- Email: [email protected]
Next-Generation Sequencing of Circulating Tumor DNA for Early Cancer Detection
Leading Edge
Commentary Next-Generation Sequencing of Circulating Tumor DNA for Early Cancer Detection Alexander M. Aravanis,1,2 Mark Lee,1,2 and Richard D. Klausner1,* 1GRAIL,
Menlo Park, CA 94402, USA author *Correspondence: [email protected] http://dx.doi.org/10.1016/j.cell.2017.01.030 2Co-first
Curative therapies are most successful when cancer is diagnosed and treated at an early stage. We advocate that technological advances in next-generation sequencing of circulating, tumor-derived nucleic acids hold promise for addressing the challenge of developing safe and effective cancer screening tests. Cancer-specific mortality from most types of solid tumors has barely decreased in decades, despite an exponential increase in our knowledge about cancer pathogenesis and significant investments in the development of effective treatments. The past few years have witnessed a dramatic success of immunotherapies in treating a subgroup of patients with a variety of tumor types, including lung, bladder, and kidney, as well as Hodgkin’s lymphoma and melanoma. While such breakthroughs offer the hope of prolonged survival for some patients with advanced cancers, finding cancers earlier would still afford the greatest chance for cure, given that the survival rates for patients with early diagnoses are five to ten times higher compared with late stage disease (Cho et al., 2014). By enabling diagnosis and localized treatment of early stage invasive cancers (and, in some cases, pre-invasive states), screening for cervical and colorectal cancers has contributed to significant declines in mortality from these diseases. Similarly, effective population screening paradigms for many other common and deadly tumor types are clearly needed to broadly reduce cancer-specific mortality. Challenges for Cancer Screening Despite the promise, the field of early cancer detection is filled with cautionary tales that highlight the challenges. Early stage solid tumors are small and are thereby difficult to be non-invasively distinguished from normal anatomic and biochemical variation. This ambiguity leads to detec-
tion algorithms that either miss a large number of invasive cancers or make the costly trade-off of over-diagnosing and consequently over treating. For instance, high false-positive rates from mammography in breast cancer screening, lowdose CT in lung cancer screening, and prostate-specific antigen (PSA) screening (Nelson et al., 2016a; Aberle et al., 2011; Chou et al., 2011) represent a significant cost to the healthcare system, with resulting mental and physical morbidity, and even mortality in some cases (Nelson et al., 2016b). Even where cancer screening has produced significant stage shifts, as with breast and prostate cancer screening, the impact on cancer-specific mortality has not been a predictable outcome (Berry, 2014). Multiple explanations may account for this phenomenon. First, early stage cancers detected by mammography and PSA are likely to be clinically insignificant or ‘‘non-lethal,’’ meaning that, in the absence of screening and treatment, these tumors would not have become clinically evident, metastatic, or contributory to patient mortality. Second, some lethal tumors diagnosed in an apparently localized state may already have disseminated with occult metastatic disease, making local treatment with surgery or radiation non-curative. Earlier diagnosis of these tumors would result in a ‘‘lead-time bias’’ without impacting cancer-specific mortality. Third, mammography and PSA may be sensitive for indolent tumors, but not sufficiently sensitive for lethal cancers, particularly when these
cancers are in a pre-metastatic state and thus still curable. This kinetic aspect of cancer progression is poorly understood, but it is essential to informing effective screening intervals. It is worth noting that mammography and PSA are only surrogate measures of cancer, which have poor specificity and provide little insight into tumor biology. We would argue that for successful screening, we need a platform that provides direct, sensitive, and specific measures of cancer and its attributes, which have bearing on clinical behavior. Circulating Tumor DNA Profiling of a tumor’s somatic alterations has become routine, and many clinical tests are now available that interrogate anywhere from a few genes to the whole human genome (Cheng et al., 2015). For a given tumor, its unique set of somatic alterations creates a biological signature that is highly specific to the individual tumor and appears to be specific to cancer in general when compared to normal tissues. Extensive research has demonstrated that tumor-derived somatic alterations in DNA can be detected in the plasma of cancer patients in the form of cell-free DNA (cfDNA) (Bettegowda et al., 2014). This circulating tumorderived DNA (ctDNA) is presumably shed from tumors, either through necrosis or apoptosis. Commercial ‘‘liquid biopsy’’ tests performed on patients with metastatic disease interrogate the ctDNA in order to identify clinically actionable tumor genotypes in lieu of a tissue biopsy.
