Cancer drug resistance is multifactorial, driven by heritable (epi)genetic changes but also by phenotypic plasticity. In this study, we dissected the drivers of resistance by perturbing organoids derived from patients with colorectal cancer longitudinally with drugs in sequence. Combined longitudinal lineage tracking, single-cell multiomics analysis, evolutionary modeling, and machine learning revealed that different targeted drugs select for distinct subclones, supporting rationally designed drug sequences. The cellular memory of drug resistance was encoded as a heritable epigenetic configuration from which multiple transcriptional programs could run, supporting a one-to-many (epi)genotype-to-phenotype map that explains how clonal expansions and plasticity manifest together. This epigenetic landscape may ensure drug-resistant subclones can exhibit distinct phenotypes in changing environments while still preserving the cellular memory encoding for their selective advantage. Chemotherapy resistance was instead entirely driven by transient phenotypic plasticity rather than stable clonal selection. Inducing further chromosomal instability before drug application changed clonal evolution but not convergent transcriptional programs. Collectively, these data show how genetic and epigenetic alterations are selected to engender a “permissive epigenome” that enables phenotypic plasticity. Significance: Drug resistance is driven by genetic–epigenetic memory that enables cancer cells to adopt multiple phenotypic states depending on environmental conditions, supporting integration of evolutionary principles into biomarker discovery and personalized treatment strategies. This article is part of a special series: Driving Cancer Discoveries with Computational Research, Data Science, and Machine Learning/AI.

In the absence of both positive and negative selections, coding sequences evolve at a neutral rate (R = 1). Such a high genomic rate is generally not achievable due to the prevalence of negative selection against codon substitutions. Remarkably, somatic evolution exhibits the seemingly neutral rate R ∼ 1 across normal and cancerous tissues. Nevertheless, R ∼ 1 may also mean that positive and negative selections are both strong, but equal in intensity. We refer to this regime as quasi-neutral. Indeed, individual genes in cancer cells often evolve at a much higher, or lower, rate than R ∼ 1. Here, we show that 1) quasi-neutrality is much more likely when populations are small (N < 50); 2) stem-cell populations in single normal tissue niches, from which tumors likely emerge, have a small N (usually <50) but selection at this stage is measurable and strong; 3) when N dips below 50, selection efficacy decreases precipitously; and 4) notably, N is smaller in the stem-cell niche of the small intestine than in the colon. Hence, the ∼70-fold higher rate of phenotypic evolution (observed as cancer risk) in the latter can be explained by the greater efficacy of selection, which then leads to the fixation of more advantageous and fewer deleterious mutations in colon cancers. In conclusion, quasi-neutral evolution sheds a new light on a general evolutionary principle that helps to explain aspects of cancer evolution.

Despite the concern of within-tumor genetic diversity, this diversity is in fact limited by the kinship among cells in the tumor. Indeed, genomic studies have amply supported the ‘Nowell dogma’ whereby cells of the same tumor descend from a single progenitor cell. In parallel, genomic data also suggest that the diversity could be >10-fold larger if tumor cells are of multiple origins. We develop an evolutionary hypothesis that a single tumor may often harbor multiple cell clones of independent origins, but only one would be large enough to be detected. To test the hypothesis, we search for independent tumors within a larger one (or tumors-in-tumor). Very high density sampling was done on two cases of colon tumors. Case 1 indeed has 13 independent clones of disparate sizes, many having heavy mutation burdens and potentially highly tumorigenic. In Case 2, despite a very intensive search, only two small independent clones could be found. The two cases show very similar movements and metastasis of the dominant clone. Cells initially move actively in the expanding tumor but become nearly immobile in late stages. In conclusion, tumors-in-tumor are plausible but could be very demanding to find. Despite their small sizes, they can enhance the within-tumor diversity by orders of magnitude. Such increases may contribute to the missing genetic diversity associated with the resistance to cancer therapy.

