitochondrial disruptors [33,34,35]. Further examination of gene expression patterns via heatmap analysis showed a selective mechanism. Rather than broad upregulation of all P450 genes, P. atrilineata induced a limited subset, while most genes were downregulated or unchanged [36,37]. A small, specific subset of P450s was powerfully and consistently induced, while the majority of family members were either repressed or unresponsive. Consistent with prior research on P450 genes [4,38,39], these findings indicate an energy-efficient strategy that focuses on specific enzymes for metabolizing various substrates, avoiding the activation of redundant pathways [40,41].

By leveraging supervised machine-learning algorithms, we moved beyond conventional differential expression analysis to identify the most functionally significant P450s with high statistical confidence. The striking convergence of results from Random Forest (RF) and Support Vector Machine (SVM) models with our transcriptomic data provides multi-faceted, robust evidence for the central role of a few key players. Two genes, CYP6(09521) and CYP6(04876), emerged as archetypal “generalist” detoxifiers, exhibiting strong upregulation across all five insecticide treatments and ranking as top biomarkers in both machine-learning models. Their profiles suggest they are central hubs in the insect’s chemical defense system [42,43]. In contrast, CYP4(04803) was strongly induced by four compounds but repressed by phoxim, suggesting specificity to certain substrates and regulated expression. Homology modeling confirmed the typical P450 structure for these proteins, with variations in substrate recognition sites that may account for their differing specificities [40,44,45].

Overall, these findings have implications for sustainable agriculture. The identified P450 genes, including CYP6(09521), CYP6(04876), and CYP4(04803), could be targeted for molecular diagnostics to track resistance alleles in field populations, facilitating early intervention [4,46]. Insights into their generalist and specialist functions may also guide insecticide rotation to minimize cross-resistance and extend the utility of current controls. Future work should validate these genes functionally through RNA interference (RNAi) or CRISPR-Cas9 knockout to assess their contribution to resistance [46,47]. Heterologous expression systems could further characterize their substrate preferences and kinetics against additional insecticides [48,49]. These efforts will help apply our genomic and transcriptomic results to practical pest management.