Retrospective Metabolomics Profiling of Clinical Urine Drug Screen Samples Reveals Features Associated With Opiate Exposure

1. Introduction

Opiates are prototypal opioids with a long history of use and abuse, extending from ancient opium consumption and the nineteenth-century Opium Wars to the contemporary opioid crisis involving prescription opiates, heroin, and synthetic opioids [1,2,3,4]. Opiates comprise naturally occurring opium alkaloids and their semisynthetic derivatives.

Morphine is the principal opiate alkaloid in Papaver somniferum and the prototypic opioid receptor agonist, whereas heroin, or 3,6-diacetylmorphine, is a diacetylated morphine derivative and oxycodone is a semisynthetic derivative of thebaine, another opium alkaloid [5,6,7]. These compounds share a morphinan/phenanthrene scaffold that contributes to opioid receptor binding [7] and, in part, to the antibody cross-reactivity exploited by the immunoassays targeting opiates, although individual immunoassays differ in target specificity and cross-reactivity [8]. In the United States, consumed opiates are largely prescription-derived; however, illicit opiates, such as heroin and prescription-diverted oxycodone, have also been consumed recreationally, causing positive results in these immunoassays [9,10].

These conventional immune-based urine drug tests only capture a limited view of opiate-associated biology, unable to broadly characterize co-exposures, downstream metabolic perturbations, or endogenous biochemical responses. In contrast, untargeted metabolomics applied to clinical specimens can provide an opportunity to explore metabolic features associated with opiate-related immunoassay positivity, potentially leading to the identification of novel biomarkers and altered pathways by opiate exposure. The growing adoption of liquid chromatography–high-resolution mass spectrometry (LC-HRMS), including liquid chromatography–quadrupole time-of-flight mass spectrometry (LC-qToF-MS), has expanded the role of untargeted acquisition in clinical toxicology [11,12,13,14].

As part of routine clinical care, the Clinical Toxicology Laboratory at the University of Pittsburgh Medical Center performs comprehensive urine drug screening (CUDS; >500 specimens per month) to evaluate suspected intoxication and medication adherence. This LC-qToF-MS workflow, with an all-ion fragmentation scan, generates untargeted mass spectral data across a broad range of detected analytes [10,12,15,16]. Although each specimen typically contains thousands of molecular features, routine clinical interpretation focuses on a limited set of known xenobiotics, primarily drugs and their metabolites. As a result, most detected features remain unannotated and are not incorporated into routine reporting. These residual feature-level data represent an underused resource for investigating co-exposures, drug-associated metabolic perturbations, and broader biochemical signatures linked to drug exposure.

In this study, we repurposed data generated through our CUDS workflow, which uses LC-qToF-MS for qualitative toxicological testing [12]. Although this platform is primarily intended for routine drug detection, its untargeted acquisition produces feature-rich datasets suitable for secondary metabolomics analysis. Using archived mass spectrometry data, we performed a retrospective metabolomics study in which specimens were classified based on qualitative opiate, oxycodone, and 6-monoacetylmorphine enzyme immunoassay (OPIA-EIA, OXY-EIA, and 6MAM-EIA, respectively) results as metadata. By comparing EIA-positive and EIA-negative specimens, we sought to characterize the distinct and overlapping metabolomic signatures associated with opiate-related immunoassay positivity.

This study also has translational significance because it repurposes data generated from a clinically accessible specimen type and an analytical workflow already embedded in routine toxicology practice. Compared with metabolomics studies using specimens that are difficult to obtain or implement clinically, such as brain tissue homogenates, urine-based CUDS data provide a more practical bridge between discovery and clinical validation. Features identified in this setting may therefore be prioritized as candidate biomarkers, co-exposure indicators, or interpretive adjuncts for future testing, while also supporting laboratory-based sentinel surveillance for evolving drug exposures and emerging adulterants.

