Homologous recombination repair creates mutations in non-coding genome that alter Topoisomerase-1 cleavage sites & orchestrates irinotecan resistance

Selection of drug-resistant mammalian cell mutants requires multiple drug exposures. When cloned genetically identical cells are exposed to the drug, resistance is unlikely to result from selection of pre-existent mutations. Therefore, adaptation must involve generation of drug-resistant mutations de-novo. Understanding how adaptive mutations are generated and protect cells is important for our knowledge of cancer biology and evolution. Here, we studied adaptation of cancer cells to topoisomerase (Top1) inhibitor irinotecan, which triggers DNA breaks, resulting in cytotoxicity. Resistance mechanism was based on gradual accumulation of hundreds of thousands of recurrent mutations in non-coding DNA at sequence-specific Top1 cleavage sites. Repair of DSBs at these sites following initial irinotecan exposures created mutant sequences that were resistant to further Top1 cleavage. Therefore, by creating DNA breaks Top1 increases the rate of highly protective mutations specifically at such spots, thus explaining a puzzling need of dose escalation in resistance development.


Graphical abstract Introduction
In development of antibiotic resistant bacterial forms, resistant colonies emerge following plating of cells on a solid medium with an antibiotic, because of selection of preexistent mutations (1-6).
The process usually takes several months and provides resistance to somewhat higher than initial drug concentrations, but not to full drug resistance.
The forces driving drug-resistance under the dose-escalation setting have not yet been investigated.
Thus, exploring this practically untouched question will be important for understanding the difference in evolutionary processes between lower organisms and mammalian cells. Furthermore, such exploration may help to uncover novel mechanisms of development of drug resistance in cancer patients (10,11,(17)(18)(19)(20)(21)(22)(23)(24)(25)(26). Indeed, multiple administration of drug doses in a clinical setting appears to mimic the in vitro scheme of drug resistance selection. Thus, understanding the mechanisms of the development of drug resistance can also provide novel insights to guide the design of drug combinations and treatment strategies.
Here, we investigated how these adaptive mutations may emerge in colon cancer cells with the example of resistance to a widely used anti-cancer drug irinotecan. Irinotecan is a pro-drug, which is converted into the active drug SN-38, which binds to Top1 (27)(28)(29)(30). The binding, in turn, allows Top1 to make DNA breaks, but prevents re-ligation (28). Top1-mediated single strand breaks may facilitate double strand DNA breaks that, if unrepaired, cause cancer cell death (27). Top1 works both during transcription and replication (28). A number of mechanisms of resistance to Irinotecan have been described, including mutations in Top1, upregulation of multidrug-resistance pumps and associated enzyme system (30)(31)(32)(33)(34). It is difficult, however, to understand why a dose escalation scheme could be important for development of drug resistance based on these mechanisms. Here, we evaluated the efficiency of the development of irinotecan resistance and uncovered a novel resistance mechanism based on the generation of a large number of mutations in Top1-dependent DNA breaking sites that reduce the chances of double strand breaks upon consequent irinotecan exposures in the process of dose escalation.

Experimental design with multiple irinotecan treatments
Since irinotecan is widely used against colon cancer, we investigated its effects on colon cancer cell line HCT116. In order to achieve genetic uniformity of the population, we cloned the cells and isolated seven independent clones. Genetic uniformity within each clone indicates that any drugresistance mutation selected in our experiments occurs either in the process the colony growth from a single cell prior to the drug treatment or is actively generated in the process of drug treatment. Sensitivity to irinotecan differed dramatically between the clones, with minimal toxic concentration between 1nM and 40nM. We chose two clones, SSC1 and SSC7, with high sensitivity for further experiments.
To understand development of drug resistance, SSC7 cells were exposed to 4nM irinotecan, which led to the death of a significant fraction of the population, and the cell cycle arrest of the rest of the population. Cells remained in the arrested state without divisions for 14 days, and then resumed growth. Second treatment with 4nM irinotecan led to a shorter period of the cell cycle arrest. Such a treatment cycle was conducted five times in total. At the fifth cycle, practically no cell death or growth inhibition was seen, indicating that cells became fully adapted to this concentration of irinotecan (Fig. 1A, B).
Further, we tested if a similar pattern occurs when cells develop adaptation to high doses (40nM) of irinotecan. The fraction of cells that survived 40nM stayed in a senescent state for more than three months, after which cells resumed propagation and filled the plate. Upon subsequent exposure of the recovered population to 40nM irinotecan the period of the growth arrest was only about one month. The process was repeated three more times, and each time the fraction of dying cells was lower and the time period of growth arrest was shorter compared to the previous round of selection, so that following the fourth exposure, cells spent approximately one week in the arrested state (Fig 1A, B). Altogether, these findings indicate similarities of adaptation to low and high doses of irinotecan.

