Staining performed according to standard protocols, as detailed in the Methods section. A, HC10-negative tumor. B, HC10-positive tumor. C, HCA2-negative tumor. D, HCA2-positive tumor. E, Tumor with weak PTGS2 expression. F, Tumor with strong PTGS2 expression. G, Enlarged view of sample in F. See the Tissue Microarray Production and Immunohistochemistry subsection of the Methods section for details of the immunohistochemical staining methods used. Original magnifications ×20 (A-F) and ×40 (G).
A, Overall survival among patients with loss of HLA class I antigen in tumor sections. B, Overall survival among patients with expression of HLA class I antigen in tumor sections.
eTable 1. Baseline characteristics of the patients included in the population-based registry and the patients included in this study
eTable 2. Primer overview of the primers used for the PIK3CA mutation analysis
Customize your JAMA Network experience by selecting one or more topics from the list below.
Reimers MS, Bastiaannet E, Langley RE, et al. Expression of HLA Class I Antigen, Aspirin Use, and Survival After a Diagnosis of Colon Cancer. JAMA Intern Med. 2014;174(5):732–739. doi:10.1001/jamainternmed.2014.511
Use of aspirin (which inhibits platelet function) after a colon cancer diagnosis is associated with improved overall survival. Identifying predictive biomarkers of this effect could individualize therapy and decrease toxic effects.
To demonstrate that survival benefit associated with low-dose aspirin use after a diagnosis of colorectal cancer might depend on HLA class I antigen expression.
Design, Setting, and Participants
A cohort study with tumor blocks from 999 patients with colon cancer (surgically resected between 2002 and 2008), analyzed for HLA class I antigen and prostaglandin endoperoxide synthase 2 (PTGS2) expression using a tissue microarray. Mutation analysis of PIK3CA was also performed. Data on aspirin use after diagnosis were obtained from a prescription database. Parametric survival models with exponential (Poisson) distribution were used to model the survival.
Main Outcomes and Measures
The overall survival benefit associated with aspirin use after a diagnosis of colon cancer had an adjusted rate ratio (RR) of 0.53 (95% CI, 0.38-0.74; P < .001) when tumors expressed HLA class I antigen compared with an RR of 1.03 (0.66-1.61; P = .91) when HLA antigen expression was lost. The benefit of aspirin was similar for tumors with strong PTGS2 expression (0.68; 0.48-0.97; P = .03), weak PTGS2 expression (0.59; 0.38-0.97; P = .02), and wild-type PIK3CA tumors (0.55; 0.40-0.75; P < .001). No association was observed with mutated PIK3CA tumors (0.73; 0.33-1.63; P = .44).
Conclusions and Relevance
Contrary to the original hypothesis, aspirin use after colon cancer diagnosis was associated with improved survival if tumors expressed HLA class I antigen. Increased PTGS2 expression or the presence of mutated PIK3CA did not predict benefit from aspirin. HLA class I antigen might serve as a predictive biomarker for adjuvant aspirin therapy in colon cancer.
There is a significant body of preclinical, epidemiologic, and randomized data demonstrating that aspirin has anticancer effects.1-7 Several studies2,4,8-10 have shown that aspirin use after a diagnosis of colorectal cancer improves colorectal cancer–specific and overall survival. Randomized trials7 designed to assess the cardiovascular benefits of aspirin demonstrate that allocation to aspirin reduces the risk of distant metastasis when cancer is diagnosed (hazard ratio [HR] 0.69; 95% CI, 0.5-0.95; P = .02) and on subsequent follow-up in patients without metastasis at diagnosis (0.45; 0.28-0.72; P < .001), with the largest effects seen for colorectal cancer (HR at diagnosis, 0.36; 0.18-0.74; P = .005; and HR at follow-up, 0.26; 0.11-0.57; P < .001). Although questions remain about the optimal dose and duration of aspirin use, its efficacy in prediagnostic users, and the localization of tumors most likely to benefit, the data suggest that aspirin is a potential adjuvant therapy to prevent distant metastasis in colorectal cancer, and possibly other tumors.
