Sex Differences in Cancer Genomes: Much Learned, More Unknown

Abstract

Cancer is a leading cause of death worldwide. Sex influences cancer in a bewildering variety of ways. In some cancer types, it affects prevalence; in others, genomic profiles, response to treatment, or mortality. In some, sex seems to have little or no influence. How and when sex influences cancer initiation and progression remain a critical gap in our understanding of cancer, with direct relevance to precision medicine. Here, we note several factors that complicate our understanding of sex differences: representativeness of large cohorts, confounding with features such as ancestry, age, obesity, and variability in clinical presentation. We summarize the key resources available to study molecular sex differences and suggest some likely directions for improving our understanding of how patient sex influences cancer behavior.

Cancer is a leading cause of death worldwide, especially in more developed countries and regions (1). It is a group of diseases categorized largely on the basis of the organ and cell-type within which they arise. This leads to substantial heterogeneity between and within cancer types. Some cancer types, such as pancreatic and brain cancers, have a very low survival rate. Other cancer types, like prostate and thyroid cancers, have very good outcomes (2). Cancers ultimately arise from the activation of oncogenes and deactivation of tumor suppressor genes. These can occur through multiple mechanisms: some can be inherited, and others are generated through somatic mutational processes (3). Point mutations and chromosomal translocations can alter the amino acid sequence and protein structure of the protein encoded by proto-oncogenes, activating them (4). Amplifications and translocations in certain situations can also affect the RNA and protein abundance of proto-oncogenes (4). Many cancer genomic studies have been conducted over the past 2 decades, and these have started to quantify the heterogeneity in genomic landscapes between and within cancer types, identifying many previously unrecognized cancer subtypes (5).

In a largely independent line of research, epidemiological studies have uncovered a large number of variables associated with risk of a cancer diagnosis, including sex, age, obesity, and others (6-9). This information can be useful in developing screening protocols or in designing interventions to reduce disease incidence with preventative strategies. It also provides insights for new therapeutic targets because subpopulations with altered cancer risks often have unique genomic alteration patterns. Sex is one of the variables that has a large impact on cancer risk. The prevalence and mortality of cancer in men is higher than women for 9 of the 10 most commonly diagnosed cancers in the United States; thyroid cancer is the sole exception (2). The combined cancer mortality is higher in men than women in most countries and regions of the world, with many African regions and Micronesia/Polynesia as notable exceptions (1).

We start by summarizing the current resources available for research into sex differences in cancer genomics. Next, we review the key limitations to the knowledge we have derived from studies of those resources. Finally, we suggest some likely directions for improving our understanding of how patient sex influences the molecular behavior of cancer.

Is Big Data Big Enough?

A series of multi-institutional cancer genomics projects, some completed and some ongoing, have generated tremendous resources. They are being broadly used by the community to generate novel hypotheses about cancer and its behavior in specific groups of patients. The Cancer Genome Atlas (TCGA) included more than 11 000 primary cancers and paired normal tissue samples derived from cancers of 33 types (10,11). The International Cancer Genome Consortium (ICGC) is an ongoing project with teams in the United States, Canada, European, Asia, and Australia—including TCGA projects. With the underlying idea that no single country can cover all cancer types in depth, ICGC brings together scientists in the world and set a target to study 50 cancer types, with at least 500 patients of each cancer category in its first phase (12). To understand the genomic features and abnormalities across cancer types, the Pan-cancer Analysis of Whole Genomes (PCAWG) project included 2658 tumor samples with whole-genome sequencing data from TCGA and ICGC and preformed comprehensive analysis (13). The Therapeutically Applicable Research to Generate Effective Treatments (TARGET) initiative is aiming at understanding the molecular characteristics of pediatric cancers and currently has studied over 6000 children and adolescents at the time of this writing (14). With the decreased price of sequencing technology, genomic profiling is becoming a routine procedure during diagnosis and decision-making of treatments. The Genomics, Evidence, Neoplasia, Information, Exchange (GENIE) project of American Association for Cancer Research comprises genomic profiling data from over 40 000 clinical patients contributed by 8 institutes from North America and Europe (15). Proteomics is largely absent from the previously mentioned projects. To complement them, the US Clinical Proteome Tumour Analysis Consortium has thus far performed comprehensive proteomic and genomic analysis on over 1300 patients with 12 cancer types, with ongoing efforts to expand this (16). Its international counterpart, the International Cancer Proteome Consortium is rapidly growing.

