The Chemopreventive Efficacies of Nonsteroidal Anti-inflammatory Drugs: The Relationship of Short-term Biomarkers to Long-term Skin Tumor Outcome

Exposure to UV light is the major cause of nonmelanoma skin cancer (NMSC) in humans (1). Intense UV exposure is usually marked by an inflammatory response and subsequent hyperplasia. Although the mechanism(s) is not entirely clear, it has been well established that UV elicitation of the prostaglandin products of arachidonic acid plays a significant role in both UV-induced inflammation and UV-induced skin tumors (2, 3). The enzymes responsible for the production of prostaglandins are referred to as the prostaglandin H (PGH) synthases, of which there are 2 isoforms, COX-1 and COX-2. Both isoforms exhibit similar COX activity and both are expressed in mouse and human epidermis. COX-1 and COX-2, however, are differentially regulated such that COX-1 is usually constitutively expressed in most tissues, including the epidermis, whereas COX-2, which is generally expressed in unperturbed tissues at low to undetectable levels, is highly inducible by a number of irritating agents, including UV (4).

The relative contribution of each COX isoform to UV tumorigenesis has recently been elucidated. Using mice deficient in either isoform, we showed that while COX-2 deficiency dramatically reduced tumor development, a reduction in COX-1 had little effect (5). The reason for this is not entirely clear but is likely related to the much greater production of prostaglandin E₂ (PGE₂) from COX-2 due to the high level of COX-2 expression induced by UV (6). The genetic studies were complemented with pharmacologic studies using the nonspecific COX-1/COX-2 inhibitor indomethacin, one of the nonsteroidal anti-inflammatory drugs (NSAIDs), as well as COX-2 selective inhibitors, referred to as coxibs, for example, celecoxib. The study with celecoxib showed that it markedly reduced UV-induced skin tumors in mice, as did indomethacin (6). Because pharmacologic inhibition of COX-2 with celecoxib mimicked genetic deficiency of COX-2 in the SKH hairless mouse, this model offers an opportunity to identify agents, including other NSAIDs, for their ability to reduce UV-induced PGE₂ synthesis and to correlate that with long-term skin tumor development.

The choice of drugs used in the present study was based on several considerations. The coxibs were originally developed to avoid the gastrointestinal irritation caused by the classical NSAIDs, based on the premise that the prostaglandins from COX-1 help maintain the integrity of the mucosa while the prostaglandins from COX-2 are proinflammatory. Because of their cardiovascular side effects, however, the use of coxibs, despite clear efficacy, is problematic for long-term chemoprevention (7). The original concerns that NSAIDs cause an increased risk of ulcers and bleeding reduced enthusiasm for considering them as chemopreventive agents. The incidence of death is roughly 1/8,000 for most NSAID users, although the incidence of hospitalization for bleeds may be 1/1,000, suggesting that from a benefit versus risk perspective they may be of great value for the majority of high-risk individuals (8). Recently, a new class of NSAIDs was developed in which a nitric oxide (NO) moiety was added to the native NSAID. The rationale was that NO would be released and have a beneficial effect on the gastrointestinal mucosa that counteracts the effects of reduced prostaglandin synthesis (9). Both NO-aspirin and NO-naproxen were reported to reduce the macroscopic mucosal damage observed with aspirin and naproxen in rats (10). A comparison of naproxen and NO-naproxen in preventing large urinary tract tumors in rodents showed that they were equally effective, although neither was effective in a model of mammary cancer (11). There thus seems to be tissue specificity, although the basis for the difference is unknown.

As mentioned above, we previously showed the effectiveness of celecoxib and indomethacin in significantly reducing both UV-induced PGE₂ levels and skin tumor development (6). In the current study, we compared aspirin, NO-aspirin, naproxen, NO-naproxen, and sulindac, for their relative abilities to inhibit UV-induced short-term PGE₂ synthesis and proliferation, and long-term tumor development in mice. The goal is to determine whether PGE₂ or proliferation inhibition is a predictive biomarker for the chemopreventive efficacy of NSAIDs and other agents. We report here that, using human equivalent doses (HED), there is a strong correlation between the extent to which an NSAID or NO-NSAID inhibited acute UV-induced PGE₂ synthesis and UV-induced keratinocyte proliferation, and skin tumor development.

