The role of APOBEC3B in lung tumor evolution and targeted cancer therapy resistance

In this study, the impact of the apolipoprotein B mRNA-editing catalytic subunit-like (APOBEC) enzyme APOBEC3B (A3B) on epidermal growth factor receptor (EGFR)-driven lung cancer was assessed. A3B expression in EGFR mutant (EGFRmut) non-small-cell lung cancer (NSCLC) mouse models constrained tumorigenesis, while A3B expression in tumors treated with EGFR-targeted cancer therapy was associated with treatment resistance. Analyses of human NSCLC models treated with EGFR-targeted therapy showed upregulation of A3B and revealed therapy-induced activation of nuclear factor kappa B (NF-κB) as an inducer of A3B expression. Significantly reduced viability was observed with A3B deficiency, and A3B was required for the enrichment of APOBEC mutation signatures, in targeted therapy-treated human NSCLC preclinical models. Upregulation of A3B was confirmed in patients with NSCLC treated with EGFR-targeted therapy. This study uncovers the multifaceted roles of A3B in NSCLC and identifies A3B as a potential target for more durable responses to targeted cancer therapy.

A3B promotes tyrosine kinase inhibitor (TKI) resistance.Next, the impact of A3B on tumor evolution with EGFR TKI therapy was examined.Subclonal expression of A3B in TKI-treated EA3Bi mice drove a significant increase in tumor grade, tumor nodules per lung section and tumor area per tissue area compared with TKI-treated Ei control mice (Fig. 3a-d).Heterogeneous A3B tumor positivity (Fig. 3e) and a significant increase in A3B positivity with TKI therapy compared to untreated EA3Bi mice were observed (Fig. 3f).In an additional experiment, tumor growth and progression with TKI treatment were associated with a significant increase in tumor nodules and a substantial increase in tumor grade in EA3Bi mice compared with Ei control mice (Fig. 3g-i).Based on previous work illustrating an important role for uracil DNA glycosylase (UNG) in repairing APOBEC-induced uracil lesions 28 , we evaluated UNG expression in A3B-expressing EA3Bi tumors.Staining for UNG revealed a significant decrease in UNG-positive cells per tumor in EA3Bi mice compared with Ei mice treated with TKI therapy (Fig. 3j,k).Taken together, these findings suggest that subclonal A3B expression with TKI therapy in conjunction with UNG downregulation contributes to increased tumor growth and TKI resistance.
Next, whole-exome sequencing (WES) was performed on TN and matched TKI-resistant mouse tumor cell lines (Extended Data Fig. 3a,b and Supplementary Table 1).A significantly higher number of mutations, as well as mutations in an APOBEC context, were detected in TKI-resistant A3B-expressing EGFRmut tumor cell lines (EPA3B) compared with control TKI-resistant EGFRmut tumor cell lines (EP), and compared with both control (EP) and A3B-expressing TN EGFRmut tumor cell lines (Extended Data Fig. 3a,b).Two unique de novo putative loss-of-function mutations in the protein tyrosine phosphatase receptor type S (Ptprs) gene were identified in an APOBEC context (Extended Data Fig. 3c).Loss of PTPRS function through mutation or deletion has been shown to increase TKI resistance in multiple human preclinical cancer models and has been linked with worse overall survival and more rapid disease progression in patients with EGFR-driven lung cancer [29][30][31] .The equivalent of the A3B-driven mutation in humans (Ptprs_mut1, D138N; Extended Data Fig. 3c) was identified in tumors of patients with lung, colorectal and bladder cancer from The Cancer Genome Atlas (TCGA) and in one EGFR L858R TRACERx patient with NSCLC (Extended Data Fig. 3d).
To validate our findings from mouse models, long-term cell viability with targeted therapy was assessed in established human cell line models of oncogenic EGFRmut and echinoderm microtubule-associated protein-like 4-anaplastic lymphoma kinase (EML4-ALK) lung adenocarcinoma with CRISPR-mediated A3B depletion.Under EGFR TKI treatment (osimertinib), A3B-depleted PC9 and HCC827 lines (harboring EGFR exon19del ; Extended Data Fig. 4a-d) showed significantly reduced cell viability compared to A3B-competent control lines (Fig. 3l,m).Similarly, a significant reduction in cell viability was observed in an A3B-knockout (KO) EML4-ALK cancer cell line (H3122; Extended Data Fig. 4e,f) treated with the Food and Drug Administration-approved ALK TKI alectinib (Fig. 3n).KO of A3B had no effect on cell viability in untreated PC9, HCC827 or H3122 cell lines (Extended Data Fig. 4g-i).These data suggest that A3B expression confers enhanced cell survival with targeted therapy.
Targeted therapy induces A3B expression and UNG downregulation.Our mouse lung cancer models demonstrated that A3B expression is associated with targeted therapy resistance.We hypothesized that (Fig. 1d).The programmed cell death marker caspase-3 was significantly higher in tumor cells of EA3B mice compared with E mice (Fig. 1e,f).
We hypothesized that A3B expression at tumor initiation in EA3B mouse models might induce increased chromosomal instability (CIN), p53 pathway activation and tumor cell death based on previous work 4 .In our models, a significantly higher fraction of lagging chromosomes and chromatin bridges were observed in anaphase tumor cells of EA3B mice compared with E mice 4 (Fig. 1g).There was also a significant increase in p53 nuclear positivity in tumors of EA3B mice compared with E mice that was not present at later stages (Extended Data Fig. 1b-d).No difference was observed in proliferation (Ki67) or DNA damage (γH2AX; Extended Data Fig. 1e,f).To assess if APOBEC activity contributes to increased tumor cell death at initiation, an EGFR L858R mouse model combined with a catalytically inactive form of A3B (E(CAG) A3B E255A ) 22,23 was generated (Fig. 1h).The decrease in EGFR L858R+ cells at 3 months postinduction observed with wildtype (WT) A3B was no longer observed in the enzyme inactive A3B mouse model (E(CAG) A3B E255A ) compared with E control mice (Fig. 1h-j), suggesting that the increase in tumor cell death with A3B expression is at least in part due to the enzymatic activity of A3B.
We hypothesized that A3B expression could drive increased tumor cell death through enhanced immune surveillance in response to increased A3B activity 24 .A significant increase in both CD4 and CD8 T cells in EA3B mice was observed at 3 months postinduction (Extended Data Fig. 1g-i).Transplantation of an EPA3B mouse tumor cell line into WT C57BL/6J or EPA3B C57BL/6J transgenic mice resulted in the growth of EGFR L858R+ A3B + tumors in EPA3B C57BL6/J transgenic mice but not WT C57BL/6J mice (Extended Data Fig. 1j-m), suggesting a level of immune tolerance to both the EGFR L858R and A3B transgenes.
Tumors were induced in an EGFR L858R p53-deficient mouse model either with or without A3B (EP and EPA3B; Fig. 1k and Extended Data Fig. 1j).No difference in the number of tumors at 3 months postinduction (Fig. 1l) or in overall survival (Fig. 1m) was observed in EP versus EPA3B mice, suggesting that A3B expression is tolerated in a p53-deficient model of EGFR-driven lung cancer.Thus, p53 in this model limits the tolerance of cancer cells to A3B expression at tumor initiation.
Next, CIN was assessed in systemic treatment-naïve (TN) patients with lung adenocarcinoma from the TRACERx421 (Tx421) cohort, confirming and expanding on previous findings from Tx100 (ref.4).Tracking NSCLC evolution through therapy (TRACERx) is a prospective multicenter cancer study designed to delineate tumor evolution from diagnosis and surgical resection to either cure or disease recurrence.Tx100 was the analysis of the first 100 patients enrolled 9 , while Tx421 was the analysis of the first 421 patients enrolled 25 .We considered the following three orthogonal approaches to estimate the extent of CIN in tumors: chromosome missegregation errors captured during anaphase; the amount of somatic copy-number alteration (SCNA) intratumor heterogeneity (ITH) between tumor regions (SCNA ITH) 25 and expression-based 70-gene CIN signature (CIN70) 4,26 .We observed a significant correlation between all three measures of CIN and A3B expression in both a subset of EGFRmut patients with lung adenocarcinoma in the Tx421 dataset (Fig. 2a-d) and patients with lung adenocarcinoma in the Tx421 dataset (Fig. 2e-h).Focusing on the genomic data, we observed a significant correlation between SCNA ITH and mutations in an APOBEC context (TCN/TCW C>T/G; Fig. 2i).These data together suggest that the increased CIN observed with A3B expression in EGFRmut mouse models is reflected in human NSCLCs in the Tx421 dataset.