The proportions of the two main biological sexes (defined by chromosome conformation: XX vs XY) in the previously mentioned major cancer genomic projects are relatively representative of the sex distribution of cancer incidence. The American Cancer Society estimates the cancer incidence and mortality in a yearly bases (2). It was estimated that there will be 1 898 160 total cases of all cancers in the United States in 2021, of which 51.1% (970 250) will arise in males and 48.9% (927 910) will arise in females (2). Excluding cancers arising in the reproductive systems and in breast tissue, there are projected to be 707 390 (57.2%) arising in men and 529 600 (42.8%) arising in women. The sex ratio of cases studied in the previously noted projects are all relatively close to this estimate (Table 1).

Table 1.

Patient number in total and by sex in major cancer genomic projects

Project	All cancers			Cancers not arising in sex organsa
	Both sexes	Male, n (%)	Female, n (%)	Both sexes	Male, n (%)	Female, n (%)
TCGA	11 167	5346 (47.9)	5817 (52.1)	7906	4699 (59.4)	3203 (40.5)
ICGC	24 289	13 135 (54.1)	10 856 (44.7)	19 301	11 700 (60.6)	7303 (37.8)
TARGET	4866	2650 (54.5)	2177 (44.7)	4718	2568 (54.4)	2111 (44.7)
GENIE	44 756	20 713 (46.3)	24 025 (53.7)	32 512	17 991 (55.3)	14 510 (44.6)
CPTAC	1121	507 (45.2)	602 (53.7)	794	507 (63.9)	287 (36.1)

Abbreviations: CPTAC, Clinical Proteome Tumour Analysis Consortium; GENIE, Genomics, Evidence, Neoplasia, Information, Exchange; ICGC, International Cancer Genome Consortium; TARGET, Therapeutically Applicable Research to Generate Effective Treatments; TCGA, The Cancer Genome Atlas.

^aAll cancer types except those originating in reproductive organs (eg, prostate, uterine, ovary) or in the breast.

Surprisingly, despite this relatively balanced sex representation, even the largest public genomic databases provide very modest statistical power. For individual cancer types, excluding cancers of the breast and reproductive organs, the median number of patients in TCGA is 231. ICGC has similar numbers: a median of 205 patients in each study. These numbers are quite modest when one considers the large number of confounding variables at play. Cancers vary in their histopathological and clinical features—stage, grade, subtype, treatment status, etc. Patients vary dramatically in their epidemiologic and lifestyle characteristics—age (as seen in breast and prostate cancer), ancestry, smoking history, obesity (as seen in multiple myeloma, colorectal carcinoma, thyroid carcinoma, carcinoma of the pancreas), mutagen exposure (such as exposure to radioactive substances, X-rays, ultraviolet radiation), exercise and diet patterns, and so forth (17-19).

Controlling for these large numbers of confounders is extremely challenging with current cohort sizes. But the situation is further complicated by the relatively low frequency of many somatic drivers and other mutational features. Many specific mutations only occur in 15% to 30% of patients. Consider a large cohort of 500 tumors arising in men and 500 arising in women; this is much larger than a typical ICGC or TCGA data set. For a driver mutation that occurs in 20% and 30% of male- and female-derived tumors, respectively; that means 100 to 150 patients with the mutation would be available. However statistical tests need to be adjusted for at least a dozen potential confounders, such as tumor stage. Finally, because several hundred driver mutations are now known (20,21), discovery studies incur large multiple hypothesis testing penalties, further reducing statistical power. Thus, differences of sex in cancer genomic abnormalities are highly diverse in different cancer types, strongly influenced by other clinical, epidemiologic, and molecular features, and thus even the largest sample sizes available today remain a critical limitation.