Female SKH-1 hr/hr (hairless) mice 3 and 4 weeks old were purchased from Charles River Laboratories and were housed in climate-controlled quarters (22° ± 1°C at 50% humidity) with 12/12 hour light/dark cycle using yellow fluorescent lights. Animals were allowed free access to water and diet and were observed daily. The following NSAIDs were supplied by the Division of Cancer Prevention, NCI: aspirin (lot# TW0592), 3-nitrooxymethylphenyl aspirin (lot# 0612105), sodium naproxen (lot# 080204), NO-naproxen (lot# 0901002), and sulindac (lot# UD0534). Indomethacin was purchased from Sigma Chemical Co. and celecoxib was purchased from LKT Laboratories. Powdered AIN-76 diet was purchased from Research Diets; the experimental diets were prepared weekly by mixing the NSAID into the diet with an electric mixer. The diets were stored at 4°C and fresh diet was supplied 3 times weekly in clean glass jars with stainless steel lids. Individual body weights were determined weekly for 20 weeks or more.

The UV apparatus and spectral irradiance used were previously described (6). For tumor studies, groups of 20 (unless noted otherwise) mice were fed their experimental diet starting 1 week before thrice weekly UV irradiation, starting with an initial dose of 90 mJ/cm² for the first week, followed by a weekly 10% increase until a dose of 175 mJ/cm² was reached. Weekly tumor counts were carried out after the appearance of the first tumor and were continued until the termination of the experiment. The tumor data are expressed both as multiplicity (i.e., mean number of tumors per mouse) and incidence (i.e., percent of mice with tumors). At the termination of the experiment, the diameters of the tumors were measured and the tumors were assigned to size categories. All tumors were processed for histologic analysis for determination of tumor type. The gastrointestinal tracts of all mice were also removed, washed with PBS, and processed for histologic staining with hematoxylin and eosin (H&E.) All sections were assessed microscopically for irritation or erosion of the epithelial lining.

Groups of 6 to 8 mice were killed 6 hours after a single UV treatment (220 mJ/cm²), their dorsal surfaces were quickly frozen on dry ice, and the animals were immersed in liquid nitrogen and stored at −70°C. A 1.5 cm² area of epidermis was chipped from the frozen skin and processed as previously described (12). The PGE₂ levels were measured by enzyme immunoassay (Cayman Chemical Co.) according to the manufacturer's instructions.

For acute treatment studies, groups of 3 and 4 mice were fed control or NSAID-containing diet for 7 to 10 days before UV irradiation with 220 mJ/cm² and killed at 24 and 48 hours. Mice were injected intraperitoneally with a sterile solution of 20 mg/mL 5-bromo-2-deoxyuridine (BrdUrd; Sigma Chemical Co.) at 0.1 mg/g body weight in PBS 1 hour before killing. Three to six 5 mm × 1.5 cm sections of skin were excised and fixed in 10% formalin before embedding in paraffin. Tissue sections (4 μm) were stained either with H&E or were immunohistochemically stained for BrdUrd incorporation using a monoclonal rat anti-BrdUrd antibody, (diluted 1:1; Accurate Chemical and Scientific Corp.). The bound antibody was visualized with 3,3′-diaminobenzidene (Sigma Chemical Co.) using avidin–biotin horseradish peroxidase (HRP; Vectastain Elite ABC kit; Vector Laboratories) linked to an affinity-purified biotin-labeled rabbit antirat immunoglobulin G (IgG). The labeling index was calculated as the percentage of basal cells staining positive for BrdUrd on at least 8 random areas for each of 3 sections from each mouse and the mean percentage and SE for each treatment group was determined. Tissue sections stained with H&E were used to measure the thickness of the epidermis (μm) using Nikon NIS-Elements software (Nikon Instruments Inc.).

For the determination of UV-induced apoptosis, 3 mice per experimental group were sacrificed 24 hours after 220 mJ/cm² UV-irradiation, skin sections were fixed in formalin, and were processed for paraffin embedding. Sections were stained for apoptosis using an antibody against cleaved capsase-3 at a 1:500 dilution, according to the manufacturer's protocol (R&D Systems). Biotinylated goat antirabbit IgG was used at 1:500 dilution as the secondary antibody (Vector Laboratories). The number of positive cells per μm length of basement membrane was counted on 10 to 12 random areas for each section (3/mouse) and the mean and SE was determined.

