**1. Introduction**

In Japan in the years before World War I and continuing through the 1920s, there were human illnesses associated with consumption of moldy, yellow, rice. The illness, classified at the time as Shoshin-kakke (acute cardiac beriberi) decreased in incidence significantly around 1910, a fact attributed to increased inspection of rice by Japanese authorities [1]. The disease was related to beriberi, now known to be caused by thiamine deficiency. By 1930 Shoshin-kakke had almost completely disappeared from Japan [1]. Subsequent investigations in Japan led to the di fferentiation of the broad category of yellow-colored rice into five groups, of which four are caused by Penicillium spoilage and one by *Eurotium amstelodami* [2]. The four caused by Penicillia are each associated with a di fferent species of fungus and a di fferent causative agent. The type of yellow rice known as Ou-hen-mai is infested

with *Penicillium citreonigrum* and has been associated with Shoshin-kakke. In 1964 the structure of the mycotoxin believed to be the causative agen<sup>t</sup> (citreoviridin, CTV) was reported [3]. Another type of yellow rice is Citrinum yellow rice (Citrinum ou-hen-mai) where the causative fungus is *P. citrinum* and the associated mycotoxin is citrinin. In 2006–2008 an outbreak of beriberi occurred in the Maranhão state of Brazil. Despite the presence of a few samples contaminated with CTV, the cases appear to have been predominantly a result of thiamine deficiency, as many were reversed following administration of thiamine [4–6]. An excellent summary of the history of yellow rice and the classification of rice infested with fungi was provided by Kushiro [2].

The connection of Shoshin-kakke to moldy rice has been confounded by the multiple types of "yellow rice", the low incidence of the disease in modern times, and the extent to which thiamine deficiency is required to produce symptoms. In research conducted in the 1960s and early 1970s, crude alcohol extracts of moldy rice were tested in 14 vertebrate species [1,7]. Symptoms included paralysis of the legs, vomiting, convulsions, and respiratory arrest [7]. Purified CTV given to mice, cats, and dogs reproduced these symptoms, with an LD50 of 20 mg/kg in mice [1]. Purified CTV given to mice and rats was lethal, with LD50s ranging from 3.6 to 11 mg/kg [7]. Reproducing symptoms of Shoshin-kakke with purified CTV was important for distinguishing intoxication due to consumption of yellow rice, from disease caused directly by thiamine deficiency. As such, Shoshin-kakke is considered to be a mycotoxicosis [2].

On the molecular level, CTV inhibits the mitochondrial adenosine triphosphatase (ATPase) [8,9]. When given to rats CTV altered the pattern of transketolase (EC 2.2.1.1) in liver, and in vitro experiments suggested an anti-thiamine e ffect of the toxin [10]. A mechanism that involves the exacerbation or causation of thiamine deficiency would be consistent with the involvement of CTV in Shoshin-kakke. Recently, the toxicokinetics of CTV was determined in swine [11]. Results suggested that following oral exposure, CTV was readily absorbed by swine, and slowly metabolized, with a half-life of 21 h [11].

CTV (Figure 1) is produced by a variety of fungi including *P. citreonigrum*, *Aspergillus terreus*, and *Eupenicillium ochrosalmoneum* (actual identity *P. ochrosalmoneum*) [12–14]. The fungi that produce CTV were summarized recently by Peterson et al. [15]. While it is known primarily for its association with rice in Asia, CTV has been found in maize and pecan nuts in the United States [16,17]. It has also been found in rice and wheat-based products in Brazil [18] and in grain dust in Belgium [19].

**Figure 1.** Structure of citreoviridin (CTV) and related compounds.

The structure of CTV was first determined in 1964 [3]. Many of the physical properties of CTV were reported by Ueno and Ueno [7] and were summarized by Cole and Cox [20]. Purified CTV is bright yellow and, in methanol, has absorption maxima in the ultraviolet at 383 nm (ε 44,925), 294 nm (ε 24,725), 285 nm (ε 23,343), 238 nm (ε 10,383), and 203 nm (ε 15,388) [21]. Similar absorption maxima, but lower molar absorptivity coe fficients were reported more recently, also in methanol: 387 nm (ε 31,590), 294 nm (ε 21,960), and 285 nm (ε 20,060) [13]. The latter also reported Fourier transform infrared (FTIR), 1H-, and 13C-nuclear magnetic resonance (NMR) spectra of CTV. Early reports established that CTV was fluorescent [7,22].