Most of the survived cells recover from the senescence like growth arrest
A large fraction of cells in the population could adapt to the treatment and resume growth after the cell cycle arrest. Alternatively, a small fraction of cells that could be originally resistant to the drug continued to propagate, which became detectable only when they began outgrowing the rest of the arrested population. To distinguish between these possibilities, SSC7 cells were infected with the cell cycle reporter virus (35) and exposed to 4nM irinotecan. Survived treated cells stopped dividing, acquired senescence-like phenotype, and according to the reporter, underwent G1 growth arrest (Fig.1C, D). Cells remained in G1 for four days, after which a majority of cells exited G1 and entered the cell cycle (Fig.1C, D). Entering the cycle following the senescence-like arrest was surprisingly slow, and only by day 8 almost 100% of cells reached G2. There were no localized shifts of cells to G2, indicating the lack of clonal expansion. This observation indicates that cells underwent true adaptation to irinotecan, rather than reflects the expansion of a small fraction of initially resistant cells.

A large fraction of cells gets adapted to irinotecan
To quantitatively assess the process of adaptation, 10 million SSC1 cells were individually barcoded using Cellecta 50M barcodes lentiviral library (36). When cells filled the plate, half of them were collected for the barcode analysis, and half were used for further dose-escalation experiments with sequential passaging over 2, 4, 6, 8, and 15nM of irinotecan. With each passage, we observed a reduction of the population of dying cells, and a reduction of the period of growth arrest ( Fig. 1E). At different stages of the experiment, we collected cells for the analysis of barcodes, as described in Materials and Methods.
Comparison of barcodes that were detected in the control population and population treated with 2nM showed that about 4 * 10 -3 of the original clones survived the selection (Fig. 1E). Analysis of barcodes in the population of cells after the 15nM irinotecan selection demonstrated that 2 * 10 -3 of original clones survived the entire series of selections, which is only two times lower than the number of clones survived the first round of selection at 2nM. These findings indicate that (a) the probability of the survival of clones is high compared to the usual probability of spontaneous or even mutagen-induced mutations (according to classical studies usually not higher than 10 -4 (3,4,7,8,(11)(12)(13)(14)(15)37)), and (b) almost 50% of clones that survive 2nM selection survived the entire series of selections, suggesting that if cells are able to survive initial treatment they can also survive the dose escalation treatment.
To test if the dose escalation process is critical for the development of the resistant forms, barcoded SSC1 cells were exposed directly to 15nM irinotecan. Analysis of barcodes indicated that only 8 2 * 10 -5 of clones survived (Fig. 1E), which is 100X lower than the survival rate of 15nM irinotecan in the dose escalation experiment. These data indicate that the dose escalation protocol is important for effective development of resistant variants.

Changes in transcriptome may not be involved in the resistance development
To explore the mechanisms of the adaptation, the survived population of multiple rounds of treatment with 40nM irinotecan, was cloned again. Barcodes from several clones were isolated and sequenced. Three clones with different barcodes (MSC1, MSC2 and MSC3) were chosen for further analysis. Since the barcoding of cells was done prior to the entire series of irinotecan treatments, the fact that these clones carry different barcodes indicated that they did not split from the same clone somewhere in the middle of the irinotecan treatment. In other words, they represent the progeny of cells that underwent the entire series of the treatments independently of each other.
We would like to reiterate that the original barcoded population was genetically homogenous because of the initial cloning.
To uncover potential mechanisms related to changes in gene expression, parental SSC1 clone and individual mutant isolates MSC1, MSC2 and MSC3 were exposed to 10nM irinotecan and their transcriptome was analyzed by RNAseq. A number of differentially expressed genes was observed in the survived clones. Importantly, we did not observe changes in either MDR1 or other ABC transporters involved in drug efflux, or Top1, or DNA repair genes (Table S1), suggesting that the mechanism of resistance in these clones may not be related to expected changes in the transcriptome.

