Aristolochic acid (AA) refers to a group of naturally occurring nitrophenanthrene carboxylic acids produced by plant species of the family Aristolochiaceae, and specifically of the Aristolochia and Asarum genera1,2,3,4. Use of AA-containing plants in traditional medicinal preparations date back to the fourth century bc and their use has been documented in the Greco-Roman period5 and eighteenth-century Europe6. They continue to be used in Ayurvedic medicine and traditional Chinese medicine7.

Research in the 1990s and the early 2000s led to the recognition of AA as a potent carcinogen and nephrotoxin, culminating in its eventual classification as a human carcinogen (group 1) in 2012 by the International Agency for Research on Cancer (IARC) following the discovery of DNA adducts and mutations in the tumour suppressor gene TP53 in AA-exposed humans8. In 2001, the US Food and Drug Administration (FDA) issued warnings and an import alert stating that herbal products containing AA are considered unsafe9; this was promptly followed by the implementation of restrictions, withdrawals or import bans on AA-containing raw herbs and herbal medicine products in Germany, the UK, the US, Canada, Australia, New Zealand, Japan, Malaysia, Taiwan, China and Singapore during the early 2000s10,11,12,13,14,15. Despite restrictions, products containing Aristolochia or Asarum plants continue to be used as supplements and remedies for various indications, possibly owing to the strong cultural and historical context associated with their use. Raw herbs and herbal medicine products containing AA remain available for purchase in the US16,17, Martinique18,19, the Netherlands12, Switzerland20, Australia21, Romania22, China23 and Bangladesh24 and are available online16,25 with unrestricted access.

More than 180 naturally occurring AA analogues have been reported, with aristolochic acid I (AA-I) and its demethoxylated derivative, aristolochic acid II (AA-II), being the most common3. AA-I and AA-II have similar genotoxic potential in terms of DNA adduct formation (Fig. 1a), although AA-I is considered solely responsible for the nephrotoxicity associated with AA exposure (also known as AA-associated nephropathy (AAN))26 and is a proven mutagenic carcinogen as discussed below. The mutagenic potential of AA-II appears to be weaker than that of AA-I when studied in cell model systems27 and its carcinogenicity warrants further investigation. Most uses of the term aristolochic acid or AA, unless specified, refer to mixtures of AA-I and AA-II, attributing observations of cytotoxicity, genotoxicity or nephrotoxicity to one or both compounds. Notably, less abundant AA analogues such as aristolactam BI, aristolochic acid D (AA-D) and aristolochic acid IIIa (AA-IIIa) have also demonstrated genotoxicity and cytotoxicity in vitro3, and warrant further investigations.

Fig. 1: Mechanistic underpinnings of the mutagenicity of AA.
figure 1

a | Aristolochic acid I (AA-I) and aristolochic acid II (AA-II) are enzymatically activated and converted to cyclic nitrenium/carbenium ions that form covalent DNA adducts with adenine (deoxyadenosine (dA)) and, to a lesser extent, guanine (deoxyguanosine (dG)) bases. dG–AL-I and dG–AL-II adducts are repaired by both global genome nucleotide excision repair (GG-NER) and transcription-coupled nucleotide excision repair (TC-NER). dA–AL-I and dA–AL-II adducts on the transcribed strand are repaired by TC-NER and, to a lesser extent, by GG-NER. However, dA–AL adducts on the untranscribed strand entirely evade TC-NER, resulting in the increased formation of A:T to T:A transversions characteristic of aristolochic acid (AA) exposure. b | Top panel: AA mutational signature, also known as SBS22, characterized by predominance of A:T to T:A substitutions, particularly in 5′-PyrAPur-3′ contexts. The canonical, more commonly used 96-channel histogram is shown, with C > A denoting C:G to A:T substitutions, C > G for C:G to G:C, C > T for C:G to T:A, T > A for T:A to A:T, T > C for T:A to C:G and T > G for T:A to G:C. Bottom panel: extended depiction of the signature indicating transcriptional strand asymmetry (bias) in genic regions. A strong preference of A:T to T:A mutations to the transcribed DNA strand indicates the compromised TC-NER of pre-mutagenic bulky DNA adducts on adenine residues located on the coding strand. Reference mutational profiles are as described by the Catalogue of Somatic Mutations in Cancer (COSMIC)202, including further details on the AA signature SBS22. AL, aristolactam. Part b, image courtesy of COSMIC Wellcome Sanger Institute.

Here we review the evidence that led to the establishment of AA as a pervasive nephrotoxin and mutagenic carcinogen, from early epidemiological clues to the discovery of the genome-scale AA mutational signature in numerous cancer types. We also discuss the apparent tissue-specificity of AA-associated carcinogenicity, the challenges of assessing the scale of AA-related toxicity and the obstacles that must be overcome to effectively identify and limit sources of AA exposure.

