24/n We are excited about what comes next, both biologically and clinically. If you are interested in addressing fundamental biologic questions based on deep analysis of cancer genomes, please consider joining this journey with me. My lab is hiring at multiple levels. Thanks for reading!
23/n This story is also built on the scientific foundation and mentorship of Sohrab Shah at MSK Computational Oncology! Thorough discussions around data and unwavering support enabled our multi-year journey involving >86K HQ cancer cells across >100 samples.
22/n These are what we’ve learned about ecDNA and ICamps through the scope of scWGS. This story was only possible through close collaboration with Sohrab Salehi, who drove this ship with me. He’s about to launch his group. Stay tuned!
21/n Bulk cancer genomes are history books: they preserve early events well, but they cannot report current status or ongoing dynamics. Bulk WGS cannot distinguish historical vs. current ecDNAs. Long reads? I am not sure. Historical footprint doesn’t go anywhere.
20/n However, they now live within the chromosomes, so the scWGS shows peaks, unlike their double minute (DM) counterparts. Some cell lines that were claimed to be ecDNA+ by genome graph approach needs validation. NCI-H82 is one example, turns out HSR.
19/n There are “zebras”: convincing cyclic paths + abrupt junctional CN transition, yet staggering CN peaks in scWGS. They are cell lines with HSR. In the bulk level, cells retain the cyclic SV footprint that had generated ecDNA in the past.
18/n In ovarian/TNBCs, however, cyclic paths often correspond to sharp peaks shared by cells, indicating symmetric division. Their genome graphs often lack abrupt junctional CN transition at borders, consistent with stepwise amplification than ecDNA.
17/n Last, we created pseudobulk BAMs from scWGS and ran AmpliconArchitect, comparing bulk-based cyclic paths to single-cell CN distribution-based classes. In GBM/SCLC/EGFR+ LUAD, cyclic paths generally agree with heavy-tail CN distributions of ecDNA.
16/n By contrast, TERT and CDK4 are also ecDNA-amplified in ~50% of cells here, but each in only one form. Another GBM shows a similar history (ecEGFR x3 and ecMDM4 x1). This suggests that GBM repeatedly acquires EGFR ecDNA, late in their evolution.
15/n Another pattern notably in GBM is convergent evolution: recurrent, independent generation of ecDNA targeting EGFR. In this tumor, clonal EGFR amplification exists in three distinct ecDNA forms created by distinct SVs! Multiple “solutions” converging on the same driver.
14/n Applying ECADeMix across ecDNA+ cases, we asked how ecDNAs evolve. One pattern is branched/divergent evolution: ecDNA becomes smaller via complex deletions, often trimming gene deserts while sparing driver oncogenes, consistent with purifying selection.
13/n Realizing that loci on the same ecDNA species should co-vary across the cells, we developed a computational framework to deconvolve ecDNA heteroplasmy into species from scWGS data. It’s called ECADeMix, pioneered by my colleague Matt Myers.
12/n In GBM, the broad “cloud-like” correlation between MDM4 and EGFR reflects two separate ecDNA species, supported by single-cell SVs. In contrast, the strictly linear correlation btw EGFR (chr7) and CCND1 (chr11) indicates a chimeric ecDNA carrying both loci.
11/n Now ecDNA. Many ecDNA+ cases carry multiple oncogenes on ecDNA. How do these oncogenes segregate? scWGS lets us to read CN correlations across cells – Here are examples from a glioblastoma (left) and an EGFR-mutant lung cancer (right).
10/n Second: structural modulation. Ongoing BFB, chromothripsis, etc. can remodel the ICamp locus. See this case where subclonal chromothripsis plus aneuploidy jointly diversifies subclones with HLAMPs. ICamps are not inert. They can be major engines of heterogeneity.
9/n Strikingly, numeric modulation can also go “negative”. See this case where CCND1 amp-containing allele is subclonally eliminated by arm-level aneuploidy. In matched scRNA-seq, we detect the clone with minimal CCND1 expression, confirming transcriptional impact.
8/n Mechanistically, we see two modes: numeric and structural. Numeric modulation by aneuploidy and WGD. Once an oncogene is amplified intrachromosomally, the chromosome or arm could be doubled or tripled, dramatically increasing oncogene dosage.
7/n First we focus on ICamps and asked: are they static end-products or dynamic substrates for evolution? Surprisingly, >50% ICamps show multiple CN peaks corresponding to subclones, evidence that ICamps can actively evolve and drive heterogeneity. Then how?
6/n Here are all HLAMP regions in our cohort, classified by two key ecDNA metrics (ecDNA score and mass-in-window). One thing struck me: tissue-type specificity. ecDNA is prevalent in GBM, SCLC, EGFR+ lung cancer, while ovarian and TNBC cases are dominated by ICamps.
5/n In others, HLAMP CN varied dramatically across cells, producing heavy-tail distribution with extreme outliers, strong evidence for asymmetric cell division, expected for ecDNA. In experimental models, we validated ecDNA status by metaphase FISH.
4/n Two prominent patterns stood out. In some cases, cells share same high CN, forming discrete sharp peak(s), consistent with symmetric inheritance, expected for intrachromosomal amplification (ICamp).
3/n To do this, we applied genome amplification-free single-cell WGS (DLP+) to clinical tumors and experimental models across 4 major cancer types, reading CN distributions across cells for each high-level amplification (HLAMP) region in each sample.
2/n www.biorxiv.org/content/10.6... fuels intratumoral heterogeneity via asymmetric segregation at mitosis. We asked a simple question: can we directly measure cell-to-cell variability in copy-number (CN) using single-cell DNA sequencing, and use it to define and learn more about ecDNA?
1/n Have you ever wondered whether an amplified oncogene in a patient’s tumor (HER2/EGFR/CCND1/…) is carried on extrachromosomal DNA (ecDNA) vs embedded within a chromosome? Bulk WGS + genome graph can hint, but in practice it’s often hard to resolve. Here is our new preprint.
Today marks a meaningful milestone -- I am starting as a lab investigator and thoracic oncologist at MSKCC, focused on improving treatment through a deep understanding of cancer evolution. I will soon be looking for fellow travelers on this mission. Stay tuned!
An amazing work utilizing Cre-LoxP system creating oncogene amplification by ecDNA. Fascinating parallels with human liposarcomagenesis. Huge congratulations to the Ventura lab and the whole team!
Thanks for making this. I would love to be added!