Proteins were isolated by scraping the epidermis from removed skin into RIPA buffer (50 mmol/L Tris pH 7.5, 150 mmol/L NaCl, 2 mmol/L EDTA, 1% Triton X-100, 0.5% Na deoxycholate, and 0.2% SDS). Following sonication, protein concentrations were measured using the BCA Protein Assay Kit (Thermo Scientific). Samples were electrophoresed on 10% or 4% to 15% gradient SDS-polyacrylamide gels and electroblotted onto nitrocellulose. COX-2 protein was detected with an anti-COX-2 polyclonal antibody (1:1,000; Millipore). A HRP-conjugated anti-rabbit IgG antibody (1:10,000; Jackson ImunoResearch Laboratories) was used as the secondary antibody. Actin, used as a loading control, was detected with an anti-actin HRP antibody (1:5,000; Santa Cruz Biotechnology Inc.). COX-2 and actin were identified after chemiluminescence detection (Amersham Corp.) of HRP by comparison with molecular weight markers. Densitometry was conducted using ImageQuantTL (GE Healthcare Life Sciences); the values for the UV bands were normalized to 1.0 and the NSAID bands were expressed as a percentage of the UV-only bands. Triplicate experiments were conducted, using 3 mice each and the data were expressed as a mean ± SE.

Mann–Whitney and unpaired t tests were done using GraphPad InStat, ver. 3.0 software to determine statistical significance for the proliferation, caspase-3, and PGE₂ data. Correlation coefficients were also determined with this software. The tumor multiplicity data were evaluated using Poisson regression analysis; differences in tumor latency were assessed using the Cox proportional hazards model.

The effect of 2 weeks of NSAID administration on body weights, as a measure of toxicity, was determined. There were no significant differences in the naproxen, NO-naproxen, aspirin, and NO-aspirin groups from the control group fed diet without NSAID (mean weight 24.2 ± 1.34 g). For sulindac, the initial doses given to the mice were 150, 300, and 600 ppm, however, the 300 and 600 ppm doses resulted in the death of several mice [these doses also caused death in FVB mice (data not shown)]. For this reason, lower doses (25 ppm and 150 ppm in the diet) of sulindac were used in subsequent experiments. Mice in the tumor experiments were also weighed at the termination of the study, after approximately 30 weeks on their NSAID diets, with no significant differences between them and the control diet group (mean weight of 30.2 ± 3.6 g). With the exception of high-doses of sulindac, it was concluded that the doses of NSAIDs administered in this study were not toxic. In addition, histologic examination of the gastrointestinal tracts of NSAID-fed mice revealed no signs of irritation or ulceration (data not shown).

As we have previously shown (6), acute exposure to UV causes high levels of PGE₂ to be synthesized in the epidermis. As shown in Table 1, in this study UV caused a nearly 5-fold increase in PGE₂ levels. Because indomethacin has previously been shown to prevent PGE₂ synthesis after UV, it was included here as a positive control for comparative purposes. Celecoxib, which is a selective COX-2 inhibitor (13), was also previously shown to reduce UV-induced PGE₂ by approximately 60% at the 500 ppm dose level (6); similar decreases were observed in this study (Table 1). The 250 ppm dose of celecoxib, which is roughly equal to the HED, was nearly as effective as the 500 ppm dose, whereas the 125 ppm dose was completely ineffective. The observation that celecoxib is not as effective as indomethacin in reducing PGE₂ is likely due to the inability of celecoxib to inhibit COX-1, whereas indomethacin is a dual COX-1/COX-2 inhibitor. All of the other NSAIDs tested also reduced PGE₂ synthesis in a dose–responsive manner, although some were more effective than others. In addition, for naproxen, the addition of a nitroso moiety diminished its COX inhibitory activity. Naproxen was very effective in inhibiting PGE₂, with the 200 and 400 ppm doses equivalent to 4 ppm indomethacin. A 400 ppm dose translates to a HED of 320 mg (7). Over-the-counter preparations of naproxen are usually 250 or 550 mg tablets to be consumed 4 times or twice daily, respectively. The nitroso form of naproxen seems less effective than the parent form, however, the higher molecular weight of NO-naproxen would partially correct for these differences, since the molecular weight of NO-naproxen is roughly 55% greater than the naproxen. Interestingly, aspirin seemed significantly less effective. Low-dose aspirin, which approximated the HED for the heart dose of aspirin, had limited effects on the induced levels of PGE₂. The median dose of aspirin, 250 ppm, reduced levels by roughly 60%, whereas the highest dose reduced levels by roughly 80%. The nitroso form of aspirin at equimolar doses was very effective at the higher doses in decreasing PGE₂ levels by almost 90%. The 150 ppm dose of sulindac, which has a HED of 159 mg/day (human doses are usually ≥300 mg/day), was remarkably effective in that it reduced PGE₂ by almost 95% compared with UV controls.