2. Materials and Methods

2.1. Specimens

The CUDS datasets were originally derived from the urine specimens from 363 patients, as described previously [12](Table 1). These urine specimens were from emergency departments of UPMC hospitals and clinics in western PA. These specimens underwent the initial EMIT-II-based qualitative drug screening panel including for opiate (OPIA-EIA, with the cutoff at 300 ng/mL for morphine), oxycodone (OXY-EIA, with the cutoff at 100 ng/mL for oxycodone), and 6-monoacetylmorphine (6MAM-EIA, with the cutoff at 20 ng/mL for 6-monoacetylmorphine) (Siemens Healthineers USA, Malvern, PA) before mass spectrometry-based analysis. OPIA-EIA can detect opiates and their metabolites broadly (e.g., 6-monoacetylmorphine at 435 ng/mL, codeine 102 - 306 ng/mL, heroin 396 ng/mL, hydrocodone at 247 ng/mL, hydromorphone at 498 ng/mL, morphine-3-glucuronide at 626 ng/mL, nor-oxycodone at 100 µg/mL, oxycodone at 1500 ng/mL, and oxymorphone 9300 ng/mL, according to the product insert). One volume of urine specimen was mixed with 4 volumes of distilled water spiked with nitrazepam and tenoxicam as internal standards (40 ng/mL final). A diluted urine specimen was directly injected into the LC-qToF-MS (3 μL).

2.2. LC-qToF-MS Assay Conditions/Settings in CUDS

A dilute-and-shoot method with 1:4 volume mixture was employed for LC-qToF-MS analysis with the ESI positive mode. The acquired high-resolution mass spectrometry (HRMS) datasets were initially analyzed using UNIFI® Scientific Information System (Waters) to screen compounds through retention time, monoisotopic mass accuracy, and fragment ions match and the CUDS reports with identified drugs were released as CUDS reports as part of clinical testing previously. Detailed information for LC-qToF-MS assay conditions was given previously [12,15].

2.3. Overall Data Processing and Analytical Workflow

The overall data processing and analysis workflow, from raw mass spectrometry data to metabolomics feature annotation, is summarized in Figure 1. Detailed information for overall data processing and analytical workflow was described previously [12].

2.3.1. Data Processing and Statistical Analysis

De-identified MS datasets in (.uep) files were exported, converted to (mzML) files, and processed using MS-DIAL v4.9 for peak detection and peak alignment and using MetaboAnalyst (both the web version and R-version or MetaboAnalystR) for normalization with log-transformation and autoscaling, unsupervised data filtering to reduce the number of the features to 5000, and various statistical analyses including including point-biserial correlation analysis, volcano plot analysis (fold change ≥2, p <0.05), significance analysis of microarrays and metabolites (SAM), empirical Bayesian analysis of microarrays and metabolites (EBAM), partial least squares discriminant analysis (PLS-DA), and random forest (RF) classification to identify metabolic features significantly associated with OPIA-, OXY-, and 6MAM-EIA results. The features selected by ≥1 statistical analysis were designated as discovery feature sets, whereas the features selected by ≥2 statistical analyses were designated as consensus feature sets. An Euler diagram was generated from the EIA-associated discovery feature sets to summarize shared and group-specific features across EIA-defined groups (Figure 2, Table S1).

2.3.2. Correlation-Based Hierarchical Clustering Within EIA-Associated Discovery Feature Sets

Hierarchical correlation clustering analysis was conducted by computing Pearson correlation coefficients between the features within the EIA-associated discovery feature set. The resulting matrix was visualized as a heatmap using the pheatmap package in R and hierarchical clustering was performed using Ward’s method. Features were grouped into seven clusters by high-level dendrogram slicing (k = 7). Additionally, middle-level (k = 14) and low-level (k = 50) dendrogram slicing were also applied to identify subclusters of co-varying features (Figure 3, Figure 4 and Figure 5, Table S2).

2.3.3. Feature-to-Feature Correlation Matrix Including all Features

The feature-to-feature correlation matrix for all 14,883 features across 363 specimens, rather than the post-filtered 5000 features, was computed to identify the most correlated features to further facilitate the subsequent feature annotation process.

2.3.3. Feature Annotations

Selected features were annotated using MS-FINDER version 3.61 [17,18], which predicts candidate molecular formulas and putative chemical structures in silico from MS/MS spectra exported from MS-DIAL. The output may include candidates from Metabolic In silico Network Expansions (MINE) databases, which comprise computationally predicted biochemically plausible metabolites rather than experimentally confirmed compounds [19]. The results of correlation-based hierarchical clustering and the feature-correlation matrix were also used to assist the feature annotation process by evaluating potential in-source fragmentation, different adducts, and the biomedical plausibility of the annotation candidates. Annotations were categorized based on Chemical Analysis Working Group Metabolomics Standards Initiative (MSI) [20].