Heterogeneity of Cancer Types in Sex Differences

Men have higher incidence and mortality compared to women in many cancer types, such as colon, lung, liver, and gastric cancer (2). A small number of cancer types show the inverse trend. Thyroid cancer is 2.9 times more frequent in women than in men in general; however, in more aggressive histological thyroid cancer subtypes (eg, anaplastic and medullary thyroid cancer), disease prevalence is comparable between the sexes (22). Gallbladder cancer is 2 to 6 more frequent in females than in males (23). The estimated incidence and mortality of the most popular cancer subtypes in men and women in United States of 2021 are shown in Table 2 [data adopted from American Cancer Society’s estimates (2)].

Table 2.

Estimated incidence and mortality of 10 most common cancer sites/tissues of nonreproductive and nonbreast organs/tissues in men and women in United States, 2021

Site/tissue	Incidence, n		Mortality, n
	Male	Female	Male	Female
Lung and bronchus	119 100	116 660	69 410	62 470
Colon and rectum	79 520	69 980	28 520	24 460
Melanoma of the skin	62 260	43 850	4600	2 580
Urinary bladder	64 280	19 450	12 260	4 940
Non-Hodgkin lymphoma	45 630	35 930	12 170	8 550
Kidney and renal pelvis	48 780	27 300	8790	4 990
Leukemia	35 530	25 560	13 900	9 760
Pancreas	31 950	28 480	25 270	22 950
Oral cavity and pharynx	38 800	15 210	7620	3230
Thyroid	12 150	32 130	1050	1150

Data adapted from Siegel et al (2). With permission from American Cancer Society, New York.

This variability in the clinical manifestations of sex differences between cancer types is mirrored in the cancer genome. Somatic mutation density is elevated in tumors arising in males from stomach and esophageal cancer, hepatocellular carcinoma, bladder urothelial carcinoma, melanoma, and kidney renal papillary cell carcinoma; by contrast, it is elevated in glioblastomas (GBMs) arising in females (24,25). The proportion of the genome with a copy number aberration (PGA) is a measure of total copy number burden in a tumor genome and is a surrogate for genomic instability (26). PGA was higher in tumors arising in males from stomach and esophageal cancers, head and neck squamous cell tumors (HNSCs), and kidney renal clear cell carcinomas. PGA was higher in sarcomas and hepatocellular carcinomas arising in females (24,25).

A trio of comprehensive pan-cancer analyses have been done to evaluate the sex differences in cancer genome using TCGA and PCAWG data (24,25,27). Table 3 summarizes key findings from these 3 studies. Two of the studies used multistage multivariate modeling to explicitly control for clinico-epidemiologic features, while the third used propensity score modeling. In all cases, this multivariate modeling implicitly reduced statistical power to help reduce false positives caused by chance unequal distributions of tumors across various covariates. Nevertheless, despite relatively low statistical power that was variable across cancer types, numerous sex differences were detected. Bladder, skin cutaneous melanoma, kidney papillary, hepatocellular carcinoma, stomach and esophageal cancers (STES), and kidney clear cell tumors all had higher somatic single-nucleotide variant (SNV) density in tumors arising in males, while GBM had higher SNV density when arising in females (24). Stomach and esophageal cancers, HNSC, and kidney clear cell tumors had higher PGA in tumors arising in males, while sarcoma and hepatocellular carcinoma had higher PGA in tumors arising in females. Microsatellite instability is a marker for impaired DNA mismatch repair and is observed in numerous cancer subtypes. It was found that in colon cancer and STES, tumors arising in females had higher microsatellite instability frequency than those arising in males (24).

Table 3.