The precise biological mechanisms underlying the anticancer effects are unknown. Prostaglandin endoperoxide synthase 2 (PTGS2; HGNC 9605), also known as cyclooxygenase 2, overexpression has been associated11,12 with a poor prognosis in colorectal cancer. Aspirin inhibits PTGS; at low doses (75-300 mg once daily), and given the short half-life of aspirin (approximately 30 minutes), this effect is manifested as permanent inhibition of PTGS1 in the anucleate platelet, which is unable to resynthesize the enzyme. Higher and more frequent dosing, for example, 600 mg once daily, would be required to constantly inhibit PTGS2 in systemic tissues.13-16 Despite this dose-dependent observation, data from 2 observational cohorts (the Nurses’ Health Study and Health Professionals Follow-up Study)17 have indicated that the survival benefits of regular use of low-dose aspirin after a diagnosis of colorectal cancer are associated with the molecular characteristics of the tumor, particularly mutations in the gene PIK3CA (HGNC 8975) (a component of the PTGS2 pathway), with a multivariable HR for aspirin users compared with nonusers in tumors with mutated PIK3CA of 0.18 (95% CI, 0.06-0.61; P < .001) for cancer death and 0.54 (0.31-0.94; P = .01) for death from any cause.
The metastatic potential of cancer cells that are shed into the bloodstream can be modified by environmental conditions, including platelets and bone marrow–derived cells in the vasculature.18 Platelets are thought to protect disseminating tumor cells from natural killer cells, which preferentially recognize and eliminate cells with low or absent expression of HLA class I antigen.19 We hypothesized that the survival benefit associated with low-dose aspirin use after a cancer diagnosis would be associated with tumors that have low or absent HLA class I antigen expression. We analyzed tumors from a cohort in which an association was shown8,10 between overall survival and low-dose aspirin use after diagnosis (adjusted rate ratio [RR], 0.65; 95% CI, 0.50-0.84; P = .001) with an even larger effect in older (>70 years) patients with colon cancer (0.59; 0.44-0.81; P = .001) for HLA class I antigen and PTGS2 expression as well as PIK3CA mutations.
The Eindhoven Cancer Registry was initially used to identify patients with colorectal cancer and linked to data on aspirin use from the PHARMO database network (PHARMO, the Netherlands). This research was performed according to the code of conduct for responsible use. Patient records information was made anonymous and deidentified prior to analysis according to national ethics guidelines (Code for Proper Secondary Use of Human Tissue, Dutch Federation of Medical Scientific Societies). As previously reported,8 compared with survival in aspirin nonusers, aspirin initiated or continued after diagnosis was associated with improved survival for patients with colon cancer, but not for those with rectal cancer. Paraffin-embedded tissue blocks were retrieved from 1026 patients with colon cancer who had a surgical resection between 2002 and 2008. For this study, 27 patients with more than 1 colon tumor at the time of diagnosis were excluded; thus, 999 patients were included in the analysis. There were no significant demographic differences between the patients included in the present study and the whole colon cancer cohort in the registry (N = 3586) (Supplement [eTable 1]).
Three 1.0-mm-diameter cores were obtained from formalin-fixed, paraffin-embedded tumor blocks using hematoxylin-eosin–stained sections for tumor identification (with a qualified pathologist [H.M.] confirming the identification of the tumor) and transferred into a receiver paraffin block using tissue microarray (TMA Master; 3DHistech Ltd). Immunohistochemical staining was performed on 4-µm sections cut from each receiver block and mounted on glass. For each primary antibody, all slides were stained simultaneously to avoid interassay variation.