3. Results

From the 14,883 features extracted across all specimens, 138 features were significantly associated with OPIA-EIA positivity (OPIA-associated discovery feature set), 121 features with OXY-EIA positivity (OXY-associated discovery feature set), and 104 features with 6MAM-EIA positivity (6MAM-associated discovery feature set), by ≥ 1 statistical method (Figure 1).

Multiple features are found to be associated with more than one EIA result (Figure 2). The number of the shared features is minimal between OXY-EIA and 6MAM (10 features in total, as compared to 44 features between OPIA-EIA and 6MAM-EIA and 75 features between OPIA-EIA and OXY-EIA (Table S1).

We next conducted cluster analyses on the normalized data matrix to classify EIA-associated features into groups, better understand the metabolomic signatures of these groups, and facilitate interpretation of unannotated or partially annotated features. The features were grouped into seven high-level clusters for each EIA result (Clusters A–G for each EIA) (Figure 3, Figure 4 and Figure 5, Table S2).

We also evaluated the top 20 most correlated features based on the feature correlation matrix for the entire 14,883 features across 363 specimens (Table S3) to support the ensuing feature annotation process, as the features derived from structurally and metabolically related chemicals and analytically related ion signatures, including different ion adducts and in-source ion fragments, are presumed to be correlated best.

Based on these analyses, we attempted to annotate the features selected by multiple statistical analyses for each EIA (21 features in the OPIA-associated consensus feature set, 17 features in the OXY-associated consensus feature set, and 9 features in the 6MAM-associated consensus feature set). Both cluster analysis and correlated features were also taken into consideration to generate putative annotations. These most significant features and their annotations and clusters are summarized in Table 2, Table 3 and Table 4.

3.1. OPIA-Associated Consensus Feature Set

We annotated the 21 features selected by more than one analysis for OPIA-EIA (OPIA-associated consensus feature set) (Table 2). These features include oxycodone/opiate metabolites (Cluster E3-1), acetaminophen metabolites (Cluster A1-1), and others.

3.1.1. Oxycodone Metabolites (Cluster E3-1)

Cluster E3-1 contains multiple oxycodone metabolites. These features have been validated and already included in the CUDS reports. These oxycodone metabolites can cause positive results in OPIA-EIA, though in a less efficient manner than OXY-EIA.

Feature 286.15695_1.378 is likely a composite feature of noroxycodol [M+H-H2O]+ (RT 1.34 and 1.55), norcodeine [M+H]+ (RT 1.60), and hydromorphone [M+H]+ (RT 1.20). Other opiates and their metabolites, including morphine (feature 286.15637_0.97), morphine-3-glucuronide (462.18497_0.782), hydrocodone (300.15921_2.15), norhydrocodone (286.14444_2.138), and dihydrocodeine-6-glucuronide (478.21341_1.086) have been filtered out by MetaboAnalyst as less contributory features before application of multiple statistical analyses.

Feature 241.11029_1.961 also belongs to cluster E3-1; however, it does not appear to be a morphinan (morphine/codeine/hydrocodone)–type metabolite. Thus, feature 241.11029_1.961 cannot be associated with oxycodone metabolism, and its chemical identity is unknown.

3.1.2. Acetaminophen Metabolites (Cluster A1-1)

Cluster A1-1 contains multiple acetaminophen metabolites. These features have been validated and already included in the CUDS reports. The features in cluster A1-1 are found only in the OPIA-associated discovery feature set, not in the OXY- or 6MAM-associated discovery feature sets. OPIA-EIA does not cross-react with acetaminophen and its metabolites, indicating that acetaminophen is likely co-ingested with opiates by patients with positive OPIA-EIA specimens.

3.1.3. Fentanyl Metabolite (Cluster D2-2)

Feature 233.17049_3.226 is annotated as norfentanyl, and this annotation has been validated and already included in the urine comprehensive drug screening reports. OPIA-EIA does not cross-react fentanyl or norfentanyl, but recreational fentanyl is often co-ingested with heroin (diacetylmorphine), which is metabolized to 6-monoacetylmorphine, morphine, and its metabolites, causing positive OPIA-EIA results.