Cell 168, February 9, 2017 ª 2017 Published by Elsevier Inc. 571
Table 1. Comparison of ctDNA Liquid Biopsy Test to Potential Cancer Screening Test Indication
Tumor Liquid Biopsy (Genotyping, Monitoring)
Early Cancer Detection
Target population
Patients with known diagnosis of cancer
Asymptomatic individuals
Tissue reference
Can be informed by tissue analyses
No prior knowledge of tissue
Key performance characteristics
Sensitivity and specificity for specific actionable genotypes
d
d
d
Sensitivity and specificity for clinically detectable cancer Premium on specificity in individuals without detectable cancer Tissue of origin needed to guide workup
Clinical Endpoint for Utility
Therapeutic benefit with specific therapies
Net outcome improvement with early detection and local treatment of cancer
Genes Covered
10-50
100-1000s
ctDNA Limit of Detection
0.1%
<0.01%
Importance of Novel Variant Detection
Low
High
Amount of Sequencing
1x
100X
Study Size for Clinical Validity and Utility
100’s
10,000 - 100,000 s
Importantly, the data provided by these tests indicate that these genotypes are not common in the plasma of individuals that are presumably cancer-free (Thress et al., 2015). It is worth noting that tumor-derived RNA and DNA methylation patterns can also be detected in the plasma (Chan et al., 2013) and provide complementary information to the somatic alterations detected in ctDNA. At present, relatively little is known about the biology of cfDNA, such as the mechanism through which it is produced and comes into systemic circulation, and the physiologic variability (e.g., diurnal variation) and kinetics of clearance. Further, the quantitative relationship between the relative and absolute level of ctDNA in the blood and biological features of a tumor (genotype, size, vascularity, anatomy) are not known. Basic research into the fundamental biology of ctDNA is needed to interpret ctDNA signals and their associations with clinical manifestations of cancer. ctDNA Detection Solid tumors are biologically diverse. Data from The Cancer Genome Atlas (TCGA) indicates immense variation between cancers at the DNA, RNA, and epigenetic levels, not just between tissues, but even within a given tissue type (Cancer Genome Atlas Research Network et al., 2013). Moreover, the patient-specific tumor microenvironment and immune system impact the evolution and lethality of a tumor. Compared to infectious disease, which has well-defined causative agents 572 Cell 168, February 9, 2017
with highly conserved biology, a population of cancer patients behaves as a heterogeneous collection of many diseases, each of which carries additional heterogeneity in its own right. Therefore, identifying a finite number of protein or nucleic acid biomarkers that are highly sensitive and specific to even a single cancer is difficult, let alone multiple cancers. There is evidence for the presence of ctDNA in early cancers (Bettegowda et al., 2014). However, the fraction of tumors that shed detectable levels of ctDNA, by tumor type and stage, is not well studied. The studies to date have small numbers of samples and use a variety of measurement techniques that are often not comparable. Despite these caveats, ctDNA is a fundamentally different type of cancer ‘‘biomarker’’ than most that have been used for cancer detection. Most importantly, ctDNA detection leverages the hallmarks of cancer as a disease of genomic alterations and is thus a direct measurement of the tumor. As such, it has the potential to be more specific to the presence of the tumor than other surrogate and downstream measurements of proteins and metabolites. However, the implementation of such a test would be technically challenging, since many genes would have to be simultaneously queried for alterations in order to cover enough of the known diversity in cancer genomes to see most tumors. While next-generation DNA sequencing technology does enable high degrees of target multiplexing, the depth of sequencing would also have to be very high to sample enough