Similar to previous reports (6), UV irradiation increased the BrdUrd-labeling index by over 6-fold at 24 hours, as shown in Table 1. All of the NSAIDs studied reduced the labeling index, although there were differences in their efficacies. The highly effective agents, indomethacin and sulindac (150 pm), strongly decreased proliferation by roughly 70%. Overall, there is a correlation, although not significant, between the ability of a NSAID to inhibit COXs and their ability to inhibit UV-induced DNA synthesis. When modulation of epidermal thickness, which occurs as a result of both proliferation and edema, was used as a potential endpoint we observed maximally a 38% increase, but this did not correlate with the BrdUrd-labeling index, apoptosis, or PGE₂ levels.

The apoptotic response of keratinocytes to UV exposure is believed to be at least partly caused by reactive oxygen species (ROS) generated in response to UV. Because a recent study indicated that inhibiting inflammation with thioredoxin significantly reduced apoptosis (14), the ability of NSAIDs to reduce UV-induced apoptosis was tested (Supplementary Table S1). While all of the NSAIDs significantly reduced apoptosis, there was no correlation with the extent of tumor development and is thus not a useful short-term biomarker.

Because we have previously shown that PGE₂ can induce the expression of COX-2 in keratinocytes (15), we were interested in determining whether inhibition of PGE₂ synthesis would reduce the level of UV-induced COX-2 expression. As shown in Fig. 1, COX-2 was not expressed in untreated mice or in mice fed the high-dose NSAIDs without exposure to UV. Naproxen, however, enhanced UV-induced COX-2 by approximately 2-fold; NO-naproxen had less of an enhancing effect. Aspirin and NO-aspirin produced minor enhancement, whereas sulindac at 150 ppm caused a reproducible 3-fold enhancement. Thus there seems to be a negative correlation between the ability of an agent to decrease PGE₂ synthesis and the levels of COX-2 expression.

We have previously shown that indomethacin and celecoxib were very effective in preventing UV-induced skin tumors. Indomethacin at 4 ppm reduced tumor yield by 78%, whereas mice fed celecoxib at 150 or 500 ppm showed a dose-dependent reduction of 60% and 80%, respectively (6). As shown in Fig. 2, in a UV tumor experiment testing the efficacy of naproxen and NO-naproxen, tumors were first observed between 10 and 11 weeks, which is consistent with our previous study on indomethacin and celecoxib (6). Naproxen at 100 ppm significantly reduced tumor yield by approximately 26% and at 400 ppm by 63%. NO-naproxen at 400 ppm, which is roughly 2/3 of the molar dose of naproxen, was relatively effective in reducing tumor yield by 52% (Fig. 2A). The doses of naproxen and NO-naproxen that were effective in reducing tumor number, also very significantly increased latency, and the time to 100% tumor incidence.

Differences in the size distribution of the tumors were also observed, as shown in Supplementary Table S2. The groups having the largest percentage of tumors with diameters greater than 7 mm were also the groups with the most tumors, i.e., the control and 100 ppm NO-naproxen groups. Both naproxen groups had the smallest number of large (>7 mm) and medium (3–7 mm) size tumors and thus the greatest percentage of very small (<3 mm) tumors. There is thus a correlation between latency, tumor size, and tumor yield, suggesting that naproxen and NO-naproxen inhibit proliferation within the tumors.

Histologic analysis of the tumors showed that most of them were papillomas, although several squamous cell carcinomas (SCC) were found. No differences in histologic appearance were noted among the different treatment groups. An assessment of the tumors showed that the UV control group had the highest number of SCCs and the highest conversion rate of papillomas to SCCs. The naproxen and NO-naproxen groups had both fewer SCCs and reduced rates of conversion, indicating that their antitumorigenic activity was not restricted to papillomas (Supplementary Table S2).