3.2. OXY-Associated Consensus Feature Set

We annotated the 17 features selected by more than one analysis for OXY-EIA (OXY-associated consensus feature set) (Table 3). These features include oxycodone metabolites (clusters B3-1, E2-3, and G1-1), dipeptide (α-phenylalanylaspartic acid) and others (clusters G3-6 and G3-7), vitamin B6 metabolite (4-pyridoxic acid) (cluster C2-1), and others.

3.2.1. Oxycodone Metabolites (Clusters B3-1, E2-3, and G1-1)

Cluster E2-3 contained multiple oxycodone reductive metabolites, including oxycodols and noroxycodol. These metabolites had been previously validated and were already included in routine CUDS reports. In contrast, feature 288.12936_1.054, annotated as noroxymorphone, was assigned to cluster B3-1, whereas feature 286.15695_1.378, a composite feature representing noroxycodol, norcodeine, and/or hydromorphone, was assigned to cluster G1-1. Thus, these features were separated from the main cluster of oxycodone reductive metabolites.

Most of these features are also found in the OPIA-associated consensus feature list, except for feature 288.12936_1.054 (noroxymorphone), presumably because the OPIA-EIA used in this study shows very weak reactivity to noroxymorphone.

3.3. 6MAM-Associated Consensus Feature Set

We annotated the nine features selected by more than one analysis for 6MAM-EIA (6MAM-associated consensus feature set) (Table 4). These features include norfentanyl and 3-hydroxycotinine artifact (cluster B1-5) and 6-acetylmorphine itself (cluster C1-4), all validated by spike studies and already included in the CUDS reports. These analytes are all related to recreational chemical usage. These analytes are all co-exposed by 6MAM-EIA positive patients secondary to recreational chemical usage.

4. Discussion

This retrospective untargeted metabolomics analysis was conducted with routine CUDS data to identify the features significantly associated with opiate exposure using OPIA-EIA, OXY-EIA, and 6MAM-EIA results as metadata. The strengths of this study include (i) the large-scale, real-world clinical dataset based on the untargeted urine LC-qToF-MS analyses yielding 14,883 features for 363 routine urine toxicology specimens, (ii) pre-specified, rigorous statistical pipeline applied independently to each EIA endpoint (not post-hoc cherry-picking) for systematic feature selection, (iii) correlation-based clustering and MS/MS inspection to prioritize coherent, biologically interpretable feature groups rather than isolated peaks, and (iv) direct translational relevance to everyday clinical toxicology, since all data are derived from routine urine drug screening specimens in a clinical laboratory of an academic medical center in the US.

Feature annotation remains a major intrinsic challenge in untargeted metabolomics studies [21,22]. To support candidate annotation, we generated a feature correlation matrix from normalized peak-intensity profiles spanning 14,883 features across 363 specimens (Table S3) and applied hierarchical cluster analysis to identify co-varying feature groups (Table S2). This correlation-guided, orthogonal approach supported recognition of analytically related ions, including isotopologues, adducts, and in-source fragments, and helped place candidate annotations within biomedically coherent drug/metabolite clusters. These relationships were used to strengthen the plausibility of candidate annotations suggested by MS-FINDER, which incorporates accurate mass and MS/MS spectral information [17,18].

4.1. Target Metabolites of OPIA-EIA, OXY-EIA, and 6MAM-EIA

The three EIAs used to define opioid exposure represent distinct “exposure selectors” rather than interchangeable opioid markers. The OXY-EIA (100 ng/mL cutoff) is comparatively sensitive and specific for therapeutic oxycodone exposure, whereas the morphine-calibrated OPIA-EIA (300 ng/mL cutoff) has a broader opiate coverage, including hydromorphone, codeine, and 6-monoacetylmorphine; however, the OPIA-EIA requires substantially higher concentrations of oxycodone and oxymorphone (1500 ng/mL and 9300 ng/mL each) to cross-react and become positive. In contrast, the 6MAM-EIA (20 ng/mL cutoff) is highly specific for the heroin-specific metabolite (6-monoacetylmorphine), which reveals very recent heroin exposure. Because of the short detection window of 6-monoacetylmorphine, 6MAM-EIA may capture an acute-use phenotype that differs from the chronic opioid-treated population enriched in OXY-EIA and many OPIA-EIA positive specimens.