ctDNA molecules to reliably measure them in a background of mostly non-tumor-derived cfDNA. We estimate that such a broad and deep sequencing approach could require orders of magnitude more sequence data than liquid biopsy assays currently use (Table 1). To generate such a large amount of DNA sequence data today in the performance of a routine cancer screening test would be cost-prohibitive, but it is likely to become more technically and economically feasible as the cost of computation and sequencing continues to decrease. Challenges of Developing a ctDNA-Based Screening Test Clinical development of a ctDNA-based approach for broadly applicable cancer screening requires very large-scale clinical evidence for both test development and demonstration of clinical utility. Test development needs to address the dual challenges of sensitivity for early stage disease and the need for exquisite specificity. For sensitivity, the low level of signal in early stage disease, as well as the heterogeneity of cancer genomes present in the population, needs to be overcome to achieve efficacy. For each targeted tumor type, it is expected that at least several hundred of examples of cancer patients, particularly early stage patients, are required to adequately characterize the potential cancer-defining variants observable in plasma. To confirm that variants are indeed cancer-defining, test development must also evaluate specificity in plasma-cell-free DNA profiles from large
numbers of healthy individuals without a known diagnosis of cancer and who are representative controls for the cancer population. Meanwhile, longitudinal clinical follow-up of these non-cancer subjects for future cancer diagnoses is important for distinguishing false positive signals from true positives. As true cancer events are relatively rare within a general screening population (<<1% per year for a given tumor type), the demands on specificity will be particularly high to minimize harm due to false positives and compatibility with practical application in the healthcare system. Adding to the specificity challenge are recent reports of somatic alterations accumulating in both solid tissues and the hematopoietic system as a function of age (Genovese et al., 2014; Alexandrov et al., 2015). To achieve high clinical specificity, a ctDNAbased screening test must be capable of distinguishing between the background signal originating from such non-cancer or pre-cancerous processes and the invasive malignancy of real interest. Additional challenges for clinical development arise from the prevailing anatomic and pathologic paradigm for cancer diagnosis and staging. First, an ideal non-invasive screening test would also provide information on the tissue of origin to streamline the downstream workup, including imaging and tissue diagnosis. This information may be possible through ctDNA, given the distinct differences in the patterns of somatic alterations between different tumor types, at least at a population level (Ciriello et al., 2013). Second, a proportion of screen-detected early stage tumors may prove to be clinically insignificant, as has been described for screen-detected prostate and breast cancers. These entities may be biologically distinct from lethal variants or potentially held in check by the host immune system, and finding and treating these cancers could lead to harm through over-diagnosis and over-treatment. It is possible that mutational signatures in ctDNA could distinguish clinically insignificant biological processes from malignant and lethal biological processes. In addition, serial measurements of the ctDNA signal may identify distinct trajectories with different kinetics for indolent versus lethal disease, thus functionally stratifying these patient populations. Third and
finally, early stage tumors, as currently defined by anatomic staging, are not uniformly cured with local therapy. For some tumor types such as lung cancer and pancreatic cancer, the majority of ‘‘localized’’ tumors eventually relapse and thus could contribute to lead-time bias for a screening program. These individuals clearly have occult metastatic disease, which is not accurately captured with current staging. Post-operative ctDNA levels have now been shown to predict relapse (Tie et al., 2016). By identifying ‘‘early stage’’ patients who have been understaged, ctDNA could minimize the effect of lead-time bias by directing aggressive post-operative systemic therapy to a population where the benefit would be greatest. Ultimately, to demonstrate the clinical utility of a ctDNA cancer screening test, a prospective clinical trial, comparing the strategy of early detection based on the ctDNA test with the standard of care, is necessary. Given the potential biases in cancer screening studies, including lead time bias (Berry, 2014), a randomized design is likely required to achieve a practice-changing level of evidence. Because the yearly incidence rate for cancer is low (1%–2% in aggregate across tumor types), and because differences in cancer-specific mortality can take years to manifest, it is anticipated that such a study would require hundreds of thousands of participants to be appropriately powered. Early assessment of efficacy would be possible with large effect sizes and by evaluating surrogate