Molecular mechanisms of the preventable causes of cancer in the United States.

The idea that cancer risk can be reduced based on modification of behavior or the environment or screening for specific risk-associated mutations is predicated on the idea that these factors are major or predominant contributors to the absolute incidence of cancer. However, an opposing viewpoint is that the major source of cancer risk is biologically programmed and cannot be avoided. This idea is based on the relationship between three key observations. It has long been known that some organ systems are more prone to cancer than others. It has also been long known that some organ systems have a greater regenerative capacity than others, based on increased proliferative potential of individual cells. Finally, cancer risk correlates with age. The significance of the relationship between proliferation rate, organ specificity of cancer risk, and aging has been discussed for more than a century (for review, see Goss 1966 ; Tomatis 1993 ). The idea linking these observations is that specific organs undergo replicative replacement throughout the life span of an individual, and this replacement process is marked by an unavoidable error rate, resulting in the gradual acquisition of sets of cancer-promoting mutations ( Armitage and Doll 1954 ; Knudson 1971 ). Adding relevance to this otherwise philosophical debate is that if a stochastic process is the major driver of cancer risk, the rationale and motivation for devoting significant efforts to prevention are undercut; in contrast, if biological programming is a minor risk factor or can be modified, prevention is strongly justified. In a highly provocative report, Tomasetti and Vogelstein (2015) used a statistical approach to correlate available information about the stem cell complement of individual tissues in 31 distinct tissue types to estimate the total number of stem cell divisions possible for that organ type and then plotted the results against age of incidence for cancers affecting each of the tissue types for all tumors reported in the U.S. in the Surveillance, Epidemiology, and End Results (SEER) database. This resulted in an extremely high linear correlation of >0.8 by Spearman's ρ or Pearson's linear determination ( P < 5 × 10 −8 ), leading the investigators to assert that 65% of the difference in the risk of cancer among distinct tissues related to stem cell divisions over time. They then defined an extra risk score (ERS) as a measurement of overall cancer risk across a lifetime. Based on these calculations, they estimated that replication rate was sufficient to explain risk for a large number of cancer types that were described as replicative and that prevention was unlikely to be productive for these tumors, which could be ascribed to “bad luck.” In contrast, prevention would be useful for a smaller group of deterministic tumors, where a contribution of environmental or hereditary risk factors could be inferred. This latter group included lung cancer in smokers, cancers associated with HCV or HPV infection, and gastrointestinal tumors associated with hereditary mutations. In a follow-up 2017 study, Tomasetti et al. (2017) extended their work to analyze cancers worldwide using statistical methods and analysis of driver mutations to separate the relative contribution of environmental, replicative, and hereditary effects. Although finding trends similar to those in their earlier analysis, this more comprehensive study led to a more nuanced conclusion, maintaining the emphasis on replicative effects for many tumors but also noting that, for certain tumor types, environment made a major contribution. In sum, the investigators estimated that ∼29% of cancers arise from environmental mutations and are potentially preventable. The arrival of high-throughput sequencing techniques, making large numbers of cancer genomes available for inspection, allows a reformulation of this debate in molecular terms. Alexandrov and colleagues ( Alexandrov et al. 2013 ; Alexandrov and Stratton 2014 ) have developed algorithms that assess distinct categories of somatic mutation events to identify underlying signatures that characterized individual tumors. The initial study, analyzing 4,938,362 mutations from 7042 cancers, identified 20 distinct signatures, of which a subset was associated with the age of the patient at cancer diagnosis. A subsequent study focused specifically on these “clock-like” mutational signatures, now analyzing 7,329,860 somatic mutations from 10,250 cancer genomes ( Alexandrov et al. 2015 ). This expanded analysis now identified 33 mutational signatures. For two of these signatures (nos. 1 and 5), which represented 23% of total mutations detected, the number of mutations increased with age ( P < 10 −253 ) in 26 out of 36 types of cancer assessed. Cancer types marked by this signature included stomach, colorectal, glioblastoma, esophagus, medulloblastoma, and pancreatic, which include several tissue types associated with high replication rates, in accord with the idea that replication-associated defects are particularly relevant in these tumors. Signature 1 appears to be associated with deamination of 5-methylcytosine at CpG dinucleotides, which causes T:G mismatches that are not effectively repaired at replication. In contrast, signature 5 is elevated in other tumor types, including kidney papillary and clear cell cancers and neuroblastoma, and involves C > T and T > C transitions with a transcriptional strand bias, suggesting a possible link to transcription-coupled repair. The investigators hypothesize that the specific elevation of signature 5 in specific kidney tumor types may reflect exposure to a metabolism-associated mutagen abundant in renal tissue, but the mechanism is currently unclear. Both of these age-associated signatures are present at a relatively low level in other common cancers, including breast, melanoma, ovarian, and AML. Furthermore, the fact that even the two aging-associated signatures do not correlate with each other suggests the involvement of some tissue-specific component exclusive of aging. Importantly, after removal of these two signatures from the overall data set, there was no significant correlation between age and number of mutations, which reflects the remaining 77% of mutations detected ( Alexandrov et al. 2015 ). This suggests that, for these sources, the contribution of replicative effects (replication error) is low and that most of the risk is associated with either hereditary or environmental effects. Although it is likely that further genes associated with hereditary risk will emerge, this pattern suggests a potentially large contribution of environmental factors and a similarly large role for prevention. While it is somewhat difficult to discern the most “important” cancer-inducing mutational source for any given tumor, given the simultaneous presence of multiple signatures (for review, Alexandrov and Stratton 2014 ), it is clear that there are specific signatures that are associated with specific environmental or behavioral factors, including tobacco smoke ( Alexandrov et al. 2016 ), aristolochic acid ( Hoang et al. 2013 ), and others. Hence, the idea of “bad luck” due to factors such as