4.2. Oxycodone Metabolites-Related Features

Oxycodone metabolites are enriched not only in the OXY-associated consensus feature set, but also in the OPIA-associated consensus feature set, likely reflecting a combination of OPIA-EIA broad cross-reactivity and overlap between OPIA-EIA–positive and oxycodone-exposed specimens. Among these metabolites, α-/β-oxycodol was detected in both the OXY- and OPIA-associated consensus feature sets. α-/β-oxycodol is formed through 6-ketoreduction of oxycodone, a metabolic route distinct from the ones mediating the formation of noroxycodone and oxymorphone [23,24,25]. Although α-/β-oxycodol has often been omitted from simplified oxycodone metabolic schemes or treated as a minor reductive metabolite [26,27], its reproducible association with both OXY-EIA and OPIA-EIA positivity highlights its value as another analytically informative biomarker of oxycodone exposure.

The separation of noroxymorphone from the oxycodone reductive metabolite cluster with oxycodol and noroxycodol in the OXY-associated discovery feature set (Table 3) may reflect its distinct metabolic position at the intersection of N-demethylation and O-demethylation pathways by CYP3A and CYP2D6, as well as variability in CYP-mediated metabolism and subsequent conjugation [28,29]. Therefore, noroxymorphone may behave as a less consistent oxycodone-associated marker than the major reductive metabolites in untargeted urine metabolomics. Even though noroxymorphone is chemically identical to nor-naloxone, a naloxone metabolite [30], the absence of naloxone (Feature 328.16931_1.645) within the same cluster B (Table S2) or among the noroxymorphone-correlated features (Table S3) is against the idea that feature 288.12936_1.054 is secondary to the naloxone exposure.

4.3. Fentanyl Metabolite-Related Features

Fentanyl is a synthetic opioid with no shared chemical structures with opiates; thus, fentanyl and its major metabolite norfentanyl do not cross-react with the antibodies used in OPIA- and 6MAM-EIA kits. Nevertheless, norfentanyl is found in both the 6MAM- and OPIA-associated consensus feature sets. Because fentanyl has been replacing heroin as a recreational opioid [4], fentanyl co-exposure is expected to take place simultaneously with heroin, providing a biomedical justification for norfentanyl, a major metabolite of fentanyl, to be included in the 6MAM- and OPIA-associated consensus feature sets.

4.4. Recreational Chemicals-Related Features

The features associated with opioid abuse (e.g., norfentanyl, 6-monoacetylmorphine) and smoking (e.g., 3-hydroxycotinine artifact) are found in the 6MAM-associated consensus feature set (Table 5). Furthermore, the 6MAM-associated discovery feature set contains at least six additional features associated with smoking (e.g., 193.10092_0.876 for 3-hydroxycotinine, 177.10405_0.988 for cotinine, 179.12364_0.927 for nicotine-N-oxide, 163.12866_0.849 for nicotine), eight additional features associated with cocaine use (e.g., 200.12935_0.795 for methyl ecgonine, 304.15652_4.399 for cocaine, 214.15097_0.935 for ethyl ecgonine), and one additional feature associated with stimulant usage (e.g., 91.05488_2.605 for the in-source fragment of methamphetamine), all in cluster B (Table S2). Most of these features were also found in the cocaine-EIA-associated feature set, as reported previously [12]. Overall, recreational chemical-related features are found predominantly among the 6MAM-EIA-associated features.

Among these recreational chemical-related features, norfentanyl is included in the OPIA-associated consensus feature set (Table 2), whereas 6-monoacetylmorphine is included in the OPIA-associated discovery feature set (Table S2). Thus, the OPIA-associated feature set occupies a heterogeneous and intermediate position between 6MAM- and OXY-associated feature sets, shaped by the broad OPIA-EIA reactivity, heroin metabolism into morphine, fentanyl co-exposure, and real-world opioid polysubstance use profile.

Overall, these metabolomic patterns support a dominant exposure-gradient model among 6MAM-, OXY-, and OPIA-associated feature sets rather than mutually exclusive exposure classes. In other words, these dominant exposure axes with overlap, rather than three mutually exclusive exposure classes.

4.5. Acetaminophen-Related Features

Acetaminophen-related features are only found among the OPIA-EIA-associated features, not among the OXY-EIA- or 6MAM-EIA-associated features, through co-exposure of acetaminophen with opiates for the patients with OPIA-EIA positive specimens. Even though acetaminophen was reported as a cutting agent for heroin before [31], the acetaminophen-related features are not associated with 6MAM-EIA, making a cutting agent improbable as a source of acetaminophen.

