Genomic and Functional Approaches to Understanding Cancer Aneuploidy

Authors

Alison M. Taylor

Juliann Shih

Gavin Ha

Galen F. Gao

Xiaoyang Zhang

Ashton C. Berger

Steven E. Schumacher

Chen Wang

Hai Hu

Jianfang Liu

Alexander J. Lazar

The Cancer Genome Atlas Research Network

Andrew D. Cherniack

Rameen Beroukhim

Matthew Meyerson

Doi

10.1016/j.ccell.2018.03.007

PMID: 29622463 · DOI: 10.1016/j.ccell.2018.03.007 · Journal: Cancer Cell (2018)

TL;DR

Taylor et al. defined an arm-level aneuploidy score for 10,522 TCGA tumors across 33 cancer types using ABSOLUTE-derived absolute copy number, then correlated the score with somatic mutation, gene expression, and immune-infiltrate features. Across the pan-cancer cohort, aneuploidy was strongly enriched in TP53-mutant samples (linear-model coefficient 0.13, the only positive outlier), positively correlated with somatic mutation rate after excluding hypermutators (microsatellite-instability and POLE-mutant tumors), positively correlated with proliferation hallmarks (E2F targets, G2M checkpoint, mitotic spindle), and negatively correlated with leukocyte-fraction-driven immune signatures within tumor types. Squamous tumors of multiple tissues clustered together by arm-level pattern, defined by chr_3p loss and chr_3q gain — a pattern strongest in LUSC, ESCC, and HPV-negative HNSC. The authors then engineered a CRISPR-Cas9 + artificial-telomere strategy to delete chr_3p in human immortalized lung epithelial AALE cells, observed an initial proliferation defect rescued by duplication of the remaining chromosome 3, and reproduced the in-cis down-regulation of 3p genes (STAC, ROBO1) and up-regulation of interferon/immune signatures predicted by the pan-cancer model.

Cohort & data

  • Pan-cancer cohort: 10,522 TCGA tumors spanning 33 cancer types (TCGA PanCancer Atlas), with arm-level somatic copy number called for >400,000 chromosome arms and >175,000 non-acrocentric whole chromosomes PMID:29622463.
  • Copy-number platform: Affymetrix SNP 6.0 arrays processed with the ABSOLUTE algorithm (Carter et al., 2012) to produce segmented absolute total copy number, sample purity, ploidy, and whole-genome-doubling calls; SCNAs called as deviations from each sample’s euploid level PMID:29622463.
  • Expression data: Illumina mRNA-seq RSEM expression for 9,670 of the 10,522 samples used for aneuploidy/expression linear modeling and GSEA against MSigDB Hallmark gene sets PMID:29622463.
  • Mutation data: Whole-exome mutation calls for 9,756 of the 10,522 samples from the TCGA MC3 pipeline PMID:29622463.
  • Recurrence statistics: Recurrent SCNA scores derived from GISTIC2.0; recurrent mutation scores from MC3 PMID:29622463.
  • Immune deconvolution: Leukocyte fraction estimated from methylation signatures (TCGA PanCancer immune subgroup) and from RNA expression (Aran et al., 2017); the two estimates correlate at Spearman ρ = 0.706 PMID:29622463.
  • Datasets used: TCGA PanCancer Atlas pan-cancer cohort. Per-tumor-type analyses include the squamous cluster contributors (LUSC, ESCC, HPV+ and HPV− HNSC, CESC), GBM, LGG, LUAD, BRCA, HCC, gastrointestinal cluster (COAD, READ, STAD, PAAD), gynecological cluster (UCEC, UCS), TGCT, AML, THYM, and UVM. Canonical PanCancer-Atlas study slugs include gbm_tcga_pan_can_atlas_2018, cesc_tcga_pan_can_atlas_2018, and coadread_tcga_pan_can_atlas_2018.
  • Functional model: Human immortalized lung epithelial AALE cells (SV40 large-T antigen, p53/RB inhibited; Lundberg et al., 2002), maintained in Lonza SAGM. Chr_3p arm deletion engineered via CRISPR-Cas9 double-strand breaks centromeric to all 3p genes plus a co-transfected artificial-telomere/puromycin plasmid carrying ~1 kb of homology to the centromeric chr_3p break (Uno et al., 2017). Recombination confirmed by Sanger sequencing, 3p hemizygosity by qPCR, whole-genome sequencing, and karyotyping PMID:29622463.
  • In vitro readouts: Cell proliferation (CellTiter-Glo over 6 days), cell-cycle and apoptosis (propidium iodide flow cytometry), RNA-seq (HiSeq 2500 PE100 → STAR + RSEM + edgeR), low-coverage WGS (MiSeq) processed with HMMCopy and IchorCNA for subclonal copy number PMID:29622463.