DSBs significantly contribute to irinotecan-induced cell death
To further explore potential mechanisms of the resistance, we performed a pooled shRNA screen to identify genes important for survival of irinotecan treatment. SSC1 cells were infected with focused lentiviral shRNA library Decipher Module-1(38) that targets signaling pathways. This library covers about 20% of human genes. Cells were treated with 10nM irinotecan for 24 hours and cells that survived the treatment after 5 days were collected and the barcodes isolated, sequenced and analyzed. As control, we used the same population of infected cells but without irinotecan treatment. Among genes, depletion of which showed sensitizing effects, was a group of genes that plays a role in DNA double strand breaks repair, predominantly representing the homologous recombination (HR) and nonhomologous end joining (NHEJ) DNA repair pathways, including POLE, POLE3, POLE4, KAT5, RAD51C, RAD54L, RAD1, RAD9A, H2AFX, LIG4 and PARP2 (Table S2). We also observed a number of genes involved in translesion DNA synthesis (Table S2). These data reinforce the understanding that (a) generation of double strand DNA breaks is critical for cell death caused by irinotecan and (b) HR and to some extent NHEJ repair pathways play an important role in the irinotecan survival in naïve cells (though these pathways are not upregulated in the resistant clones (Table S1)).

Development of irinotecan resistance associates with emergence of multiple non-random mutations
To study mutations that emerged in the survived clones, we performed whole genome sequencing.
Genomes of the survived clones MSC1, MSC2 and MSC3 were compared with the genome of original population SSC1, and each of them with the Reference Genome (GRCh38/hg38) in the UCSC (39) database. We observed that the parental SSC1 genome had hundreds of thousands of SNPs and InDels compared to the Reference Genome.
These mutations may reflect the fact that HCT116 cells were isolated from a different individual than a group of individuals whose sequence compose the Reference Genome. Alternatively, these mutations could arise in the process of cancer development and/or further culturing of HCT116 cells in the laboratory conditions. Overall analysis of mutation in SSC1 indicated that 93% of them do not correspond to known SNPs in the human population, suggesting that the vast majority of the mutations simply reflect either the cancer nature of these cells or genetic instability upon culturing. Accordingly, they will be further called "cancer alleles". When genomes of the irinotecan-resistant isolates MSC1, MSC2 and MSC3 were compared with the genome of SSC1, we identified hundreds of thousands of InDels and SNPs that arose in the process of adaptation to irinotecan.
Strikingly a very large fraction of these mutations was common between the independently isolated clones ( Fig. 2A, B). If compared by pairs, i.e. MSC1/MSC2, MSC1/MSC3 and MSC2/MSC3, in each pair between 17% and 45% of mutations were common, and between 7% and 15% were common between all three independent isolates (46,099 mutations, of which 28.4% were InDels and 71.6% were point mutations (Table S4)). Even considering that irinotecan may have a high mutagenic activity and triggers protective mutations with the rate as high as 10 -4 per nucleotide, the probability of overlap of a mutation in three independent clones will be 10 -12 (i.e. way less than one triple mutation per 3*10 9 bp genome), which is many orders of magnitude lower than seen in the experiment. Thus, the overlapping (common between three clones) mutations clearly point to a non-random mutation mechanism. Generation of these mutations seem to be guided by a mechanism, understanding of which may clarify the adaptation pathway.
Analysis of the ENCODE (promotors and enhancer datasets (40)), GEO (GSE57628, mapping of Top1 sites in human HCT116 cells) (41) and UCSC (39) database datasets uncovered that the vast majority of the mutations are present in Top1 cleavage sites, heterochromatin (high content of histone H3K9me3) in the noncoding regions (Fig. 2C). There was a small fraction of mutations present in promotors/enhancers (<1%) and the coding regions (2%), most of them in the exons ( Fig. 2B, C). Importantly, the genes that have mutations in their coding and regulatory regions did not belong to known pathways associated with either cell survival or DNA repair, and therefore are unlikely to be involved in the adaptation process (Table S3, also see mutation landscape section in supplement for description on an exception case). To avoid dilution of the focus of this paper, we present more detailed analysis of mutations in the Supplement information (see supplement section Mutation Landscape, mutation mapping to regions in genome and its correlations to distinct signatures such as nature of repetitive elements (Fig. S1), identification of pericentromeric and heterochromatin region based on methylation signatures of H3k9me3 and H3k27me3 (Fig.   S2), Genome wide density plot for triple mutations (Fig. S3)). Altogether, these data suggest that adaptation to irinotecan was not associated with either mutations in functional genes or changes of expression of these genes.