Summary of genomic features with sex differences identified in TCGA and PCAWG

Cancer subtype	Mutation density	PGA	Genes with sex-different SNV
Bladder urothelial	Higher in male
Melanoma	Higher in male
Liver hepatocellular	Higher in male		CTNNB1, BAP1
Glioblastoma	Higher in female
Stomach and esophageal	Higher in male (MSI adjusted)	Higher in male	PGM5, GTF3C1, LTBP1, MEGF8, RNF213, EP400, OBSCN, ZFHX3, SLIT2, ZBTB20
Head and neck		Higher in male
Kidney renal cell		Higher in male	BAP1
Sarcoma		Higher in female
Thyroid	Higher inmale (marginal)		TERT
Kidney renal papillary cell			TTN
Long adenocarcinoma			STK11, SMG1, CNTN5, ZNF521, COL21A1, RBM10, ABCB5, FAM47A, DMD, F8, MED12

Data adapted from Li et al (24,25) and Yuan et al (27) with permission from Elsevier.

Abbreviation: MSI, microsatellite instability; PACWG, Pan-cancer Analysis of Whole Genomes; PGA, proportion of the genome with a copy number aberration; SNV, single-nucleotide variant; TCGA, The Cancer Genome Atlas.

Some cancer types have not displayed significant sex-differences in these large surveys. In acute myeloid leukemia, low-grade glioma (LGG) and pheochromocytoma and paraganglioma, none of the genomic alterations tested were significantly different between tumors arising in males and those arising in females—these included somatic SNV burden, PGA, and specific driver genes (25,27). Interestingly, men also have a higher risk of diagnosis for cancer types without genomic sex-differences such as acute myeloid leukemia and LGG (28,29).

Seeing Through the Variability: Known Sex Differences

Using data collected by the SEER (Surveillance, Epidemiology, and End Results) program has shown that across the 10 most common nonsex-specific cancers, men had more advanced stage disease at diagnosis than women in most. These included lung, melanoma, non-Hodgkin lymphoma, and kidney, thyroid, pancreatic and liver cancer but not bladder, colorectal, and brain cancer (30). In lung cancer, women were more likely to have adenocarcinoma, and men, squamous-cell carcinoma (30). Data collected in England also showed a slightly higher proportion of men diagnosed with stage 1 colorectal cancer and women with stage 2 (31). The same study also observed that the anatomical site tended to be higher (eg, caecum, ascending, transverse colon) in women, and lower (eg, sigmoid colon and rectum) in men (31). It is unclear whether these differences primarily reflect differences in health-seeking behavior, in other lifestyle or social characteristics, in the underlying molecular biology of the tumor, or some other features.

As previously noted, many factors contribute to the tumor development and progression, and could influence our understanding of the sex-difference in cancer genomics. Cancer prevalence and genomic features are associated with ethnic and geographic diversity. In TCGA data, the genome instability of African American patients was higher in breast (BRCA), head and neck and endometrial cancers [uterine corpus endometrial carcinoma (UCEC)], but lower in patients with kidney cancers (32). The elevated genomic instability of African ancestry in BRCA, HNSC, and UCEC was also associated with increased TP53 mutation and CCNE1 amplification (32). Similarly, another study observed correlations between African ancestry and aneuploidy and East Asian ancestry and mutation load (33). Another recent study integrated TCGA and PCAWG data and observed that African ancestry was associated with higher PGA (proportion of genome with a copy number alteration) in HNSC and UCEC, while East Asian ancestry was associated with higher PGA in hepatocellular carcinoma and STES (34). East Asian and African ancestry were also associated with lower SNV density than European ancestry in melanoma (34). It is yet difficult to assess the race disparities in large cancer genomic studies such as TCGA and ICGC. In TCGA studies, somatic mutations with mutational frequency below 10% are not detectable in most cancer types for most non-Caucasian ancestry groups (35). The underrepresentativeness of nonwhite patients makes it difficult to study the genomic disparities in sex while trying to adjust for effects of ancestry. Thus, our understanding of sex differences in cancer will remain at least partially incomplete and uncertain until there is more equal representation in our databases of genomic profiles of patients of all ancestries.