CTV has been extracted from commodities and fungal cultures with a variety of solvents, including ethanol (used in the early toxicity testing), chloroform [13,17], dichloromethane [21] and aqueous methanol [23–25]. Before 1988 most analytical methods for detecting CTV relied upon thin-layer chromatography, however, in that year a normal phase liquid-chromatography method with fluorescence detection (LC-FLD) was reported [21]. With a mobile phase of ethyl acetate-hexane (3+1) the fluorescence was detected at 480 nm using an excitation of 388 nm. More recently, normal phase LC-FLD was used to detect CTV in rice [4]. CTV has also been measured using reverse phase LC with photodiode array detection (LC-PDA) [13]. Several laboratories have reported reverse phase LC-tandem mass spectrometry (LC-MS/MS) methods [13]. The molecular ion in positive mode was observed at *m*/*z* 403.2, with the main fragments at *m*/*z* 315 and 139 [13,18]. Detailed mass spectra were reported by Rebuffat et al. [26].

The total synthesis of CTV gave a product that upon exposure to ambient light yielded a mixture of two major components. This led to the realization authentic CTV, which exists as the all-trans form, undergoes photoisomerization, with the product termed "iso-CTV" [27]. The ratio of CTV:iso-CTV has been reported as 7:3 [28] and 3:2 [4,27]. Handling the purified CTV only under red light minimized the isomerization, however handling the toxin under typical laboratory ambient light resulted in the mixture reaching a photostable state within only 1 to 9 h [27]. Even when stored under frozen conditions and protected from light, CTV has been reported to isomerize [13]. As such it should be considered that, under typical laboratory conditions, preparations of "citreoviridin" will likely exist as mixtures that at equilibrium have ratios of CTV:iso-CTV ranging from 1.5:1 to 2.3:1.

Antibodies against CTV have been reported previously. These included polyclonal antibodies [23], a monoclonal antibody (mAb) [24], and a single chain variable fragment (scFv) antibody [25]. While these represented important efforts, the sensitivities of the immunoassays allowed for improvement. In this report we undertook to develop monoclonal antibodies for the detection of CTV, to improve immunoassays for CTV, and to apply such immunoassays to spiked white rice. In an attempt to yield improved antibodies, CTV-protein conjugates were prepared using a different approach from that described previously. Following the development of the mAbs, an interesting effect of the antibodies on the fluorescence of CTV was observed and was likewise studied.

#### **2. Results and Discussion**

#### *2.1. Antibody Development, and Tolerance to Solvents*

Because of its low molecular weight, CTV was first conjugated to bovine serum albumin (BSA) prior to immunization of mice. To facilitate the identification of CTV-binding antibodies, an ovalbumin (OVA) conjugate was also prepared and used as the immobilized antigen in an indirect competitive enzyme-linked immunosorbent assay (CI-ELISA). Previous attempts to produce CTV antibodies either oxidized the hydroxy groups on CTV and then linked them to proteins [23], or formed hemisuccinate derivatives that were then linked to the proteins [24]. Both routes involved linkage through the hydroxy groups of CTV. In our attempt, the linkage was also made through the hydroxy groups, but using a carbodiimide-based chemistry that did not involve oxidation of the hydroxy groups first or require the introduction of a long linker group. When reacted with hydroxy groups, the reagen<sup>t</sup> used, 1, 1-carbonyldiimidazole, forms a carbamate linkage with primary amines of the proteins, adding only one carbon to the length of the linkage [29]. The BSA conjugate (CTV-BSA3) was administered to 10 mice and their sera was evaluated for binding to the OVA conjugate (CTV-OVA2) and for response to free CTV. Two of the sera showed binding to the immobilized antigen (CTV-OVA2) and were selected for sacrifice and splenocyte fusions. No viable products were collected from the first fusion, while the second fusion yielded only five products. Of the five, three were able to bind free CTV. Attempts to subclone these three yielded two stable cell lines designated herein as "2-2" and "2-4". Antibodies

produced by these two cell lines were evaluated in a CI-ELISA format for their sensitivity towards free CTV, cross-reactivity towards small molecules with similar structures, tolerance to solvents, and ability to detect CTV in spiked rice.