AA-induced nephropathy and cancer

Evidence from human studies

Modern-day records of Aristolochia-related toxicity in humans began in the 1960s, when ten cases of acute renal failure were attributed to Aristolochia manshuriensis exposure by Wu28, Hong et al.29 and Zhou et al.30, as cited in Poon et al.31. An overlooked link between AA exposure and nephrotoxicity was made in a phase I clinical trial conducted in 1964 to evaluate AA-I as an anticancer agent32 following observations made by the US National Cancer Institute about AA-I having antitumour activities in experimental models33. Drug development efforts were terminated owing to the profound nephrotoxicity of AA-I in patients with advanced tumours.

In 1969, Ivić hypothesized that AA was the causal factor underlying Balkan endemic nephropathy (BEN), a progressive renal disease first identified in communities around the Danube River and its tributaries in present-day Serbia, Bulgaria, Romania, Bosnia and Herzegovina, and Croatia, which came to be known as the ‘endemic region’. BEN is strongly associated with urothelial carcinoma of the renal pelvis and the upper ureter, and is hypothesized to arise in humans due to the ingestion of bread made from Aristolochia clematitis-contaminated wheat flour34.

The safety of Aristolochia-containing products came under international scrutiny in the early 1990s, as reports emerged linking the accidental administration of supplements containing Aristolochia fangchi to approximately 1,800 healthy Belgian women to more than 100 cases of a unique tubulo-interstitial renal disease. Initially described as Chinese herbs nephropathy, the disease was characterized by extensive renal injury and hyperplasia of the urothelium, and rapid progression to end-stage renal failure35,36,37,38. In a study of 39 of these patients, 18 cases presented with upper-tract urothelial neoplasia, whereas 19 of the 21 remaining patients without carcinoma were diagnosed with urothelial dysplasia39,40,41. This unusually high prevalence of urothelial neoplasia in patients with AAN contributed to the IARC Monographs Working Group’s conclusion of sufficient evidence for human carcinogenicity of AA8.

Cases of pathologies consequent to the intake of Aristolochia-containing herbal remedies later emerged in other countries, including urothelial malignancies in China42 and nephrotoxicity observed in China43,44,45, Spain46, France47, Germany48, Japan49, the US50 and Korea51. These findings were supported by mechanistic evidence providing a strong and probable causal link between AA exposure and urothelial malignancies and nephropathy39,52,53,54,55,56,57,58,59, then termed AAN60,61 (Box 1). For example, aristolactam (AL)–DNA adducts originating from metabolites of AA-I and AA-II (see below and Fig. 1a for a description of AA metabolism and AL–DNA adduct formation) were detected in the kidneys and ureters of patients with documented AA exposure. In parallel with these early epidemiological findings, investigators noticed striking clinical and morphological similarities between AAN and BEN38. Reports demonstrating the presence of AL–DNA adducts in the renal cortex of patients with BEN and in associated urothelial tumours, and their absence in patients from non-endemic regions, pointed to AA as a causative agent of BEN62,63,64,65. These findings corroborated the earlier hypothesis by Ivić on the AA exposure route for patients with BEN, and established BEN as the environmental form of AAN34,62,66, resulting from chronic exposure to low doses of AA in contrast to the more acute exposure and higher doses involved in the iatrogenic contexts.

In Taiwan, the widespread use of AA-containing herbal products prior to 2003 and a policy of systematically recording prescriptions for herbal remedies provided a unique opportunity for assessing the pathological effects of AA. Studies using data from this period showed increased risks of urinary tract cancer, kidney failure and renal cell carcinoma (RCC) associated with consumption of AA-containing herbal products67, with local case–control studies and molecular investigations in the region providing further evidence57,68,69,70. Further studies showed that Taiwanese traditional herbalists had a greater standardized mortality ratio for urological cancers compared with the general population, suggesting that occupational exposure can increase the risk for urological cancers71. This was supported by a separate study showing that the standardized incidence ratio of urological cancer in traditional herbalists in Taiwan was significantly higher compared with the general population72. The authors of this study postulated that the herbalists could have been ingesting herbal preparations or may have inhaled and swallowed ground particles of AA-containing herbs during processing. This hypothesis was supported by two case–control studies of Chinese herbalists, which showed that a history of processing, selling or dispensing herbal medicines containing the AA-containing drug fangchi (also known as fangji) is associated with a significantly increased risk of kidney disease73 and urothelial carcinoma74.