The rapidity with which the mutation rate evolves could greatly impact evolutionary patterns. Nevertheless, most studies simply assume a constant rate in the time scale of interest (Kimura 1983; Drake 1991; Kumar 2005; Li 2007; Lynch 2010). In contrast, recent studies of somatic mutations suggest that the mutation rate may vary by several orders of magnitude within a lifetime (Kandoth et al. 2013; Lawrence et al. 2013). To resolve the discrepancy, we now propose a runaway model, applicable to both the germline and soma, whereby mutator mutations form a positive-feedback loop. In this loop, any mutator mutation would increase the rate of acquiring the next mutator, thus triggering a runaway escalation in mutation rate. The process can be initiated more readily if there are many weak mutators than a few strong ones. Interestingly, even a small increase in the mutation rate at birth could trigger the runaway process, resulting in unfit progeny. In slowly reproducing species, the need to minimize the risk of this uncontrolled accumulation would thus favor setting the mutation rate low. In comparison, species that starts and ends reproduction sooner do not face the risk and may set the baseline mutation rate higher. The mutation rate would evolve in response to the risk of runaway mutation, in particular, when the generation time changes. A rapidly evolving mutation rate may shed new lights on many evolutionary phenomena (Elango et al. 2006; Thomas et al. 2010, 2018; Langergraber et al. 2012; Besenbacher et al. 2019).

Historically, DNA sequence mutability has been considered relatively uniform and low in tumors with chromosomal instability (CIN), based on the assumption that high mutability would be detrimental in karyotypically aberrant contexts. Recent in silico analyses have challenged this view, suggesting some heterogeneity in mutation rates across CIN tumors; however, these predictions lack experimental validation. It also remains unclear how the intertumor variability of mutation rates compares to intratumor diversification and evolves along disease progression, whether mutation rates are functionally relevant in CIN cancers, and which mutational processes shape mutational accrual during CIN tumor onset and evolution. To address these gaps, we performed mutation accumulation experiments using clonal populations of patient-derived tumoroids from seven CIN, microsatellite-stable colorectal cancers (CRCs), and one microsatellite-unstable CRC. Each tumor exhibited a distinctive mutation rate footprint that was conserved among different clones from the same ancestor. In contrast, mutation rates diverged markedly across different tumors, with variations in magnitude within microsatellite-stable tumors as prominent as those distinguishing them from microsatellite-unstable tumors. New mutations reflected mutational processes associated with defective DNA replication and repair, which were not detected in normal tissues. Last, both mutation accumulation assays and high-depth whole-exome sequencing of subclonal variants showed higher mutation rates in metastatic lesions compared with matched primary tumors, suggesting positive selection for cells with increasing mutability during cancer dissemination. By providing an empirical assessment of mutation rates in human cancer, our data delineate heterogeneity, heritability, and progression-associated evolvability of DNA mutational instability as hallmarks of microsatellite-stable CRC.

Oestrogen receptor (ER) activation leads to the formation of DNA double strand breaks (DSB), promoting genomic instability and tumour heterogeneity. The single-stranded DNA cytosine deaminase APOBEC3B (A3B) serves as a co-activator of ER and is implicated in inducing DSBs at transcriptional enhancers regulated by ER. Using whole-genome sequencing in an engineered cell model lacking base excision repair (BER) function, we demonstrate that A3B preferentially targets transcriptionally active regulatory regions in an R-loop-dependent manner. Strand-specific DNA:RNA immunoprecipitation sequencing (ssDRIP-seq) and ssDNA-associated protein immunoprecipitation sequencing (SPI-seq) confirm that A3B binds to and deaminates ssDNA within R-loops, a process facilitated by ER transactivation. Furthermore, BER-mediated processing of A3B-induced uracil bases contributes to the formation of R-loop-associated DSBs, which are essential for ER-regulated gene activation. These findings establish a role for A3B in R-loop homeostasis and transcriptional regulation, with implications for understanding ER-driven genomic instability and potential therapeutic targeting of A3B.