Research ResourcesCancer

A method of high-throughput functional evaluation of EGFR gene variants of unknown significance in cancer

See allHide authors and affiliations

Science Translational Medicine  15 Nov 2017:
Vol. 9, Issue 416, eaan6566
DOI: 10.1126/scitranslmed.aan6566

No longer unknown

Mutations in the epidermal growth factor are present in a variety of tumor types and are often responsible for driving cancer progression and drug resistance. A variety of mutations in this gene have been reported in genomics studies, but their effects on the tumor phenotype were often uncertain. To identify and classify these potentially pathogenic mutations, Kohsaka et al. developed a high-throughput method for assessing the effects of epidermal growth factor mutations alone and in combinations, which should help predict tumor behavior and treatment response for each mutation pattern found in patients.


Numerous variants of unknown significance (VUS) have been identified through large-scale cancer genome projects, although their functional relevance remains uninvestigated. We developed a mixed-all-nominated-mutants-in-one (MANO) method to evaluate the transforming potential and drug sensitivity of oncogene VUS in a high-throughput manner and applied this method to 101 nonsynonymous epidermal growth factor receptor (EGFR) mutants. We discovered a number of mutations conferring resistance to EGFR tyrosine kinase inhibitors (TKIs), including gefitinib- and erlotinib-insensitive missense mutations within exon 19 and other gefitinib-resistant mutations, such as L833V, A839T, V851I, A871T, and G873E. L858R-positive tumors (12.8%) harbored compound mutations primarily in the cis allele, which decreased the gefitinib sensitivity of these tumors. The MANO method further revealed that some EGFR mutants that are highly resistant to all types of TKIs are sensitive to cetuximab. Thus, these data support the importance of examining the clinical relevance of uncommon mutations within EGFR and of evaluating the functions of such mutations in combination. This method may become a foundation for the in vitro and in vivo assessment of variants of cancer-related genes and help customize cancer therapy for individual patients.


Since transforming mutations in epidermal growth factor receptor (EGFR) were first identified in non–small cell lung carcinoma (NSCLC) (13), advancements in the diagnostics for such mutations and the evolution of targeted therapeutics against EGFR have greatly improved the management and outcome of patients with this lethal disease (4). However, extensive next-generation sequencer–driven analyses of the NSCLC genome have revealed a large number of variants of unknown significance (VUS) in EGFR and other regions in the cancer genome that await further investigation (5).

The approval of gefitinib, erlotinib, or afatinib as first-line treatments for advanced lung cancers requires the presence of classical/sensitizing EGFR mutations, such as exon 19 deletions or the L858R point mutation (6, 7). Furthermore, thoracic oncologists may generally agree with the clinical usage of EGFR tyrosine kinase inhibitors (TKIs) for uncommon sensitizing mutations, including exon 18 insertions/deletions (indels), E709 mutations (8, 9), G719 mutations (10, 11), exon 19 insertions (12), the insertion of A763_Y764insFQEA (13), the S768I mutation (14), and the L861Q mutation, although preclinical and clinical trial data suggest that uncommon EGFR mutations are frequently less sensitive to first-generation EGFR TKIs (11, 15). The main mechanism underlying resistance to gefitinib/erlotinib is the acquisition of the T790M mutation in EGFR (16, 17), which can be overcome by the third-generation EGFR TKI osimertinib (18), but such efficacy is further bypassed by C797 compound mutations (19). Another class of EGFR TKI–insensitive mutations includes exon 20 insertions (20, 21).

In addition to these genomic alterations, the cancer genome contains a large number of nonsynonymous mutations with unknown clinical significance. In the COSMIC (Catalogue of Somatic Mutations in Cancer) database of somatic mutations (v78;; for example, a total of 770 nonsynonymous mutations have been reported for EGFR. Similarly, 442 of such mutations have been deposited for the ALK gene. However, the clinical relevance remains unknown for the vast majority of such alterations. Thus, we designed an approach, termed the mixed-all-nominated-mutants-in-one (MANO) method, to evaluate the transforming ability and drug sensitivity of hundreds of such VUS.