A retrospective, population-based cohort study in patients infected with hepatitis B virus (HBV) in which the authors estimated the cumulative dose of AA for each subject based on prescription records revealed a significant dose-dependent relationship between AA consumption and hepatocellular carcinoma (HCC) risk75. The authors also reported higher incidences of chronic kidney disease and upper tract urothelial carcinoma (UTUC) and bladder cancers in patients regardless of HBV infection status, corroborating previous reports. A similar prospective study involving patients infected with hepatitis C virus (HCV) reported that patients exposed to AA had an increased risk of developing liver cancer76. Most recently, a large epidemiological study of more than 800,000 patients with type 2 diabetes of whom 37,554 men and 31,535 women had been diagnosed with cancer revealed that the documented use of AA-containing herbal products was significantly associated with a higher, dose-dependent risk of cancer of the liver, colorectum, kidney, bladder, prostate, pelvis and ureter77, with an increased risk of extrahepatic bile duct cancer observed among women exposed to higher doses of AA77.

The genotoxicity of AA in humans became apparent when multiple reports observed atypically enriched A:T to T:A transversions in AA-affected tissues, a mutation pattern then recognized as one of the hallmarks of AA exposure57,59,62,63,78 (see Box 2 and Table 1) and later characterized as a key feature of a highly specific genome-wide mutational signature (Box 3 and Fig. 1b, discussed below). The causal role of AA in inducing AAN has subsequently been validated in experimental studies revealing the key mechanisms of AA-induced toxicity while also allowing for improved sample processing and analytical techniques for the detection of AA-induced mutations and DNA adducts.

Table 1 Summary of evidence for AA-induced carcinogenicity in humans

Evidence from animal carcinogenicity studies

Animal exposure studies provided an early indication of the carcinogenicity of AA and were considered in the IARC’s evaluation of AA carcinogenicity in humans8. Most studies involved the exposure of various strains of mice and rats to extracts from Aristolochia spp., to a mixture of AA-I and AA-II or to AA-I alone (Table 2). Significantly increased tumour incidence was observed for forestomach, kidney and lung carcinomas in mice upon oral administration of AA79, and for forestomach, renal pelvis and intestine carcinomas in exposed rats80,81,82,83. Other instances of cancer formation were reported in animals administered with AA through oral or subcutaneous routes, including urinary bladder, ear duct, thymus, pancreas and urothelial carcinomas and fibrohistiocytic sarcomas in rats, and cancers of the kidney, urothelium and peritoneal cavity in rabbits injected peritoneally with AA, as reviewed by the IARC Working Group on the Evaluation of Carcinogenic Risks to Humans8. A recent study in mice revealed that AA causes liver tumours with molecular features essentially identical to those found in human liver tumours associated with AA exposure84.

Table 2 Summary of evidence for AA-induced carcinogenicity in animals

AA-induced genotoxicity and mutagenicity

The nephrotoxic, genotoxic and cytotoxic properties of AA have been extensively and systematically reviewed2,8,85,86. Here, we review key findings on the mechanisms underlying AA-induced genotoxicity and mutagenicity (summarized in Table 3) and the evidence for differentiating these mechanisms from those of AA nephrotoxicity (Box 1).

Table 3 Summary of experimental evidence for AA-induced genotoxicity and mutagenicity

The mutagenic nature of AA is inextricably linked to AA-derived AL–DNA adducts (Fig. 1a). These adducts form following the bioactivation of AAs in target organs and the liver to active AA metabolites87,88,89,90,91,92,93, which undergo decomposition to cyclic carbenium–nitrenium ions94 that bind DNA covalently. DNA adduction sites of AA are preferentially located on the exocyclic amino groups of adenine and, to a lesser extent, guanine bases, and adduction is enhanced in both cases when the sites are flanked by pyrimidine nucleotides. Deoxyguanosine–AL (dG–AL) and deoxyadenosine–AL (dA–AL) adducts cause the incorporation of deoxyadenosine monophosphate (dAMP) during DNA synthesis, resulting in G:C to T:A and A:T to T:A substitutions, respectively95,96. Although dG–AL adducts eventually disappear from rodent and human tissues, dA–AL adducts can persist for more than 20 years after exposure to AA has ceased55,97,98,99 owing to their reduced susceptibility to DNA repair (Fig. 1a). The dG–AL adduct is recognized by Xeroderma pigmentosum group C (XPC) protein and removed by global genome nucleotide excision repair (GG-NER), which monitors transcribed and untranscribed DNA strands. This adduct can also be removed from the transcribed DNA strand by transcription-coupled nucleotide excision repair (TC-NER) when encountered by the transcription machinery during RNA synthesis. By contrast, the dA–AL adduct evades GG-NER and is removed mainly from the transcribed strand by TC-NER (Fig. 1a). This model has been built on mechanistic in vitro studies that revealed that dG–AL adducts induce structural disturbances to the DNA that are recognized by XPC, whereas dA–AL lacks these characteristics and escapes lesion recognition unless placed in the mismatched DNA bubble100,101. Translesional DNA polymerases are important for the generation of 5ʹ-pyrimidine-A-purine-3ʹ (5ʹ-PyrAPur-3ʹ) mutation hotspots as bypass of the dA–AL adduct by DNA polymerase-ζ (Polζ) strongly associates with the sequence context surrounding the adduct and with the conversion of 5′-CAG-3′ to 5′-CTG-3′ (ref.102). Activation and distribution of AA species in circulation to different organs, sequence-specific adduct formation, differential repair by NER pathways and translesional DNA synthesis across dA–AL adducts contribute to the formation of the hallmark AA mutational pattern, which is characterized by A:T to T:A substitutions mainly in the 5ʹ-PyrAPur-3ʹ context, with the mutations preferentially located in the untranscribed (coding) DNA strand of genic regions — a feature underlying the genome-wide mutational signature of AA57,62,78,103,104,105,106 (Box 3 and Fig. 1b).