Subclonal A3B inhibits tumorigenesis.Analysis of TN patients in the Tx421 cohort revealed that APOBEC-mediated mutagenesis is enriched subclonally in EGFRmut disease (Fig. 2j,k) and the wider cohort 9 .Mice in which A3B expression could be temporally separated from EGFR L858R expression (EA3Bi), allowing for induction of A3B expression in a subset of tumor cells within the already proliferating EGFRmut tumor, were generated to mirror subclonal APOBEC induction and to assess if subclonal  targeted therapy may induce adaptations that increase the expression of A3 family members and decrease the expression of UNG in human models.Based on current literature 4,5,32,33 , mRNA expression levels of A3A, A3B, APOBEC3C (A3C) and APOBEC3F (A3F) were measured.In PC9 cells, a significant increase in all four members was observed with osimertinib, with A3A being the most significantly elevated (Fig. 4a).In HCC827 cells, A3A and A3B were the most significantly elevated, with both induced to similar levels with osimertinib (Fig. 4b).A significant increase in overall APOBEC activity (Fig. 4c,d) and A3B protein levels (Fig. 4e,f) were also observed.Each A3 gene was then silenced using small interfering RNAs (siRNAs) specific for each family member (Extended Data Fig. 5a-i), and APOBEC activity was assessed.
Only knockdown of A3B resulted in a significant decrease in APOBEC activity with TKI therapy in PC9 and HCC827 cell lines (Fig. 4g).These data suggest that while several A3 family members likely contribute to the increased APOBEC activity observed with TKI therapy, A3B appears to be a major contributor.Targeted therapy-induced transcriptional changes of A3B and UNG were assessed in established human lung cancer cell line data from publicly available datasets (Gene Expression Omnibus (GEO) database, GEO2R).Treatment of EGFRmut cell lines (HCC827, PC9 and HCC4006 harboring EGFR L747-E749del,A750P ) with the EGFR TKI erlotinib was associated with transcriptional upregulation of A3B both acutely (6-h to 1-d treatment) and at later timepoints (8-d treatment; Fig. 5a).These transcriptional changes were confirmed in an independent RNA-seq (RNA sequencing) dataset 34 with a significant upregulation of A3B and downregulation of UNG following osimertinib treatment (Fig. 5b), suggesting a conserved effect of EGFR pharmacologic inhibition independent of the generation (evolution of targeted therapy development leading to more specific and effective molecules) of EGFR inhibitor.
Transcriptional upregulation of A3B and downregulation of UNG were subsequently validated in multiple oncogenic EGFR-driven cellular models of lung adenocarcinoma at both the RNA (Fig. 5c,d) and protein levels (Fig. 5e).To rule out off-target pharmacological effects of EGFR TKIs, A3B expression was examined with siRNA-mediated silencing of EGFR and also led to A3B upregulation and UNG downregulation (Fig. 5f).Induction of A3B was also observed upon treatment with an inhibitor of mitogen-activated protein kinase kinase (MAP2K or MEK1 (selumetinib; Fig. 5a).The induction of A3B by different inhibitors of oncogenic receptor tyrosine kinases (RTKs) and their downstream signaling components, such as MEK1, indicates that upregulation of A3B is likely a consequence of oncogenic signaling inhibition, and not specific to EGFR TKIs.
Consistent with RNA and protein level changes, TKI treatment resulted in a significant increase in nuclear APOBEC activity 35 and decrease in nuclear uracil excision capacity of UNG in multiple EGFR-driven cell line models, including EGFR exon19del cells (PC9 and HCC827) and EGFR L858R+T790M cells (H1975; Fig. 4c,d and Extended Data Fig. 6a-e).Increased A3B expression and APOBEC activity as well as decreased UNG expression and uracil excision activity were also observed in EML4-ALK-driven cellular models (H3122 and H2228) during ALK TKI treatment (Extended Data Fig. 6f-i).
A3B was then stably knocked down using small hairpin RNA (shRNA) in PC9 cells, and rescue experiments with expression vectors containing either WT A3B tagged with human influenza hemagglutinin (HA) (A3B WT-HA tagged) or catalytically inactive A3B tagged with HA (A3B E225A-HA tagged) were performed.APOBEC activity with A3B knockdown was significantly reduced with TKI treatment versus A3B-proficient lines with TKI treatment (Extended Data Fig. 6j).Expression of the WT catalytically active, but not the mutant catalytically inactive A3B, rescued the decline in nuclear APOBEC activity caused by A3B depletion (Extended Data Fig. 6j-l).While knockdown of A3B induced no off-target reductions in any other A3 family members, significant increases in A3A, A3G and A3H expression were detected (Extended Data Fig. 6m), corroborating previous reports in human breast and lymphoma cancer cell lines showing increased A3A expression with A3B loss 36 .These data suggest that A3B is a substantial contributor to the increased APOBEC activity observed with TKI treatment.
To exclude an indirect effect of targeted therapy on cell cycle arrest that might alter APOBEC enzyme expression, EGFRmut NSCLC PC9 cells were treated with the CDK4/6 cell cycle inhibitor palbociclib 37 .Palbociclib treatment-induced G0/G1 cell cycle arrest with a comparable arrest measured with osimertinib (Extended Data Fig. 6n).UNG expression decreased upon palbociclib treatment; however, there was a significant decline in A3B expression (Extended Data Fig. 6o), contrasting with the increased expression observed upon TKI therapy and suggesting that TKI-mediated induction of A3B is unlikely to be a consequence of TKI treatment-induced cell cycle inhibition.
A3B and UNG expression levels were then examined in multiple human tumor xenograft models.An increase in A3B and a decrease in UNG protein levels were detected in EGFR TKI-treated tumor tissues from three distinct oncogenic EGFR-driven CDX models of human lung adenocarcinoma (Extended Data Fig. 7a-f).Additionally, RNA-seq analyses from an EGFR L858R -harboring patient-derived xenograft (PDX) model of lung adenocarcinoma 38 revealed a nonsignificant increase in A3B mRNA and a decrease in UNG mRNA levels upon treatment with erlotinib (Extended Data Fig. 7g), and significant increase in A3B and a nonsignificant decrease in UNG with osimertinib 34 (Extended Data Fig. 7h).These findings support a model whereby EGFR oncoprotein inhibition induces increased A3B expression and decreased UNG expression.
To investigate the clinical relevance of these findings, we examined single-cell RNA-seq data in an established dataset obtained from clinical specimens of NSCLC procured from patients at the following three timepoints: (1) treatment naïve before initiation of systemic targeted therapy (classified as TN), (2) while on targeted therapy when the tumor was regressing or at stable state as evaluated by standard clinical imaging (classified as residual disease (RD)) and (3) at clear progressive disease (PD, acquired resistance) as determined by standard clinical imaging (classified as PD).The classification of response was based on Response Evaluation Criteria in Solid Tumors (RECIST) criteria 41 .In total, 66 samples obtained from 30 patients with lung cancer pre-TKI or post-TKI therapy (erlotinib (EGFR), osimertinib (EGFR) and crizotinib (ALK) being the most frequent targeted therapies) were analyzed (Supplementary Table 2a).We observed that mRNA expression of A3B and NF-κB components RELA and RELB, as well as an NF-κB gene signature 42 , were significantly increased in tumors exposed to EGFR TKI treatment, in particular at tumor progression with therapy (Extended Data Fig. 8h-k).