Key findings

  • Aneuploidy score construction. A per-sample arm-level aneuploidy score (range 0–39, summing altered p and q arms across non-acrocentric chromosomes plus q arms of 13–15, 21, 22) was generated by clustering arm-level SCNAs with a Gaussian mixture model and BIC selection (2–9 clusters); arms with mean SCNA length ≥80% of the arm were called positive, ≤20% negative, intermediates left uncalled. The score correlates almost perfectly with fraction of genome altered by aneuploidy (Spearman ρ = 0.975) PMID:29622463.
  • Aneuploidy frequency varies dramatically by tumor type. 88% of pan-cancer samples had at least some detectable aneuploidy (mean score 10.0; s.d. between tumor-type means = 4.8). Lowest: thyroid carcinoma (26% any arm event, mean score 0.87), AML (mean 1.6), THYM (mean 3.8). Highest: GBM (99% any event, mean 8.2), UCS (96%, mean 17.2), TGCT (99%, mean 18.7) PMID:29622463.
  • Whole-genome doubling drives aneuploidy. WGD samples (per ABSOLUTE) have higher aneuploidy (Spearman ρ = 0.55 between WGD status and score). Within a fixed genome-doubling status, increased aneuploidy associates with decreasing ploidy — i.e. the arm-level events that contribute to aneuploidy are predominantly losses, not gains PMID:29622463.
  • TP53 is the only positive mutation outlier. In a multivariable linear model controlling for cancer type and per-sample mutation count, TP53 was the single mutation enriched among aneuploid samples (coefficient = +0.13, the highest in magnitude and significance). Thirty-four other genes reached significance but all had negative coefficients with magnitudes ≤0.027 (i.e., enriched in low-aneuploidy samples) PMID:29622463.
  • Mutation rate and aneuploidy: positive once hypermutators are excluded. Pan-cancer the relationship is anti-correlated, but the anti-correlation is driven by MSI-high and POLE-mutant cancers (mostly COAD, READ, UCEC). Excluding hypermutators yields Spearman ρ = 0.38 between aneuploidy and mutation frequency, and ρ = 0.34 between recurrent SCNAs and recurrent mutations. The exceptions to this positive correlation per tumor type are COAD, READ, UCEC, UCS, STAD, and UVM — all but UVM contain MSI/POLE cases PMID:29622463.
  • Aneuploidy positively correlates with sample impurity but negatively with leukocyte fraction. Cancer impurity (1 − ABSOLUTE purity) rises with aneuploidy across the cohort, but leukocyte fraction (methylation-based) negatively correlates with aneuploidy — strongest in PAAD (ρ = −0.428) and HNSC (ρ = −0.312). The pan-cancer-level slight positive correlation between leukocyte fraction and aneuploidy (ρ = 0.0568) is a Simpson reversal driven by between-tumor-type differences; within-tumor-type correlations are negative. The implication: high-aneuploidy tumors have a non-leukocyte stromal fraction that inflates impurity PMID:29622463.
  • Immune signatures track leukocyte fraction, not aneuploidy directly. Univariate GSEA against the aneuploidy regression coefficient enriched both proliferation hallmarks (E2F targets, G2M checkpoint, mitotic spindle) and immune hallmarks (IFNγ response, allograft rejection, immune response) at FWER p < 0.01. Adding tumor type as a covariate flipped the immune signal to negative; adding leukocyte fraction abolished it entirely. Proliferation enrichment remained significantly positive in every model — i.e. it is intrinsic to aneuploidy, while the immune signal is leukocyte-fraction-mediated PMID:29622463.
  • Arm-level alteration frequencies are asymmetric. Most frequently deleted: 8p (33%) and 17p (35%). Most frequently gained: 8q (33%). Least altered: 2p (18% total) and 2q (16%). Arms 6p, 12q, 17q, 19q are gained and lost at near-equal rates (Δ < 0.03); 1q, 7p, 8q, 20q are predominantly gained; 3p and 17p are predominantly lost PMID:29622463.
  • Tumor types cluster by arm-level pattern. Hierarchical clustering of mean arm-level calls produces a squamous cluster (CESC, HPV+ and HPV− HNSC, ESCC, LUSC) defined by chr_3p loss + chr_3q gain; a gastrointestinal cluster (COAD, READ, non-squamous esophageal, STAD, PAAD) co-gaining 8q, 13q, and chromosome 20; a gynecological cluster (ovarian, copy-number-high UCEC/UCS); an epithelial cluster (LUAD, BRCA, HCC) defined by 1q gain; and a neural-lineage cluster (GBM, LGG, melanoma) with chromosome 7 gain. Molecular subtypes generally cluster within their tumor type, with copy-number-high serous-like endometrial as a notable exception PMID:29622463.
  • Squamous chr_3 signature is dominant in lung. In TCGA LUSC, chr_3p was deleted in ~80% of tumors and chr_3q was gained in >60%. Co-occurrence of 3p loss + 3q gain was significantly above chance (chi-square p = 0.0386); the reverse pattern (3p gain + 3q loss) was not observed. In LUAD by contrast, <50% of tumors had 3p loss, only 13% had 3q gain, and the two events did not significantly co-occur (p = 0.0626) PMID:29622463.
  • Chr_3p loss has tumor-type-independent transcriptional consequences. Linear modeling of expression vs arm status (controlling for total aneuploidy and tumor type) showed that chr_3p loss reduces 3p gene expression in cis and is anti-correlated with cell-cycle hallmarks (E2F targets, G2M checkpoint), epithelial-mesenchymal transition, IFNγ response, and TNFα signaling (FWER p < 0.01 for each). The polarity flips when 3p is gained PMID:29622463.
  • CRISPR-Cas9 deletion of chr_3p in AALE cells works. Co-transfection of (i) a CRISPR plasmid generating double-strand breaks centromeric to all 3p genes and (ii) a plasmid bearing ~1 kb of homology to the centromeric break plus an artificial telomere and puromycin selection cassette, followed by puromycin selection and single-cell cloning, produced clones with verified recombination (Sanger), hemizygous 3p loss (qPCR, WGS, karyotype) PMID:29622463.
  • Initial 3p-deleted clones proliferate slowly, then evolve. Ten passages post-cloning, chr_3p-deleted AALE clones proliferated more slowly than non-deleted siblings (p < 0.05) with G1 cell-cycle accumulation (p < 0.001) and no apoptosis change. After ~10 additional population doublings and one freeze-thaw, the proliferation defect resolved. In two of three deleted clones, subclones with duplication of the remaining wild-type chromosome 3 emerged — converting “3p loss” to “3q gain” and recapitulating the squamous-cancer 3q-gain signature, which contains oncogenes SOX2, PIK3CA, and TERC PMID:29622463.
  • Engineered model mirrors TCGA expression program. Pan-cancer prediction (3p loss → IFN/immune up-regulation) was reproduced experimentally: 3p-deleted AALE clones up-regulated interferon/immune-response pathways by GSEA (FWER < 0.01) at early passage. The 3p-encoded ubiquitin enzyme UBA7, which targets the interferon gene ISG15, was paradoxically up-regulated more than two-fold (along with LMCD1) at early passage but lost its up-regulation at late passage; meanwhile STAC and ROBO1 were the most strongly down-regulated 3p genes (>15-fold at early passage; >13-fold at late passage). Sixty-four percent of 3p genes were significantly down-regulated at early passage (FDR < 0.05); 53% remained down at late passage PMID:29622463.