The antibodies were relatively sensitive to CTV in buffer (0.1% OVA-phosphate buffered saline (PBS)), with IC50s in the range of 11 to 18 ng/mL. The analytical standard used to determine the IC50s contained predominantly CTV, with a small proportion of iso-CTV (7%, Figure 2A).

**Figure 2.** Liquid chromatography-tandem mass spectrometry (LC-MS) traces of protonated CTV and iso-CTV (*m*/*z* = 403.2). (**A**) The analytical standard used for establishing the response of the competitive enzyme-linked immunosorbent assays (CI-ELISAs) under various solvent conditions (93% CTV, 7% iso-CTV). (**B**) The standard mixture used for examining cross-reactivity to iso-CTV and for spiking rice samples (64% CTV, 36% iso-CTV).

Three compounds having structures that resembled portions of CTV were examined for cross reactivity, including 4,6-dimethyl-α-pyrone (DMP), iso-dehydracetic acid (IDHA), and L-(+)-threose (Figure 1). DMP and IDHA have some similarity to the lactone portion of CTV. L-(+)-threose has some similarity to portions of the substituted furan of CTV. Neither IDHA nor L-(+)-threose were recognized by either of the mAbs, even at levels as high as 500 μg/mL, indicating a cross-reactivity of less than 0.004%. There was a slight inhibition with DMP at 500 μg/mL, indicating a slight cross reaction on the order of 0.04%. Obtaining IC50 data for DMP would have required concentrations above the solubility limit for this compound in the test buffer, hence the upper limit on the cross-reactivity. Of course, the most interesting compound to test for cross-reactivity was iso-CTV. This material was not available commercially as an analytical standard, likely because of the photoconversion between iso-CTV and CTV. However, some insights were obtained by comparing responses of standard curves prepared with a small proportion of iso-CTV (7%, Figure 2A) to those prepared with a higher proportion of iso-CTV (36%) (Figure 2B). The comparison was made by accounting for the total of the CTV and iso-CTV present. As indicated in Figures S1 and S2, when the total of CTV and iso-CTV was accounted for, the calibration curves obtained with both mixtures were nearly superimposable, with nearly identical IC50s. Put another way, if the calibration curves were not corrected for the presence of the iso-CTV, the preparation in Figure 2B (high iso-CTV) gave a much better ELISA response than the preparation in Figure 2A (low iso-CTV). The strong similarities between the two curves, when iso-CTV was accounted for, provided indirect evidence that the antibodies also recognize iso-CTV. Unfortunately, the absence of a relatively pure iso-CTV standard prevented us from establishing this directly.

Citreoviridin is relatively hydrophobic, with a predicted Log P of 3.04, compared to 2.68 for toluene and 3.94 for n-hexane [30]. Because of this, the conditions to extract CTV and iso-CTV from commodities use organic solvents or aqueous mixtures of organic solvents. For this reason, it was

important to determine the impact of solvent concentration upon the CI-ELISAs of the two antibodies. Both antibodies demonstrated good tolerance to methanol and poorer tolerance to acetonitrile (Table 1). There was not a significant impact of methanol until the concentration used to prepare the standards reached 30% (*v*/*v*). At that point, the solubility of the OVA in the test buffer began to fail, and the variability amongs<sup>t</sup> replicates worsened. The proteins began to precipitate at a lower concentration with acetonitrile (20%), which limited the concentration tested here to 15%. For the best results the methanol concentration should be kept at or below 20% and the acetonitrile concentration at or below 10%.