UNG downregulation is associated with c-JUN suppression during TKI treatment
We next investigated the mechanism of UNG downregulation during targeted therapy.UNG gene promoter analysis (using PROMO) 43 revealed the presence of predicted JUN consensus binding sites.RNA-seq data from EGFR TKI-treated PC9 cells indicated that like UNG, c-JUN was also transcriptionally downregulated upon treatment, which was validated using RT-qPCR (Extended Data Fig. 8l).This aligns with the expected downregulation of c-JUN upon inhibition of the mitogen-activated protein kinase (MAPK) pathway during EGFR inhibition by TKI treatment 44 .We hypothesized that TKI treatment-induced UNG downregulation could be caused by c-JUN downregulation.Silencing of c-JUN by siRNA was sufficient to suppress UNG expression, suggesting that UNG downregulation could be a consequence, in part, of the c-JUN suppression that occurs during TKI-mediated MAPK signaling suppression (Extended Data Fig. 8m).

A3B is required for APOBEC mutation signature accumulation during targeted therapy
To examine the role of A3B expression on mutagenesis during targeted therapy, A3B-deficient and A3B-proficient single-cell cloned PC9 cells (Extended Data Fig. 4a,b) were treated with osimertinib using a dose-escalation protocol to resistance (3 months; Fig. 6a).
To further explore this hypothesis, we analyzed sequencing data for potential TKI resistance mutations in A3B-proficient PC9 TKI-resistant clones and found an acquired early stop codon mutation in the tumor suppressor gene NRXN3 (Q54*) 46,47 in an APOBEC-preferred context (T(C>T)A).The potential impact of this loss-of-function mutation was validated by depleting NRXN3 (given the early stop codon mutation detected, which is likely a loss-of-function event) in a naïve PC9 lung cancer cell line, which increased levels of phosphorylated AKT, a previously identified convergent feature of EGFR TKI resistance 48 , and conferred resistance to EGFR TKI treatment (Extended Data Fig. 9b-d).
A3B expression and APOBEC-associated mutations are elevated with targeted therapy in NSCLC.To verify the clinical relevance of our findings, A3B expression was examined in several NSCLC clinical datasets (Supplementary Table 2b) 41,[49][50][51][52] .Bulk RNA-seq of 32 pre-TKI and 42 post-TKI treated (osimertinib/erlotinib/crizotinib/alectinib) clinical tumor samples revealed a significant increase of A3B expression post-TKI relative to pre-TKI samples (P = 0.011; Fig. 7a).A3B was the only A3 family member with significantly increased expression post-TKI treatment (Extended Data Fig. 10a).Stratification at TN, RD and PD timepoints revealed a significant expression increase from TN to RD (P = 0.02) and an increase approaching significance from TN to PD (P = 0.057; Extended Data Fig. 10b).Further validating these observations, single-cell RNA-seq data revealed that A3B expression, specifically in tumor cells isolated from clinical specimens, was significantly increased from TN to PD (P < 0.001) and from RD to PD (P < 0.001; Fig. 7b).Compared to the other A3 genes, A3B expression had the second highest effect scores of all A3 family members as calculated using Cohen's d method (TN to PD, d = 1.048;RD to PD, d = 0.953; Extended Data Fig. 10c).A3C expression exhibited the highest effect scores; however, APOBEC activity assays revealed A3C did not contribute to overall activity with TKI treatment (Fig. 4g).Immunohistochemical (IHC) analyses, as performed previously 4 , on clinical samples also revealed a significant increase in A3B nuclear protein levels in EGFR TKI-treated tumor samples both at RD and PD timepoints (Fig. 7c,d and Supplementary Table 2c).Demonstrating the clinical effect of TKI treatment on the proportion of mutational signatures, a recently published dataset shows that APOBEC-associated mutation signatures (SBS2 and SBS13) were dominant, defined as the mutational signature with the highest fraction of mutations, in a significantly higher number of osimertinib-resistant samples when compared with naïve samples 53 .To independently test this observation with our own data, WES was performed on paired pre-and post-TKI treated samples obtained from 32 patients (Supplementary Table 4) to quantify mutations acquired following TKI treatment in NSCLC EGFRmut (treated with erlotinib/osimertinib) and ALK fusion (treated with alectinib) clinical samples.This analysis revealed that both the overall mutation burden (SNV count; Fig. 7e) and number of APOBEC-associated mutations (C>T or C>G mutations in a TCN context; Fig. 7f) increased post-treatment.