The effectiveness of aspirin and NO-aspirin on tumor development was also determined (Fig. 3). Aspirin at 60 ppm and 750 ppm in the diet significantly reduced tumor yield by 19% and 50%, respectively. Low-dose NO-aspirin was ineffective in reducing tumor yield, whereas the high-dose NO-aspirin (1,350 ppm) that is equimolar to 750 ppm of aspirin reduced tumor numbers significantly by 33% (Fig. 3A). There was, however, no suggestion of a dose–response reduction in tumor size (Supplementary Table S2) with either aspirin or NO-aspirin, although the number of SCCs and the conversion rates were about half of that of the UV control group.

The third study examined whether oral sulindac was an effective inhibitor of UV-induced skin tumorigenesis. As shown in Fig. 4A, low-dose sulindac (25 ppm) reduced the number of tumors by 50%, whereas high-dose sulindac (150 ppm) inhibited significantly by approximately 94%. The high dose also significantly delayed the appearance of tumors (Fig. 4B). Both doses also had a marked effect on the size of the tumors in that neither sulindac group had tumors larger than 7 mm in diameter (Supplementary Table S2). Histologic analysis of the tumors showed that only 3 tumors in the 25 ppm group were SCCs, whereas no SCCs were found in the 150 ppm group. Thus of all the NSAIDs tested here, sulindac had the greatest chemopreventive efficacy by far with regard to tumor multiplicity, incidence, size, and progression to SCC.

The data in Table 1 and Figs. 2–4 were used to calculate the percentage reduction from the UV control for PGE₂ levels, BrdUrd-labeling index, and tumor multiplicity (Table 2). These values were then subjected to calculations for coefficients of correlation. Although it is recognized that the relationship between the early biomarkers, PGE₂, and proliferation is not likely to be linear, significant correlations were found for both PGE₂ and BrdUrd labeling with tumor outcome. These correlation coefficients indicate that either PGE₂ or BrdUrd labeling are excellent short-term predictors for tumor outcome (multiplicity), with PGE₂ a slightly better predictor. Multiple regression analysis gave an r² value of 84.80% (P = 0.0014) suggesting that use of the 2 short-term biomarkers together are even better predictors of tumor outcome.

NMSC is most commonly associated with excessive exposure to the UVB component of sunlight and is the most common form of cancer in the world (16, 17). The use of mouse models, most notably those employing the SKH-1 hairless mouse, have been instrumental in identifying the critical molecular and biologic changes that are required for skin tumor development (5). Thus the SKH mouse model employs both the same carcinogenic insult and results in the same driving p53 mutations, as in human squamous cell skin cancer. Exposure to UVB results in both DNA damage and inflammation (16, 17), with the latter characterized by an infiltration of inflammatory cells into the dermis and the upregulation of genes and proteins producing soluble mediators of inflammation, including cytokines and prostaglandins. The upregulation of COX-2, with consequent elevation of prostaglandins, is crucial to the development of skin tumors (5). This is based on a number of observations, including pharmacologic studies showing that mice fed celecoxib or indomethacin had a significantly reduced tumor multiplicity (6). Because many drugs, including the NSAIDs and coxibs, have off-target effects (7), genetic approaches were also used. Although the loss of 1 allele of COX-1 had no effect on UV-induced skin tumor development, the loss of only one allele of COX-2 significantly reduced (50%–65%) tumor multiplicity (18). COX-2 transgenic SKH-1 mice were shown to respond to UV with a decreased tumor latency and increased tumor multiplicity (12). These genetic studies have clearly shown that susceptibility to UV-induced skin cancer is correlated with COX-2 gene copy number and expression levels.

The goals of the study presented here were 2-fold. The first goal was to compare NO-NSAIDs with their parent NSAID counterpart for efficacy in reducing short-term biomarkers, particularly inhibition of PGE₂ synthesis. The second goal was to determine whether efficacy against one or more short-term biomarkers induced by UV correlated with long-term efficacy against skin tumor development. Establishing the strength of such a correlation could be useful in designing long-term prevention studies for the human population in which such short-term biomarker and long-term tumor endpoint studies are not feasible.