In a similar way, age is associated with the evolution and somatic mutation features of cancer and might interact with sex differences. In TCGA data, genome instability increased with age at both the pan-cancer level as well as in specific cancer types including LGG, ovarian cancer, UCEC and sarcoma (36). Somatic point mutation burden also increased with age at the pan-cancer level and in 18 of 33 TCGA cancer types (36). Driver gene mutations such as mutations in IDH1 and ATRX were negatively associated with age, while PIK3CA were positively associated with age in pan-cancer data (36). These trends were confirmed in ICGC PCAWG whole-genome sequencing data (37). In addition, a positive association between clonal SNVs and indels were also observed pan-cancer and in 2 specific cancer types: medulloblastoma and melanoma (37). Several mutational signatures were associated with age, suggesting different exogenous and endogenous mutational oncogenic processes in older patients relative to younger ones (37). These data highlight that age significantly affects the somatic mutational evolution of cancer and needs to be considered in the context of sex differences.

Besides innate and inheritable factors such as race and age, environmental factors also contribute to cancer risks and are associated with specific patterns of genomic abnormalities. It is well known that smoking and alcohol consumption are associated with mutation rates and types in multiple cancer types such as lung, colon, and oral cancer (38-41). The consumption of tobacco and alcohol itself also shows sex differences, as both smoking and alcohol exposure was higher in men than women in lung and liver cancers (42,43). Diet is another important environmental factor that has a strong association to cancer, but one that is not yet well understood. A study in the Netherlands reported that dietary heme consumption is strongly associated with KRAS and APC mutations in colorectal cancers arising in males but not in those arising in females (44). The same team reported that dietary folate intake was associated with a decreased risk for rectal cancer in men, particularly for patients carrying KRAS point mutations. By contrast, it showed an association with increased risk in female patients independent of KRAS mutation status (45). Another study reported that fish oil consumption was significant associated with a lower risk for colorectal cancer in men but not in women (46). Similarly, dietary fiber intake was associated with a lower risk for recurrence of colorectal adenoma in men but not women (47). Those findings indicated potential sex differences in diet-mutation interactions in cancer that will need to be extensively studied moving forward. Unfortunately, dietary profiling for patients with sequenced cancer genomes has not been a major focus to date. The lack of dietary data in large cancer genomic projects such as TCGA and ICGC makes it almost impossible to achieve a comprehensive understanding of the relationship between diet and cancer genome abnormality. Similarly, physical exercise has protective effect in several caner types, especially in breast and colon cancer (48,49) and has a positive effect when undertaken with or after treatment (50). Growing evidence suggests that exercise alters DNA methylations in patients with different cancer types (51). Obesity, which can be the consequence of unhealthy dietary patterns (high calorie, high fat, high refined sugar, and low fiber), is associated with cancer in a complex way. While obesity is associated with a higher risk for many cancer subtypes (52-54), it was observed that overweight and obese patients experienced statistically significantly higher survival in renal cell cancer, colorectal cancer, and acute myeloid leukemia (55-57). This obesity paradox was also observed in other diseases such as cardiovascular disease (58,59), diabetes (60), and chronic kidney disease (61) but not in prostate and breast cancer (62). Importantly, these lifestyle factors of diet, exercise, and obesity themselves vary with an individual’s age. It is unclear whether their impact is most acute in certain age windows—either in absolute time or relative to a specific landmark in tumor evolution such as the first whole-genome duplication.

Thus, a growing body of evidence shows that mutational processes and specific driver somatic mutations are affected by numerous endogenous and exogenous variables. When evaluating sex differences, it is necessary to consider those variables properly. Sex differences in TMN stage, histology, and anatomical site of various cancer highlight the importance of exogenous factors and also suggest societal and behavioral difference in men and women could underlie at least some proportion of sex differences. To highlight the importance of these differences, imagine that tumors arising in men and women differ in either their stage at diagnosis or their age at diagnosis. If stage or age is itself associated with somatic mutational features, then a univariate uncontrolled analysis will reveal sex differences in cancer genomics, while a statistical control for these features will remove them. It is unclear how we might best understand sex differences that disappear after controlling for other clinico-epidemiologic features, when those clinico-epidemiologic features themselves might differ across the sexes in specific cancer types. Effectively communicating complex multivariate models appears to be an increasingly important skillset in cancer genomics.