replication error during cell turnover should be interpreted holistically as one cancer-predisposing element in addition to, or complemented by, preventable procarcinogenic factors. Another intriguing observation that suggests the importance of modifiable environmental factors is the fact that, even for tumors where there is currently strong evidence for a correlation between abundant stem cell population, high replicative potential, and an age-associated signature of mutations, the pattern of tumor incidence is changing in the general population. For example, while the incidence of colorectal cancer is decreasing overall in the U.S., it is increasing among younger adults, with individuals born in 1990 having twice the risk of colon cancer and four times the risk of rectal cancer as those born in 1950 ( Bailey et al. 2015 ; Siegel et al. 2016 , 2017a , b ). Such an observation is difficult to explain solely through a stochastic model based on stem cell pools unless one assumes that the size of the stem cell pool is itself affected by factors such as environmental toxins and obesity. This is not inconceivable; a number of proteins that support stem cell self-renewal potential have been shown to be up-regulated and promote aggressive tumor aggressiveness in cancers (e.g., Kudinov et al. 2017 ) and may have altered expression based on such modifiable factors. Furthermore, as deep genome analysis now begins to address clonality as a critical feature of the emergence of tumors, it is becoming recognized that mutational patterns within a single tumor mass—or even within morphologically normal tissue—can be highly complex, creating uncertainty about absolute mutation rates ( Cooper et al. 2015 ). Deep comprehensive molecular surveys of tumor genomes are an emerging field; ultimately, the ability to compare the mutational spectrum and incidence patterns of tumors diagnosed over multiple decades should definitively inform this debate. At present, the sum of the data available supports the idea of an important role for environmental and behavioral contribution to cancer risk. While multistep models may explain the dramatic increase in cancer incidence with age, they do not preclude the possibility that complex and potentially reversible aging-related processes might contribute to cancer through systemic changes that favor tumor growth ( Campisi 2003 , 2013 ). These aging-related changes may differ from organ to organ and can include diverse processes such as impaired immune response, defects in DNA repair, and altered hormonal environment. As data in support of this idea, mutations that extend life span in mice delay the onset of diseases of old age, including cancer. Furthermore, senescent cells that accumulate with age secrete factors that can be inflammatory, can promote angiogenesis, and can favor the growth of cancer in mouse models ( Campisi 2013 ). In the immune system, the phenomenon of age-related clone hematopoiesis (ARCH), also known as age-related clonal expansion, describes the reduction in clonal diversity among hematopoietic stem and progenitor cells that gradually reduces the functionality of the immune system ( Shlush 2018 ). The molecular basis for ARCH is unclear but is likely to represent a combination of cell-intrinsic mutations and microenvironmental effects; individuals with ARCH are at higher risk for some forms of cancer, including nonhematological cancers ( Forsberg et al. 2014 ). Importantly, the presence of ARCH is correlated with diabetes ( Bonnefond et al. 2013 ), although whether this correlation reflects causation (in either direction) is not yet clear, and ARCH is also correlated with smoking ( Coombs et al. 2017 ). Hence, this aging-related deficiency may be at least partially controllable through prevention methods. Such observations have stimulated efforts by companies to develop drugs to prevent cancer by delaying aging itself or at least eliminate aged cell populations. Some agents, broadly termed “senolytics” ( Zhu et al. 2015 ), focus on selective removal of senescent cells by various mechanisms; for example, ABT-263/navitoclax, an inhibitor of the anti-apoptotic BCL2 and BCLXL, selectively removes cells that have senesced in response to irradiation or due to normal aging, causing apparent rejuvenation of the hematopoietic system ( Chang et al. 2016 ; Zhu et al. 2016 ). Similar senolytic effects were seen for other targeted therapies, including dasatinib, HSP90 inhibitors, and other agents ( Zhu et al. 2015 ; Fuhrmann-Stroissnigg et al. 2017 ).

Cancer 5-yr survival rates vary substantially between anatomic sites and depend on the size, grade, and stage of the tumor. Stage refers to whether the tumor is local (confined to the organ of origin) or has metastasized regionally (has extended beyond the organ of origin to surrounding tissues or lymph nodes) or distantly (to remote tissues). For all tumors, 5-yr survival rates are best if the tumor is detected at the local stage, although, for some tumors, even early detection is associated with poor survival because of the current lack of effective treatment strategies. For instance, for breast cancer, survival rates are 99% (local), 85% (regional), and 26% (distant) because of excellent therapeutic options, whereas for pancreatic cancer, these rates are 29%, 11%, and 3% (The American Cancer Society 2016, https://www.cancer.org/content/dam/cancer-org/research/cancer-facts-and-statistics/annual-cancer-facts-and-figures/2016/cancer-facts-and-figures-2016.pdf ). Thus, a major means of lowering cancer mortality for many cancers is to detect them at the local (or even regional) stage and treat them promptly. For many cancers detected at the local stage, surgical resection may be the only treatment recommended and may be curative. The chief means of early detection for many cancers is the recognition by the patient that something is awry following the appearance of characteristic signs or symptoms (e.g., the appearance of unusual moles) ( Swetter et al. 2016 )—hence, campaigns such as the American Cancer Society's “seven warning signs of cancer,” designed to alert patients to visit their doctors and request diagnostic assessment while the cancer is at an early stage. This awareness alone, combined with increased access to health care or the means to pay for it such as Medicare and Medicaid, in the U.S. led to decreases in the proportion of cancers that was detected as distant or late stage, so-called “down-staging.” This is still an underused strategy, particularly in less developed countries ( Sankaranarayanan and Boffetta 2010 ). Other routes to early detection include “opportunistic screening,” in which recommendations for cancer testing are made on a national level but the actual action on the recommendation is up to the individual (the approach used in the U.S.), or “organized screening,” in which testing is systematically offered to asymptomatic high-risk individuals or the general population based on a national program (for example, in Scandinavia). A broad portfolio of evidence-based tests proven to reduce cancer-associated mortality and suitable for application in large populations is currently available to screen for cancer. Additional options ranging from classic medical approaches to