AA mutational signature in human cancers

Advancements in genomic sequencing and computational algorithms led to the identification of the AA mutational signature (Box 3 and Fig. 1b), which has been detected in numerous human cancer types107 including urothelial103,106,108,109,110,111, renal cell70,112,113,114, hepatocellular11,84,103,115, biliary116,117 and oesophageal84,118,119 tumours (Fig. 2).

Fig. 2: Summary of all whole-genome or whole-exome sequencing reports discussed in this Review.
figure 2

Proportion of tumours reported to harbour aristolochic acid (AA) mutational signature (red bubbles). Data included for renal cell carcinoma (RCC)112, oesophageal squamous cell carcinoma (ESCC)84,119, hepatocellular carcinoma (HCC)11,84,103,126,128, intrahepatic cholangiocarcinoma (ICC)116,117, upper tract urothelial carcinoma (UTUC)109,123 and bladder cancers110,123. Total number of tumours indicated for each tumour type. Studies in which participants had been preselected for likely or confirmed AA exposure and studies in which the number of observed AA-associated cancers was not explicitly reported were excluded from this analysis.

Urothelial cancers

The first description of the genome-scale AA mutational signature in human cancers was made in tumours of Taiwanese patients with UTUC103,106. Exome sequencing of UTUC cells from individuals with documented AA exposure showed that strand-biased A:T to T:A transversions in the 5′-PyrAPur-3′ context were predominant in the cancers of 17 of the 19 studied individuals106. Poon et al.103 similarly profiled eight AA-associated UTUCs and observed a strikingly high proportion of A:T to T:A transversions in the characteristic trinucleotide context and with transcription strand bias.

Improved techniques applicable to formalin-fixed paraffin-embedded (FFPE) samples allowed for a retrospective mutational signature analysis of archived pathological specimens108. This analysis showed the presence of the AA mutational signature in 77% (10/13) of studied tumours from cases of BEN and a lack of this signature in non-BEN tumour samples. Interestingly, tumour pairs from the same patient arising in different parts of the urinary tract shared overlapping mutation patterns, suggesting they had originated from a common precursor108 — a finding later corroborated by a study of Chinese patients with AA-associated recurrent urothelial cell carcinomas120.

In a study on UTUCs from patients treated in China, 30% (27/90) of the analysed tumours harboured the AA mutational signature109. The authors had not explicitly selected for patients with documented or suspected exposure to AA, which explains the lower — albeit still substantial — occurrence rate. Notably, the AA mutational signature was detectable in morphologically normal urothelium specimens of patients with multifocal disease, suggesting that AA exposure may contribute to field cancerization.

A study by Poon et al.110 detected the AA mutational signature in two bladder tumours obtained from patients with documented exposure to AA. Sequencing bladder tumour cells from a group of Singaporean patients where no information on AA exposure was available showed the AA mutational signal present in only 1 out of 11 tumours; data from a publicly available dataset revealed that the signature was also present in 9 of 99 and 1 of 194 bladder tumours from patients treated in China121 and North America122, respectively.

In a study from China, the AA mutational signature was detectable in 36 morphologically normal samples and 40 tumour samples among a collection of 287 samples collected from 120 patients with urothelial cancer123. By contrast, a similar analysis by Lawson et al.124 did not detect the AA mutational signature in 2,097 micro-biopsies of histologically normal bladder from 20 individuals residing in Europe.

Renal cancers

Molecular profiling of clear cell renal cell carcinoma (ccRCC) from the Czech Republic, the UK, Romania and Russia reported striking differences in somatic mutation frequencies between patients of the listed participating countries112. In this study, increased levels of A:T to T:A transversion within the 5′-PyrAPur-3′ trinucleotide context on the untranscribed strand were observed exclusively in tumour samples originating from Romanian patients112. Unexpectedly, none of the patients resided in the endemic region of Romania112,113. Assessment of non-tumour renal cortices from the above patients and 15 cases from the UK, the Russian Federation and the Czech Republic113 revealed the presence of AA-associated DNA adducts in all of the Romanian cases and not in the other cases, indicating the existence of an unknown, possibly iatrogenic, source of AA exposure in Romania113. In a separate cohort of patients with either ccRCC or chromophobe RCC from the BEN endemic region, 62.5% (five of eight) of the tested FFPE tumour specimens harboured the AA mutational signature based on strand-biased A:T to T:A mutations in the 5′-PyrAPur-3′ context114. By contrast, unremarkable A:T to T:A mutation rates were observed in control cases from the non-endemic, metropolitan area of Croatia.