Immunohistochemical analyses to detect HLA class I antigen expression with mouse monoclonal antibodies HCA2 and HC10 (heavy chains of HLA-A), using diaminobenzidine solution (Dako) for visualization of the antibodies, were performed by 2 independent observers (M.S.R. and R.L.P.v.V.), as previously described,20 with good interobserver agreement (κ = 0.5-0.7). The mouse monoclonal antibodies HCA2 and HC10 that were used recognize the heavy chains of HLA class I antigen. Their reactivity spectrum has been described in detail.18 The HLA class I antigen expression status was determined according to the International HLA and Immunogenetics Workshop,21 with tumor cell HLA class I antigen status defined as follows: loss of HLA class I antigen: less than 5% expressing both HCA2 and HC10 or less than 5% expressing either of the markers, and expression of HLA class I antigen: 5% or more expressing both markers. Normal epithelial, stromal, or lymphoid cells served as positive internal controls. Expression of PTGS2 was analyzed automatically with a double staining to separately visualize stromal cells (using diaminobenzidine solution for visualization of anticollagen I, anticollagen VI, and elastin [all polyclonal rabbit antibodies obtained from Abcam]) and positive tumor cells (with the monoclonal mouse antibody anti-PTGS2 [Cayman Chemical Co]), using Vector Blue (Vector Laboratories) for visualization of the PTGS2 antibody. Slides were scanned (Pannoramic Midi; 3DHistech) and PTGS2 expression was assessed (AxioVision, 4.6; Zeiss) using the criteria proposed by Buskens et al22 and comparable to the scoring method used by Chan et al.2 Representative images of the immunohistochemical staining are shown in Figure 1.
Microsatellite stability status was determined by immunohistochemical analyses as previously described.23 In short, 4 antibodies directed against mutL homologue 21 (MLH1, clone ES05; DakoCytomation), mutS homologue 2 (MSH2, clone g219-1129; BD Biosciences), mutS homologue 6 (MSH6, clone EPR3945; Epitomics), and postmeiotic segregation of Saccharomyces cerevisiae 2 (PMS2, clone A16-4; BD Biosciences) were used. The criteria used to confirm microsatellite instability in the tissues are described elsewhere.23,24
All slides were stained simultaneously to avoid interassay variation. Slides that underwent the entire immunohistochemical staining procedure without primary antibodies served as negative controls. The quality of the staining, the scoring method, and discrepancies between the 2 observers were checked by a pathologist (H.M).
Samples of DNA were extracted from 1 to 2 variable-length, 2.0-mm-diameter cores taken from 663 of the 999 blocks randomly chosen, with a 1:2 ratio for aspirin user to nonuser, using a fully automated system (Tissue Preparation System with VERSANT Tissue Preparation Reagents; Siemens Healthcare Diagnostics) as described previously.25
Hydrolysis probe assays were performed for the major known mutations (hotspots) in exon 9, c.1624G>A; p.E542K, c.1633G>A; p.E545K as well as in exon 20, c.3140A>G; and p.H1047R as previously described.26 Hydrolysis probe assays were analyzed using quantitative polymerase chain reaction analysis software (CFX Manager, version 3/0; Bio-Rad). To identify additional nonhotspot mutations, Sanger sequencing was performed on exon 9 and exon 20 of all samples. Mutation detection was performed by 2 observers independently (M.S.R. and R.v.E.) using DNA variant analysis software (Mutation Surveyor, version 4.0.9; Softgenetics). All primers and probes used for the assays are listed in the Supplement (eTable 2).
The vital status of patients (alive vs dead) was established from medical records or through linkage of cancer registry data with the municipal population registries. Follow-up started 30 days from diagnosis of colorectal cancer (T0), because information on hospital prescriptions was not available, and was continued until the last contact date (January 1, 2012) or the date of death. Patients who died within 30 days of diagnosis were excluded from the survival analyses (2.4% for colon cancer). Nonusers were classified as patients who never had a prescription for aspirin or had a prescription for less than 14 days after diagnosis of colon cancer. Users were defined as those who had been given a prescription for aspirin for 14 days or more after a colon cancer diagnosis. The median duration of prescriptions was 30 days and the mean number of prescriptions was 12 (range, 1- 220). Nonusers were defined from T0 to first use and users from first use to the end of the follow-up in the time-dependent exposure survival analysis.
Because the data were split into 2 episodes for users (multiple identification rows for 1 patient), we were not able to model a Cox proportional hazards regression model and instead used a parametric survival model with an exponential (Poisson) distribution after the data were declared as survival-time data and split at the time to first prescription.