**Table 1.** Effects of methanol and acetonitrile on the CI-ELISAs based upon mAbs 2-2 and 2-4.

a Concentration of solvent in which the CTV standards were prepared (percentage, *v*/*v*). The concentration of solvent in the competition mixture is half of this value. b Number of replicate plates used in the statistic. c 0.1% ovalbumin-phosphate buffered saline (OVA-PBS).

As noted previously, the mAbs 2-2 and 2-4 are not the first antibodies that have been applied to immunoassays for CTV detection. The development of mouse polyclonal antibodies (pAb), mouse mAbs, and a single chain variable fragment (scFv) antibody were reported by researchers at the Fujian Agricuture and Forestry University [23–25]. These are the only known reports of CTV antibodies, and the response of the mAbs 2-2 and 2-4 compared well to them, with lower IC50s and working ranges (Table 2).


**Table 2.** Immunoassays for CTV.

#### *2.2. Application of mAb 2-4 Competitive Enzyme-Linked Immunosorbent Assay (CI-ELISA) to Spiked Rice*

CTV has been extracted from commodities using a wide range of solvents, from pure dichloromethane to 5% methanol [13,17,23–25]. To ensure compatibility with the ELISAs, a mixture of 80% methanol/20% water was used for the extractions reported here. The extracts were filtered and diluted 1:8 yielding test solutions containing 10% methanol, a level within the range that was found to be acceptable in the solvent-tolerance tests (Table 1). Yellow rice has typically been described as rice that has molded following dehulling and polishing. For this reason, the rice used for the recovery experiments was polished, dehulled, white rice. Because it was also expected that yellow rice samples would likely have been exposed to ambient lighting, and CTV is known to form an equilibrium with iso-CTV, a spiking solution was chosen that contained both CTV and significant iso-CTV (Figure 2B). Previous literature indicated that under ambient lighting CTV and iso-CTV reach an equilibrium with a ratio of approximately 1.5:1 to 2.3:1 CTV:iso-CTV. In the recovery studies reported here, the ratio of CTV:iso-CTV was 1.8:1. Rice was spiked with this mixture over the range of 0.36 to 7.24 mg/kg "total" (CTV + iso-CTV). Matrix matched calibration with the same stock solution was used to quantify the concentrations of total CTV + iso-CTV in the spiked rice.

While mAb 2-2 was the more sensitive of the two antibodies, in terms of IC50s (Table 1, Figure S2), it also tended to give greater variability in response when toxin was absent (data not shown). For this reason, mAb 2-4 was selected as the basis of the screening assay for rice. Performance characteristics such as the limit of detection (LOD), IC20, IC50, and IC80 were determined from the calibration curve (Figure 3).

)

**Figure 3.** Calibration curve in rice matrix. The standard used for this curve contained a mixture of CTV and iso-CTV in the ratio of 1.8:1. Data are the mean values from 6 replicate plates with mAb 2-4. Error bars are ± 1 standard deviation (SD). The midpoint was 21.7 ng/mL, equivalent to 694 μg/kg in rice.

The LOD, calculated as the response three standard deviations below the response of the toxin-free samples, was estimated to be 0.13 mg/kg. The IC20 and IC80 were used to establish the dynamic range of 0.23 to 2.22 mg/kg. To improve quantification, spiked samples that contained greater than 2 mg/kg were additionally diluted to keep the resulting signals within the dynamic range. The recoveries observed were excellent (Table 3). From 30 samples covering the range of five spiking levels, the recoveries averaged 96.7% with individual recoveries that ranged from 74.0% to 118.8%. The relative standard deviation was 10.2%. We were not able to obtain any naturally contaminated samples of yellow rice that would permit evaluation beyond our recovery studies. CTV has been found at levels of up to 254 μg/kg in rice bran [4] which is at the lower end of the dynamic range, and in maize kernels at up to 2790 μg/kg [16], which is above the dynamic range of the mAb 2-4 CI-ELISA. Currently there are no regulatory levels established for CTV in commodities or foods, which makes determining a target level at which the assay should perform difficult. However, the data from oral toxicity testing provide a context for what levels may be important in foods. Recently a neurogenesis model was described wherein maternal mice were exposed to CTV through the diet from gestation day 6 through postnatal day 21. Male offspring were analyzed for effects on hippocampal neurogenesis. The no-observed-adverse-effect level was determined to be 1 mg/kg [31]. This level fits within the dynamic range for the CI-ELISA (0.23 to 2.22 mg/kg). Because of this, it appears that the CI-ELISA would be applicable to the screening of CTV and iso-CTV at levels that are toxicologically relevant.