Next, mutations in an APOBEC-preferred context were identified in genes previously associated with TKI resistance in tumors from patients who had progressed on or shown incomplete response to EGFR inhibitor therapy (Fig. 7g and Supplementary Table 5).These mutations include activating mutations in PIK3CA (E545K) 54 , WNT signaling-activating mutations in β-catenin at a glycogen synthase kinase-3β (GSK-3β) phosphorylation site 55 , MAPK pathway reactivating-mutations through inactivation of PP2A, a negative regulator of MAPK signaling 56,57 , an activating mutation in MET tyrosine kinase domain (H1095Y) 53,58 , as well as an ALK inhibitor desensitizing mutation in ALK (E1210K) 59 in the tumors of some patients who had progressed on or shown incomplete response to EGFR or ALK inhibitor therapy.AKT, WNT and MAPK pathway activation have previously been shown to cause EGFR and ALK inhibitor resistance [60][61][62][63][64][65] .All but one of these APOBEC-associated putative resistance mutations were detected selectively post-treatment, suggesting not only that these mutations are induced by APOBEC (itself engaged) during targeted therapy but also that these variants could promote resistance.All samples containing these APOBEC-associated mutations, except for one, did not harbor a detectable EGFR T790M mutation, which has been reported to be present in ~50-60% of first-and second-generation EGFR TKI-resistant cases 66,67 and arising from a non-APOBEC clock-like mutation signature (SBS1 (ref.68); Fig. 7g).Altogether, of the resistance mutations in this cohort, 53% (8/15) of mutations were associated with clock-like mutation signature SBS1 and 46% (7/15) of mutations with the APOBEC signatures SBS2 + SBS13, with no other mutational signatures contributing to putative resistance mutations.In total, 8/32 tumors have APOBEC-associated putative resistance mutations.The observation that APOBEC-mediated mutations in resistance-associated genes detected in post-treatment samples and the EGFR T790M mutation appear to be mutually exclusive suggests that these APOBEC-mediated mutations could be the potential mechanism of resistance to targeted therapy in these patients.These data suggest that APOBEC signatures are a complementary route to acquired TKI therapy resistance, contributing to the diverse mechanisms of resistance that exist [69][70][71] .
Taken together, these data illustrate the diverse effects of A3B at different stages of tumor evolution with or without the selective pressure of therapy.The findings demonstrate multiple roles of A3B, as an inhibitor of tumor progression at initiation, an inducer of APOBEC mutations and a contributor to targeted therapy resistance (Fig. 8).

Discussion
Our collective findings shed light on the important, context-specific roles of A3B on lung cancer pathogenesis and tumor evolution.Along with other recent findings in the field 5 , our data reinforce the concept that targeted therapies can induce adaptive changes that promote resistance 72 , including those that are APOBEC-mediated and that may involve multiple APOBEC family members.This A3 induction during therapy might contribute to the development of treatment resistance and appears to be clinically relevant based on our clinical datasets obtained from targeted therapy-treated patients.Additional clinical cohort analyses will be important to conduct as further human tumors obtained from patients on targeted therapy become available.
We demonstrate that the expression of A3 family members might contribute to resistance in preclinical human and mouse models of lung adenocarcinoma.Although we focus on oncogenic EGFR-driven lung adenocarcinomas, our findings appear to extend to other molecular subsets such as EML4-ALK-driven lung cancer (Fig. 3l and Extended Data Fig. 6b-d) and likely reflect a more general principle of targeted therapy-induced adaptability.While APOBEC has been implicated in drug resistance previously 33,73 , our study reveals a distinct mechanism by which targeted cancer therapy is actively responsible for the upregulation of APOBEC via NF-κB-mediated transcriptional induction in response to therapy.Our study further explains the enhanced efficacy of cotreatment with an NF-κB inhibitor compared to EGFR inhibition alone at preventing the emergence of resistance 38 .There are however caveats to our findings (further discussion in Supplementary Note).The mouse models, although helpful for a deeper understanding of the biological effects of enforced A3B expression, are imperfect as A3B is expressed from a transgene promoter system.APOBEC3 enzyme expression has also been shown to occur episodically 32 , which differs from the constitutive expression of our mouse models.Future studies that reveal the upstream regulators of endogenous mouse APOBEC enzymes could help in the development of better models in future studies.
Our work expands upon prior studies suggesting a potential association between APOBEC-mediated mutagenesis and acquisition of putative resistance mutations in the APOBEC-preferred context during the treatment of EGFR-driven lung cancers 74,75 .Our data suggest that inhibition of APOBEC3 family members could suppress the emergence of one pathway to resistance and thereby improve response to targeted therapy, consistent with the work of others in the field that suggests that multiple APOBEC3 family members including A3B contribute to targeted therapy resistance 5,32 , with both A3A and A3B shown to be contributors of mutagenesis 6,32,36,76 .The role of A3B in promoting resistance to TKI is likely multifaceted, and our data do not discount the contribution of other possible parallel cytosine deaminase-independent mechanisms, such as induced CIN 4,77 , regulation of cell cycle 22 and regulation of the DNA damage repair pathway 78,79 .Our evidence here and these emerging collective findings 5,33,80,81 suggest that endogenous drivers of mutagenesis have diverse roles that are both detrimental and beneficial to tumor evolution depending on the context of tumor pathogenesis and treatment.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41588-023-01592-8.