new tests based on molecular signatures and other recently established biomarkers are in development. Methods for population testing include fecal occult blood or immunochemical screening or endoscopy of the lower gastrointestinal tract (colorectal cancer), visual inspection of the cervix or cytology (cervical cancer), detection of an oncogenic virus (e.g., HPV for cervical cancer), mammography (breast cancer), radiology (e.g., spiral computed tomography [CT] for lung cancer), or blood-based biomarkers (e.g., PSA [prostate-specific antigen] for prostate cancer) ( https://www.uspreventiveservicestaskforce.org/Page/Name/recommendations ). These screens fall into two fundamental classes: those that identify premalignant lesions and remove the damaged tissue (e.g., removing adenomas at colonoscopy) and those that indicate that a cancer may be found on further searching and/or biopsy (e.g., fecal occult blood screening). Notably, there are multiple options for some types of cancer, such as breast, colorectal, lung, and cervical cancers, which account for a high proportion of cancers globally. However, there are no feasible options available to screen at the population level for most other types of cancer, highlighting an area where investment in test development might lead to major public health dividends. Also notable is that the equipment and expertise needed to screen varies substantially according to the organ site, and thus screening programs tend to be site-specific, and there are few economies of scale across sites. Screening tests are generally evaluated in terms of their sensitivity (the percent of true disease positives, who are called as positive by the screen) and their specificity (the percent of true disease negatives, who are called as negative by the screen). In general, sensitivity needs to be high (e.g., 60%–80%) such that a high proportion of cases is detected by the test. Specificity needs to be even higher (e.g., ≥98%) when the probability of disease is low, as when most cancer screening tests are applied in the general population. This is in part so that the proportion of true negatives (i.e., healthy individuals) who are screen-positive is small, and thus few people suffer the anxiety of a cancer concern and the potential morbidity associated with further diagnostic testing, which is costly and often invasive. Also, at a population level, high specificity is important so that the health system is not overwhelmed by expensive and unjustified follow-up testing (although here the specificity requirement depends on whether the consequences of being labeled positive are relatively minor [e.g., referral to a dermatologist for a skin biopsy] or burdensome [e.g., laparoscopy to diagnose or exclude ovarian cancer]). Another key performance characteristic is the positive predictive value (PPV); i.e., the proportion of screen positives that are true positives. A good screening test will have a high PPV. An important fact about this metric that is not necessarily intuitively understood is that the PPV of a test varies with the prevalence of the disease being tested for. One way of understanding this is that, at the extremes, the PPV will be zero (if there are no people with the disease in the population tested, then all of the screen positives will be false positives) or 100% (if all people tested have the disease, then all of the screen positives will have the disease). Thus, the PPV varies according to whether the disease is rare or common in the population screened; the rarer the disease is in the population, the higher the fraction of test positives who are false positives will be. This obviously has implications for test development for cancers that are relatively common (e.g., breast, prostate, and lung) versus rare cancers. A technically excellent screening test is no use if it is (1) too expensive to justify, (2) needs to be repeated too frequently to be feasible, (3) takes too long to generate a result, (4) identifies cancers at such a late stage that treatments are ineffective, (5) or is followed by confirmatory tests that do more harm than good or (6) if people cannot be convinced that they should be screened with the test. Thus, the development of a cancer-screening program involves many actors—the government or insurance companies, the people administering the test and following up the results, and the population being tested being prepared to submit to the test. Because the characteristics of each screening test are different and vary according to the population being screened and the level of development of the health system, no cancer-screening test is universally applied worldwide. One of the most successful screening tests has been the Papanicolaou (Pap) smear for prevention of cervical cancer. In this test, cells are collected from the opening of the cervix and stained with a mixture of five dyes selected to highlight cellular features, including the keratins found in squamous cell carcinomas. Subsequently, a pathologist evaluates slides to determine whether there is evidence of abnormalities that are characterized as low- or high-grade squamous intraepithelial lesions (LSIL or HSIL, respectively), which represent ∼2%–5%, and 50% over the last 40 yr; most of this decline is attributed to the Pap smear ( https://www.cancer.org/cancer/cervical-cancer/about/key-statistics.html ). By detecting both premalignant and malignant lesions, the Pap smear both reduces the risk of cancer by leading to the removal of premalignant tissues, which are scraped off by colposcopy subsequent to an abnormal test result, and reduces risk of cancer mortality by leading to early diagnosis of curable cancers. In many less developed countries, however, efforts to introduce the Pap smear have failed, due in part to the time it takes to get an answer (usually the slides are sent to a central laboratory for microscopy, by which time the patient may not be available for follow-up colposcopy). Other issues include the difficulty of getting results back to women who may not have a telephone, postal service, or other means of communication; high costs; the requirement for skilled personnel for interpretation of results; and the lack of personnel for follow-up treatment of positive or suspicious smears (as surgeons and operating theaters may ultimately be needed if a cervical amputation or hysterectomy is required). Fortunately, developments in molecular biology have enabled advances in cervical cancer prevention and screening in two ways. The finding that oncogenic strains of HPV are a necessary cause of cervical cancer ( Walboomers et al. 1999 ) led to the development of an HPV vaccine that protects women against the majority of the oncogenic HPV strains, including the common HPV16 and HPV18 strains. The two first developed and commercially available vaccines (Gardisil and Cervarix) were virus-like particles (VLPs) based on expression of the major L1 capsid protein derived from multiple oncogenic HPV strains. Initial clinical trials showed 100% efficacy of these vaccines in preventing cervical dysplasia in young women who were HPV-naïve at the time of vaccination 4 yr after administration of a single vaccine dose (for review, see Schiller and Lowy 2012 ). A current