Rather, these findings likely reflect the co-medicated acetaminophen within the opiate prescriptions in the U.S. because the ClinCalc DrugStats database about the U.S. outpatient drug usage statistics in 2023 shows that hydrocodone-acetaminophen was the most prescribed opiate analgesics (over 21.5 million prescriptions), followed by oxycodone (over 13.5 million prescriptions), oxycodone-acetaminophen (over 7.2 million prescriptions), morphine (over 3.5 million prescriptions), codeine-acetaminophen (over 2.1 million prescriptions), and hydrocodone (79,852 prescriptions)[32].

4.6. Vitamin B6

Prior metabolomics studies have reported perturbation of vitamin B6-related metabolism in opioid/opiate exposure. In a human urine metabolomics study of Golestan Cohort opium users, pyridoxine and pyridoxal were increased, whereas 4-pyridoxic acid, the major urinary catabolite of vitamin B6, was modestly decreased [33]. In a separate mouse heroin-exposure model, 4-pyridoxate were also down-regulated in the hepatic tissue [34]. Thus, the positive association between urinary 4-pyridoxic acid and OXY-EIA+ in the present study appears discordant with the direction of the prior studies

One plausible explanation is the use of over-the-counter vitamin supplements. OXY-EIA positive specimens in our clinical cohort may be enriched for patients receiving prescribed oxycodone for chronic pain, a population in which over-the-counter vitamin supplement use is common [35,36]. Because urinary 4-pyridoxic acid is responsive to recent vitamin B6 intake rather than being a specific marker of hepatic B6 stores [37], increased 4-pyridoxic acid may reflect recent vitamin B6 or multivitamin exposure rather than opioid-mediated hepatic B6 depletion.

4.7. α-Phenylalanylaspartic Acid

α-phenylalanylaspartic acid is a dipeptide found among the human fecal metabolome [38,39], but it has not been reported as part of human urine metabolome before. On the other hand, an isomeric dipeptide β-phenylalanylaspartic acid is a known endogenous dipeptide detectable in human urine and plasma [40], and another isomeric dipeptide α-aspartylphenylalanine is known as a metabolic byproduct of aspartame, an artificial sweetner [41]. Even though these isomeric dipeptides have similar MS2 mass spectra consistent with features 264.08627_1.057 and 281.11124_1.08, α-phenylalanylaspartic acid is the only one eluting around 1.05 min in our LC-qToF-MS conditions and thus, these features are annotated as α-phenylalanylaspartic acid. To our knowledge, this represents the first report of α-phenylalanylaspartic acid as a urinary metabolomic feature/metabolite in human urine.

The association of α-phenylalanylaspartic acid with OXY-EIA and OPIA-EIA positive specimens may reflect altered dipeptide or amino-acid metabolism secondary to chronic opioid exposure. Plausible contributors include opioid-induced androgen leading to altered protein turnover in chronic pain populations ([42,43,44,45]), and possible comorbid renal dysfunction [46,47,48]. But urinary dipeptides can also be influenced by other factors such as diet, nutritional status, or comorbid illness, thus, further investigation is warranted for these confounding factors.

4.8. Potential Role of Routine Clinical HRMS Data in Laboratory-Based Drug and Chemical Surveillance

Although not designed to estimate population prevalence, this workflow provides a scalable, laboratory-based sentinel surveillance framework by leveraging archived LC–qToF-MS data from routine clinical toxicology testing to detect exposure-associated molecular signatures and to generate early hypotheses about regional drug use and co-exposure trends. The relevance of this retrospective metabolomics framework for surveillance is further illustrated by the temporal pattern of emerging adulterants in the dataset.

The feature corresponding to xylazine (221.12518_3.759, filtered out in the statistical analyses in this study) was included in the datasets collected during 2021–2022, consistent with reports that xylazine was already increasingly present in the U.S. illicit opioid supply during this period [15,49]. In contrast, the features corresponding to medetomidine-related metabolites were not observed, which is also consistent with reports that medetomidine became more widely recognized as an illicit opioid adulterant later, with sporadic detections beginning around 2022–2023 and broader public health concern emerging since 2024 [50,51,52].

Thus, archived clinical LC–qToF-MS dataset may provide a chemically resolved historical baseline against which newly emerging adulterants can be retrospectively queried.