Mutations result from repair of Top1-generated DSBs
To understand how hundreds of thousands of mutations in repeats and untranslated regions (see Mutation landscape section in the Supplement) could be involved in adaptation to irinotecan, we proposed that they could result from DNA breaks generated by Top1. Indeed, there are several lines of evidence that most of these mutations were generated either by HR or NHEJ repair of DSBs. 16% of the mutations in the isolates were clustered (7, Tables S5,   S6). Since random positioning predicts that mutations should be on average separated by about 10,000 bp (about 300,000 mutations per clone distributed over 3 billion base pairs of the total genome), such clustering demonstrated their non-random appearance and suggested a mechanism of their generation. In all these cases, these were loss of heterozygosity (LOH) mutations. These LOH mutations did not result from deletions of one of the alleles (since the number of reads corresponding to these regions were similar to the average number of reads along the genome), but by copying of an allele from one chromosome to another, including copying of the entire mutation cluster (Fig. 3B). Such copying could result only from the HR repair of DSBs. LOH occurred in 78% of common mutations (36,228 mutations out of 46,099), and similar fraction of LOH was found with the overall set of mutations in resistant clones (Table S7), suggesting that the majority of mutations were generated by the HR repair system.
In the resistant clones 28% of de-novo mutations that were not LOH resulted from NHEJ (9,871 de novo common mutations, and similar fraction of NHEJ was found in the overall set of mutations). They resulted from NHEJ because in case of insertions, these mutations have a very specific signature of duplication of a neighboring region. This duplication results from pairing of broken ends at the terminal nucleotides and filling the gaps on both strands via translesion DNA synthesis (Fig. 3B). This structure allows precise identification of the site of the DNA break, i.e. at the site of pairing between the duplication regions ( Fig. 3B) (see also Mutation landscape section and Table S8). Overall, de novo appearing InDels can be used as hallmarks of DNA breaks that were repaired via NHEJ, while LOH mutations can be used as hallmarks of DNA breaks repaired via HR.
Very importantly, a high fraction of the overlapping mutations between independent resistant clones suggests that the irinotecan-inhibited Top1 generates DNA breaks at specific sites in the chromatin (possibly specific Top1 binding or activation sites), which further leads to generation of mutations upon HR or NHEJ repair of DSBs.