endpoints, including cumulative incidence of de novo and recurrent metastatic disease. In this regard, GRAIL, a company recently created by some of the authors, was formed to develop ctDNA-based cancer screening tests, is conducting a study to create a reference library of the cancer mutations in the blood for the most common cancers and the background mutations found in matched healthy subjects. This 10,000-plus subject study, called the Circulating CellFree Genome Atlas (CCGA) (clinicaltrials. gov identifier: NCT02889978), should complete enrollment within one year and will be the largest database on mutations found in the blood of cancer patients. The Cancer Moonshot Initiative recently announced the Blood Profiling Atlas
Project, which also aims to compile data on cancer signals in the blood. GRAIL intends to apply a next-generation sequencing approach, combining sequencing depth and breadth of genomic coverage, as well as machine learning, to develop models based on cell-free DNA for the accurate classification of subjects with and without cancer. Tests based upon these models, derived from CCGA and additional studies, will be prospectively evaluated in observational cohorts and eventually in clinical utility studies as described above. Conclusions The biological, technical, and clinical obstacles to developing a safe and effective pan-cancer screening test are significant. However, we have new tools that directly measure the sin qua non of cancer (its somatic alterations) in ways not previously possible. Low-cost next-generation sequencing has enabled the sequencing of tumor-derived nucleic acids circulating in the blood. This non-invasive characterization of a tumor’s genome—combined with new methods of processing complex data, including machine learning—may enable sensitive and specific detection of curable, lethal cancers. Finally, new adaptive approaches to population-scale screening trials are emerging that could be employed to test the clinical utility of this approach. Given these advances, the time has come for a new and unprecedented level of effort to develop lifesaving tests that can detect cancer early, when it can be cured. ACKNOWLEDGMENTS A.M.A. and M.L. are employees and shareholders of GRAIL. R.D.K. is a member of GRAIL’s board of directors and scientific advisory board and is also a shareholder. REFERENCES Aberle, D.R., Adams, A.M., Berg, C.D., Black, W.C., Clapp, J.D., Fagerstrom, R.M., Gareen, I.F., Gatsonis, C., Marcus, P.M., and Sicks, J.D.; National Lung Screening Trial Research Team (2011). N. Engl. J. Med. 365, 395–409. Alexandrov, L.B., Jones, P.H., Wedge, D.C., Sale, J.E., Campbell, P.J., Nik-Zainal, S., and Stratton, M.R. (2015). Nat. Genet. 47, 1402–1407. Berry, D.A. (2014). Cancer 120, 2784–2791. Bettegowda, C., Sausen, M., Leary, R.J., Kinde, I., Wang, Y., Agrawal, N., Bartlett, B.R., Wang, H.,
Cell 168, February 9, 2017 573
Luber, B., Alani, R.M., et al. (2014). Sci. Transl. Med. 6, 224ra24-224ra24. Cancer Genome Atlas Research Network, Weinstein, J.N., Collisson, E.A., Mills, G.B., Shaw, K.R., Ozenberger, B.A., Ellrott, K., Shmulevich, I., Sander, C., and Stuart, J.M. (2013). Nat. Genet. 45, 1113–1120.
Cho, H., Mariotto, A.B., Schwartz, L.M., Luo, J., and Woloshin, S. (2014). J. Natl. Cancer Inst. Monogr. 2014, 187–197.
Nelson, H.D., O’Meara, E.S., Kerlikowske, K., Balch, S., and Miglioretti, D. (2016a). Ann. Intern. Med. 164, 226–235.
Chou, R., Croswell, J.M., Dana, T., Bougatsos, C., Blazina, I., Fu, R., Gleitsmann, K., Koenig, H.C., Lam, C., Maltz, A., et al. (2011). Ann. Intern. Med. 155, 762–771.
Nelson, H.D., Pappas, M., Cantor, A., Griffin, J., Daeges, M., and Humphrey, L. (2016b). Ann. Intern. Med. 164, 256–267.
Chan, K.C.A., Jiang, P., Chan, C.W.M., Sun, K., Wong, J., Hui, E.P., Chan, S.L., Chan, W.C., Hui, D.S.C., Ng, S.S.M., et al. (2013). Proc. Natl. Acad. Sci. USA 110, 18761–18768.
Ciriello, G., Miller, M.L., Aksoy, B.A., Senbabaoglu, Y., Schultz, N., and Sander, C. (2013). Nat. Genet. 45, 1127–1133.
Thress, K.S., Brant, R., Carr, T.H., Dearden, S., Jenkins, S., Brown, H., Hammett, T., Cantarini, M., and Barrett, J.C. (2015). Lung Cancer 90, 509–515.
Cheng, D.T., Mitchell, T.N., Zehir, A., Shah, R.H., Benayed, R., Syed, A., Chandramohan, R., Liu, Z.Y., Won, H.H., Scott, S.N., et al. (2015). J. Mol. Diagn. 17, 251–264.
Genovese, G., Ka¨hler, A.K., Handsaker, R.E., Lindberg, J., Rose, S.A., Bakhoum, S.F., Chambert, K., Mick, E., Neale, B.M., Fromer, M., et al. (2014). N. Engl. J. Med. 371, 2477–2487.
Tie, J., Wang, Y., Tomasetti, C., Li, L., Springer, S., Kinde, I., Silliman, N., Tacey, M., Wong, H.L., Christie, M., et al. (2016). Sci. Transl. Med. 8, 346ra92.