The NSAIDs used in this study were chosen for several reasons. They are among the most commonly consumed NSAIDs and have been shown to have cancer preventive activity in a number of rodent tissues and are associated with reduced incidence of several human cancers (reviewed in 7). Naproxen is a member of the 2-arylproprionic family of NSAIDs and inhibits COX-1 and COX-2 at comparable IC₅₀ levels. The molecular basis for naproxen inhibition of COX, which is reversible, has recently been described (19). naproxen also seems to have the best CV profile of any NSAID other than aspirin (20). Aspirin is a salicylate drug that inhibits COX-1 to a greater extent than COX-2 and does so through irreversible acetylation of the enzyme (21). Sulindac belongs to the arylalkanoic acid class of NSAIDs. Following absorption, it undergoes 2 biotransformations, to a reversible reduction of the sulfide metabolite and an irreversible oxidation to the sulfone metabolite. The sulfide form is the active metabolite with regard to inhibiting COX, although the proapoptotic effect of the sulfone metabolite may also contribute to its antitumor activity (22). This agent in conjunction with the ornithine decarboxylase inhibitor difluoromethylornithine was profoundly effective in inhibiting colon adenomas in a human trial (23). Although we conducted studies with NO-naproxen and NO-aspirin, we were unable to carry out studies with NO-sulindac. The NO-sulindac was not soluble or readily dispersed in a variety of vehicles (R.A. Lubet and C.J. Grubbs, unpublished data) and we were therefore unable to test it properly.

One of the major limitations to the use of NSAIDs is their damaging effect on the gastrointestinal mucosa and exacerbation of pre-existing gastric lesions, which occurs as a result of the loss of the cytoprotective prostaglandins. Like prostaglandins, NO is cytoprotective to the gastric mucosa. Thus, NO-releasing NSAIDs were developed on the basis of the beneficial effects of NO on the gastric mucosa (9), although in human trials the decrease in gastric toxicity seems imperfect.

The NO–NSAIDs have been reported to produce lower gastric toxicity while still inhibiting PGE₂ synthesis to a level comparable with the parent NSAID (24). NO-aspirin and NO-naproxen in particular were reported to inhibit PGE₂ synthesis to essentially the same level as aspirin and naproxen in rat gastric mucosa (10). In the SKH-1 skin model used here, we found that naproxen and its NO derivative had similar effects given that the NO was administered at roughly 2/3 of the molar dose of the parent compound. Low-dose aspirin and NO-aspirin gave comparable, albeit weak, inhibition. In contrast, the higher doses of aspirin or equimolar NO-aspirin (750 or 1350 ppm) were relatively effective. However, the dose of aspirin which was effective is 4 and 5 times higher than the cardiovascular dose of aspirin (<100 mg) based on standard FDA scaling factors (7). The high dose of all the NSAIDs tested inhibited PGE₂ synthesis by more than 80%, which is a significant reduction. The relative lack of efficacy of aspirin is consistent with data in rat bladder, rat colon, and the Min mouse (7). The apparent discrepancy with regards to the human, where aspirin is effective is somewhat puzzling. It is presumably not totally a species-dependent result as this low activity is seen both in mice and rats. Furthermore, the data with low-dose aspirin or NO-aspirin are certainly in line with the probability that COX-1-mediated PGE₂ production is inhibited.

Increased proliferation is one of the hallmarks of tumor promotion. Several mechanisms have been proposed for this increased proliferation of keratinocytes in skin, including dysregulated growth factor signaling and overproduction of prostaglandins (25–27). Thus, the inhibition of prostaglandin synthesis by NSAIDs would be expected to reduce UV-induced proliferation. This was observed for all the NSAIDs tested here and in a previous study (6). What was not clear was whether the degree to which inhibition of proliferation after acute UV would correlate with tumor outcome, due in part to the unlikely linear relationship between the two and possible (probable) off-target effects of specific NSAIDs that affect proliferation and/or cell death. Given these caveats, the observation that there is a significant correlation suggests that this short-term biomarker should be of value in screening for the most efficacious chemopreventive drugs.

The effect of the NSAIDS studied here on COX-2 expression levels was also investigated. Although we previously showed that neither celecoxib nor indomethacin reduced UV-induced COX-2 expression (6), another type of anti-inflammatory drug, 5-aminosalicylic acid, suppresses COX-2 expression (28). In a study on UV phototoxicity, Athar and colleagues (29) found that sulindac markedly enhanced UV-induced COX-2 expression even though PGE₂ synthesis was suppressed. We observed a similar enhancement in that the higher doses of sulindac and naproxen, both of which strikingly decreased PGE₂ levels (>90%), increased protein levels of COX-2.