Sex Differences and the Selective Pressure of Treatment

Beyond the clinical and molecular differences between tumors arising in males and females, there are also sex differences in drug responses. Females have a slower clearance and have a stronger side effects from many chemotherapy drugs (63). For example, clearance of 5-florouracil is slower in females, often leading to elevated toxicity (64-67). This sex disparity can be explained by the decreased activity of dihydropyrimidine dehydrogenase, which is responsible for degrading 5-florouracil (68,69). Female patients with colorectal cancer suffered from more frequent and severe adverse side effects with 5-florouracil treatment, including stomatitis, leukopenia, alopecia, and diarrhea (66). Paclitaxel, another chemotherapy drug, was shown to have a 20% lower elimination rate in women with solid tumors and causes more toxicity in female patients, including leukopenia (70-72). A pharmacokinetic study of normal liver function showed that females have slower doxorubicin clearance than males (73). In patients receiving cisplatin-based chemotherapy, women had a higher incidence of vomiting and nausea than men, and the side effects were also harder to control (74).

Sex differences also exist in responses to immunotherapy. Cancer cells can harbor immunosuppression ability through pathways such as cytotoxic T-lymphocyte protein-4 and programmed death receptor 1 (75). Immune checkpoint inhibitors are drugs that inhibit these pathways and are emerging as a breakthrough in the management of multiple cancer types (75). Monoclonal antibodies targeting immune checkpoint inhibitors (eg, ipilimumab, tremelimumab, nivolumab) have shown survival benefits for several solid cancer types (76-78). A recent meta-analysis that reviewed Phase 3 clinical studies in melanoma, lung cancer, head and neck carcinoma, kidney, stomach and urothelial cancers concluded that men had a greater improvement in overall survival with immune checkpoint inhibitors relative to women, suggesting a better drug response in men (79). Another meta-analysis that included advanced or metastatic patients reported a better overall survival of male melanoma patients on immune checkpoint inhibitors alone or with other drugs (chemotherapy or other immunological compounds); an inconsistent result was found in non-small cell lung carcinoma under the same setting (80). This topic is now the subject of intense investigation (81,82).

Sex Discordance in Tumor Evolution

One hypothesis for the sex-differences in response to immunotherapy is that sex hormones such as estrogen mediate the activity of immune cells and overall immunity that could impact drug response (83). But, more broadly, immunotherapy resistance has been linked to tumor evolution: under the selective pressure of the treatment, tumor lineages evolve fitness to the their local environment (84). Thus sexual dimorphism in cancer evolution may play a role in the drug response to immunotherapy.

Evidence that sex differences in cancer evolution may come from study of metastasis. A pair of studies have reported different metastatic patterns in mem and women. In colon and rectal cancer, men had more frequent liver metastases while women had more frequent lung metastases (85,86). Cancer metastasis is an evolutionary process in which tumor cells must acquire the ability to disseminate from primary tumor, survive in the circulation, and ultimately adhere to and colonize the target organ or tissue (87). It is possible that the difference in metastasis pattern in men and women indicates a sex difference in cancer evolution and mutation timing.

Although differences in cancer evolution in sex are largely unstudied outside of the broad differences in driver gene frequency and mutation density reported previously, some initial steps have been taken by mining PCAWG and TCGA data (25,88). Using PCAWG data, it was found that, in biliary adenocarcinoma, females had more frequent polyclonal tumors than males, while males had more clonal somatic structural variants (25). In esophageal cancer, structural variants are more frequently clonal in male-derived samples than in female-derived ones (25). Sex differences were also seen in mutation signatures, with different proportions of single and double base substitution signatures observed in male and female tumors (25). In another report, female carried more subclonal but not clonal mutations in both GBM and LGG, and this sex dimorphism is associated with more malignant tumors (88). Females also tend to have more clonal CDH18 mutations in GBM, and males tend to have more clonal ATRX mutations in LGG, suggesting an interplay between sex-biased driver acquisition and evolutionary timing.