Adjustments for potential confounders were made for sex, age (continuous), stage (pathologic stage and clinical stage if pathologic stage was unknown), adjuvant chemotherapy (yes/no), comorbidity (yes/no), tumor grade, and year of diagnosis. Stratified analyses were performed for HLA class I antigen expression, weak or strong PTGS2 expression, and wild-type PIK3CA and PIK3CA mutation.
Of the 999 patients, 182 (18.2%) were defined as aspirin users, and there had been 465 deaths recorded until January 2012. There were 396 deaths in 817 aspirin nonusers (48.5%) and 69 deaths in 182 aspirin users (37.9%) after diagnosis. In this cohort, aspirin use after diagnosis was associated with improved overall survival (RR, 0.64; 95% CI, 0.49-0.83; P = .001) compared with nonuse.
Thirty-six of the 999 tumors could not be analyzed for HLA class I antigen expression because of staining artifacts or loss of material. Table 1 summarizes the clinical characteristics of the patients presented by HLA class I antigen expression and according to aspirin use and nonuse after diagnosis. Of the 963 remaining samples, loss of HLA class I antigen expression was found in 33.2% (320) and expression was noted in 66.8% (643) in accord with results from previous studies.27,28 Aspirin use was similar: in patients with loss of HLA antigen class I tumors, 18.0% (57 of 320) were aspirin users, and of the patients with HLA antigen class I–expressing tumors, 19.0% (122 of 643) were aspirin users. Aspirin users were older and more likely to have comorbidities compared with nonusers. In the HLA class I antigen expression group, there were more lower-stage tumors in aspirin users compared with nonusers (P < .001).
The effect of HLA class I antigen expression status on the survival benefit associated with aspirin use after diagnosis was examined (Table 2 and Figure 2). For patients whose tumors expressed HLA class I antigen, aspirin use after diagnosis was associated with a significantly longer overall survival (RR, 0.61; 95% CI, 0.44-0.85; P = .003), and when adjusted for potential confounders, this effect remained with an adjusted RR of 0.53 (0.38-0.74; P < .001). In contrast, for patients whose tumors had loss of HLA class I antigen expression, aspirin use after diagnosis was not associated with a survival benefit (adjusted RR, 1.03; 95% CI, 0.66-1.61; P = .91).
Twenty-five of 999 samples could not be analyzed for PTGS2 expression because of staining artifacts or loss of material. Weak expression of PTGS2 was seen in 43.7% of the samples (426 of 974 samples), and strong PTGS2 expression was noted in 56.3% (548 of 974) in accord with the literature.2,29 Use of aspirin after diagnosis was significantly associated with a survival benefit when tumors showed weak PTGS2 expression (adjusted RR, 0.59; 95% CI, 0.38-0.91; P = .02) as well as when they showed strong PTGS2 expression (0.68; 0.48-0.97; P = .03) (Table 2).
We extracted DNA from 663 tumor blocks, and PIK3CA mutation status (wild-type/mutation) was established in 95.2% of the samples (631 of 663). Baseline characteristics among participants with colon cancer who underwent analysis for PIK3CA were largely similar to the baseline characteristics of the PTGS2/HLA class I antigen cohort (mean age at inclusion, 70.4 vs 69.0 years; male sex, 53.9% vs 50.6%; stage I, 15.4% vs 13.8%; stage II, 40.3% vs 40.2%; stage III, 29.3% vs 28.7%; stage IV, 14.7% vs 16.9%; presence of comorbidity, 59.9% vs 55.7%; adjuvant chemotherapy, 28.1% vs 30.8%; histologic grade I, 10.6% vs 10.0%; grade II, 64.5% vs 63.0%; and grade III, 17.6% vs 19.3%; P > .09 for all comparisons). A PIK3CA mutation was found in 15.1% of the patients (100 of 663), also in accord with what has been reported previously.30 Aspirin use was 27.0% (27 of 100) among patients with a mutated PIK3CA tumor and 27.7%% (147 of 531) in patients with a PIK3CA wild-type tumor. Aspirin use after a colon cancer diagnosis was significantly associated with better overall survival among patients with a wild-type PIK3CA tumor (adjusted RR, 0.55; 95% CI, 0.40-0.75; P < .001). In patients with a PIK3CA mutation, aspirin use after diagnosis showed no association, with an adjusted RR of 0.73 (0.33-1.63; P = .44). The small number of deaths (9) among patients with mutated PIK3CA tumors who took aspirin precludes robust statistical assessment (Table 2).