Cell line and growth assays
Cell lines were grown in Roswell Park Memorial Institute-1640 medium (RPMI-1640) with 1% penicillin-streptomycin (10,000 U ml −1 ) and 10% FBS or in Iscove's modified Dulbecco's medium (IMDM) with 1% penicillin-streptomycin (10,000 U ml −1 ), l-glutamine (200 mM) and 10% FBS in a humidified incubator with 5% CO 2 maintained at 37 °C.Drugs used for treatment except PBS-1086 (ref.38) were purchased from Selleck Chemicals or MedKoo Biosciences.For growth assays, cells were exposed to DMSO or the indicated drugs for indicated durations in six-well or 96-well plates and assayed using crystal violet staining or Celltiter-Glo luminescent viability assay (Promega) according to the manufacturer's instructions.

Deriving clonal populations and generating APOBEC3B KO cells
Clonal cells were derived by sorting single cells into 96-well plates and expanding them over a few weeks.We then derived pools of one of the clones expressing either a green fluorescent protein (GFP)-targeting or A3B-targeting guide along with CRISPR/Cas9 by lentiviral transduction as done in a previously published study 82 .A3B gRNA target sequences, designed by the Zhang Lab 83 , were subcloned into the lentiCRISPR v2 plasmid (Addgene, 52961; a gift from F. Zhang) 83 and the one that showed better A3B depletion was selected for further analysis.

Transductions and transfections
Hek293T cells were cotransfected with lentiviral packaging plasmids pCMVdr8 and pMD2.G plasmid, along with the plasmid of interest using FuGENE 6 Transfection Reagent (Promega).APOBEC3B shRNA was purchased from Sigma (TRCN0000142875).Cells were transduced with 1:1 diluted lentivirus for 1-2 d and selected with antibiotic marker (puromycin).siRNAs were purchased from GE Healthcare Dharmacon and transfected using Lipofectamine RNAi Max according to the manufacturer's protocol, and the cells were collected within 48 h of transfection for subsequent assays.

Enzymatic assays
APOBEC assays were performed by incubating nuclear extracts from rapid efficient and practical (REAP) method 58 or whole-cell extracts with the following DNA oligo substrates (Integrated DNA Technologies, IDT): 5′-ATT ATT ATT ATT CAA ATG GAT TTA TTT ATT TAT TTA TTT ATT T-FAM-3′ using established protocols 28,35 .Upon completion of the reactions, they were heated at 95 °C for 5 min after the addition of TBE-urea buffer (Novex) and immediately run on a 15% TBE-urea gel (Bio-Rad) and imaged using Cy2 filter on ImageQuant LAS 4000.

Subcutaneous tumor xenografts and PDX studies
All animal experiments were conducted under University of California, San Francisco (UCSF) Institutional Animal Care & Use Committee (IACUC)-approved animal protocols.PC9 and H1975 tumor xenografts were generated by injection of 1 million cells in a 1:1 mixture of matrigel and PBS into 6-to 8-week-old female non-obese diabetic/severe combined immunodeficiency disease (NOD/SCID) mice.Once the tumors grew to ∼100 mm 3 , the mice were treated with vehicle or 5 mg kg −1 osimertinib once daily by oral gavage and the tumors were collected on day 4 for western blot analysis.PDX was generated as indicated in a previous study 38 .Tumors were passaged in SCID mice, treated with 25 mg kg −1 erlotinib once daily by oral gavage once they reached ~400 mm 3 and collected on day 2.

Mouse strains and tumor induction and treatment
The Cre-inducible Rosa26::LSL-APOBEC3Bi mice and Rosa26::CAG-LSL-APOBEC3Bi-E255A are described in refs.20,23.The TetO-EGFR L858R ; Rosa26 LNL-tTA (E) and CCSP-rtTA;TetO-EGFR L858R ;Rosa26 CreER(T2) mice have been described in refs.11,12,87,88.All mice were purified C57BL/6J mice, aged between 8 and 20 weeks, with a mixed sex ratio for each experiment (Supplementary Table 6).Tumors were initiated in E, EA3B, EP and EPA3B mice by intratracheal infection with adenoviral vectors expressing Cre recombinase as described 89 .Adenoviral-Cre (Ad-Cre-GFP) was from the University of Iowa Gene Transfer Core.Tumors were initiated in EA3Bi mice using chow containing doxycycline (625 ppm) obtained from Harlan-Teklad.All animal-regulated procedures were approved by the Francis Crick Institute BRF Strategic Oversight Committee that incorporates the Animal Welfare and Ethical Review Body and conformed with the UK Home Office guidelines and regulations under the Animals (Scientific Procedures) Act 1986 including Amendment Regulations 2012.To assess the recombination efficiency of the LSL allele upstream of APOBEC3B, PCR primers targeting the R26 site, the LSL cassette and the APOBEC3B transgene were used as described 20 .
Erlotinib was purchased from Selleckchem (erlotinib, Osi-744), dissolved in 0.3% methylcellulose and administered intraperitoneally at 25 mg kg −1 , 5 d a week.Tamoxifen was administered by oral gavage three times in 1 week at 2-4 d intervals (three injections total).Mice received tamoxifen at 150 mg kg −1 dissolved in sunflower oil.

Assessment of recombination efficiency
PCR was performed to assess the recombination of the LSL cassette upstream of the A3B allele in six tumors collected at progression.Five of six (5/6) of the tumors had a recombination efficiency above 90%, and one tumor of six was unrecombined.This rate of recombination aligns https://doi.org/10.1038/s41588-023-01592-8 with the rate of recombination observed by IHC staining at 3 months and at termination and suggests that a lack of recombination of the LSL cassette upstream of the A3B transgene explains A3B-negative tumors.

Micro-computed tomography (micro-CT) imaging
Mice were anesthetized with isoflurane/oxygen for no more than an hour each and minimally restrained during imaging (~8 to 10 min).Mice were then observed and, if necessary, placed in cages in a recovery chamber/rack until they regained consciousness and started to feed.Tumor burden was quantified by calculating the volume of visible tumors using AnalyzeDirect.

