Personalized Prostate Cancer Screening: Improving PSA Tests with Genomic Information

See allHide authors and affiliations

Science Translational Medicine  15 Dec 2010:
Vol. 2, Issue 62, pp. 62ps55
DOI: 10.1126/scitranslmed.3001861


The use of a prostate-specific antigen (PSA) test to screen for prostate cancer is controversial because of its modest predictive value and the potential overdiagnosis and over-treatment of the disease. A research article in this issue of Science Translational Medicine describes single-nucleotide polymorphisms (SNPs) in or near six genes that are independently associated with serum PSA concentrations and that help to explain interindividual PSA variation. Three of these SNPs are also associated with prostate biopsy outcomes. These findings are an important step toward incorporating genetic markers into PSA screening, with the ultimate goal of devising personalized PSA tests for use in the clinic.

Ninety-nine percent of men diagnosed with prostate cancer in the United States survive for at least 5 years (1). This high survival reflects improvements in treatment options and extensive prostate cancer screening by measuring serum concentrations of prostate-specific antigen (PSA)—a protein produced by prostate cells. Support for widespread PSA screening is mixed, however, because PSA testing has only moderate accuracy. In some men, PSA testing misses disease; in others, it gives false-positive results or detects indolent or early-stage tumors that may not require immediate treatment (2). Using more individualized thresholds based on PSA velocity; molecular measures; and patient age, ethnicity, and family history of prostate cancer can improve the predictive ability of this test (3, 4). Information from genetic markers that correlate with higher or lower basal PSA concentrations could further enhance the utility of the test. Now, Gudmundsson et al. have identified such PSA-associated markers in a genome-wide association study (GWAS) described in the current issue of Science Translational Medicine (5). This study represents a noteworthy advance toward clarifying the quandary surrounding PSA screening for prostate cancer.


The decisions to test serum PSA concentrations and what to do with the results are complicated and controversial ones. If a man’s serum PSA is elevated or rapidly increasing from one year to the next, he is generally referred for further evaluation and possible needle biopsies of the prostate, which are assessed for cancer (Fig. 1). PSA screening is especially helpful in detecting early-stage prostate cancer that responds well to surgery or radiotherapy and has resulted in a migration of tumors to lower-stage disease among U.S. men since the introduction of this test in the early 1990s (6). Although cancer screening is generally considered to be a good thing, serum PSA testing using conventional cutoffs to decide whether a man should be evaluated further (for example, PSA concentrations above 2.5 to 4 ng/ml) has mixed sensitivity and specificity for the diagnosis of clinically relevant prostate cancer (7).

Fig. 1 Making PSA screening selective.

Genes and the continuum from PSA testing to prostate cancer diagnosis. A large majority of older men in the United States have their PSA tested, and a reasonable proportion of such men have abnormal PSA levels and undergo needle biopsies of the prostate. Only about 20 to 30% of men with abnormal biopsies are diagnosed with prostate cancer, and roughly 3% of all men with this disease die from it (8). In an attempt to improve the PSA screening process, Gudmundsson et al. report SNPs in or near six genes associated with different steps in the sequence of PSA testing, needle biopsy of the prostate, and prostate cancer diagnosis. SNPs in the TERT, MSMB, and HNF1B genes were associated with both serum PSA concentrations and prostate cancer. SNPs in the KLK3 (which produces PSA), FGFR2, and TBX3 genes were associated with serum PSA concentrations and negative prostate biopsies. The SNPs in these latter three genes plus TERT were further evaluated as potential germline genetic markers for determining personalized PSA concentration cutoffs for follow-up and possible prostate biopsy (four genes highlighted in yellow). The authors did not include MSMB or HNF1B in these prediction models because of their strong associations with prostate cancer.


On the basis of data from the European Randomized Study of Screening for Prostate Cancer (ERSPC), ~10 to 20% of men who undergo PSA screening are recommended for follow-up, of which ~80 to 90% undergo needle biopsy of the prostate (8). Of the men biopsied, only ~20 to 30% will be diagnosed with prostate cancer, because an elevated or increasing PSA concentration can also reflect other conditions, such as prostatitis or benign prostatic hyperplasia (Fig. 1) (8, 9). Many of the PSA-detected early-stage prostate cancers will not cause ill health or require immediate treatment—which can have substantial side effects. Thus, between 20 and 60% of PSA-screened cases might be considered “overdiagnoses” (10, 11). On the flip side, low PSA concentrations do not rule out prostate cancer, whereas a high PSA concentration is correlated with more aggressive disease (9). These ambiguities could partially explain the findings of several large randomized clinical trials that have shown minimal or no mortality benefit from PSA screening (8, 12, 13).