Mechanism of adaptation to irinotecan
The following considerations brought a framework for understanding the mechanism of adaptation. At each genome location, the parental SSC1 population could have alleles either with no mutations compared to the Reference Genome (0/0), with mutation in one allele (0/1) (heterozygosity), or both alleles (1/1) (Fig. 4A, B). Sites where two alleles in SSC1 had different mutations compared to the Reference Genome (1/2) were extremely rare. Accordingly, mutations that appear in MSC clones compared to parental SSC1 could be mutations de-novo generated by NHEJ (e.g. 0/0→0/1) or loss of heterozygosity generated by HR (0/1→0/0; or 0/1→1/1) (Fig. 4C, In sites with LOH repaired by HR, a heterozygous allele could revert to either the Reference Genome allele (0/1 to 0/0) or to the "cancer" allele seen in the parental clone SSC1 (0/1 to 1/1). A surprising key observation that led to understanding the mechanism of adaptation was that the frequency of 0/1 to 0/0 shifts was 5.44 times higher than 0/1 to 1/1 shifts (Fig. 5A). 0/1 to 0/0 shift means that the allele from the Reference Genome was copied to the DSB at the "cancer" allele (an allele in SSC1 parental cells that differs from the Reference Genome), which ultimately means that the probability of double strand DNA breaks in "cancer" alleles is 5.44 time higher compared to the Reference Genome allele. The probability of breaks in "cancer" allele was even higher in the pericentromeric regions, where the ratio of breaks in Reference Genome allele to "cancer" allele was 1/20 ( Fig. 5B). In the chromosome arms this ratio was 1/3.4. Therefore, surprisingly, alleles that acquired mutations in the course of cancer development were significantly more prone to Top1-induced double strand breaks than normal human genome alleles ( Fig 5B).
This unexpected finding provides the mechanism of gradual adaptation to irinotecan. Initial exposures to irinotecan generate reversion of a number of "cancer" alleles to the Reference Genome alleles, which are more resistant to Top1-induced DSBs, which in turn creates a protective mechanism against following exposures. In other words, with each exposure there will be fewer and fewer potential Top1-cleavage sites, which will lead to stronger and stronger adaptation. This novel mechanism of adaptation to irinotecan does not involve expression of any protective proteins or mutations in them, but rather involves a high number of changes in the DNA structure that make it less prone to Top1-induced breaks.

Adaptation associates with reduced ability of irinotecan to trigger DNA breaks
This mechanism predicts that in the process of adaptation, following multiple exposures to the same concentration of irinotecan, cells should experience fewer DSBs, while the rate of repair of DSBs remains the same. To test this prediction, we took cells that have not been drug exposed, and that underwent five cycles of exposure to 4nM of irinotecan. Both populations were subjected to 4nM of irinotecan for 24 hours, and the number of g-H2AX foci was assessed by immunofluorescence. While irinotecan exposure of naïve cells caused a dramatic increase in the number of foci, exposure of cells that were preadapted by five cycles of irinotecan treatment barely caused foci formation (Fig. 6A, B). On the other hand, the rate of DSB repair (recovery of gH2AX foci) was not accelerated (Fig. 6C). Importantly, lower foci formation correlated with the lack of cell death and growth arrest. Similarly, there was a lower overall number of DNA breaks in adapted cells, as judged by the comet assay ( Fig. 7A-E). Therefore, indeed reversion of "cancer" alleles to the Reference Genome alleles appears to be associated with fewer DSBs.