Establishment of a high-throughput functional assay

The MANO method uses a retroviral vector that enables the stable integration of individual genes into the genome of assay cells [such as mouse 3T3 fibroblasts or the interleukin-3 (IL-3)–dependent, murine pro-B cell line Ba/F3] along with 6–base pair (bp) bar code sequences (Fig. 1). Individually transduced assay cells are subsequently pooled and cultured in a competitive manner to evaluate their transforming potential or drug sensitivity in vitro or in vivo. At the end of the expansion period, genomic DNA (gDNA) is extracted from the assay cells to polymerase chain reaction (PCR)–amplify the bar code sequences, which are subsequently subjected to deep sequencing with the Illumina MiSeq platform to quantify their relative abundance.

Fig. 1. Schematic representation of the MANO method.

Mouse 3T3 or Ba/F3 cells were infected with recombinant retrovirus expressing oncoproteins with corresponding 6-bp bar codes. Equal numbers of the stably transduced cells were mixed and cultured with different types of medium and/or treated with TKIs or vehicle. gDNA was harvested from the mixture of the remaining viable cells at the appropriate periods for each assay. The bar code sequences were PCR-amplified and subjected to deep sequencing on MiSeq sequencers to quantitate their relative abundance (a direct reflection of the cell number). To evaluate the transforming potential of each oncoprotein, the read number for each bar code was normalized to that of day 0. To evaluate the inhibition profile of test compounds across transduced clones in the mixture, the read number for each bar code was normalized to that of the vehicle-treated control. WT, wild-type.

To determine the feasibility and the sensitivity of the MANO method, we transduced the cDNA of NSCLC-related oncogenes, including EML4-ALK, KIF5B-RET, KRAS(G12V), CD74-ROS1, EGFR(E746_A750del), or EGFR(L858R) (2225), with individual bar codes into Ba/F3 cells. A total of 20,000 each of the transduced cells with mutations other than L858R was mixed together with a different number (100 to 20,000 cells) of the EGFR (L858R) mutant cells. gDNA was subsequently isolated from the pools and subjected to bar code sequencing. As shown in Fig. 2A, the number of bar code reads corresponding to the former five cells was constant in all mixtures, whereas that corresponding to Ba/F3 cells expressing EGFR(L858R) varied proportionally to the initial input number (r = 0.99). Thus, the MANO method is highly sensitive, enabling the quantitative detection of as little as 0.1% of the initial input.

Fig. 2. The MANO method in vitro and in vivo.

(A) Mouse 3T3 cells were infected with a retrovirus expressing the oncoprotein shown at the top. The indicated number of cells was mixed together in the wells of a six-well tissue culture plate, and gDNA was prepared from the mixture the next day. The 6-bp bar codes were PCR-amplified and subjected to deep sequencing. The relative read numbers of bar codes in the mixtures were compared to those of the initial input of cells infected with a retrovirus for EGFR(L858R) (n = 4). (B) Mouse 3T3 cells were infected with a retrovirus expressing an oncogene or the corresponding wild-type gene (indicated on the right) and subjected to a focus-formation assay. On day 14, the cells were stained with Giemsa solution. A retrovirus expressing GFP was used as a negative control. (C) Temporal changes in the proportion of 3T3 or Ba/F3 cells expressing each oncoprotein, the corresponding wild-type protein, or GFP are shown with different colors as indicated in the lower right panel. Mouse 3T3 cells were either cultured in DMEM containing 10% FBS (upper left) or subcutaneously injected into nu/nu mice (upper right), and the relative proportion of cell clones was assessed at the indicated times using the MANO method. The read number for each cell clone was normalized to that of day 0. Ba/F3 cell clones expressing oncoproteins were cultured in vitro in RPMI 1640 without IL-3 and similarly analyzed using the MANO method (lower left). The yellow arrowhead indicates the complete depletion of several cell clones at day 6. (D) The correlation between the focus-formation assay and the MANO method. The Giemsa-stained area (%) in the 3T3 focus-formation assay in (B) was compared to the fold change of the read number on day 8 relative to day 0 in 3T3 cells cultured in vitro using the MANO method in (C). The Pearson’s correlation coefficient (r) was calculated.

We next tested whether the MANO method could evaluate the transforming potential of cDNAs. First, we used the standard 3T3 focus-formation assay for measuring the transforming activity of 14 EGFR mutants and their wild-type counterparts (Fig. 2B and fig. S1). Other transforming cDNAs, such as BRAF(V600E), MET exon 14 skipping, ERBB2(V777L) (2628), and those shown in Fig. 2B, were similarly investigated.