challenge is extending the use of the vaccine, including to young men as well as young women, which is particularly important given the rapidly growing incidence of cancers at noncervical anatomical sites, such as the head and neck ( Chaturvedi et al. 2011 ). Screening techniques have also improved based on exploitation of the obligate infection of the cervix with high-risk HPV strains prior to detectable Pap abnormalities. HPV testing can entail the use of PCR probes for high-risk HPV strains to amplify viral DNA or a hybrid capture approach to detect the oncogenic strains ( Clavel et al. 1998 ). With respect to screening, several randomized trials have shown that HPV-based screening is superior to Pap/cytology-based screening in protecting against invasive cervical cancer ( Ronco et al. 2014 ; Schiffman et al. 2017 ). Colorectal cancer is the second leading cause of death in the U.S. ( U.S. Preventive Services Task Force 2016 ) Early detection has a huge impact on preventing death from this disease, and fuller uptake of proven methods of early detection represents an opportunity to realize large gains from population-level screening. Tumors found through screening approaches at an early stage have a 5-yr survival rate of ∼90%; for metastatic colon cancers, the survival rate is 11%. The gold standard method of screening for colon cancer for the past several decades has been by colonoscopy, with a standard recommendation of commencing such testing at age 50 and then testing every 10 yr for members of the general population lacking known risk factors. For individuals with known hereditary risk factors for colorectal cancer (e.g., individuals with Lynch syndrome), similar tests are used but beginning at a much earlier age and with testing performed at frequent intervals ( Stoffel et al. 2010 ). Colonoscopy has been particularly successful because the vast majority of colorectal tumors has a similar life history, progressing through polyps to noninvasive adenomas to adenocarcinomas over many (≥10) years and rarely metastasizing until late stages of tumor growth; concurrent with detection of early stage polyps, these early premalignancies can be readily removed during screening. However, colonoscopy is often avoided due to annoying physical effects associated with clearing the gastrointestinal tract for scoping and due to a lack of access or affordability, with as many as 50% of individuals who would benefit not actually being tested. Alternative approaches, including flexible sigmoidoscopy, CT colonography (CoTCo), the guaiac-based fecal occult blood test (gFOBT), a fecal immunochemical test (FIT; also known as the immunochemical fecal occult blood test [iFOBT], which assesses hemoglobin), and a multitargeted stool DNA test, have become available ( U.S. Preventive Services Task Force 2016 ; Knudsen et al. 2016 ). Comparative assessments of these tests have indicated that for the general population, colonoscopy every 10 yr, annual FIT, sigmoidoscopy every 10 yr with annual FIT, and CoTCo every 5 yr yielded similar benefits in improving life span if the tests are applied between the ages of 50 and 75. The fact that noninvasive FOBT testing is performing so well suggests that this type of testing may be more readily adopted by individuals unwilling to undergo more invasive screening approaches. Although mammography has become established as a means to reduce breast cancer mortality, the data regarding the value of this approach to prevention are surprisingly controversial, with large disagreement on the extent of any reduction in mortality. Most consensus estimates suggest that regular mammography screening reduces breast cancer mortality by ∼20% ( Myers et al. 2015 ). However, this may come at a price of overdiagnosis of breast cancers (i.e., diagnosis of precancerous lesions that would not have been diagnosed as progressive tumors in a woman's lifetime) that some suggest has increased breast cancer incidence by ∼20% ( Kalager et al. 2012 ). The original rationale for mammography at the population level has been the idea that tumor progression goes through a set progression, where tumors reach a minimal size that is detectable by mammography before metastasis occurs. This paradigm has been challenged by a number of recent studies using genomic analysis, analysis of circulating tumor cells (discussed below), and sophisticated imaging techniques to analyze the timing of tumor dispersion. Emerging data suggest that early stage tumors secrete small extracellular vesicles that condition niches to enhance the growth of metastasizing cancer cells ( Peinado et al. 2017 ), with work in mouse models indicating that even very early stage mammary tumors shed circulating tumor cells that can seed such niches ( Harper et al. 2016 ; Hosseini et al. 2016 ). Similar early dissemination has been seen for other solid tumor types, such as pancreatic tumors ( Rhim et al. 2012 ). Further studies of pancreatic cancer have suggested that such metastases may arise from clonal populations within the larger tumor mass, further uncoupling measurement of tumor size from propensity to metastasize ( Yachida et al. 2010 ). This evolving understanding of metastasis ( Massague and Obenauf 2016 ) supports the idea that prevention approaches focusing on detection and genomic characterization of circulating tumor cells in the peripheral blood may significantly augment mammography in limiting breast cancer mortality (for instance, if effective interception strategies are developed that can control the establishment of metastases before primary tumors are detectable by imaging). A highly contentious screening test is regular testing of men for PSA in order to detect prostate cancer. PSA screening, which became popular in the U.S. and was recommended as an annual test for middle-aged and older men by several expert bodies, was not widely introduced in most European countries. In 2008, the U.S. Preventive Services Task Force recommended against screening men >75 yr and concluded that the evidence was insufficient to the balance of benefits and harms ( U.S. Preventive Services Task Force 2008 ). The results of randomized trials of screening compared with usual care have been much debated, with many men in the nonscreening arm of the major U.S. trial being screened as part of usual care. The most solid conclusions are about the difficulty of doing randomized trials of a screening method that has achieved broad popularity. There is no doubt that PSA testing results in the detection of lesions that pathologists label as cancer in a high proportion of men. Problems with this test include the fact that heterogeneity of prostate cancer aggressiveness among individuals makes it difficult to predict which individuals will develop a life-threatening disease. Many of the lesions detected following a positive PSA test may not have been symptomatic during a man's life, and the follow-up procedures and treatments lead to morbidity (urinary incontinence and/or impotence) in a substantial proportion of treated men. Even if screening results in a net reduction of site-specific cancer mortality, the trade-off against decreased quality of life in treated survivors is a difficult balance for any screening test for which the