Developmental mosaicism underlying EGFR-mutant lung cancer presenting with multiple primary tumors

Abstract

Although the development of multiple primary tumors in smokers with lung cancer can be attributed to carcinogen-induced field cancerization, the occurrence of multiple tumors at presentation in individuals with EGFR-mutant lung cancer who lack known environmental exposures remains unexplained. In the present study, we identified ten patients with early stage, resectable, non-small cell lung cancer who presented with multiple, anatomically distinct, EGFR-mutant tumors. We analyzed the phylogenetic relationships among multiple tumors from each patient using whole-exome sequencing (WES) and hypermutable poly(guanine) (poly(G)) repeat genotyping as orthogonal methods for lineage tracing. In four patients, developmental mosaicism, assessed by WES and poly(G) lineage tracing, indicates a common non-germline cell of origin. In two other patients, we identified germline EGFR variants, which confer moderately enhanced signaling when modeled in vitro. Thus, in addition to germline variants, developmental mosaicism defines a distinct mechanism of genetic predisposition to multiple EGFR-mutant primary tumors, with implications for their etiology and clinical management.

Fig. 1. Genetic analysis of familial lung cancer caused by inherited T790M mutation in the EGFR gene.

a, Pedigree of a family with multiple cases of lung adenocarcinoma, in which the index case (III-1) was diagnosed with six primary carcinomas (first resection), followed by resection of seven tumors 10 years later. Individuals shown in black have a confirmed or obligate germline T790M-EGFR mutation and those who have developed lung adenocarcinoma are denoted LUAD. The pedigree was minimally altered to preserve confidentiality (males, square; females, circles). b, CT scans of two tumors from patient III-4, one in the right middle lobe (RML) and the other in the right lower lobe (RLL). c, Histology of tumors from T790M-EGFR family patients showing the range of invasiveness encountered in our cohort, from precancerous AAH (patient III-4 lesion T2), to AIS (patient III-4 lesion T2), to MIA (patient III-1 lesion T2), to invasive adenocarcinoma (patient III-1; lesion T12). Panels are at ×40 magnification, with insets at ×200. Scale bars, 1 mm. d, Schematic of the tumor locations in patient III-1, at the first resection (left) and the second resection 10 years later (right). e, Copy number data for two tumors from the first resection of patient III-1. f, Phylogenetic lineage tracing of multiple tumors from patient III-1 based on WES. The tumors from the first resection, T1–T6, share no mutations outside of EGFR, as represented on the tree by no intersection point for clones 1, 6, 7, 12, 13 and 14 and, in the pie charts, by no colors shared between them. In contrast, the tumors from the second resection, T7–T13, share 29 mutations, as represented by the long trunk leading from cl1 to cl2 before branching into cl4, -5, -8 and -10. In addition, the pie charts for these tumors are complex mixtures of these four clones and clones are shared among multiple tumors. Numbers on branches are mutations that accumulated between two nodes, which represent distinct clones identified by WES. Numbers in parentheses are exonic mutations that are not in the FFPE context.

Fig. 2. Mosaic somatic EGFR mutations mediate multiple primary lung tumors resulting from shared common ancestors.

WES-derived phylogenetic trees of four cases with multiple tumors that share a common somatic ancestor. The shared somatic mutations, including EGFR, are shown in magenta. Numbers on branches are mutations that accumulated between two nodes, which represent distinct clones identified by WES. Numbers in parentheses are exonic mutations that are not in the FFPE context. In comparison with the previous cases, the branches of these trees do intersect at the pink clones, indicating some shared genetic ancestry that is not observed in completely independent or germline tumors. However, the number of shared mutations and the trunk of shared ancestry are very small relative to the total number of mutations in each clone. This is distinct from the patients with metastatic cancer. The pie charts of these tumors all exhibit the pink clone but are otherwise relatively simple and do not share clones between tumors. a, In case 7, the two geographically distinct tumors share an extremely rare somatic EGFR mutation, SPKANTKEI752del, and then acquire 236 and 328 separate exonic mutations. b, In case 8, two geographically distinct contralateral tumors share the recurrent mutation L858R, in addition to two somatic mutations, before acquiring 111 and 83 separate mutations each. c, In case 9, two tumors (T2 and T6) share the L858R mutation before acquiring between 38 and 69 private mutations. T4 and T7 were too early stage and of too low purity to assess by WES whether they also carry the mutation; however, clinical sequencing confirmed the L858R mutation in both. d, In case 10, six tumors share XKR6 P580Q, ZBTB16 (intron) and ARID3B (3′-UTR) mutations. ^*EGFR L858R was found in three of the six tumors (T1, T6 and T7) by WES and a fourth (T4) by clinical genotyping. ^**ZBTB16 and AIRD3B mutations were observed in multiple tumors but not normal tissue samples, therefore they are borderline for FFPE filtering. The other two tumors are early stage and low purity. These tumors went on to acquire 45–409 private mutations.