Discussion
Here, we addressed why the approach towards the selection of drug-resistant mutants in cancer cells requires multiple exposures to drugs and dose escalation. Such a selection scheme suggests that (a) either cells are somehow adapted to the low concentrations of drugs (e.g., via epigenetics mechanisms), and this adaptation guides further selection of the resistant mutant forms; or alternatively (b), development of drug resistance involves acquiring of a large number of mutations, each of which provides a fraction of the resistance, but gradually they accumulate and get selected in the process of drug escalation. At least with an inhibitor of Top1 irinotecan, we show that the second possibility is correct. It appeared that a very high number of mutations is generated in the process of selection of irinotecan-resistant mutants in the dose escalation experiment. Surprisingly, the absolute majority of them were in the non-coding and silenced regions of the genome, suggesting that these mutations do not affect expression or function of specific genes involved in the irinotecan resistance. Accordingly, this mechanism of adaptation is fundamentally different from previously known mechanisms related to changes in drug target, drug metabolism or transport.
The key to understanding the nature of the resistance was the observation that these mutations result from the repaired DSBs via the HR and NHEJ pathways. Though cuts by Top1 generate single strand breaks (SSB), these breaks can develop into DSBs (28). In HR-dependent DSB repair, we observed loss of heterozygocity associated with copying of intact alleles to the allele with DSB, which allowed precise identification of the allele with DSB. Strikingly, more than 80% of DSBs took place in the alleles with mutations associated with the cancer nature of the parental cells, and accordingly, the HR repair led to restoration of the original "normal" alleles. As an example of such sequences, we found a large fraction of breaks in the polyC sequences of 30-40bp, which were present in the parental cancer cells (See Mutation landscape, Supplement Information). As a result of HR repair, these alleles were changed to alleles with interrupted polyC regions which were present in the reference human genome. Accordingly, at these sites, DSBs appear to require extended polyC, and thus an allele that has interruption of the polyC tract must have a lower probability of breaks. Therefore, reversion of extended polyC to the interrupted tract of polyC protects this site from further breaks, and thus contributes to the overall development of resistance to irinotecan (Fig. S4). This observation ultimately means that upon the first exposure to the drug, a fraction of sites with high probability of Top1-induced breaks will be reverted to sites with low probability of breaks. Therefore, with each cycle of exposure to irinotecan fewer and fewer sites with high probability of DSBs will remain in the genome, which ultimately must increase the chance of survival. Indeed, we demonstrated that the number of DSB is reduced following cycles of exposure to irinotecan, and this effect was associated with lower probability of breaks rather than with more efficient DNA repair. Therefore, development of resistance to irinotecan involves acquiring a large number of mutations, each of which provides a fraction of the overall resistance, which gradually accumulate and are selected in the process of drug escalation.
This adaptation mechanism that associates with gradual reduction of the number of DNA breaks by Top1, suggests lower efficiency of Top1-dependent relaxation of DNA supercoils in selected clones. Possibly, the number of mutations that provide adaptation to irinotecan may be limited by the necessity to carry out the DNA relaxation activity. Alternatively, in these clones Top2 can take over essential DNA relaxation. This work also illuminates novel aspects of function of Top1. Though it was reported that Top1 associates with active RNA polymerase to relieve DNA supercoils generated in the process of transcription (41)(42)(43), our data suggests that Top1 can also functions in a transcription-independent relief of supercoils since a majority of mutations was seen in heterochromatin (H3K9me3 and H3K27me3 patterns, see Fig. S2). The fact that there was a very large fraction of DSBs common between the three-independent irinotecan-resistant isolates indicates that there are preferable sites of breakage. Indeed, when mutation sites were compared with sites of DNA cleavage by Top1 (41), we observed a strong (19.8%) overlap. This percent of overlap is probably an underestimation since experimental conditions in two studies were different. Possibly, these mutation sites are preferable sites of binding of Top1 or some Top1 activating factors to DNA, suggesting that Top1 or its activators have a sequence binding preference. Alternatively, Top1 may bind anywhere on the DNA, but moves together with RNApol and possibly DNApol, and stalls at these regions to increase the probability of cuts. Another attractive possibility is that these repeat regions are preferable sites where SSBs are converted to DSBs.
The novel drug resistance mechanism may have interesting implications for understanding evolutionary processes. Indeed, it is possible that DSBs generated by Top1 take place predominantly at the sites of mutations that deviate from "normal" genomes. Accordingly, these DSBs can be repaired by the HR pathway, which will lead to the restoration of normal genome homozygocity. In other words, a stabilizing evolutionary selection may take place even in the absence of the selection pressure, simply as a result of the Top1 and HR repair function.
Accordingly, the overall diversity of SNPs and InDels in the plurality of normal genomes has limitations that are shaped by the function of the Top1 and HR repair systems. Amplification of the barcodes was carried out by nested PCR. Detailed procedure of PCR and primer details for shRNA screen is available in Supplementary datasheet "Data S1" (Tables S11& S12). Briefly, 1-st PCR (PCR 1) was performed using Titanium Taq DNA Polymerase (# 639209, Takara Bio, CA, USA). Separation of the PCR products from primers and gel purification was done by QIAquick PCR & Gel Cleanup Kit (Qiagen, Germany). 2-nd PCR (PCR 2) was carried out using nested primers either generic or having unique sample barcodes. PCR 2 was performed using Phusion High-Fidelity PCR Master Mix (Thermo Scientific, MA, USA). Samples were multiplexed by adding an additional sample barcode during the second round of PCR. Samples were normalized individually, then pooled together, and purification of the PCR products was completed using AmpureXP magnetic beads (Beckman Coulter, CA, USA) following manufacturer protocols. Next, we sequenced the barcodes using Ion Torrent.