follow-up confirmation of the diagnosis and/or the treatment of the disease is burdensome. Blood-based tests such as PSA screening start with the advantage that the initial screening test does not require expensive equipment (such as spiral CT machines for lung cancer screening), unpleasant preparation (such as the bowel prep for colonoscopy), exposure to radiation (such as mammography), or clinical skills (such as the Pap smear). Thus, there is enthusiasm for the concept that finding tumor markers in the blood may provide a method of routinely screening large populations and potentially replace some of the more burdensome methods ( Aravanis et al. 2017 ). The use of NGS to detect tumor mutations in circulating tumor DNA (ctDNA) that is released by apoptosis or necrosis of tumor cells and/or techniques to detect circulating tumor cells have been shown to identify minimal residual disease and indicate prognosis and treatment response in some types of cancer ( Bardelli and Pantel 2017 ). The sensitivity of these tests may be limited by the fact that most cell-free DNA is derived from normal cells, and the specificity may be limited by the fact that cancer-associated mutations increase with age in tissues that are not yet cancerous ( Bardelli and Pantel 2017 ). Other liquid biopsies focus on the analysis of extracellular vesicles (EVs; comprising exosomes, microvesicles, and oncosomes): small membrane-encased vesicles shed from normal and tumor cells that can transfer nucleic acids and proteins between cells ( Strotman and Linder 2016 ). EVs derived from tumors have attracted great interest, based on evidence that they can perform functions ranging from conditioning the premetastatic niche to immune suppression to control of angiogenesis ( Sato and Weaver 2018 ). In spite of these challenges, there have been some convincing recent studies that tests integrating analysis of tumor proteins and DNA will significantly augment the ability to use “liquid biopsy” to detect early tumors noninvasively and to provide information on the specific mutations present in tumors in cancer patients ( Cohen et al. 2017 , 2018 ). Such testing is likely to become routine for some cancers, although the hurdles to converting this information into a test for early detection that could be applied to individuals in the general population remain high. It will be imperative to develop more selective and efficient capture techniques, incorporate advances in bioinformatics and computation, and obtain greater insights and context from cancer biology to further research in this high-priority area of cancer prevention (see also Lewis et al. 2018 ; Neoh et al. 2018 ).

When estimating cancer risk and developing strategies for prevention and early detection, it is important to consider that within a general population, the risk of individuals for specific types of cancer can vary significantly based on inherited factors. Over the past two decades, the general understanding of factors causing genetic predisposition to cancer has increased significantly. Earlier studies focused on a small number of genes that were discovered through observations in multiple affected family members (“loaded pedigrees”) and settings with cases enriched in specific populations or ancestries (e.g., Ashkenazi Jews). The variants discovered in such studies were often highly penetrant (forms of genetic variation strongly predisposing to cancer) for a limited number of cancer sites. Examples of genes where inactivating or hypomorphic mutations lead to such strong signals include the BRCA1 and BRCA2 genes, variants in which predispose strongly to breast and ovarian cancer ( Welcsh and King 2001 ), and the set of mismatch repair genes associated with Lynch syndrome ( MSH2 , MLH1 , MSH6 , and PMS2 ) ( Jasperson et al. 2010 ). Another example for colon cancer involves inherited genetic variants in the APC gene that lead to familial adenomatous polyposis (FAP), affecting one in 10,000 individuals and conferring ∼95% risk of colorectal cancer by the age of 50 ( Jasperson et al. 2010 ). Rare mutations in the Fanconi anemia (FA) genes are associated with very high risk of AMLs and HNCs and are at particular risk from UV and other forms of irradiation and tobacco smoke ( Romick-Rosendale et al. 2013 ). For the breast and colon cancers, it is estimated that up to ∼5% of total cancers are associated with such penetrant forms of genetic variation. For carriers of these variants, active surveillance starting at an early age, coupled in some cases with management of lifestyle exposures and prophylactic surgery (e.g., mastectomy/oopharectomy for BRCA mutation bearers), is routine ( Wei et al. 2010 ; Kanth et al. 2017 ). It is now clear that such relatively common highly penetrant damaging genetic variants represent only a small subset of cancer-predisposing inherited variants. For example, studies of Scandinavian twins and other population-based studies have estimated that up to 30% of the risk for colorectal cancer (including “sporadic” or nonfamilial cases) can be attributed to inherited genetic variation but that these variants may be rare and/or of modest penetrance ( Lichtenstein et al. 2000 ). Indeed, the majority of risk-associated gene variants in an entire population is expected to be of low to intermediate penetrance. Many of these genes are dispersed across a network of proteins associated with control of the DNA damage machinery, directly targeting either proteins involved in DNA repair ( Srivas et al. 2013 ; Nik-Zainal et al. 2014 ; Arora et al. 2015 ; Nicolas et al. 2015 ; Shlien et al. 2015 ) or proteins regulating the activity of DNA repair proteins. For example, panel testing for a signature of BRCA gene deficiency identified such a signature in 22 tumors with BRCA1/BRCA2 genetic lesions and 47 tumors without such lesions, likely due to defects in BRCA regulatory proteins ( Davies et al. 2017 ). Some predisposing variants are associated with specific types of cancer and sensitivity to specific controllable factors; for instance, a variant in the BAP1 gene is predisposing to a small group of cancers, including mesothelioma ( Murali et al. 2013 ) and greater cancer risk from exposure to asbestos ( Testa et al. 2011 ) and other environmental carcinogens ( Bononi et al. 2017 ). Many variants of unknown significance (VUSs) occurring in the general population remain to be assigned for function. Ongoing efforts to characterize the import of such VUSs combine exome analysis and functional testing for phenotypic effects on DNA repair (e.g., see Arora et al. 2015 ). Large databases, such as ClinVar and others, are systematically compiling information found in affected individuals, the general population, and other populations of interest (for example, the healthy elderly), with the goal of generating a resource that can be used to support statistical estimations of gene–risk correlation ( Bodian et al. 2014 ; Erikson et al. 2016 ; Rancelis et al. 2017 ). The NGS technological revolution has driven down the costs required to discover the complete set of genetic variants in large cohorts of individuals, allowing large surveys of variation in the exomes of cancer cases and associated population-based controls. Such studies may yield discoveries