Genes & alterations

  • TP53 — somatic missense/truncating mutation. Only mutation positively associated with high aneuploidy in the pan-cancer linear model (coefficient +0.13). Authors note even with p53 inhibited via SV40 large-T in AALE cells, chr_3p deletion still slowed proliferation initially, suggesting non-p53 mechanisms also restrain aneuploidy PMID:29622463.
  • POLE — hypermutator phenotype. Together with MSI, POLE-mutant tumors drive the apparent pan-cancer anti-correlation between mutation rate and aneuploidy; excluding them flips the correlation positive (Spearman ρ = 0.38) PMID:29622463.
  • IDH1 — defines the LGG subgroup that clusters separately from GBM by aneuploidy pattern, characterized by 1p loss and 19q gain. IDH-wildtype GBM is characterized by chromosome 7 gain and chromosome 10 loss PMID:29622463.
  • SOX2, PIK3CA, TERC — chr_3q oncogenes whose gain authors invoke to explain the squamous chr_3q-gain signature and the rescue of proliferation observed when AALE 3p-deleted subclones acquire chromosome-3 duplication PMID:29622463.
  • STAC, ROBO1 — 3p genes most strongly down-regulated in cis after engineered chr_3p deletion (>15-fold early passage; >13-fold late passage) PMID:29622463.
  • UBA7 — 3p-encoded E1 ubiquitin enzyme that targets ISG15. Up-regulated >2-fold in early-passage 3p-deleted clones but no longer up at later passage; consistent with the IFN-pathway up-regulation observed by GSEA PMID:29622463.
  • LMCD1 — 3p gene also up-regulated >2-fold in early-passage 3p-deleted clones PMID:29622463.
  • TERT — invoked in the discussion as a potentially insufficient activation context (alongside p53/RB inhibition) for selection of broad aneuploidies in vitro PMID:29622463.