Of the 3 NSAIDs tested here, sulindac was by far the most potent, with the 150 ppm dose resulting in a 94% reduction in tumor number. One of the most striking effects was the complete prevention of SCC development. Although we are unaware of epidemiologic studies in humans to support this finding, sulindac has previously been reported to inhibit experimentally induced skin cancer. Kim and colleagues (30) found that a 320 ppm dose reduced chemically induced skin tumors in mice by approximately 50%. Oral sulindac (160 ppm in drinking water) was also shown to significantly reduce markers of photodamage, including edema, hyperplasia, and inflammatory cell infiltration, in SKH-1 hairless mice exposed to UV (29). Using a xenograft model with human A451 carcinoma cells, Cheng and colleagues (31) found that oral dosing (150 mg/kg/d) with a novel phosphoderivative of sulindac reduced tumor growth by 43%. Collectively, these findings suggest that sulindac has significant potential as a NMSC prevention agent.

In this study, oral naproxen was found to significantly reduce skin tumor development (56% reduction at the 400 ppm dose). Using topical treatment, naproxen was reported to suppress UV-induced skin tumors by 63% (32). In comparing naproxen with NO-naproxen, we found that the NO derivative was less effective than naproxen. In the azoxymethane-induced aberrant crypt foci assay in rat colon, oral naproxen was also found to be more effective than its NO counterpart, whereas in a rat urinary bladder model both forms were similarly very effective (11).

Aspirin has been previously tested for efficacy against UVB-induced skin cancer in mice. Using topical applications of 10 or 40 μmol/L doses before each UV exposure, Bair and colleagues (33) reported that the 40 μmol/L dose reduced tumor multiplicity by approximately 25% but had a greater effect on tumor size. Topical treatment with aspirin was also shown to markedly block UVB-induced AP-1 transactivation in murine epidermis, an event required for skin tumor development (34). With regard to comparing the efficacy of aspirin with its NO derivative, we found that NO-aspirin was less effective, which differs from observations by others in other organ models of cancer. In the hamster model of pancreatic cancer, NO-aspirin was found to significantly reduce cancer development, whereas aspirin had no chemopreventive effect (35). NO-aspirin also significantly inhibited colon carcinomas in azoxymethane-treated rats (36), however, no side-by-side comparison was made with the parent NSAID. The reason for the difference between skin and other tissues is not known. There is also conflicting evidence in the human population on the efficacy of aspirin in preventing NMSC. A recent large prospective study found that aspirin and other NSAIDs did not reduce the risk of NMSC or melanomas (37), although 2 recent population-based case–control studies found that NSAID use, including aspirin, reduced risk of SCC but not basal cell carcinoma (38, 39). Furthermore, there was a clinical trial of the COX-2 inhibitor celecoxib which caused a roughly 60% decrease in SCCs (40). Although this was not the primary endpoint of the trial, the placebo-controlled design and the consistency of the COX-2 inhibitor, with regard to both dose and frequency of administration, is a major strength when compared with most epidemiologic studies. In addition, the fact that topical treatment with NSAIDS is effective in blocking SCC formation also makes these results particularly promising.

In summary, this study showed that different classes of NSAIDs, at HEDs, have chemopreventive activity against NMSC induced by exposure to UV light, with sulindac being the most potent. The addition of a NO moiety did not significantly alter the efficacy. These data suggest that daily consumption of NSAIDs by the human population could reduce the incidence, and possibly severity, of developing NMSC. An important facet of this study was the determination that the short-term biomarkers, PGE₂ and proliferation levels, are excellent predictors for the long-term tumor response. This suggests that these biomarkers should have utility in screening agents for their chemopreventive activity. This would be both a cost and time effective approach to identifying the best drugs available.

No potential conflicts of interest were disclosed.

Conception and design: R.A. Lubet, S.M. Fischer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.D. Mikulec, M.S. Simper, S.M. Fischer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.D. Mikulec, J.E. Rundhaug, M.S. Simper, S.M. Fischer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.D. Mikulec, S.M. Fischer

Writing, review, and/or revision of the manuscript: J.E. Rundhaug M.S. Simper, R.A. Lubet, S.M. Fischer

Study supervision: J.E. Rundhaug, S.M. Fischer

The authors thank Joyce Rundhaug for her statistical analyses of the data.

This work was supported by funds (N01-CN-53300) from The National Cancer Institute, The University of Texas MD Anderson Cancer Center Support Grant CA16672, and the National Institute of Environmental Health Science Center Grant ES07784.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.