A study on Taiwanese patients with RCC detected AA-derived DNA adducts in the non-neoplastic renal cortical tissue of 76% (39/51) of patients, supporting previous assertions of widespread AA exposure amongst the Taiwanese population and AA-associated DNA damage occurring in non-tumour tissue125. Exome-wide sequencing of RCC tumours from ten of the patients exposed to AA revealed that the fraction of strand-biased A:T to T:A mutations in the 5′-PyrAPur-3′ context and a bias towards splice acceptor (CAG-based) sites were significantly higher in the AA exposure-associated tumours compared with ccRCCs of The Cancer Genome Atlas (TCGA)70.

Hepatocellular cancers

The AA mutational signature was first reported in HCCs by Poon et al.103; analysis of a published dataset of HCC genomic data126 revealed the AA mutational signature in 11 of 93 combined HCC genomes and exomes103,126. It was further described by Ng et al.11, who reported that 78% (76/98) of HCC samples collected from two Taiwanese hospitals harboured the signature. The authors mined public data from 1,400 HCC samples from China, Japan, Korea, southeast Asia, North America and Europe, and reported marked disparities in occurrence rates of the AA mutational signature across geographical boundaries, ranging from 56% of cases from southeast Asia (5/9 cases) to 1.7% (4/230) of cases from Europe11. These findings were corroborated by Letouzé et al.115, reporting that the AA mutational signature was present in ≤5% of 44 tumours sampled from patients treated in France and 264 HCC samples from the International Cancer Genome Consortium (ICGC) Japan series. A separately conducted reanalysis of public HCC datasets reaffirmed these findings by reporting the AA mutational signature in 26% (133/510), 44% (4/9) and 7% (16/231) of tumours from patients treated in China, Singapore and Korea, respectively84. By contrast, the signature was observed in <1% of HCCs from patients treated in Japan and France. Notably, in HCCs derived from patients treated in the US127, Lu et al.84 observed higher rates of detection for the AA mutational signature in patients of Asian descent (24/160, 15%) compared with non-Asian patients (5/204, 2.5%).

The AA mutational signature has also been reported in 9% of tumours (7/76) studied in Mongolian patients with HCC128, and in three of five morphologically normal liver samples from organ donors recruited in China129. In a study of HCC patients from four countries in the Asia-Pacific region (Singapore, Thailand, Malaysia and the Philippines), the AA signature accounted for 8% of single-base mutations130. Interestingly, the detected AA mutational signature was often shared by multiple samples of the same tumour, suggesting that these mutations are early events in HCC tumorigenesis130.

Biliary tract cancers

A study of 103 patients with intrahepatic cholangiocarcinoma (ICC) treated in China reported a pattern of A:T to T:A mutations117 occurring preferentially on the untranscribed strand, reminiscent of the AA mutational signature. A subsequent study of 803 patients with biliary tract cancer, including 92 from the dataset described by Zou et al.117, reported the AA mutational signature in 35.8% (53/148) of cases, indicating that AA exposure may be implicated in the aetiology of ICCs116. It remains to be determined whether similar findings result from other cohorts — including those from other geographical regions — and whether the pervasiveness of the AA mutational signature in hepatobiliary tumours tracks the likelihood of exposure to AA as seen for UTUCs.

Gastrointestinal tract cancers

Detection of the AA mutational signature in gastrointestinal tract cancers is rare. A study of oesophageal tumours from China using the ICGC dataset reported the signature in <1% of cases84 and another study involving 508 Chinese patients with oesophageal squamous cell carcinoma (ESCC) found ~20 (~4%) AA signature-positive tumours119. Similarly, a recent study of 552 ESCC cases from eight countries including China detected the signature in less than 5% of cases118. Further, a study of 54 oro-gastrointestinal cancers from Taiwan found the signature in a single case of gastric adenocarcinoma131. Potential AA-related mutations have been detected in a study investigating 1,737 morphologically normal tissue biopsies of 9 organs from 5 donors129. The AA signature was observed in morphologically normal oesophagus, duodenum and stomach tissue in two donors for whom AA mutations comprised ~50% and ~75% of liver mutations, respectively; in these cases, the AA signature accounted for <10% of the mutations in these tissues129.