of rare forms of variation that confer intermediate (and potentially actionable) levels of risk; such forms of variation would have been missed by large-scale surveys of common variation. Aside from the daunting number of candidate cancer risk genes to be assessed, a number of confounding factors complicate risk prediction based on analysis of genes that function in an autocrine manner to prevent normal cells from undergoing transformation. For example, some known somatic driver mutations in the genes ARID1A , PIK3CA , KRAS , and PPP2R1A have been found in the endometriotic lesions of 19 of 24 patients with deep-infiltrating endometriosis even though this form of endometriosis almost never undergoes malignant transformation ( Anglesio et al. 2017 ; Dawson et al. 2018 ). How tissue microenvironment versus cell-intrinsic factors restrains the transforming effect of these mutations remains to be established. In contrast to the longtime focus on cancer risk arising from inherited gene variants affecting the cell that becomes the tumor, a growing field of research addresses non-tumor-intrinsic inherited and noninherited features that influence the ability of a mutated cell to progress. Broadly speaking, these changes affect the tumor microenvironment—a compartment composed of multiple untransformed cell types, including both immune system and stromal cells, as well as secreted insoluble proteins of the ECM and associated soluble factors. This vast topic cannot be summarized in any depth here (for reviews, see Lu et al. 2012 ; Gajewski et al. 2013 ; Faurobert et al. 2015 ). We focus on two examples. A growing body of data indicates an important role for the host immune system in the surveillance and elimination of cancer cells, with the recognition that escape from immune restriction is an essential transition at early stages of tumor growth (for review, see Dunn et al. 2002 ; Koebel et al. 2007 ). This role for the immune system accounts for the well-known elevated rates of multiple forms of cancer in immunosuppressed patients, such as solid organ transplant recipients, where 32 distinct malignancies occur at elevated rates ( Engels et al. 2011 ). In an exciting recent study, Marty et al. (2017) explored the hypothesis that one inherited factor regulating the emergence of tumors is individual variation in the ability of the immune system to recognize specific common transformation-associated mutations (for example, the G12V and G12D mutations of KRAS or the R175H mutation of TP53 ). By modifying existing algorithms to study the ability of MHC proteins to present peptides to the immune system, they were able to rank >300 common MHC-1 alleles dispersed in the population for their ability to present peptides bearing such common oncogenic driver mutations. They subsequently compared the co-occurrence of MHC-1 HLA-A, HLA-B, and HLA–C alleles with high or low presentation capacity with specific common oncogenic mutations across 1018 likely driver mutations found in a set of 9176 tumors in The Cancer Genome Atlas. This led to the conclusions that some common drivers are uniformly poorly presented by MHC-1 alleles and also that there existed a strong correlation between an MHC-1 profile associated with poor presentation and the likelihood of the mutation being present in a patient's tumor. This offers a new strategy to qualify the risk associated with specific inherited genetic variants or somatic mutations ( Marty et al. 2017 ), which is potentially relevant to other immune system antigen recognition components. However, in assessing genetic contributions, it is also important to keep in mind the fact that immune system contributions can be potently influenced by the behavioral and environmental factors discussed above. As only one example, inflammatory signals associated with obesity alter the landscape of immune cells in a manner that promotes metastasis ( Quail et al. 2017 ). Finally, the immune system is subject to declining or aberrant function in age (e.g., Palmer et al. 2018 ; Shlush 2018 ), making the efficiency of immune surveillance inextricably linked to the process of aging. The roles of stromal tissue and the ECM in regulating tissue and tumor growth have been long appreciated. In 1889, Paget's proposal ( Paget 1889 ; for review, see Fidler 2003 ; Oskarsson et al. 2014 ) that “soil” is as important as “seed” in targeting the growth of cancer metastases first laid out the concept that spatially restricted extratumoral signals may be essential for specifying niches capable of supporting tumor growth. As early as 1911, Peebles’ studies ( Peebles 1910 ) of limb bud engrafting between different sites during chick embryogenesis indicated that the extracellular environment could profoundly influence the fate of tissue differentiation, converting the limb bud fate to that specified by the new environment. Exploration of these observations led to the recognition that the ECM per se could profoundly affect the gene expression and signaling properties of associated cells in a process termed “dynamic reciprocity” ( Bissell et al. 1982 ; Lin and Bissell 1993 ). This influence of the ECM represents the contribution of both specific proteins present in the tumor microenvironment and also the overall architecture of the tissue matrix and degree of rigidity. These contributions are highly relevant to all stages of cancer growth. For example, it is well established that mammographic density, which reflects ECM rigidity, is one of the strongest predictors of breast cancer risk ( Burton et al. 2017 ). This density differs between distinct populations, being, for example, higher in Asian and lower in European women. The reasons for these differences are thought to involve height, weight, and parity ( Rajaram et al. 2017 ), although twin studies have indicated at least some genetically heritable component that is not yet well understood ( Boyd et al. 2009 ). However, regardless of the basal density of the mammary tissue, within genetically homogeneous populations, higher breast density is associated with greater cancer risk ( Bae and Kim 2016 ). These relationships clearly imply a role for breast density and ECM rigidity in creating a microenvironment that elevates the risk associated with any tumor-intrinsic initiating mutation. Regional matrix stiffness is significantly increased as tumors begin to grow beyond microscopic precursor lesions, based on a feedback between nascent tumors and surrounding stromal cells that increases tension; these changes promote aggressive growth ( Paszek et al. 2005 ). Some specific somatic mutations that occur in tumors, such as disruptions affecting the TGF-β pathway in pancreatic tumors, act in part by increasing ECM rigidity and fibrosis in the microenvironment, increasing tumor aggressiveness ( Laklai et al. 2016 ). As a further point of complexity, ECM rigidity varies dynamically over the life span, influenced by secreted signals from the senescent cells that accumulate in aging individuals ( Lecot et al. 2016 ) and also by DNA damage response ( Rodier et al. 2009 ). Taken in sum, these studies clearly indicate that cancer risk arising from behavioral or environmental factors likely influences the tissue “soil” as much as the tumor “seed.”