Materials and Methods
Analysis of the barcode data. We used a combination of custom-tailored applications to analyze sequencing reads along with the R programming language. Data were first checked for quality of reads through FastQC (v0.11.7, RRID:SCR_014583)(46), further using barcode-splitter (v0.18.6, barcode splitter, RRID:SCR_021825)(47) reads were demultiplexed based on sample barcodes (1 error as mismatch or deletion was allowed for sample barcodes while demultiplexing). Obtained Transcriptome analysis. RNA was extracted from cells using the RNeasy Mini kit (Cat#74104, Qiagen). Library preparation strategy (BGISEQ-500, RRID:SCR_017979) was adopted and performed by BGI, China. Briefly, mRNA molecules were purified from total RNA using oligo(dT) attached magnetic beads and fragmented into small pieces using fragmentation reagent after reaction a certain period at proper temperature. First-strand cDNA was generated using random hexamer-primed reverse transcription, followed by a second strand cDNA synthesis. The synthesized cDNA was subjected to end-repair and then was 3' adenylated. Adapters were ligated to the ends of these 3' adenylated cDNA fragments. This process amplified the cDNA fragments with adapters from previous step. PCR products were purified with Ampure XP Beads (AGENCOURT), and dissolved in EB solution. Library was validated on the Agilent Technologies products were heat denatured and circularized by the splint oligo sequence. The single strand circle DNA (ssCir DNA) were formatted as the final library. The library was amplified with phi29 to make DNA nanoball (DNB) which had more than 300 copies of one molecular. The DNBs were load into the patterned nanoarray and single end 50 (pair end 100) bases reads were generated in the way of sequenced by synthesis. For in house pipeline was developed to analyze the data where reads were first trimmed and clipped for quality control in trim_galore (v0.5.0, RRID:SCR_011847)(51) and checked for each sample using FastQC (v0.11.7, RRID:SCR_014583)(46). Data was aligned by Hisat2 (v2.1.0, RRID:SCR_015530)(52) using hg38, GRch38.97. High-quality reads were then imported into samtools (v1.9 using htslib 1.9, RRID:SCR_002105)(53) for conversion into SAM files and later to BAM file. Gene-count summaries were generated with featureCounts (v1.6.3, RRID:SCR_012919)(54): A numeric matrix of raw read counts was generated, with genes in rows and samples in columns, and used for differential gene expression analysis with the Bioconductor, RRID:SCR_006442 (edgeR v3.32.1, RRID:SCR_012802 (55), Limma v3.46.0, RRID:SCR_010943(56)) packages to calculate differential expression of genes. For normalization "voom" function was used, followed by eBayes and decideTests functions to compute differential expression of genes.

Assessment of cluster mutations in resistant mutants.
For estimating the number of mutations present in close proximity or whether it is clustered, we took 100bp small windows and calculated whether the mutations were present in close proximity in these small intervals. We used custom written python codes to first, separate the whole genome into small bins of 100bp respectively. differences was determined using unpaired Welch's correction, two-tailed t-test (*P < 0.0332, **P < 0.0021, ***P < 0.0002, ****P <0.0001).

Single cell gel electrophoresis (SCGE).
To estimate the change in break of DNA strands before and after treatment in normal cells and resistant cell population, a well-known method of single cell gel electrophoresis (SCGE) or COMET assay was performed (68). Cells were taken and embedded in 1% agarose on a microscope slide are lysed with detergent and high salt.
Glass slides were precoated with 1% agarose. Cells were treated with irinotecan or left untreated to serve as control. Preparation of sample was done by scrapping cells gently with 0.05% trypsin.   available in supplement information Table S13) is included in this paper. All list genes from transcriptome analysis and shRNA screens are available in this paper.
Raw data and files after post analysis for transcriptome analysis is deposited and is available at GEO data base with accession number GSE189366. Analyzed data is presented in supplement        Statistics were calculated as mean of n=21 comets in each using GraphPad Prism version 9.0.0, California USA. The significance of differences was determined using unpaired Welch's correction, two-tailed t-test (*P < 0.0332, **P < 0.0021, ***P < 0.0002, ****P <0.0001) denoted in above figures (C, D, E).