Subsequently, the MANO method was conducted using 3T3 cells and the same cDNA samples as in Fig. 2B (Fig. 2C). Culturing 3T3 cells under different medium conditions resulted in similar alterations of each gene proportion, although rapid alterations in the relative cell number were observed when the cells were cultured under low concentrations of fetal bovine serum (FBS) (fig. S2A). We used the MANO method to investigate the correlation of the growth of 3T3 cells between culture with 10% FBS and that with other serum concentrations and observed a good correlation among the different settings (fig. S2B). In addition, a positive correlation was observed between the focus-formation assay and the MANO method (r = 0.63; Fig. 2D and fig. S3A).

The MANO method was further conducted using 3T3 cells in a nude mouse tumorigenicity assay (Fig. 2C). Whereas cells expressing green fluorescent protein (GFP) or the wild-type forms of EGFR, ERBB2, or MET were depleted by day 11, the cells expressing EGFR(L858R) or EGFR(E746_A750del) gradually increased during the same period. This observation was consistent with the results obtained in the in vitro assay (Fig. 2D), supporting the feasibility of this method for assessing transforming activity in vivo. We also transduced the same 25 genes into Ba/F3 cells and monitored the relative proportion of cells expressing each gene. In the presence of IL-3, the proportion of all cells was relatively constant until day 20 (fig. S3B), whereas several variants were depleted at about day 6 in the absence of IL-3 (Fig. 2C). The cells whose ratio of relative occupancy at day 6 compared to day 0 was >0.01 showed IL-3–independent cell growth (fig. S3C).

We next treated a pool of 16 Ba/F3 cells expressing active EGFR mutants (n = 11) or other oncoproteins (n = 5) with various TKIs. Considering the different doubling times of the transduced cells, we compared each TKI-treated Ba/F3 cell to vehicle-treated controls to calculate the relative growth inhibition of each cell clone (Fig. 3A and table S1). Whereas treatment with the cytotoxic compound puromycin induced uniform cell death across the cell clones, treatment with EGFR TKIs (gefitinib, erlotinib, afatinib, osimertinib, and rociletinib) resulted in the dose-dependent death of cells for five TKI-sensitive EGFR mutants (L858R, E746_A750 del, G719S, E861Q, and S768I) in the pool (Fig. 3A). As expected, Ba/F3 cells expressing EGFR(T790M) were resistant to first- and second-generation EGFR TKIs (gefitinib, erlotinib, and afatinib) but sensitive to third-generation EGFR TKIs (osimertinib and rociletinib). By contrast, cells expressing EGFR (T790M_C797S) showed resistance to these third-generation TKIs. Similarly, crizotinib, a TKI for ALK and ROS1 (29, 30), suppressed the growth of cells expressing EML4-ALK or CD74-ROS1, and another inhibitor for ALK and RET, alectinib (31, 32), inhibited the growth of the cells expressing EML4-ALK or KIF5B-RET (Fig. 3A). To independently evaluate the sensitivity of the EGFR mutants to each EGFR TKI, the number of viable cells was also determined using the alamarBlue cell viability assay, a method for quantifying cell viability based on mitochondrial enzyme activity (fig. S4) (33). The relative fold changes of read counts (representing cell number) in the MANO method were consistent with those of the alamarBlue cell viability assay (r = 0.89; Fig. 3B).

Fig. 3. TKI sensitivity assessed by the MANO method.

(A) Ba/F3 cells expressing each of the 16 genes shown on the right were mixed and cultured in the presence of different concentrations of gefitinib, erlotinib, afatinib, osimertinib, rociletinib, crizotinib, alectinib, or puromycin. Bar code read numbers of the compound-treated cells were normalized to those of the dimethyl sulfoxide (DMSO)–treated mixture, and the relative viability (%) of each cell clone on day 5 is color-coded according to the indicated scheme. (B) Comparison of cell viability measured with alamarBlue cell viability assay and the MANO method for Ba/F3 cells with 10 EGFR mutants in (A). Each data point was normalized to vehicle-treated cells. Pearson’s correlation coefficient (r) was calculated as 0.89 (P < 0.0001). The low ratio area is magnified in the right panel. (C) Changes in the relative cell populations in the tumors of mice treated with TKIs. The MANO method was used to quantify the bar code read numbers of tumors in 10 erlotinib-treated and 10 vehicle-treated mice. The bar code number for each cell line was normalized to the total bar code numbers of the tumor on day 18, and the calculated number was subsequently used to determine the percentage contribution of each cell line to the tumors. The mean relative cell population (%) within the tumors is shown for mice treated with either vehicle (green circle) or erlotinib (orange circle). The blue and pink arrows represent decrease and increase in the relative cell populations, respectively, in the tumors treated with erlotinib compared to those treated with vehicle. Error bars denote SD.