Previous immunoassays for CTV have also reported good recoveries from rice and corn. In those experiments, the extraction solution was either 5% methanol (for rice powder) [23], or 10% to 20% methanol (for corn) [24,25]. Unfortunately, in all three cases the authors reported the spiking levels in units of "μg/mL", rather than in units related to the mass of the food (i.e., μg/kg or mg/kg). The units of μg/mL suggested that perhaps these referred to the spiking levels in liquid extracts, rather than to spiking levels in the solid food, a point that is unclear in the manuscripts. Because of this uncertainty, it is unclear what the real spiking ranges were in the previous work. Despite this, results from all three articles sugges<sup>t</sup> recoveries were good, with average recoveries of 90% [24] 84–90% [25] and 70.5–95% [23]. Given the greater IC50s for the previously reported antibodies (Table 2), it is likely that, on a weight basis, the spiking levels were generally higher than reported here.


**Table 3.** Recovery of a mixture of CTV and iso-CTV from spiked rice.

a Mean recovery ± one standard deviation (SD). b Number of rice samples that were spiked.

#### *2.3. E*ff*ects of mAbs on the Fluorescence of Citreoviridin (CTV)*

Molecular oxygen is known to quench most fluorophores [32]. Local environments that exclude molecular oxygen, for example the binding of a fluorophore within a protein, can increase fluorescence by reducing the inhibition caused by molecular oxygen. The previous sections have established that the mAbs selectively bound CTV. Therefore, we examined whether the fluorescence of CTV was influenced by the presence of mAbs or BSA. The concentrations of CTV and protein chosen for these experiments (1.25 and 2.0 μM, respectively) were selected so as to minimize interference from the fluorescence from the proteins themselves. The results demonstrated that the CTV mAbs significantly enhanced the fluorescence of CTV in aqueous systems (Figure 4).

**Figure 4.** Effects of two CTV mAbs, bovine serum albumin (BSA), and a control mAb on the fluorescence of CTV. ( **A**) Emission spectra with excitation at 420 nm. (**B**) Excitation spectra with emission at 570 nm. The concentration of CTV was 1.25 μM, while the concentrations of the proteins were 2.0 μM. The black and red lines represent CTV incubated with mAb 2-2 and mAb 2-4, respectively. The pink line represents the control of CTV incubated with an anti-deoxynivalenol mouse mAb (mAb 1-6.2.6). The green line represents the control of CTV in bu ffer without added mAb or BSA.

Both the excitation and emission of CTV were manifest as broad peaks. The excitation maximum was 420 nm and the emission maximum 570 nm (Figure 4). BSA also enhanced the fluorescence of CTV slightly. The concentrations of the proteins used in our experiments were relatively low. For immunoglobulin Gs 2.0 μM equates to approximately 0.3 g/L, while for BSA it equates to

approximately 0.13 g/L. For comparison, the reference range for human serum albumin in blood is 35–50 g/<sup>L</sup> (approximately 525 to 750 μM), which is substantially higher. At those levels we suspect that the binding of CTV would also be increased significantly, although quantifying this e ffect would require separating out the fluorescence of the CTV from the previously reported fluorescence of the serum albumin [33]. With regard to the antibodies, the enhancement of CTV fluorescence was seen at levels at which the fluorescence of the antibody did not interfere (e.g., Figure 4). This observation could potentially be used to determine the strength of the binding interaction, something that we have only approximated here through competitive immunoassay. The e ffect might also be used to establish a fluorescence-based immunoassay for CTV, analogous to systems that have been reported for ochratoxin A and zearalenone [34].