4.9. Relationship to Prior Opium/Opiate Metabolomics Studies

Previous metabolomics studies using urine specimens from traditional opium users in the Golestan Cohort Study have provided important reference data for opium- and opiate-associated metabolic perturbations in humans [33,53]. However, the present study differs from the Golestan studies in both study design and intended application. The Golestan studies evaluated community-dwelling adults in Northern Iran, where exposure largely reflected chronic traditional opium use, and opium use or opioid use disorder was defined by self-reported questionnaire or DSM-based interview data. In contrast, our study used clinical specimens from a U.S. hospital-based clinical toxicology population and classified specimens by routine OPIA-, OXY-, and 6MAM-EIA results. Accordingly, Golestan opium users had traditional opium exposure with tobacco-related metabolites and combustion products in Northern Iran, whereas the current study captures prescription opiates (e.g., oxycodone), illicit opioids (e.g., fentanyl, heroin) and various co-exposed substances (e.g., acetaminophen, nicotine, and their metabolites) in the clinical urine specimens in the United States. Thus, the Golestan studies primarily captured metabolic signatures associated with chronic traditional opium exposure, whereas the present study reflects EIA-associated metabolomic signatures of opiate-related and polysubstance exposures encountered in hospital-based urine drug testing in the United States. These studies are therefore complementary; the former provide epidemiologic and mechanistic insight into chronic traditional opium use and opioid use disorder, whereas the present work demonstrates how repurposed clinical LC-qToF-MS data can support biomarker discovery and sentinel surveillance for opiate- and opioid-related drug exposures in real-world clinical specimens.

4.10. Limitations of this Study

This study has several limitations, some of which were partly addressed previously [12].

One is the limited coverage of acidic compounds. The majority of drugs are basic [54], and thus our CUDS is performed predominantly with the positive ESI mode to maximize these drugs. Urine specimens contain numerous acidic metabolites, which are often only weakly ionized under positive ESI mode [55,56]. Thus, our CUDS LC-qToF-MS dataset predominantly covers basic and neutral chemicals, but its coverage for acidic metabolites might be limited.

Unsupervised feature filtering prior to MetaboAnalyst-based statistical analysis is an additional limitation of this study. MetaboAnalyst supports filtering of low-value or low-variance features to improve downstream statistical analysis and reduce the contribution of noisy signals [57]; however, feature filtering in untargeted LC-MS metabolomics can also remove high-quality or biologically informative features if filtering thresholds are not fully optimized for the specific dataset. Because the original dataset contained 14,883 aligned features, filtering was necessary to reduce computational burden and minimize baseline noise. However, some clinically relevant but sparse or intermittently detected analytes expected to be associated with opioid or polysubstance exposure, including morphine-6-glucuronide and xylazine, were excluded before MetaboAnalyst-based statistical analyses. Therefore, absence from the final statistically selected feature lists should not be interpreted as absence in the specimens or as a lack of biological relevance, but rather as a consequence of the combined preprocessing, filtering, and statistical selection workflow [58].

Finally, the number of 6MAM-EIA-positive specimens was smaller than the OPIA-EIA and OXY-EIA positive groups. Class imbalance was addressed before Random Forest analysis using over- and under-sampling methods [12]; however, this approach does not increase the number of independent 6MAM-EIA-positive specimens. Therefore, findings in this subgroup may have lower statistical precision and may be more sensitive to cohort-specific exposure patterns, including polysubstance usage. Thus, the 6MAM-EIA-associated findings should be interpreted as discovery-stage metabolomic signatures requiring validation in larger cohorts.

5. Conclusions

In this study, we identified the features associated with opiate exposures using OPIA-, OXY-, and 6MAM-EIA results by retrospective analysis of our CUDS dataset.

OXY-EIA-positive specimens are enriched for prescribed oxycodone exposure and a chronic pain/pharmaceutical-opioid context, whereas 6-MAM-EIA-positive specimens define a more acute illicit opioid exposure phenotype with stronger polysubstance/recreational-use signatures. OPIA-EIA-positive specimens occupy an intermediate and heterogeneous position, shaped by morphine-calibrated immunoassay reactivity, heroin/morphine biology, fentanyl co-exposure, and real-world opioid polysubstance use. These metabolomic patterns indicate a dominant exposure-gradient model rather than mutually exclusive exposure classes.