Furthermore, 25 independent 3T3 clones were pooled and subcutaneously xenografted into mice that were further treated for 14 days with vehicle, erlotinib, or afatinib. Erlotinib treatment induced a marked reduction in the relative abundance of six of six TKI-sensitizing EGFR mutants (L858R, E746_A750del, L861Q, G719C, G719S, and S768I), whereas the number of cells carrying both TKI-resistant EGFR mutants (T790M and T790M_C797S) and nine of nine clones harboring other oncogenes increased during the same period (Fig. 3C). Similar results were obtained with afatinib treatment (fig. S5). These experiments demonstrated the feasibility of the MANO method to assess drug sensitivity in vivo.

Functional annotation of VUS within EGFR

We subsequently used the MANO method to evaluate 101 EGFR mutants reported 4 to 7386 times in the COSMIC database (tables S2 and S3), including 13 mutations in the extracellular (EC) domain (primarily present in glioblastoma), 86 mutations in the intracellular TK domain, and 2 mutations in the C-terminal domain.

We performed paired t tests to examine the transforming potential among the 101 EGFR mutations under several serum concentrations. Compared with wild-type EGFR, 27, 28, 25, 22, and 34 EGFR mutants were predicted to confer higher transforming potential in 3T3 cells assayed under 10, 5, 2, and 1% FBS or 5% bovine serum albumin, respectively. As depicted in fig. S6A, 55.9% of the total mutations induced oncogenicity across all serum conditions, implying that oncogenic potential may partially depend on assay conditions.

We subsequently investigated the correlation of transforming ability of the 101 EGFR mutants evaluated by the focus-formation assay and the MANO method. About 70% of the mutations were predicted as oncogenic using both methods (fig. S6B). Furthermore, Ba/F3 cells transduced with the mutant cDNAs were individually examined for IL-3–independent growth. Every EGFR mutant with cytokine abrogation potential was also predicted as an oncogenic mutation using the MANO method (fig. S6C).

Sixty-two and 57 mutations within EGFR were predicted to confer transforming potential in 3T3 and Ba/F3 cells, respectively (Fig. 4A). We further examined whether the remaining 37 mutations that did not provide transforming potential in either cell line may have some functional role in TKI resistance. These mutants were individually transduced into Ba/F3 cells to investigate their TKI sensitivities. The A839T mutation showed a marked resistance to any generation of EGFR TKIs (Fig. 4B and fig. S7).

Fig. 4. Functional annotation of VUS in EGFR.

(A) Venn diagram revealing the numbers of oncogenic EGFR mutants seen in 3T3 cells using a focus-formation assay (3T3 FF), 3T3 cells by the MANO method (3T3 MANO), Ba/F3 cells using the MANO method (Ba/F3 MANO), or BA/F3 cells individually examined for IL-3–independent growth (Ba/F3 individual). The method for the evaluation of the transforming activity in each assay is described in detail in the legend for fig. S6. (B) Ba/F3 cells expressing the L858R, A839T, or T790M_C797S_Cis mutants of EGFR were treated with the indicated concentrations of gefitinib, erlotinib, afatinib, or osimertinib for 72 hours. Viable cells (%) were measured using the alamarBlue cell viability assay, and the results are shown as the means of five independent experiments. (C) Ba/F3 cells expressing 86 EGFR mutants (indicated at top) were treated with either DMSO or EGFR TKIs (gefitinib, erlotinib, afatinib, osimertinib, or rociletinib) at 0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 1, 5, or 10 μM, and the relative viability of TKI-treated cells relative to the corresponding DMSO-treated cells is color-coded according to the scheme indicated at the top left. Missense, deletion, and insertion mutations are shown in green, brown, and red, respectively.