Fig. 3. Developmental mosaicism is demonstrated by mutated normal lung cells and early common ancestors.

a, Detection of the L858R EGFR mutation using ddPCR in cases where the tumor harbors the L858R mutation (patients (Pt) 8–10) compared with cases where the tumor contains another EGFR mutation (negative control (Neg. ctrl), patients 1 and 7). The VAF for tumors is on the left y axis whereas the lower VAF in normal samples is on the right y axis (n = 2 tumor and 11 normal samples (independent microdissection regions) for patient 8; 4 tumor and 8 normal samples for patient 9; 12 tumor and 9 normal samples for patient 10; 3 tumor and 6 normal samples for patient 1; and 2 tumor and 9 normal samples for patient 7). b, A schematic showing the relationship between divergence from zygote and divergence between tumors. c, Schematic of poly(G) genotype analysis method, based on ref. . Samples collected from a single patient have poly(G) sites that may have undergone slippage due to hypermutability. The assay detects these indels and measures their mean length change compared with normal tissue. The correlation between the two tumors across all poly(G) sites is represented by Pearson’s correlation coefficient (r). d, Heatmap showing the mean distance from normal lung for each poly(G) hypermutable region, for each tumor from patient III-1. Tumors that have similar patterns are more closely related than tumors that have different patterns. e, Phylogenetic tree of patient III-1 based on the poly(G) analysis. The tree is rooted at the germline sample: i shows only samples from the first resection and ii samples from the second resection plus recurring sample T5. f, Correlation plots between tumors from patient III-1. The dots represent the mean length from normal at each poly(G) location (n = 26 poly(G) loci). The r estimates what fraction of cell divisions were shared between the tumor pair before divergence. The gray shading represents the 95% CI. g, Poly(G) relatedness between tumor pairs. Each point represents the poly(G) evolutionary distance between two tumors from cases that are unrelated (different individuals), mosaic or metastatic (classification based on WES analysis of exonic mutations). The dotted line represents that correlation coefficient >95% of the unrelated tumor pairs. Boxplot elements: center line, median; box limits, lower and upper quartiles; whiskers, lowest and highest value within 1.5× the IQR. P values are a Holm–Bonferroni-corrected, post-hoc Dunn’s test after a significant Kruskal–Wallis test. The arrow is the tumor pair from metastatic patient 3. NS, Not significant. (n = 443 interpatient tumor pair comparisons across eight patients for the unrelated analysis; n = 7 intrapatient tumor pair comparisons within four mosaic patients for the mosaic analysis; and n = 29 intrapatient tumor pair comparisons within two metastatic patients for the metastatic analysis). Source data

Fig. 4. Germline H988P and G873E mutations increase EGFR activity.

a, Position of the relevant residues within a partial EGFR protein crystal structure (aligned PDB structures EGFR 696-1022 T790M (5gty) and EGFR 703-985 (4zjv)). The dimerization domain is green and the catalytic tyrosine kinase domain blue. b, Schematic of tumor locations in patients 4 and 5. c, Lineage tracing of patient 4 and 5 tumors, derived from WES. Numbers on branches are mutations that accumulated between two nodes, which represent distinct clones identified by WES. Numbers in parentheses are exonic mutations that are not in the FFPE context. d–g,j–m, Functional effect of the H988P-EGFR mutant (d–g) or G873E mutant (j–m), compared with the WT construct. d,e,j and k are western blots and f,g,l and m are quantifications of the blots immediately above. The pY845 was normalized to vinculin, then total EGFR and finally to the average signal within an experiment for comparison across experiments. phos, phosphorylation. OE, overexpression. EGFR and pY845 are both rabbit antibodies and were run on different blots (processed in parallel), each with their own vinculin loading control, which was used for quantification. A representative vinculin blot is shown in the figure. The pS473 was normalized to a vinculin sample processing control from a matched EGFR blot, then to the average signal within an experiment. AKT and pS473 are both rabbit antibodies and were run on different blots (processed in parallel). Images vertically sliced to juxtapose nonadjacent lanes were run on the same gel (n = 10 biologically independent samples per figure). Data are presented as mean values ± s.e.m. A two-way ANOVA was performed to determine statistical significance and false recovery rate (FDR) q-values were corrected for multiple comparisons using the Benjamini, Krieger and Yekutieli procedure. h,i,n, Representative images of colony formation by NIH/3T3 cells in soft agar in cells expressing WT or mutant EGFR constructs (h). Scale bars, 100 μm. This experiment was repeated 3× with similar results, as quantified in i and n. Quantification data of colonies at least 20 μm in size is presented as mean values ± s.e.m. P values are a one-tailed, unpaired Student’s t-test not corrected for multiple comparisons. Source data