Tissue-specific roles of A:T to T:A mutations in carcinogenesis

The genome-scale AA mutational signature has been observed by independent research groups across the globe and in multiple cancer types (Fig. 2), with findings validated by controlled exposure experiments in experimental models27,84,103,132,133 (Table 3) analysed by next-generation sequencing and dedicated computational data analyses134,135,136,137.

It is an important question whether AA-induced mutations could act as an initiating factor for cancers where they occur in cancer driver genes or as a bystander in other cancer sites where mutations in driver genes are not observed89. Indeed, although AA signature mutations have been described in RCC, they make up the minority of the mutations found in the PBRM1 gene and none were detected in VHL, the two main driver genes in renal cancers112. Hoang et al.70 reported infrequent A:T to T:A mutations in PBRM1 and SETD2 — another driver gene of renal cancers — in cancer tissue of patients with renal cancers harbouring the AA mutational signature. By contrast, numerous cancer drivers affected by A:T to T:A were observed in UTUCs106,108,109,138, morphologically normal urothelium123 and liver cancers11,130, suggesting that AA exposure could be causal in these cancer types (see Box 2).

Additional biological and chemical factors may contribute to AA causal effects. For example, the risk of primary liver cancer due to AA exposure has been reported as significantly elevated in patients infected with HBV or HCV75,76. Furthermore, a recent study conducted in areas of Taiwan with endemic arseniasis suggested that combined exposure effects of AA and arsenic contributed to the risk of UTUC development139.

The findings of the genome-scale studies described in this section independently validate earlier observations of the AA mutational pattern, adding new information on additional AA-related cancers and the potential causal role of AA in certain cancers. Together, these observations further underscore the importance of identifying and curtailing routes of exposure to AA.

Monitoring AA sources and biomarkers

Detecting AA in live plants or plant products

Given the evidence that AA exposure can occur through iatrogenic and environmental routes, detecting AA in processed or dried herbal products — as well as fresh plants cultivated in locations with suspected AA contamination — can alert the relevant authorities or communities to the potential risks of AA exposure. Moreover, as the chronic effects of AA only manifest decades after exposure, biomonitoring of AA in at-risk populations could allow for early identification of individuals exposed to AA who would benefit from early interventions. Methods for detecting and quantifying AAs in plants and herbal products include those based on capillary electrophoresis, enzyme-linked immunosorbent assays and high-performance liquid chromatography (HPLC)140. HPLC is the most widely used method for analysing plant material, suspected AA-containing products or field samples in the absence of consensus guidelines for AA detection140. Similarly, HPLC-based methods with low-nanomolar sensitivity are appropriate for the quantitative detection of AA as an adulterant or contaminant in herbal products141. Some of these assays are conducted following reductive conversion of AA to fluorescent ALs to improve detection sensitivity.

The above analytical chemistry methods can be confounded by inherent intra-species variation in AA concentrations, in which seasonal or environmental conditions could cause some plant products to have concentrations of AA below the detection limit of the assays used3,4,142. To account for variation, DNA-based approaches present an alternate or complementary strategy for detecting the genetic material of plant species known to produce AA. These assays rely on DNA barcoding to discriminate plant species, relying on relative DNA content stability across plant populations of the same species142. DNA barcoding assays are based on high levels of inter-species variation of polymorphic DNA regions in the plant genome, the analyses of which allow identifying distinct plant species143.

Sequencing-based and PCR-based assays14,143,144,145,146 have been evaluated for the identification of material from several Aristolochia and Asarum spp. These assays can identify minute traces of target plant species from fresh plant material, raw herbs or minimally processed products. DNA-based methods are most useful when used in combination with appropriate chemical methods for identifying whether AA-containing species are present within tested products147, although it should be noted that they are unsuitable for analysing processed products such as tinctures and extracts, where DNA is often absent, degraded or present in insufficient quantities142,148.

Biomarkers of AA exposure in humans

AL–DNA adducts are robust biomarkers of past AA exposure due to their persistence in human tissue39,54,97. AL–DNA adducts can be detected from isolated DNA using 32P-post-labelling methods57,62,63,105, although these require potentially hazardous amounts of radioactive phosphorus and do not reveal the specific chemical identities of the analysed adducts149. Replacement techniques use ultra-high performance liquid chromatography (UHPLC) or HPLC coupled with electrospray ionization multistage mass spectrometry, which allow for sensitive and precise AL–DNA adduct identification58,62 even in FFPE tissues111,150,151,152, thereby superseding the post-labelling methods that rely exclusively on fresh frozen tissue151.