574 Cell 168, February 9, 2017
Commentary Next-Generation Sequencing of Circulating Tumor DNA for Early Cancer Detection Alexander M. Aravanis,1,2 Mark Lee,1,2 and Richard D. Klausner1,* 1GRAIL,
Menlo Park, CA 94402, USA author *Correspondence: [email protected] http://dx.doi.org/10.1016/j.cell.2017.01.030 2Co-first
Curative therapies are most successful when cancer is diagnosed and treated at an early stage. We advocate that technological advances in next-generation sequencing of circulating, tumor-derived nucleic acids hold promise for addressing the challenge of developing safe and effective cancer screening tests. Cancer-specific mortality from most types of solid tumors has barely decreased in decades, despite an exponential increase in our knowledge about cancer pathogenesis and significant investments in the development of effective treatments. The past few years have witnessed a dramatic success of immunotherapies in treating a subgroup of patients with a variety of tumor types, including lung, bladder, and kidney, as well as Hodgkin’s lymphoma and melanoma. While such breakthroughs offer the hope of prolonged survival for some patients with advanced cancers, finding cancers earlier would still afford the greatest chance for cure, given that the survival rates for patients with early diagnoses are five to ten times higher compared with late stage disease (Cho et al., 2014). By enabling diagnosis and localized treatment of early stage invasive cancers (and, in some cases, pre-invasive states), screening for cervical and colorectal cancers has contributed to significant declines in mortality from these diseases. Similarly, effective population screening paradigms for many other common and deadly tumor types are clearly needed to broadly reduce cancer-specific mortality. Challenges for Cancer Screening Despite the promise, the field of early cancer detection is filled with cautionary tales that highlight the challenges. Early stage solid tumors are small and are thereby difficult to be non-invasively distinguished from normal anatomic and biochemical variation. This ambiguity leads to detec-
tion algorithms that either miss a large number of invasive cancers or make the costly trade-off of over-diagnosing and consequently over treating. For instance, high false-positive rates from mammography in breast cancer screening, lowdose CT in lung cancer screening, and prostate-specific antigen (PSA) screening (Nelson et al., 2016a; Aberle et al., 2011; Chou et al., 2011) represent a significant cost to the healthcare system, with resulting mental and physical morbidity, and even mortality in some cases (Nelson et al., 2016b). Even where cancer screening has produced significant stage shifts, as with breast and prostate cancer screening, the impact on cancer-specific mortality has not been a predictable outcome (Berry, 2014). Multiple explanations may account for this phenomenon. First, early stage cancers detected by mammography and PSA are likely to be clinically insignificant or ‘‘non-lethal,’’ meaning that, in the absence of screening and treatment, these tumors would not have become clinically evident, metastatic, or contributory to patient mortality. Second, some lethal tumors diagnosed in an apparently localized state may already have disseminated with occult metastatic disease, making local treatment with surgery or radiation non-curative. Earlier diagnosis of these tumors would result in a ‘‘lead-time bias’’ without impacting cancer-specific mortality. Third, mammography and PSA may be sensitive for indolent tumors, but not sufficiently sensitive for lethal cancers, particularly when these
cancers are in a pre-metastatic state and thus still curable. This kinetic aspect of cancer progression is poorly understood, but it is essential to informing effective screening intervals. It is worth noting that mammography and PSA are only surrogate measures of cancer, which have poor specificity and provide little insight into tumor biology. We would argue that for successful screening, we need a platform that provides direct, sensitive, and specific measures of cancer and its attributes, which have bearing on clinical behavior. Circulating Tumor DNA Profiling of a tumor’s somatic alterations has become routine, and many clinical tests are now available that interrogate anywhere from a few genes to the whole human genome (Cheng et al., 2015). For a given tumor, its unique set of somatic alterations creates a biological signature that is highly specific to the individual tumor and appears to be specific to cancer in general when compared to normal tissues. Extensive research has demonstrated that tumor-derived somatic alterations in DNA can be detected in the plasma of cancer patients in the form of cell-free DNA (cfDNA) (Bettegowda et al., 2014). This circulating tumorderived DNA (ctDNA) is presumably shed from tumors, either through necrosis or apoptosis. Commercial ‘‘liquid biopsy’’ tests performed on patients with metastatic disease interrogate the ctDNA in order to identify clinically actionable tumor genotypes in lieu of a tissue biopsy.