Since the war on cancer began, there have been continual improvements in survival from many cancers thanks to advances in imaging, surgery, radiation, chemotherapy, and a handful of adjuvant therapies such as tamoxifen. In parallel, there have long been significant efforts to develop resources in education and infrastructure to support screening and prevention behaviors in the general population ( Engstrom 1983 ; Amsel et al. 1987 ; Fleisher et al. 1988 ). Today, thanks to stunning research progress that has led to a much better understanding of the biology and genetics of cancer, immunotherapy and some targeted therapies promise additional improvements in treatment outcomes over the next 20–30 yr. Such advances also raise the possibility that, with further research, some cancers might be prevented or treated successfully at very early stages by drugs or vaccines. Despite such success, it is probably fair to say that only by the broader use of proven methods of prevention and early detection, together with treatment, can one guarantee major reductions in current U.S. cancer death rates in the coming two decades. The fuller uptake of existing methods of prevention and early detection across the U.S. population would also contribute to decreasing disparities in health and longevity that arise from unequal access to the full benefit of these approaches. A meaningful discussion of this topic is beyond the scope of this review. For example, it is now well established in the cancer prevention community that the adverse impact of low SES on prevention is multidimensional and significant. Lower-SES individuals typically have higher rates of depression and anxiety, with attendant higher rates of tobacco and chronic/heavy alcohol use; less access to healthy food/safe environments to permit safe exercise and more obesity; more opportunities for exposure to cancer-associated microbes; and less access to medically based prevention in the form of vaccines, with the attendant lack/delays in evidence-based screening and early detection tests to relevant populations. Therefore, they have more opportunities to be infected with cancer-associated microbes, fail to undergo recommended screening, and experience delays in diagnoses (associated with later stages of disease at presentation), all of which contribute to poor outcomes. For these reasons, effective prevention will require much more than public education on the topic; indeed, most of the potential gains would require actions by governments or large social movements. As a starting point, the interested reader is directed to the following works, and references therein: Colditz and Wei (2012) , Stringhini et al. (2017) , and Marmot (2018) . However, for the molecular biology community, the topics in this review suggest some areas for productive research. For example, better understanding of critical carcinogenic effects of agents such as obesity, alcohol, and processed meats may lead to the development of targeted interventions that detoxify proximal mutagens and cancer promoters. Better understanding of the mechanism of immune surveillance in cancer may improve cancer vaccines in a manner tailored for individuals with distinct MHC haplotypes. Better understanding of the specific ways in which senescent cells negatively condition the tumor microenvironment may lead to prophylactic interventions that either eliminate specific senescent cell populations or blockade their negative effectors. Importantly, better understanding of how individual genetic variation interacts with specific environmental factors to regulate relative risk is likely to become ever more important as clinical care incorporates “precision” approaches and can help in the development of robust and personalized risk prediction models ( Kattan et al. 2016 ). Microfluidic capture approaches continue to enhance the process of early detection. With an ever clearer view of the mechanistic underpinnings of cancer risk, it becomes more possible to develop tools to counteract these processes, improving quality of life and survival.

To explore the extent of a maximal impact of prevention through uptake of idealized health-promoting behaviors and policies, we examined two recent studies that quantified these effects ( Kohler et al. 2016 ; Islami et al. 2017 ). Kohler et al. (2016) summarized modifiable risks from 10 independent large cohorts reported by groups, including the Women's Health Initiative, the National Institutes of Health–American Association of Retired Persons Diet and Health Study, and others. Cumulatively, these cohorts involved >1.5 million individuals and generated a rich data set, including hazard ratios for specific behavioral risk factors identified by the ACS and/or the World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR). From this information, they were able to estimate resulting preventable cancers and deaths. Islami et al. (2017) modeled the uptake of behavioral risk factors in the U.S. population and related this to cancer incidence and deaths reported in national registries. Notably, the two studies were roughly concordant in their findings, notwithstanding their differing methodologies. This concordance strengthened their conclusions and emphasized the promise of primary prevention in limiting cancer incidence, highlighting the potential of idealized population-wide adherence to recommended healthy behaviors and implementation of policies that reduce cancer risk. For convenience, we focus on the recent study by Islami et al. (2017) due to their use of population-attributable fractions (PAFs; the proportion of cancer incident cases and deaths that can potentially be prevented due to the elimination of risk factors) as a summary measure. Risk factors were obtained by a meta-analysis of multiple published studies to identify modifiable risk factors for which a causative role in cancer is supported by sufficient or strong evidence. Cancer occurrence and death data were reproduced from Islami et al (2017) , who obtained occurrence data from the CDC’s National Program of Cancer Registries ( http://www.cdc.gov/cancer/npcr/public-use ) and the NCI SEER program ( https://www.seer.cancer.gov ) and death data from the CDC’s National Center for Health Statistics ( https://www.cdc.gov/nchs ). Cancer incidence in the U.S. in 2014 was analyzed for 26 cancers for which contributions to incidence from potentially modifiable risk factors (including those discussed in earlier sections) have been demonstrated previously. These data were analyzed in the context of age- and sex-specific risk factor exposures to estimate PAFs; PAFs were then summarized by cancer site and risk factor as well as in aggregate. Islami et al. (2017) confirmed what has been suggested for years; namely, that a surprisingly large proportion of cancer can be prevented. Through primary prevention