Sequencing-based methods can be used for AA exposure biomonitoring. Current techniques allow for mutational signature analyses (Box 3) in FFPE tissues108,109,114, enabling systematic screens for AA exposure fingerprints in archived tumour samples from patients with cancer and thus the identification of at-risk populations. Furthermore, rare mutational signatures present in non-clonal tissue can be detected using next-generation sequencing technologies such as duplex sequencing and its variants125,153,154,155,156 as error rates are below the typical mutational load in human tissues155. Using the next-generation sequencing method BotSeqS, Hoang et al.125 found that A:T to T:A mutations made up a higher fraction of detected mutations in renal cortical tissue samples from individuals exposed to AA (29%) than in normal kidney samples (9%) and samples from heavy smokers (2%). NanoSeq, a duplex sequencing modification, further improves error rates153, allowing for reliable detection of rare mutations in various non-clonal sample types.

AA-exposure detection techniques can be effectively applied to non-invasive screening tests. Chromatography-based methods have been used to detect AL–DNA adducts in the DNA of exfoliated urinary cells149,157,158, and AA–DNA159 and AL–RNA160 adducts in cell-free urinary DNA. A non-invasive, sequencing-based test for urothelial neoplasms (UroSEEK) detected genetic abnormalities in the DNA of exfoliated urinary cells161, including a high proportion of AA mutations in the urinary cells of patients with UTUC161. The AA mutational signature has also been detected in urinary cell-free DNA, establishing a promising new screen for AA exposure, AAN or AA-associated cancers109. It should be noted that these assays are yet to be systematically evaluated in a clinical setting and it is unclear how the results of these assays are influenced by renal function, cumulative dose and time of exposure to AA149,158.

Caution is advised when interpreting mutation spectra and signatures from sequencing data, particularly if considering a single predominant mutation type162. Experimental exposures to dibenzo[a,l]pyrene, glycidamide, urethane and 7,12,-dimethylbenz[a]anthracene conducted in vitro or in rodents result in A:T to T:A-enriched mutation spectra similar to that of AA27,163,164,165,166. Additionally, new A:T to T:A-rich mutational signatures have been reported sporadically in human tissues137,167; however, unlike the AA-specific pattern (Fig. 1b), these signatures harbour contributions from non-A:T to T:A substitutions and can be mathematically separated from the AA signature. Mutational signature analysis must be always conducted with rigour to identify the AA-specific signature with all its specific features and to distinguish it unambiguously from similar patterns originating from unrelated mutagenic sources.

Excreted microRNAs can be used to discriminate between AA-related cancers, non-AA-related cancers and normal tissue. Two studies have demonstrated distinct microRNA signatures in BEN/AAN-associated UTUCs168,169, non-AA-associated UTUCs and non-tumour tissue. Partially overlapping microRNA signatures have also been observed between BEN-associated UTUCs and matched urine that could potentially allow monitoring of tumour presence and recurrence in the urinary tract138.

Impact of reducing AA exposure

Two decades have elapsed since the first regulatory measures on AA-containing products were imposed, allowing for investigations of their impact on disease trends. Records held by the Taiwanese National Health Insurance Research Database show a decline in the standardized incidence ratios of UTUC in both men and women starting from the year 2000, possibly relating to the reduced consumption of AA-containing herbal products in the aftermath of the 1990s AAN outbreak in Belgium170. Similarly, a study of the Taiwan Cancer Registry revealed statistically significant changes to trends in age-standardized incidence rates of bladder cancer, kidney cancer and carcinomas of the renal pelvis and other urinary organs between 1995 and 2013 (ref.171). For all these cancer types, an initial change of slope corresponded to the introduction of regulatory controls on AA-containing herbal products and a second change of slope was hypothesized to occur following the implementation of a nationwide ban on most AA-containing herbal products in 2003.

Intriguingly, Ng et al.11 reported no significant difference in the prevalence of the AA mutational signature or the number of AA signature mutations between patients treated for HCC in Taiwan before and after 2003 (ref.11). The authors suggest that a decline in AA-associated HCCs could lag behind AA exposure reduction as AL–DNA adducts are extremely persistent and exposure to AA years or decades prior may induce the delayed formation of the AA mutational signature. Limitations in the regulations used to reduce actual AA exposure could also explain this result.

Pervasiveness of AA exposure

AA-containing products remain available on the market despite government warnings and the evidence of AA-associated toxicity. Continued demand may stem from a belief that traditional medicine has a long history of apparently safe use6. The demand for AA-containing products and their perceived pharmaceutical value is such that the detoxification of AA-containing products has been proposed to allow for their safe use, as reviewed by Fan et al.172.