Cell 168, February 9, 2017 ª 2017 Published by Elsevier Inc. 571
Table 1. Comparison of ctDNA Liquid Biopsy Test to Potential Cancer Screening Test Indication
Tumor Liquid Biopsy (Genotyping, Monitoring)
Early Cancer Detection
Target population
Patients with known diagnosis of cancer
Asymptomatic individuals
Tissue reference
Can be informed by tissue analyses
No prior knowledge of tissue
Key performance characteristics
Sensitivity and specificity for specific actionable genotypes
d
d
d
Sensitivity and specificity for clinically detectable cancer Premium on specificity in individuals without detectable cancer Tissue of origin needed to guide workup
Clinical Endpoint for Utility
Therapeutic benefit with specific therapies
Net outcome improvement with early detection and local treatment of cancer
Genes Covered
10-50
100-1000s
ctDNA Limit of Detection
0.1%
<0.01%
Importance of Novel Variant Detection
Low
High
Amount of Sequencing
1x
100X
Study Size for Clinical Validity and Utility
100’s
10,000 - 100,000 s
Importantly, the data provided by these tests indicate that these genotypes are not common in the plasma of individuals that are presumably cancer-free (Thress et al., 2015). It is worth noting that tumor-derived RNA and DNA methylation patterns can also be detected in the plasma (Chan et al., 2013) and provide complementary information to the somatic alterations detected in ctDNA. At present, relatively little is known about the biology of cfDNA, such as the mechanism through which it is produced and comes into systemic circulation, and the physiologic variability (e.g., diurnal variation) and kinetics of clearance. Further, the quantitative relationship between the relative and absolute level of ctDNA in the blood and biological features of a tumor (genotype, size, vascularity, anatomy) are not known. Basic research into the fundamental biology of ctDNA is needed to interpret ctDNA signals and their associations with clinical manifestations of cancer. ctDNA Detection Solid tumors are biologically diverse. Data from The Cancer Genome Atlas (TCGA) indicates immense variation between cancers at the DNA, RNA, and epigenetic levels, not just between tissues, but even within a given tissue type (Cancer Genome Atlas Research Network et al., 2013). Moreover, the patient-specific tumor microenvironment and immune system impact the evolution and lethality of a tumor. Compared to infectious disease, which has well-defined causative agents 572 Cell 168, February 9, 2017
with highly conserved biology, a population of cancer patients behaves as a heterogeneous collection of many diseases, each of which carries additional heterogeneity in its own right. Therefore, identifying a finite number of protein or nucleic acid biomarkers that are highly sensitive and specific to even a single cancer is difficult, let alone multiple cancers. There is evidence for the presence of ctDNA in early cancers (Bettegowda et al., 2014). However, the fraction of tumors that shed detectable levels of ctDNA, by tumor type and stage, is not well studied. The studies to date have small numbers of samples and use a variety of measurement techniques that are often not comparable. Despite these caveats, ctDNA is a fundamentally different type of cancer ‘‘biomarker’’ than most that have been used for cancer detection. Most importantly, ctDNA detection leverages the hallmarks of cancer as a disease of genomic alterations and is thus a direct measurement of the tumor. As such, it has the potential to be more specific to the presence of the tumor than other surrogate and downstream measurements of proteins and metabolites. However, the implementation of such a test would be technically challenging, since many genes would have to be simultaneously queried for alterations in order to cover enough of the known diversity in cancer genomes to see most tumors. While next-generation DNA sequencing technology does enable high degrees of target multiplexing, the depth of sequencing would also have to be very high to sample enough