alone, they estimated that 45% of incident cancers and 45% of cancer deaths in the U.S. in 2014 were attributable to modifiable risk factors for which they evaluated effects. These factors included tobacco (cigarette smoking and secondhand smoke), “lifestyle” factors (including excess body weight; alcohol; diet, such as red and processed meats and low intake of fruits, vegetables, dietary fiber, and calcium; and physical inactivity), cancer-associated chronic infections ( H. pylori , HBV, HCV, HPV, HIV [human immunodeficiency virus, which promotes aggressiveness of several virally associated cancers], and HHV8 [human herpesvirus 8, which causes Kaposi's sarcoma]), and UV radiation (from natural and artificial sources). Figure 6 is an alternative presentation of findings from Islami et al. (2017) that simultaneously represents the scale of cancers prevented by elimination of modifiable risk factors (“attributable deaths”) and site-specific cancer mortality (annual for the U.S., 2014). Several observations jump out. First, the burden of cancer deaths of the lungs and part of the aero–digestive tract (organs heavily exposed to tobacco in smokers) is exceedingly high even compared with cancer deaths in other common cancer sites, such as the colorectum, breast, and pancreas. Second, these lung cancers reside high in the plot, illustrating that they are largely preventable—at once sobering and motivational. Third, multiple additional sites contributing large numbers of cancer deaths are highly preventable, with >50% attributable to preventable factors for colorectal, kidney, esophageal, and liver cancers as well as melanoma. Finally, cervical cancer is considered to be essentially 100% preventable (via elimination of persistent HPV infection in the population through vaccination). Cancer deaths attributed to modifiable risks. Each anatomically categorized cancer was plotted by the current annual deaths due to that cancer ( X -axis; total deaths in 2014) and the proportion of deaths attributable (and thus preventable by the elimination of risk factors) to the following modifiable risk factors: tobacco, UV exposure, infections, and Western lifestyle ( Y -axis; PAF). Circle size is in proportion to cancer-specific incidence (incident cases, 2014), and colors are assigned by SEER 5-yr relative survival estimates (2007–2013; https://seer.cancer.gov/statfacts ). Thus, although cancers of the breast and pancreas situate proximally, indicating an approximately equal number of total deaths and a similar PAF for the examined risk factors, breast cancer incidence is much higher (larger point size), and outcomes for breast cancer are far superior (bluish purple in color, indicating a >75% relative survival). Cancers (i.e., lung) shown in the top right are those for which we can achieve the greatest reduction in total cancer deaths by the population-wide adoption of healthy behaviors and policies, such as tobacco prevention/cessation or elimination. Cancers shown in the top left result in far fewer cancer-associated deaths but may be similarly profoundly reduced through population-wide adoption of healthy behaviors and policies (e.g., avoiding tobacco, cancer-associated infections, and harmful UV exposure). This figure was plotted based on data from Tables 2 and 4 of Islami et al. (2017) and from SEER ( https://seer.cancer.gov/statfacts ). We note that our presentation, as with Islami et al. (2017) , applies to cancers with nontrivial fractions of cases attributable to modifiable risk factors (i.e., primary prevention) rather than screening or early detection (i.e., secondary prevention). Prostate cancer, for example, is omitted. For attributable cancer cases (data not shown), a figure emerges that is similar to that for deaths. If considering only cancer incidence, colorectal and breast are shifted far to the right, with the large number of annual breast cancer diagnoses exceeding even those for lung cancer. The relatively lower mortality for these cancers reflects reductions achieved through treatment, early detection, and screening. Table 1 presents estimates of attributable cases and deaths by risk factor, with a total attribution from these risk factors estimated to be 660,000 cases (40% of the ∼1.6 million new cancers diagnosed in the U.S. in 2014) and >265,000 deaths (∼45% of the ∼588,000 U.S. cancer deaths in 2014). Tobacco (smoking and secondhand) is the largest contributor to both new cases and deaths, with the cumulative factors associated with an unhealthy lifestyle nearly as impactful for cancer incidence; the greatest contributors to deaths and cases among “lifestyle” factors are obesity (excess body weight; 7.8% PAF) and alcohol (6.5% PAF). The high ratio of cases to deaths for UV exposure is likely due to melanoma being amenable to earlier detection and a resulting relatively favorable survival. Annual estimates of U.S. cancer cases and deaths by attributable fraction in 2014 There exist caveats to these estimates, as acknowledged by Islami et al. (2017) , who wrote the study. Risk factors were assumed to influence cancers independently (not interacting and without correlative effects among cancers). This does not necessarily fully reflect what is known in cancer biology based on emerging evidence about the interaction of risk factors, as summarized above. We also note that the inclusion of only established effects and the incomplete nature of data in cancer registries lead to an underestimate of the total number of cancer deaths attributable to these factors. The stated percentage of “preventable deaths” or cases is thus very much a function of the cancers and risk factors included, the completeness of the surveys, and the population under study, which here is the U.S. population. Due to the specific choice of cancers and risk factors considered in this analysis, the number of deaths that we estimated as avoidable by primary prevention is an underestimate and should thus be considered as a lower bound. For example, prostate cancer is not considered here, but the deaths due to this disease are part of the whole in calculating the 45% figure; certainly, some of these deaths can be attributed to these risk factors or others with modest evidence of causality, some of which are yet to be discovered. We also note that the risk factor exposure estimates were from the most recent year available and thus not averaged over the lifetime of cases, the data for which were also measured in a single time point (i.e., 2014). While Islami et al. (2017) systematically broke out these effects by risk factor and gender and presented measures of uncertainty in their calculations, our goal here was to summarize briefly the theoretical impact of a utopian adoption of practices rather than advocate for any single behavioral change; still, the high potential impact of tobacco prevention/cessation remains stark.