Chronic exposure to AA of environmental origin has been well described owing to the prevalence of BEN173,174. Although disease-causing environmental AA exposure has not been reported elsewhere — aside from a putative link to Meso-American nephropathy (MeN) (Box 1) — AA-containing plants of the genera Aristolochia and Asarum grow as weeds in continental southeast Asia, China, tropical Africa, Oceania and the Americas175,176 (Fig. 3a and Supplementary information). An example of possible environmental contamination is a community garden in France recurrently overgrown by A. clematitis (Fig. 3b). AA has been detected in soil177 and groundwater178 samples collected from the endemic region and both AA-I and AA-II can resist degradation by soil microorganisms for at least a month179, and in groundwater samples for at least 2 months178. Further, food crops cultivated in AA-contaminated soil and water may bioaccumulate AA and, inadvertently, contaminate the food chain141,177,179,180,181. AA surveillance programmes therefore need to be conducted on foodstuffs in addition to herbal medicinal products. Given the environmental persistence of AA, further research is needed into effective strategies for bioremediation of AA-contaminated farmlands and water sources182, and future investigations are warranted to determine the overall effectiveness, feasibility and safety of such approaches.

Fig. 3: Global observations of AA-associated diseases and cancers and the global distribution of Aristolochia.
figure 3

a | Global geographical distribution of Aristolochia spp. — a major genus of aristolochic acid (AA)-containing plants — illustrated as a possible source of environmental exposure. Locations where records of Aristolochia have been deposited in the Global Biodiversity Information Facility (GBIF) are shaded yellow (see Supplementary Fig. 1 for full details). Pink areas indicate countries where cases of AA-associated cancers have been described; blue areas indicate countries where AA-associated nephropathy (AAN) and AA-associated mutagenesis, but not AA-associated cancers, have been described. b | Aristolochia clematitis, previously linked to endemic nephropathy and AA-associated cancers in the endemic regions, photographed in 2021 as a weed that massively and recurrently contaminates a community garden in Ardèche, France, illustrating a possibly common route for environmental exposure to AA. Part b, image courtesy of Ghislaine Scélo.

Conclusions and perspectives

Currently available epidemiological and mechanistic data provide sufficient evidence for AA nephrotoxicity and carcinogenicity. However, broadly orchestrated programmes to curtail existing routes of AA exposure are lacking. Inaction persists despite worldwide occurrence of Aristolochia and Asarum spp., and the continued demand for and availability of AA-containing products. In our view, this results from several unaddressed questions. First, the true environmental pervasiveness of AA is unknown. The studies demonstrating persistence of AA in groundwater, soil and food crops141,175,177,179,182,183 have been limited to small geographical areas. It remains unclear whether soil and groundwater AA contamination reflects the distribution of AA-containing plants across the globe. Second, the true scale of iatrogenic exposure to AA is unknown. Even in Taiwan, where prescriptions for herbal medicine are systematically monitored, it is estimated that half of AA exposure occurs following undocumented over-the-counter or Internet purchases, or through unknown sources139. It is unclear how effective the implementation of restrictive measures has been in preventing iatrogenic AA exposure as herbal products are not conveniently classified by plant family, complicating product identification and the enforcement of regulatory measures. Indeed, herbs containing AA are difficult to identify without specialized knowledge or laboratory analysis and may be confused with innocuous herbs that bear similar common names, leading to their inadvertent sale or use184. Along with increased pharmacovigilance, the reasons behind the continued demand for AA-containing products must be addressed by improved communication between the scientific community and the public, as well as organizations representing practitioners of traditional medicine185. Third, the overall impact of AA on human health and the pervasiveness of AA-associated disease are not fully mapped. The list of AA-associated cancers continues to grow thanks to genome profiling studies conducted in previously under-represented populations. Although the emergence of multiple studies from East Asia may lead to the perception that the problem of AA is restricted to that geographical location, the AA mutational signature has also been observed in ccRCCs from Romanian patients112,113, liver tumours from patients treated in France115, cirrhotic liver samples from the UK186 and TCGA HCC data from the US84 (Supplementary information).

Sequencing more populations and tumour types will give a more in-depth understanding of AA-associated diseases, their local sources and exposure routes and levels. An important step to achieve this will be to systematically screen individuals at risk of AA exposure using non-invasive methods or archival tissue samples if available, for example from individuals diagnosed with AA-associated diseases. Identified individuals exposed to AA could be further examined to identify previously unknown environmental or iatrogenic sources of AA. Considering that patients with AAN face an elevated lifetime risk of developing urothelial cancers41,111, effective identification of individuals at high risk could allow for the implementation of potentially life-saving prophylactic interventions or personalized screening187.

It remains unclear how previous exposure to AA might influence clinical management of AA-associated cancers. Reports on the potential effect of AA exposure on the survival of patients with UTUC have been contradictory109,188. Importantly, cancers harbouring the AA mutational signature have higher mutation loads and increased numbers of neoantigens11,109 and may be amenable to immune checkpoint blockade therapy. To our knowledge, no other reports investigating the impacts of AA exposure on clinical outcomes and management have been published.

Overall, there has been increased recognition of the nephrotoxicity and carcinogenicity of AA, particularly in East Asia where risks of iatrogenic exposure are high. Work remains on building community awareness of the hazards of AA exposure, both at the local and global levels, and understanding the pervasiveness of AA in daily life and its full impact on human health.