The limitations of PSA screening have given rise to much debate about its usefulness; major medical societies have a range of opinions about whether and when a man should undergo regular PSA screening. The American Urological Association recommends that physicians offer PSA tests to men starting at age 40 in order to determine baseline levels and velocity (3). The American Cancer Society supports physician-patient discussions about the pros and cons of annual screenings starting at age 50, whereas the U.S. Preventive Services Task Force argues that there is insufficient evidence about the potential value of PSA screening to make a recommendation. Nevertheless, PSA testing for potential prostate cancer remains the standard of care in the United States, although there is agreement that men with life expectancies of less than 10 years should not have their PSA tested because any cancer detected probably will not merit treatment (3).


The traditional single-cutpoint PSA screening test can be improved by using more personalized thresholds to determine whether a man should be further evaluated for prostate cancer. Current recommendations are to take into account PSA velocity and patient age, ethnicity, and family history of prostate cancer and then apply varying PSA thresholds when considering whether to recommend further evaluation. One might also incorporate into the decision factors that affect PSA via inflammation, such as infections and the use of anti-inflammatory drugs (14), or other molecular markers (4, 15).

Another promising prospect is to integrate information about genetic factors that influence PSA concentrations independently of prostate cancer. There is a clear genetic component to serum PSA concentrations—a twin study estimated a 40% heritability in PSA concentrations, a magnitude similar to that seen with prostate cancer (16). A previous GWAS detected an association between a single-nucleotide polymorphism (SNP) near the gene that encodes PSA [kallikrein-related peptidase 3 (KLK3)] and prostate cancer (17). This SNP was subsequently associated with PSA concentrations in unaffected men, as were other prostate cancer–associated SNPs in the hepatocyte nuclear factor-1 β (HNF1B) and β-microseminoprotein (MSMB) genes (1821). These findings suggest that these genes and others may explain some of the inherent variability in serum PSA concentrations.

Gudmundsson and colleagues’ GWAS of serum PSA levels first focused on men without prostate cancer from the Icelandic Cancer Registry (effective sample size, ~6500); they then followed up the most promising results in nondiseased men from Iceland and the UK (5). They found that SNPs in or near the following six genes were associated with serum PSA concentrations: telomerase reverse transcriptase (TERT; chromosomal region 5p15.33, SNP rs2736098); MSMB (10q11, rs10993994); fibroblast growth factor receptor 2 (FGFR2; 10q26, rs10788160); T-box transcription factor TBX3 (12q24, rs11067228); HNF1B (17q12, rs4430796); and KLK3 (19q13.33, rs17632542). These genes and gene products may provide insights into the biological basis of variations in serum PSA concentrations or of prostate cancer. For example, HNF1B specifies a transcription factor expressed in prostate adenocarcinoma (22), the MSMB gene product has tumor suppressor properties (23), the FGFR2 gene product may lead to an increase in expression of the androgen receptor (24), and TERT and TBX3 encode proteins that may affect senescence (25, 26).

Expanding their population to include prostate cancer cases and focusing on men who were undergoing needle biopsies, Gudmundsson et al. found that half of these SNPs were also associated with a cancer-negative prostate biopsy (rs10788160 at FGFR2; rs11067228 in TBX3; and rs17632542 in KLK3) (5). Moreover, the first two of these SNPs were specifically associated with serum PSA concentrations and not prostate cancer. Figure 1 summarizes the points at which these SNPs may influence the continuum from PSA screening to prostate biopsy to cancer diagnosis.


Although these SNP-PSA associations are highly statistically significant, a critical next question is how much incorporating these SNPs into the assessment of serum PSA concentrations improves the decision of whether to perform a needle biopsy and the resulting prostate cancer outcome. Gudmundsson et al. used the four SNPs most clearly associated with PSA concentrations (rs2736098 in TERT; rs10788160 at FGFR2; rs11067228 in TBX3; and rs17632542 in KLK3) to estimate personalized PSA cutoffs in two European populations from Iceland and the UK (5). These cutoffs reflect each man’s PSA concentration and genotypes at the four SNPs and distinguish whether he should be followed up further for prostate cancer. First, they characterized the substantial variability in PSA cutoffs across the study populations. Then, they investigated how well applying these cutoffs to decide on further follow-up (for example, performing a needle biopsy) predicted the ensuing prostate cancer outcome by calculating the area under the receiver operating characteristic curve (AUC). The AUC reflects the relative trade-offs between true positives (PSA test “positive,” prostate cancer) and false positives (PSA “positive,” no prostate cancer).