We found that the survival benefit associated with low-dose aspirin use after a diagnosis of colon cancer was significantly associated with HLA class I antigen–positive tumors. In contrast, in patients whose tumors had lost their HLA class I antigen expression, aspirin use did not change the outcome. In contrast to previous studies,2,17,31 PTGS2 expression and PIK3CA mutation analysis could not identify patients with a high likelihood of benefit from aspirin.
Currently, the molecular mechanisms underlying the anticancer effects of aspirin are incompletely understood. Given that colon cancer in most of our cohort (>80%) was identified as stage III or less at the time of diagnosis, the predominant effect of aspirin on cancer outcomes is likely to result from an effect on circulating tumor cells and their ability to develop into metastatic deposits. Natural killer cells play an important role in tumor immune surveillance, preferentially eliminating targets with low or absent expression of HLA class I antigen.19 Adhesion of HLA antigen–expressing platelets to tumor cells with absent or low HLA class I antigen expression is thought to result in a “pseudonormal phenotype” and reduced natural killer cell–mediated lysis.19 We originally hypothesized that aspirin might inhibit platelet adhesion to tumor cells, leaving patients with absent or low HLA class I antigen expression susceptible to immune clearance; however, we unexpectedly found that the effect of aspirin is dependent on intact HLA class I antigen expression within the original primary tumor. Assuming that circulating tumor cells retain the same HLA antigen phenotype as the original tumor does not support the hypothesis that the attenuation of metastases by aspirin and possibly other anticoagulants is a result of enhanced natural killer cell activity.19
A possible explanation for this intriguing observation is that HLA antigen expression might be necessary for platelet-mediated nuclear factor κB signaling in circulating tumor cells, resulting in an epithelial-mesenchymal–like phenotype with enhanced metastatic potential.18 In this model, direct contact of platelets and tumor cells results in secretion of tumor growth factor β and activation of the nuclear factor κB pathway, which, in synergistic action, prime circulating tumor cells for subsequent metastases. In a breast cancer model, acquisition of an epithelial-mesenchymal phenotype markedly reduced susceptibility of cancer cells to T-cell–mediated immune surveillance in vitro.32 Our data would be compatible with the hypothesis that aspirin inhibits platelet-tumor cell signaling (which is dependent on intact HLA antigen expression) and prevents epithelial-mesenchymal transition in circulating tumor cells, thereby reducing the metastatic potential.
Our data have not confirmed previous reports2,17 that the benefits of aspirin after a colorectal cancer diagnosis are associated with strong PTGS2 expression in the original tumor and the presence of mutations in PIK3CA, with no benefit for patients whose tumors had wild-type PIK3CA. Liao et al17 postulated that by blocking the PIK3CA pathway PTGS2 activity decreases, which leads to apoptosis of colon cancer cells; this was in accord with previous work2 demonstrating a clinical benefit of aspirin in patients with PTGS2-positive tumors. In a separate study,31 the benefits of aspirin in PIK3CA-mutated tumors were seen, but the correlation with strong expression of PTGS2 expression was not confirmed.