Evaluation of chromosome missegregation errors in hematoxylin and eosin (H&E)-and/or phospho-histone H3-stained samples
Lung sections were evaluated for anaphases with chromosome missegregation events using a ×100 objective light microscope.For E and EA3B mice at early and late timepoints, the percentage of missegregation errors was calculated and averaged across all mice using the harmonic mean.For EA3B mice, the percent error was normalized to an A3B recombination efficiency of 82% based on observed recombination efficiency observed (Extended Fig. 4).For E and EA3Bi mice with subclonal A3B expression, normalization for the recombination efficiency was not possible, so the percentage of missegregation errors was calculated based on the number of errors versus normal anaphases observed.

Mouse tumor processing
Frozen tumor tissue was cut into pieces and lysed in RLT Buffer with β-mercaptoethanol.TissueRuptor was used for disruption and homogenization of tissue.Lysate was added to a QIAshredder tube and centrifuged at full speed for 1 min.The homogenized solution was then added to AllPrep DNA spin columns (Qiagen AllPrep DNA/RNA Mini Kit, 80204).

Histopathological examination of mouse
Four micrometers thick, formalin-fixed, paraffin-embedded (FFPE) sections from lung lobes were stained with H&E and examined by two board-certified Veterinary Pathologists (A.S.B. and S.L.P.).Histopathological assessment was performed blind to experimental grouping using a light microscope (Olympus, BX43).Tissue sections were examined individually, and in case of discordance in diagnosis, a consensus was reached using a double-head microscope.Proliferative lesions were diagnosed as alveolar hyperplasia, bronchioloalveolar adenoma and well-differentiated, moderately or poorly differentiated bronchioloalveolar adenocarcinoma.Sections were histopathologically assessed and graded for the presence and type of proliferative epithelial lung lesions using the International Harmonization of Nomenclature and Diagnostic Criteria for Lesions (INHAND) guide for nonproliferative and proliferative lesions of the respiratory tract of the mouse 91 .

WES-mouse data
WES was performed by the Advanced Sequencing Facility at the Francis Crick Institute using the Human Core Exome Kit (Twist BioScience) for library preparation and SureSelectXT Mouse All Exon, 16, Kit (Agilent) for library preparation, respectively.Sequencing was performed on HiSeq 4000 platforms.

RNA-seq-mouse data
RNA-seq was performed by the Advanced Sequencing Facility at the Francis Crick Institute using the KAPA mRNA HyperPrep Kit (KK8581-96 Libraries) and KAPA Dual-Indexed Adapters (Roche, KK8720).Sequencing was performed on HiSeq 4000 platforms.The processed FASTQ files were mapped to mm10 reference genome using the STAR (version 2.4) algorithm, and transcript expressions were quantified using the RSEM (version 1.2.29) algorithm with the default parameters.The read counts were used for downstream analysis.
To minimize false positives, additional filtering was performed.For single-nucleotide variants (SNVs) or dinucleotides detected by VarScan2, a minimum tumor sequencing depth of 30, VAF of 5%, variant read count of 5 and a somatic P value < 0.01 were required to pass a variant.For variants detected by VarScan2 between 2% and 5% VAF, the mutation also needs to be detected by MuTect.
As for insertions/deletions (INDELs), variants need to be passed by both Scalpel (PASS) and VarScan2 (somatic P < 0.001).A minimum depth of 50×, 10 alt reads and VAF of 2% were required.
For all SNVs, INDELs and dinucleotides, any variant also detected in the paired germline sample with more than five alternative reads or a VAF greater than 1% was filtered out.
The detected variants were annotated using Annovar 95 .

Generation of EGFR L858R mutant mouse tumor cell lines
A portion of mouse lung tumor was dissected (1/3 to 1/2 of the original tumor depending on size) and cut into small pieces with scissors.Pieces were then digested for 30 min at 37 °C while rotating at full speed in digestion media (1,400 µl HBSS-free w/o Ca 2+ , 200 µl Collagenase IV and 40 U ml −1 DNase).Tumor cells were pelleted down in a centrifuge (1,100 r.p.m. for 4 min) and resuspended in IMDM supplemented with 1% penicillin-streptomycin solution (10,000 U ml −1 ), l-glutamine (200 mM) and 10% FBS.This cell suspension was then plated in a 10-cm plate and passaged over a period of 1-3 months until consistent growth was observed.

Generation of TKI-resistant mouse or human tumor cell lines
TKI naïve cell lines were cultured in increasing levels of erlotinib or osimertinib using a dose-escalation protocol from 100 nM to 1 µM when cells were growing with minimal cell death.

Mutational and SCNA ITH calculations for TRACERx data
SCNA ITH was calculated by dividing the percentage of the genome harboring heterogeneous SCNA events, that is, those events that were not present in every region, by the percentage of the genome involved in any SCNA event in each tumor 25 .

Cell line whole-genome mutational signature analysis
Sequences were aligned to the human genome (hg38) using the Burrows-Wheeler Aligner (version 0.7.17).PCR duplicates were removed using Picard (version 2.18.16).Reads were locally realigned around indels using GATK3 (version 3.6.0)tools RealignerTargetCreator to create intervals, followed by IndelRealigner on the aligned BAM files.
MuTect2 from GATK3 (version 3.6.0)was used in tumor/normal mode to call mutations in test versus control cell lines.SNVs that passed the internal GATK3 filter with read depths over 30 reads at called positions, at least 4 reads in the alternate mutation call and an allele frequency greater than 0.05 were used for downstream analysis.Mutational profile plots in Fig. 6g were plotted using the deconstructSigs R package 102 .

DNA and RNA isolation from cell line models for sequencing
DNA or RNA were extracted from frozen cell pellets using Qiagen's DNeasy Blood and Tissue Kit or Qiagen's RNeasy MINI Kit, respectively, as per the manufacturer's instructions.The isolated DNA or RNA was quantified and qualitatively assessed using a Qubit Fluorometer (Thermo Fisher Scientific) and a Bioanalyzer (Agilent), as per the manufacturer's instructions.The DNA or RNA were then sent to BGI for WGS (30×) or Novogene for mRNA or WES.