ctDNA molecules to reliably measure them in a background of mostly non-tumor-derived cfDNA. We estimate that such a broad and deep sequencing approach could require orders of magnitude more sequence data than liquid biopsy assays currently use (Table 1). To generate such a large amount of DNA sequence data today in the performance of a routine cancer screening test would be cost-prohibitive, but it is likely to become more technically and economically feasible as the cost of computation and sequencing continues to decrease. Challenges of Developing a ctDNA-Based Screening Test Clinical development of a ctDNA-based approach for broadly applicable cancer screening requires very large-scale clinical evidence for both test development and demonstration of clinical utility. Test development needs to address the dual challenges of sensitivity for early stage disease and the need for exquisite specificity. For sensitivity, the low level of signal in early stage disease, as well as the heterogeneity of cancer genomes present in the population, needs to be overcome to achieve efficacy. For each targeted tumor type, it is expected that at least several hundred of examples of cancer patients, particularly early stage patients, are required to adequately characterize the potential cancer-defining variants observable in plasma. To confirm that variants are indeed cancer-defining, test development must also evaluate specificity in plasma-cell-free DNA profiles from large
numbers of healthy individuals without a known diagnosis of cancer and who are representative controls for the cancer population. Meanwhile, longitudinal clinical follow-up of these non-cancer subjects for future cancer diagnoses is important for distinguishing false positive signals from true positives. As true cancer events are relatively rare within a general screening population (<<1% per year for a given tumor type), the demands on specificity will be particularly high to minimize harm due to false positives and compatibility with practical application in the healthcare system. Adding to the specificity challenge are recent reports of somatic alterations accumulating in both solid tissues and the hematopoietic system as a function of age (Genovese et al., 2014; Alexandrov et al., 2015). To achieve high clinical specificity, a ctDNAbased screening test must be capable of distinguishing between the background signal originating from such non-cancer or pre-cancerous processes and the invasive malignancy of real interest. Additional challenges for clinical development arise from the prevailing anatomic and pathologic paradigm for cancer diagnosis and staging. First, an ideal non-invasive screening test would also provide information on the tissue of origin to streamline the downstream workup, including imaging and tissue diagnosis. This information may be possible through ctDNA, given the distinct differences in the patterns of somatic alterations between different tumor types, at least at a population level (Ciriello et al., 2013). Second, a proportion of screen-detected early stage tumors may prove to be clinically insignificant, as has been described for screen-detected prostate and breast cancers. These entities may be biologically distinct from lethal variants or potentially held in check by the host immune system, and finding and treating these cancers could lead to harm through over-diagnosis and over-treatment. It is possible that mutational signatures in ctDNA could distinguish clinically insignificant biological processes from malignant and lethal biological processes. In addition, serial measurements of the ctDNA signal may identify distinct trajectories with different kinetics for indolent versus lethal disease, thus functionally stratifying these patient populations. Third and
finally, early stage tumors, as currently defined by anatomic staging, are not uniformly cured with local therapy. For some tumor types such as lung cancer and pancreatic cancer, the majority of ‘‘localized’’ tumors eventually relapse and thus could contribute to lead-time bias for a screening program. These individuals clearly have occult metastatic disease, which is not accurately captured with current staging. Post-operative ctDNA levels have now been shown to predict relapse (Tie et al., 2016). By identifying ‘‘early stage’’ patients who have been understaged, ctDNA could minimize the effect of lead-time bias by directing aggressive post-operative systemic therapy to a population where the benefit would be greatest. Ultimately, to demonstrate the clinical utility of a ctDNA cancer screening test, a prospective clinical trial, comparing the strategy of early detection based on the ctDNA test with the standard of care, is necessary. Given the potential biases in cancer screening studies, including lead time bias (Berry, 2014), a randomized design is likely required to achieve a practice-changing level of evidence. Because the yearly incidence rate for cancer is low (1%–2% in aggregate across tumor types), and because differences in cancer-specific mortality can take years to manifest, it is anticipated that such a study would require hundreds of thousands of participants to be appropriately powered. Early assessment of efficacy would be possible with large effect sizes and by evaluating surrogate endpoints, including