Pharmacologic data on aspirin indicate that systemic concentrations of aspirin, reached with low doses (75-325 mg once daily), are inadequate to permanently acetylate PTGS2 but are optimal for platelet inhibition.13 It is possible that there may be more than one mechanism of action that accounts for the anticancer effects of aspirin. A direct antiplatelet effect due to PTGS1 inhibition that is responsible for the reduction in metastases and only requires a dose of aspirin that inhibits platelets, and a second mechanism, possibly mediated through platelets again or perhaps activated with higher or more frequent dosing that inhibits the PTGS2 pathway in systemic tissues, which may partly explain the differences between the results of our study and those of Liao et al.17 It has also been reported33 that PTGS2 expression could not identify a subgroup of patients with breast cancer for whom aspirin decreased recurrence. Furthermore, in breast cancer, low-dose aspirin did not influence local recurrence but was significantly associated with a decrease in metastatic disease.34
Aspirin use has also been associated35 with a decreased risk of developing a colorectal tumor with an intact BRAF gene, but no association between aspirin use after diagnosis, BRAF mutation status, and clinical outcome has been found. BRAF is a member of the RAF–mitogen-activated protein kinase signaling pathway and involved in the upregulation of PTGS2, again suggesting that aspirin may have differential effects on carcinogenesis and prevention of metastatic spread.35
Strengths of our study include a more precise definition of regular aspirin use and dose with this information derived from prescriptions (rather than patient recall), noting also that low-dose aspirin is not available as an over-the-counter medication in the Netherlands, thereby minimizing nondifferential misclassification of exposure. Use of higher-dose over-the-counter aspirin is unknown, which could have biased our results toward the null hypothesis. However, it has been shown36 that pharmacy data can give valid associations even though a high proportion (25%) of the drugs are available as over-the-counter formulations. Other limitations of our study include the inherent issue that these are nonrandomized data, adherence is unknown, and some subgroups contained few events, although our series is the largest study thus far that has reported on aspirin use in patients with colon cancer.
The molecular profiling of tumors—for example, KRAS testing in colorectal cancer and ERBB2 (formerly HER-2) testing in breast cancer—has become standard clinical practice and the basis of therapeutic decisions. If the association of HLA antigen expression and benefit from aspirin is confirmed in other data sets it could be used in clinical practice, and our data may have important clinical implications for both the dose and timing of aspirin as an anticancer agent. First, low-dose daily aspirin may suffice as an antimetastatic therapy in patients with early-stage cancer. Second, because circulating tumor cells are found in the perioperative period, it could be argued that aspirin therapy should be initiated as soon as considered clinically appropriate after diagnosis.
We report the novel finding that the survival benefit associated with low-dose aspirin use after diagnosis of colon cancer is dependent on intact HLA class I antigen expression in the original tumor. Randomized trials of the use of aspirin in the adjuvant setting may provide key information about platelet-tumor interactions and the signaling pathways they elicit.
Accepted for Publication: January 31, 2014.
Corresponding Author: Gerrit Jan Liefers, MD, Department of Surgery, Leiden University Medical Center, PO Box 9600, 2300 RC Leiden, the Netherlands (firstname.lastname@example.org).
Published Online: March 31, 2014. doi:10.1001/jamainternmed.2014.511.
Author Contributions: Drs Reimers and Liefers had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Reimers, Bastiaannet, van Vlierberghe, Fodde, Kuppen, Morreau, Liefers.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Reimers, Bastiaannet, Langley, van Eijk, Lemmens, Kuppen, Liefers.
Critical revision of the manuscript for important intellectual content: Reimers, Bastiaannet, van Vlierberghe, Lemmens, van Herk-Sukel, van Wezel, Fodde, Kuppen, Morreau, van de Velde, Liefers.
Statistical analysis: Reimers, Bastiaannet, van Herk-Sukel.
Obtained funding: Kuppen, Liefers.
Administrative, technical, or material support: Reimers, van Eijk, van Vlierberghe, Lemmens, van Herk-Sukel, Morreau.
Study supervision: Reimers, Fodde, Kuppen, Morreau, van de Velde, Liefers.
Conflict of Interest Disclosures: Dr van Herk-Sukel is an employee of the PHARMO Institute. This independent research institute performs financially supported studies for government and related health care authorities and several pharmaceutical companies. The present study, however, was not supported by a pharmaceutical company. Dr Langley has received a honorarium from the Aspirin Foundation and is a consultant for Bayer. No other disclosures were reported.
Funding/Support: This work was supported by an unrestricted grant from the Sloos-Alandt family (Dr Liefers).
Role of the Sponsor: The sponsor had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Additional Contributions: We thank the pathologists in all participating hospitals for their cooperation and in providing study material.