Human participants
All patients gave informed written consent for the collection of clinical correlates, tissue collection and research testing under institutional review board (IRB)-approved protocols (CC13-6512 and CC17-658, NCT03433469).Patient demographics are listed in Supplementary Tables 2a-c, 4, and 5a,b.Patient studies were conducted according to the Declaration of Helsinki, the Belmont Report and the U.S. Common Rule.

Studies with specimens from patients with lung cancer
Frozen or FFPE tissues from patients with lung cancer for DNA or RNA sequencing (bulk and single cell) studies were processed and sequenced as described previously 41,60 .Classification of response was based on RECIST criteria.Some of these biopsies were subjected to WES at the QB3-Berkley Genomics for which library preparation was performed using IDT's xGen exome panel.For additional specimens, tumor DNA from FFPE tissues and matched nontumor from blood aliquots or stored buffy coats were collected as part of the UCSF biospecimen resource program (BIOS) in accordance with UCSF's IRB-approved protocol.DNA from blood aliquots was isolated at the BIOS.Other nontumor samples and FFPE tumor tissues were sent for extraction and assessment of quality and quantity to Novogene, and those meeting the required sample standards were subjected to WES at Novogene's sequencing facility.

Mutation analysis
Paired-end reads were aligned to the hg19 human genome using the Picard pipeline (https://gatk.broadinstitute.org/).A modified version of the Broad Institute Getz Lab CGA WES Characterization pipeline (https://docs-google-com.ezp-prod1.hul.harvard.edu/document/d/1VO2kX_fgfUd0x3mBS9NjLUWGZu794WbTepBel3cBg08) was used to call, filter and annotate somatic mutations.Specifically, SNVs and other substitutions were called with MuTect (v1.1.6) 93.Mutations were annotated using Oncotator 103 .MuTect mutation calls were filtered for 8-OxoG artifacts, and artifacts were introduced through the formalin fixation process (FFPE) of tumor tissues 66 .Indels were called with Strelka (v1.0.11).MuTect calls and Strelka calls were further filtered through a panel of normal samples (PoN) to remove artifacts generated by rare error modes and miscalled germline alterations 93 .To pass quality control, samples were required to have <5% cross-sample contamination as assessed with ContEst 93 ; mean target coverage of at least 25× in the tumor sample and 20× in the corresponding normal as assessed using GATK3.7 DepthOfCoverage and a percentage of tumor-in-normal of <30% as determined by deTiN 104 .This pipeline was modified for analysis of cell lines rather than tumor-normal pairs as follows: indels were called through MuTect2 alone rather than Strelka; deTiN was not performed and a common variant filter was applied to exclude variants present in the Exome Aggregation Consortium if at least ten alleles containing the variant were present across any subpopulation, unless they appeared in a list of known somatic sites 105,106 .

Mutational signature analysis
Active mutational processes 107 were determined using the deconstruct-Sigs R package 63 , with a signature contribution cutoff of 6%.This cutoff was chosen because it was the minimum contribution value required to obtain a false-positive rate of 0.1% and a false-negative rate of 1.4% per the authors' in silico analysis and is the recommended cutoff 102 .Samples with <10 mutations were excluded from analysis due to poor signature discrimination with only a few mutations, and a sample with less than 15 d of exposure to TKI therapy was excluded because it is too short a time to accumulate detectable mutations due to therapy.
For TRACERx data analysis, data processing was performed in the R statistical environment version ≥3.3.1.

RNA-seq analyses
PDX tissue and mouse tumor cell line RNA extractions were carried out using an RNeasy Micro Kit (Qiagen).RNA-seq was performed on PDX tissue using replicate samples on the Illumina HiSeq 4000, paired-end https://doi.org/10.1038/s41588-023-01592-8 100-bp reads at the Center for Advanced Technology (UCSF).For the differential gene expression analysis, DESeq program was used to compare controls to erlotinib samples as previously described 108 .
RNA-seq samples from patients and cell lines were sequenced by Novogene (https://en.novogene.com/)with paired-end sequencing (150 bp in length).There were ~20 million reads for each sample.The processed FASTQ files were mapped to the hg19 reference genome using the STAR (version 2.4) algorithm, and transcript expressions were quantified using the RSEM (version 1.2.29) algorithm.The default parameters in the algorithms were used.The normalized transcript reads (TPM) were used for downstream analysis.Gene set enrichment analysis was performed using GSEA software 109 .
For single-cell RNA-seq analyses, the data from a previously published study (all cancer cells from patients with advanced lung cancer) were used and analyzed in a similar manner 41 .All cells used are identified as malignant by marker expression and CNV inference and originated in from various biopsy sites (adrenal, liver, lymph node, lung and pleura/pleural fluid).Nonparametric, pairwise comparisons (Wilcoxon rank-sum test) were used to determine the statistical significance of the pairwise comparisons of different timepoints for their average scaled expression.