cumulative incidence of de novo and recurrent metastatic disease. In this regard, GRAIL, a company recently created by some of the authors, was formed to develop ctDNA-based cancer screening tests, is conducting a study to create a reference library of the cancer mutations in the blood for the most common cancers and the background mutations found in matched healthy subjects. This 10,000-plus subject study, called the Circulating CellFree Genome Atlas (CCGA) (clinicaltrials. gov identifier: NCT02889978), should complete enrollment within one year and will be the largest database on mutations found in the blood of cancer patients. The Cancer Moonshot Initiative recently announced the Blood Profiling Atlas
Project, which also aims to compile data on cancer signals in the blood. GRAIL intends to apply a next-generation sequencing approach, combining sequencing depth and breadth of genomic coverage, as well as machine learning, to develop models based on cell-free DNA for the accurate classification of subjects with and without cancer. Tests based upon these models, derived from CCGA and additional studies, will be prospectively evaluated in observational cohorts and eventually in clinical utility studies as described above. Conclusions The biological, technical, and clinical obstacles to developing a safe and effective pan-cancer screening test are significant. However, we have new tools that directly measure the sin qua non of cancer (its somatic alterations) in ways not previously possible. Low-cost next-generation sequencing has enabled the sequencing of tumor-derived nucleic acids circulating in the blood. This non-invasive characterization of a tumor’s genome—combined with new methods of processing complex data, including machine learning—may enable sensitive and specific detection of curable, lethal cancers. Finally, new adaptive approaches to population-scale screening trials are emerging that could be employed to test the clinical utility of this approach. Given these advances, the time has come for a new and unprecedented level of effort to develop lifesaving tests that can detect cancer early, when it can be cured. ACKNOWLEDGMENTS A.M.A. and M.L. are employees and shareholders of GRAIL. R.D.K. is a member of GRAIL’s board of directors and scientific advisory board and is also a shareholder. REFERENCES Aberle, D.R., Adams, A.M., Berg, C.D., Black, W.C., Clapp, J.D., Fagerstrom, R.M., Gareen, I.F., Gatsonis, C., Marcus, P.M., and Sicks, J.D.; National Lung Screening Trial Research Team (2011). N. Engl. J. Med. 365, 395–409. Alexandrov, L.B., Jones, P.H., Wedge, D.C., Sale, J.E., Campbell, P.J., Nik-Zainal, S., and Stratton, M.R. (2015). Nat. Genet. 47, 1402–1407. Berry, D.A. (2014). Cancer 120, 2784–2791. Bettegowda, C., Sausen, M., Leary, R.J., Kinde, I., Wang, Y., Agrawal, N., Bartlett, B.R., Wang, H.,
Cell 168, February 9, 2017 573
Luber, B., Alani, R.M., et al. (2014). Sci. Transl. Med. 6, 224ra24-224ra24. Cancer Genome Atlas Research Network, Weinstein, J.N., Collisson, E.A., Mills, G.B., Shaw, K.R., Ozenberger, B.A., Ellrott, K., Shmulevich, I., Sander, C., and Stuart, J.M. (2013). Nat. Genet. 45, 1113–1120.
Cho, H., Mariotto, A.B., Schwartz, L.M., Luo, J., and Woloshin, S. (2014). J. Natl. Cancer Inst. Monogr. 2014, 187–197.
Nelson, H.D., O’Meara, E.S., Kerlikowske, K., Balch, S., and Miglioretti, D. (2016a). Ann. Intern. Med. 164, 226–235.
Chou, R., Croswell, J.M., Dana, T., Bougatsos, C., Blazina, I., Fu, R., Gleitsmann, K., Koenig, H.C., Lam, C., Maltz, A., et al. (2011). Ann. Intern. Med. 155, 762–771.
Nelson, H.D., Pappas, M., Cantor, A., Griffin, J., Daeges, M., and Humphrey, L. (2016b). Ann. Intern. Med. 164, 256–267.
Chan, K.C.A., Jiang, P., Chan, C.W.M., Sun, K., Wong, J., Hui, E.P., Chan, S.L., Chan, W.C., Hui, D.S.C., Ng, S.S.M., et al. (2013). Proc. Natl. Acad. Sci. USA 110, 18761–18768.
Ciriello, G., Miller, M.L., Aksoy, B.A., Senbabaoglu, Y., Schultz, N., and Sander, C. (2013). Nat. Genet. 45, 1127–1133.
Thress, K.S., Brant, R., Carr, T.H., Dearden, S., Jenkins, S., Brown, H., Hammett, T., Cantarini, M., and Barrett, J.C. (2015). Lung Cancer 90, 509–515.
Cheng, D.T., Mitchell, T.N., Zehir, A., Shah, R.H., Benayed, R., Syed, A., Chandramohan, R., Liu, Z.Y., Won, H.H., Scott, S.N., et al. (2015). J. Mol. Diagn. 17, 251–264.
Genovese, G., Ka¨hler, A.K., Handsaker, R.E., Lindberg, J., Rose, S.A., Bakhoum, S.F., Chambert, K., Mick, E., Neale, B.M., Fromer, M., et al. (2014). N. Engl. J. Med. 371, 2477–2487.
Tie, J., Wang, Y., Tomasetti, C., Li, L., Springer, S., Kinde, I., Silliman, N., Tacey, M., Wong, H.L., Christie, M., et al. (2016). Sci. Transl. Med. 8, 346ra92.
574 Cell 168, February 9, 2017