Fig. 5. Lineage tracing of metastatic cancers.

Phylogenetic trees from WES of two patients with metastatic cancer. The shared EGFR mutation is annotated on the tree. Numbers on branches are mutations that accumulated between two nodes. Numbers in parentheses are exonic mutations that are not in the FFPE context. These trees show a long trunk of shared mutations before branching into separate clones. Consistently, the pie charts are made up of mixtures of clones that are shared between tumors.

Fig. 6. Genetic distinctions between multiple lung cancers with inherited, mosaic and metastatic origin.

Schematic representation of three distinct mechanisms underlying multiple EGFR-mutant lung tumors. The average percentage of somatic mutations shared among different tumors is shown on the right (range in parentheses). a, In cases with inheritance of an attenuated mutant EGFR allele (either familial or apparent sporadic), the mutation presenting in the germline (lightning bolt) is shared by all somatic tissues. A second canonical EGFR mutation arises somatically at high frequency in predisposed lung cells, leading to multiple tumors with no shared somatic mutations. The evolutionary distance to the most common shared ancestor between different tumors (arrow) extends to the germline; 0 nongermline mutations are shared between tumors. b, Cases with mosaic predisposition arising from acquisition of an EGFR mutation during development (lightning bolt). This timing determines the proportion of normal cells containing the variant allele and the likelihood of developing multiple tumors. In addition to the activating EGFR mutation, a small number of additional somatic genetic variants are shared before the mosaic tissues diverge and acquire independent tumor-associated mutations. c, In sporadic cancers without genetic susceptibility, a single EGFR mutation arises in a somatic lung epithelial cell (lightning bolt), generating a single tumor. Metastases from this tumor share extensive mutational profiles and the most recent common ancestor for these multiple tumors is the primary tumor. n, number of analyzed cases in cohort.

Extended Data Fig. 1. Sample locations and patient III-4 data.

a, Schematics of the relative location of all normal samples and corresponding tumors in our cohort. b, Schematic of the tumor locations in patient III-4. c, Lineage tracing of patient III-4 with a known germline T790M-EGFR mutation, identified by Whole Exome Sequencing.

Extended Data Fig. 2. Mutation signature analysis of WES data.

a, The mutational spectrum across all patients in this cohort is denoted, with each color representing one of the six potential base substitutions, and each substitution further stratified based on the flanking nucleotides. These spectra were decomposed into distinct signatures using a Bayesian NMF approach (SignatureAnalyzer) and compared to a database of known signatures (COSMIC). SignatureAnalyzer discovered signatures for aging (SBS1 and SBS5) and smoking (SBS4, cosine similarity=0.93). There is a small contribution of APOBEC (SBS2 and SBS13) in S2. b, Adding up all the probabilities associated with a particular signature (across all mutations), yields the expected mutational burden for that signature. The sum of these signature-specific mutational burdens (across all signatures) is the total number of mutations found in a given sample. The bar chart represents the mutational burden associated with each signature per sample.

Extended Data Fig. 3. Germline mutations and patient 1 and 6 WES results.

a, IGV tracks for the patients harboring germline variants, showing the germline EGFR mutations in the normal samples as well as tumors. Patient 4 also shows EGFR G873E on same read as L858R in the tumor samples, indicating that the two mutations are in cis. b, Phylogenetic tree from patient 6, with a germline variant identified by WES. c, Phylogenetic tree from patient 1 predicted to be independent by WES.