Statistical analysis
One-way or two-way ANOVA test with Holm-Sidak correction for multiple comparisons (>2 groups) or two-tailed t test (2 groups) were used to determine the statistical significance of the differences between groups for RT-qPCR, growth and enzymatic assays and bulk RNA-seq analysis.Normality of IHC and micro-CT data was determined using multiple testing methods (Anderson-Darling test, D'Agostino-Pearson test, Shapiro-Wilk test and Kolmogorov-Smirnov test).A two-sided t test or two-sided Mann-Whitney test was used for IHC and micro-CT data depending on the normality tests to determine the statistical significance of the differences between groups.Analysis for these assays was done using GraphPad Prism.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
The WES data and RNA-seq data (from the TRACERx study) used during this study have been deposited at the European Genome-phenome Archive (EGA), which is hosted by the European Bioinformatics Institute and the Center for Genomic Regulation under the accession codes EGAS00001006494 and EGAS00001006517, respectively, is under controlled access due to its nature and commercial licenses.Specifically, data are available through the Cancer Research UK and University College London Cancer Trials Center (ctc.tracerx@ucl.ac.uk) for academic noncommercial research purposes only and are subject to review of a project proposal by the TRACERx data access committee, entering into an appropriate data access agreement and subject to any applicable ethical approvals.A response to the request for access is typically provided within ten working days after the committee has received the relevant project proposal and all other required information.The WES data of tumor-derived cell lines shown in Extended Data Fig. 3 are available at the European Nucleotide Archive (ENA) with the identifier PRJEB67640 (ERP152649).The WGS data of PC9 cell lines shown in Fig. 6 are available at the ENA with the identifier PRJEB67559 (ERP152586).For the single-cell RNA-seq analyses shown in Extended Data Fig. 10b,c, the data from a previously published study (all advanced lung cancer cell data) were used and analyzed in a similar manner 41 .These data are available in the National Center for Biotechnology Information (NCBI) BioProject ID PRJNA591860.The RNA-seq data for Extended Data Fig. 10a were from a previously published study 38 .These data are available at NCBI GEO under accession GSE65420.Clinical Extended Data Fig. 6 | TKI treatment induces increased A3B and decreased UNG expression and activity in pre-clinical models of lung adenocarcinoma.a, Uracil excision capacity assay (UEC) using PC9 nuclear extracts treated with DMSO or 2 µM osimertinib (Osi) (n = 3 biological replicates, mean ± SD, two-tailed t-test, *P = 0.0275).b, UEC in HCC827 treated with DMSO or 0.4 µM osi (n = 3 biological replicates, mean ± SD, two-tailed t-test, ****P < 0.0001).c, Western blot (WB) from H1975 treated with DMSO, 0.1 µM or 0.5 µM crizotinib (CYTO: cytoplasmic; NUC: nuclear; H3: Histone H3; TUBB: beta-tubulin) (n = 3 biological replicates).d, APOBEC activity assay (AAA) using H1975 treated with DMSO or 1 µM osi (n = 3 biological replicates, mean ± SD, two-tailed t-test, **P = 0.0084).e, UEC in H1975 treated with DMSO or 1 uM osi (n = 3 biological replicates, mean ± SD, two-tailed t-test, **P = 0.0054).f, WB from H3122 treated with DMSO or 1 µM crizotinib (n = 3 biological replicates).g, AAA from H3122 treated with DMSO or 0.5 µM crizotinib (n = 3 biological replicates, mean ± SD, two-tailed t-test, *P = 0.0204).h, UEC in H3122 treated with DMSO or 0.5 µM crizotinib (n = 3 biological replicates, mean ± SD, two-tailed t-test, *P = 0.0123).

Fig. 7 |
Fig. 7 | APOBEC3B expression and APOBEC-associated mutations are elevated with targeted therapy in patients with NSCLC.a, APOBEC3B (A3B) expression levels (batch-corrected transcripts per million (TPM)) measured using RNA-seq analysis in human NSCLC specimens driven by EGFR-and ALK-driver mutations obtained before TKI treatment (pre-TKI, n = 32 samples) or post-treatment (post-TKI, n = 42 samples; all data points shown, two-sided t test, *P = 0.02).b, APOBEC family member expression measured using single-cell RNA-seq obtained from human NSCLC before TKI treatment (TN), on-treatment at RD or at PD (all data points shown, n = 762, 553 and 988 cells per group, respectively, two-sided Wilcoxon test with a Holm correction, ****P < 2.22 × 10 −16 ).c,d, Representative images of IHC analysis of A3B protein levels in patients with NSCLC at TN, RD and PD stages.Red arrows indicate positive stained cells (scale bar: 30 µM, c) with IHC quantification of human NSCLC samples pre-TKI (n = 16 samples) or post-TKI single agent (n = 15 samples; all data points shown, two-sided unpaired t test, *P = 0.0113, d).e, Total mutation burden (SNV count) in paired human NSCLC samples pre-TKI or post-TKI (n = 32, two-tailed Wilcoxon matched-pairs signed-rank test, **P = 0.0013).f, APOBEC-associated mutation count in paired human NSCLC samples pre-TKI or post-TKI (n = 32, two-tailed Wilcoxon matchedpairs signed-rank test, *P = 0.0155).g, Mutation signature associated with each putative de novo TKI resistance mutation detected in clinical samples analyzed post-TKI at PD.An asterisk denotes a sample from a patient who has received prior chemotherapy.Boxplots: middle line represents median; lower and upper hinges represent the first and third quartiles; lower and upper whiskers represent smallest and largest values within 1.5× interquartile range from hinges.

Fig. 8 |
Fig. 8 | APOBEC3B in EGFR-driven lung tumor evolution.At tumor initiation, continuous APOBEC3B expression and activity induces CIN and p53 pathway activation, resulting in cell death.With targeted therapy, NF-κB induction leads to increased A3B expression, fueling TKI resistance.Figure created in BioRender.com.

12 Extended Data Fig. 10 |
Expression of APOBEC3 enzymes in clinical samples upon targeted therapy treatment.a, Comparison of APOBEC3B (A3B) expression levels (Exp: batch corrected TPM) measured using RNA-Seq analysis in human NSCLC specimens driven by EGFR and ALK driver mutations obtained before treatment (Pre-TKI, 32 samples), or post-treatment (Post-TKI, 42 samples) (all data points shown, two-sided t-test, *P = 0.011).b, Comparison of APOBEC3 (A3) family member expression levels (Exp: batch corrected Log (TPM + 1) measured using RNA-seq analysis in human NSCLC specimens obtained at treatment naïve (TN), residual disease (RD) or progressive disease (PD) with TKI (all data points shown, 762, 553, and 988 cells per group respectively, two-sided Wilcoxon test with Holm correction, *P = 0.02).c, Boxplot of normalized A3 family member expression measured using scRNA-seq obtained from the same samples as b (all data points shown, 762, 553, and 988 cells per group respectively, two-sided Wilcoxon test with Holm correction, *P < 0.05, **P < 0.01, ****P < 0.001, d=effect size calculated using a Cohen test).Boxplots: middle line=median, lower and upper hinges=first and third quartiles, lower and upper whiskers=smallest and largest values within 1.5×inter-quartile range from hinges.

Fig. 3 | APOBEC3B drives targeted therapy resistance in mouse and human preclinical models. a, TetO
://doi.org/10.1038/s41588-023-01592-878.Nikkila, J. et al.Elevated APOBEC3B expression drives a kataegiclike mutation signature and replication stress-related therapeutic vulnerabilities in p53-defective cells.Br.J. Cancer 117, 113-123 (2017).79.Buisson, R., Lawrence, M. S., Benes, C. H. & Zou, L. APOBEC3A and APOBEC3B activities render cancer cells susceptible to ATR inhibition.Cancer Res.77, 4567-4578 (2017).80. Russo, M. et al.Adaptive mutability of colorectal cancers in response to targeted therapies.Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. https https://doi.org/10.1038/s41588-023-01592-8