Supplementary Materials
This article contains supplement information, and supplement tables that is provided separately. Other Supplementary Materials for this manuscript include the following: Data S1 spreadsheet format (.xlsx)

Supplementary Text Mutational landscape in irinotecan resistant clones.
As noted in the main text, mutations in coding and regulatory regions did not affect genes involved in cell sensitivity to irinotecan. The only exception was a mutation in the MMS22L gene that normally works in the base excision repair, but also plays a role in the DSB repair (69,70).
However, this mutation was a 1 nucleotide insertion, which most likely inactivated the MMS22L gene. In such a scenario, we expect to see sensitization rather than protection from DSBs.
Accordingly, it is unlikely that this mutation associates with adaptation to irinotecan.
To understand the relevance of the mutations to the adaptation process, we manually identified the precise positions of breaks in 1,000 out of 2,829 NHEJ-repaired common mutation sites and analyzed if there are any shared features in sequences of the sites. About 80% of these breaks took place either within or at the edge of 1-4 bp repeat clusters, and the length of these repeats usually was within the range of 5-30bp (see examples of such repeats in Table S8). Though we could not define the breaking points in HR repaired sites precisely, we observed that 75% of these sites are in similar repeats as NHEJ breaking points, and therefore assumed that breaks in these sites occurred in or at the edges of such repeats, as with NHEJ. Overall UCSC analysis showed that the mutations were mostly located in the repetitive elements such as microsatellites (Table S14), simple repeats (Table S15) and satellites (Table S16 for mutations in chromosomal arms and supplement (Table S17) for mutations in pericentromeric genomic locations). Parallel RepeatMasker analysis supports these findings (Fig S1A, Table S18 and S19 for detailed data on repetitive elements and mutation correlation) where specific satellites such as alpha satellites and human satellite-II is shown to have majority of common mutations (Fig S1B). In other words, repetitive elements accommodate majority of all mutations, including common mutations between the clones, during the adaptation process.
Upon analysis of distribution of mutations along the chromosomes, we found that there was a disproportionally around 40% of the total mutations densely populated regions adjusted to centromere (up to 5Mb depending on the chromosome) at both sides (Fig. S2), which we referred to as "pericentromeric" areas. This region is identified generally by signatures of H3K9me3 and H3K27me3 followed by richness of repetitive elements such as satellites and simple repeats ( Fig   S2A, B and Fig S3). As a representative example, Figure S2a shows correlation between pericentromeric locations of common mutations on chromosome-1, pericentromeric chromatin modification signatures, and locations of satellites. For comparison, Fig. S2B shows these distributions and parameters in a chromosomal arm of same chromosome-1. Whereas, Fig. S3 visualizes the overall distribution of common mutations in genome and show its specific enriched presence in centromere and pericentromeric regions.
A very common sequence at the breaking points in the chromosome arms was a polyC stretch (25% of all mutation sites in the arms). The length of these polyC sequences associated with breaking points usually were within the range of 20-40bp. To our surprise, in 100% cases of LOH in these regions, original long polyC was substituted with polyC stretch that was interrupted by multiple SNPs (Fig. S4). The fact that these breaks were generated in response to Top1 inhibition by irinotecan suggests that Top1 either binds to such sites specifically, or stalls there in the process of its function, or is activated at these sites.  Alleles (red and blue) in parental line represent heterozygosity (0/1) with "Cs" base. After the irinotecan adaptation, generated mutations lead to LOH, so that the polyC tract became interrupted as in normal reference genome alleles.

Data S1. (Separate file)
List of tables has been provided in excel sheet as "Data S1".