Clinical implications

  • No direct therapeutic claim. The paper is a descriptive pan-cancer/functional study and does not test or propose a therapy. Authors note that 1p/19q co-deletion in LGG (Cairncross et al., 2013) is a precedent for arm-level alterations being therapeutically relevant, and frame future work as identifying aneuploidy-specific dependencies via in vitro CRISPR/shRNA screens PMID:29622463.
  • Mechanistic implication for immunotherapy. Within tumor types, high aneuploidy tracks with reduced leukocyte infiltrate, consistent with prior Davoli et al. observations that aneuploidy could blunt immune response. The arm-resolved analysis refines this: deletions of 3p, 8p, 13q, and 17p actually correlate positively with immune signatures, while deletions of 4q, 5q, and 14q correlate negatively — implying arm-specific genetic content (not bulk aneuploidy) shapes the immune microenvironment PMID:29622463.
  • Aneuploidy score as a covariate. The ABSOLUTE-derived per-sample arm-level aneuploidy score (Table S1/S2) is a quantitative pan-cancer feature suitable for downstream biomarker analyses PMID:29622463.

Limitations & open questions

  • Aneuploidy definition is intentionally narrow. The authors restrict aneuploidy to broad arm-level and whole-chromosome SCNAs and exclude focal SCNAs, departing from definitions used in Davoli et al. (2017) and Buccitelli et al. (2017). This choice harmonizes with arm-generation mechanisms but means findings are not directly comparable to prior focal+broad analyses PMID:29622463.
  • Confounder reconciliation depends on covariate choice. The pan-cancer immune-signature direction reverses when tumor type or leukocyte fraction is added to the linear model; the authors interpret the within-tumor-type negative correlation as the truth, but acknowledge tumor-type confounding as a major source of literature discrepancies PMID:29622463.
  • Functional model uses a single immortalized line. All in vitro chr_3p deletion experiments are in AALE cells (SV40 large-T, p53/RB inhibited); generalizability to other lung lineages or other arm-level deletions is untested. The authors flag CRISPR in lung organoids and modeling stromal interactions as next steps PMID:29622463.
  • Three independent 3p-deleted clones; karyotype rescue in two of three. The duplication-of-3 rescue is observed in 2/3 clones; the third’s mechanism of accommodation is not characterized PMID:29622463.
  • Anchorage-independent growth not gained. AALE parental cells do not form colonies in soft agar (Lundberg et al., 2002), and 3p-deleted cells likewise scored negative — i.e. 3p loss in this context is not sufficient for transformation PMID:29622463.
  • The big open question. Whether aneuploidy is positively selected in cancer remains unresolved. Authors favor positive selection on epidemiological grounds (universality, tissue-specificity of patterns) but their experimental data show only neutral or negative effects on proliferation, paralleling oncogene-induced senescence and prior single-chromosome aneuploidy models (Sheltzer et al., 2017; Stingele et al., 2012) PMID:29622463.
  • Tetraploid baseline. 18 samples identified by ABSOLUTE as purely tetraploid were treated as having zero arm-level SCNAs; rare and conservative but worth flagging when re-using the score PMID:29622463.

Citations from this paper used in the wiki

  • “In 10,522 cancer genomes from The Cancer Genome Atlas (TCGA), aneuploidy was correlated with TP53 mutation, somatic mutation rate, and expression of proliferation genes.” (Summary)
  • “Aneuploidy was anti-correlated with expression of immune signaling genes, due to decreased leukocyte infiltrates in high-aneuploidy samples.” (Summary)
  • “TP53 was an outlier, with the highest coefficient in the linear model … coefficient for TP53 was positive 0.13.” (Results, Figure 2A / Table S3)
  • “When these hypermutated tumors are excluded, we observed a positive correlation between mutation frequency and aneuploidy score (Spearman’s rank correlation coefficient = 0.38).” (Results)
  • “In our lung squamous cell carcinoma dataset, chromosome 3p was deleted in almost 80% of tumors and chromosome arm 3q was gained in over 60% of tumors … chi-square p = 0.0386.” (Results, Figure 4B)
  • “STAC and ROBO1 were the most down-regulated 3p genes, decreased by more than 15-fold, and chr_3p genes UBA7 and LMCD1 were up-regulated more than two-fold.” (Results)
  • “We also observed subclones with duplication of the remaining full copy of chromosome 3 in two of the three deleted clones … 3q is gained and 3p is no longer lost.” (Results, Figure 6E)
  • “Deletions of some arms, including 3p, 8p, 13q, or 17p, are positively correlated with immune signatures, whereas deletions of other arms, including 4q, 5q, and 14q, are anti-correlated with immune signatures.” (Discussion)

This page was processed by crosslinker on 2026-05-15.