In the Icelandic and UK populations, the AUCs for PSA alone were 70.4 and 57.1%, respectively. Using the four-SNP individualized PSA measure only slightly increased these AUC values to 70.9% and 58.5%, respectively. As expected, adding 23 SNPs that are associated with prostate cancer to their outcome prediction model further increased the AUCs to 73.2 and 63.6%, respectively. The substantially higher AUCs for the Icelandic population may reflect the fact that the PSA SNPs were discovered in this population or how the UK subjects were sampled.

The limited improvement in AUCs in part reflects the modest effects of the four associated SNPs on PSA concentrations. Figure 2 depicts the potential shift in cutoff points and highlights the challenges of achieving a substantial increase in AUC values when using personalized PSA concentrations. These difficulties are not unique to this study; previous applications of results from numerous other GWASs to prediction models have resulted in little improvement in AUC values [for example, in breast cancer (27)].

Fig. 2 PSA and predicting outcome.

These graphs illustrate the impact of using a gene-based PSA cutpoint on the accuracy of the test for recommending follow-up and ultimately predicting the outcome of a prostate biopsy for cancer. Plotted is the distribution of serum PSA concentration values among prostate cancer cases and controls from the Swedish Västerbotten Intervention Project, a population not screened for prostate cancer (29). (A) Applying the traditional PSA concentration cutoff of 4 ng/ml would have resulted in approximately half of the cases not being followed up, which is a high false-negative rate (area with yellow stripes). This large percentage of cases with PSA concentration values of <4 ng/ml may reflect the lack of PSA screening in this population. (B) Using personalized PSA cutoffs based on gene variations increases the overall amount of true positives [yellow shaded area minus the yellow striped (false negative) area] and true negatives (blue area). The notched cutoff line is formed from extrapolating Gudmundsson et al.’s personalized PSA cutoff values that were calculated in an Icelandic population after incorporating information on SNPs in the four PSA-associated genes (KLK3, FGRF2, TBX3, and TERT) (Fig. 1) (5). These personalized cutoffs offer a limited improvement, however, because of the small net gain in accurate reclassification and the large overlap in case-control PSA concentration distributions.


In addition to the AUC, one might consider the ability of personalized PSA cutpoints to reclassify men into different follow-up groups and possibly improve the accuracy and clinical utility of such a test. Gudmundsson et al. showed that ~6 to 7% of Icelandic men had at least one PSA value reclassified into higher or lower prostate cancer risk categories when incorporating information from the four PSA SNPs (5). It is unclear, however, whether reclassifying a single PSA test would be cause for reclassification of a man who has multiple differing PSA results and whether this would ultimately lead to the most appropriate follow-up with regard to prostate cancer outcomes. Although this study represents an auspicious beginning, on the basis of the AUCs and reclassification analysis it does not appear that SNP-adjusted PSA values are primed for translation into clinical practice without additional follow-up studies.


The new data from Gudmundsson et al. provide an important advance toward the development of genotype-based individualized PSA concentration cutoffs for prostate cancer screening and in our understanding of the biology underlying PSA levels. Finding additional genes that independently influence PSA levels may aid in the development of personalized PSA testing with improved sensitivity and specificity over what is now available. Moreover, incorporating as much information as possible into PSA testing—genetics, molecular markers, PSA velocity, age, ethnicity, and family history—can also strengthen the predictive value of this test (4, 28). Data from well-powered clinical trials or large prospective cohorts will spur the detection of additional genetic markers and clarify the translational value of SNP-adjusted PSA testing. Ultimately, the enhancement of PSA testing with genetic and other information may substantially improve the screening, diagnosis, and treatment of prostate cancer.


  • Citation: J. S. Witte, Personalized prostate cancer screening: Improving PSA tests with genomic information. Sci. Transl. Med. 2, 62ps55 (2010).

References and Notes

  1. The author thanks I. Cheng and A. Reese for their comments and J. Nelson for help with Fig. 2. Supported by NIH grants CA88164 and CA127298.
View Abstract

Navigate This Article