It remains largely unknown whether the metastatic pattern is different in men and women in other cancer subtypes and in non-Caucasians. It is also largely unexplored whether sex-specific signatures evolve differently in different subtypes of cancer. Treatment serves as a powerful selective pressure upon the population of cancer cells and understand to what extent this selective pressure molecularly influences tumors arising in men and women differently is a major gap in our knowledge.

Perspectives: Next Steps

The rapid development of high-throughput DNA sequencing has accelerated our knowledge of cancer genomes and our ability to detect sex differences in them. Although large clinical cancer genomic studies integrating multiple institutions and countries have tremendously accelerated our knowledge, we still have only a modest understanding of sex differences in primary human cancer. Our understanding of sex differences in relapsing and metastatic tumors is negligible. Each cancer type of ICGC and TCGA studies has ~200 to 500 patients. After stratifying by sex and adjusting for cofounding variables such as age and race, statistical powering is very limited. Existing cohorts are poorly representative of the global cancer burden, being biased toward large surgically resected tumors arising in individuals of Caucasian ancestry. Our understanding the interplay of clinical, histopathologic, and epidemiologic features with molecular sex differences in cancer is in its infancy. Thus, targeted sequencing of key driver genes with larger patient number within each individual cancer types might be necessary next step. Fortunately, much of these data are starting to be collected through the routine adoption of DNA sequencing in clinical care, particularly for locally advanced and metastatic tumors.

As previously noted, environmental factors such as dietary pattern and lifestyle are associated with disease risk in several cancer subtypes, but their association with cancer genomic alteration has not been well characterized: these data are absent in most large studies. Dietary information is commonly collected through self-reporting in epidemiological and clinical studies, for both the short term (24-h dietary recall or 3-day diet record) and the long term (food frequency questionnaire). Although those measurements suffer from a questionable accuracy because of underreporting of energy intake and potentially certain categories of dietary components, we and others still recommend that these be incorporated into human studies to begin to adjust for their cofounding effects (89). The recent rapid emergence of wearable technology in healthcare is bringing new hope to this area. Several pioneering studies have used wearable cameras to record dietary data and successfully reduced dietary underreporting (90,91). The collection of other lifestyle data such as physical activity and sleeping also rely heavily on self-reporting and suffer similar biases (92). Improved collection of these data in large cohorts may become increasingly feasible in the near future.

The development of cancer is an evolutionary process consists of somatic mutations under Darwinian selection pressure. Cancer evolution and mutation timing play very important roles in aspects such as drug resistance and metastasis. To better understand sex difference in cancer genomics, evolution will increasingly need to be central to data analysis. This requires us identifying not only the specific mutations in a tumor but also when each occurred during a tumor’s life history and what endogenous or exogenous mutational process most likely caused each one. Great caution should be taken to ensure data quality in cancer evolution studies, including sequencing multiple regions of the same sample; phased or single-cell sequencing are immensely advantageous but are generally cost-prohibitive for routine clinical use at this time (93). It is also largely unknown which key genomic variants contribute to driving tumors to metastasize from their primary sites, and how sex plays a role during this process. This requires not only that we collect paired samples of primary and metastatic tumors but will also that we specifically model the selective and adaptive response to treatment, which itself might show sex differences.

Sex and age are perhaps the 2 most fundamental “subtypes” of human physiology. Understanding how sex influences cancer genomes will improve our understanding of disease initiation, progression, and response to therapy.

Additional Information

Disclosure Statement: P.C.B. sits on the scientific advisory boards of Sage Bionetworks, Intersect Diagnostics Inc. and BioSymetrics Inc. The relationship may not relate to the subject matter of this manuscript.

Data Availability

Data sharing is not applicable to this article as no data sets were generated or analyzed during the current study.