We also evaluated the TKI sensitivity of other EGFR mutants using the MANO method. As shown in Fig. 4C and table S4, the L858R and E746_A750del mutants were sensitive to any TKI, whereas the T790M, C797S, T790M_C797S_Cis, and T854A substitutions and exon 20 insertions showed resistance to some TKIs, suggesting that the MANO method well recapitulates the sensitivity data obtained in previous studies (34). We revealed that compared with L858R or the exon 19 deletion, the EC domain mutations (at exons 2 to 15), E709 mutations (exon 18), exon 19 missense mutations, V769L (exon 20), and exon 21 mutations (L833V, V851I, A871T, and G873E) were all insensitive to TKIs (Fig. 4C). Although the exon 20 insertions showed decreased sensitivity to osimertinib and rociletinib in general, the N771_P772insN change conferred higher sensitivity, and D770_N771insSVD conferred higher resistance than the other mutants (Fig. 4C).

Compound mutations in EGFR

To investigate whether infrequent mutations within EGFR play an important role in the tolerance against TKIs in clinical settings, we sequenced EGFR exons among 11 rebiopsy specimens from cases that recurred after gefitinib treatment without acquiring the T790M mutation. Three (27%) of 11 cases harbored compound mutations corresponding to L858R plus E709G or E709A (table S5). Because the allelic frequencies of L858R and E709G/A were almost identical, these compound mutations are likely present in the same subclone in tumors (table S5).

Thus, we further performed EGFR exon sequencing among 195 specimens of EGFR(L858R)-positive NSCLC and identified 39 cases harboring compound mutations (19.5%), including another 9 cases (4.6%) with compound mutations of L858R and E709A/G/K (Fig. 5A and table S6). We further identified 24 compound mutations in 195 additional cases formerly identified as positive for the EGFR exon 19 deletion, G719 mutations, or L861Q. Notably, more than 90% of the G719 mutations existed as compound mutations in the examined cohort. Furthermore, in the COSMIC database, more than 75% of the E709 mutations existed as compound mutations, primarily with L858R (35%) or G719 mutations (32%) (table S7). To investigate whether the compound mutations exist in cis or trans allele, we performed gDNA- or cDNA-based amplicon sequencing (table S6) or Droplet Digital PCR (fig. S8) and observed that all compound mutations were present in cis alleles in all cases analyzed. Thus, the transforming potential of compound mutations may be stronger than that of minor mutations alone, as determined using the MANO method (Fig. 5B) and the 3T3 focus-formation assay (figs. S9 and S10).

Fig. 5. Compound mutations in EGFR.

(A) The frequency and patterns of EGFR compound mutations. Three hundred ninety cases formerly identified as positive for EGFR mutations (L858R, exon 19 deletion, G719 mutations, and L861Q) were examined for EGFR target sequence to investigate the frequency and the patterns of EGFR compound mutations. (B) The fold change in the ratio of 3T3 cell number with each EGFR compound mutation in cis (x axis) or trans (y axis) on day 12 compared with that of the corresponding EGFR single mutation is plotted. (C) Ba/F3 cells expressing 106 EGFR mutants (indicated at top) were treated with either DMSO or EGFR TKIs (gefitinib, erlotinib, afatinib, or osimertinib) at 0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 1, 5, or 10 μM, and the viability of TKI-treated cells relative to the corresponding DMSO-treated cells is color-coded according to the scheme indicated at the top left. (D) Ba/F3 cells expressing single or compound EGFR mutations (indicated at the right) were treated with the indicated concentrations of gefitinib, erlotinib, afatinib, or osimertinib for 72 hours without IL-3. The percentage of viable cells relative to that of parental Ba/F3 cells similarly treated in the presence of IL-3 was measured using the alamarBlue cell viability assay (n = 5). The mean median inhibitory concentration (IC50) values of the EGFR mutants for each TKI were calculated from five independent experiments.

We further evaluated the TKI sensitivity of EGFR compound mutants using the MANO method. As shown in Fig. 5C and table S8, the TKI sensitivity of compound mutations was between those of the two single mutations. The IC50 of gefitinib in Ba/F3 cells expressing EGFR with L858R_E709 mutations was 20 to 80 times higher than that of Ba/F3 cells expressing EGFR with L858R alone according to the alamarBlue cell viability assay (Fig. 5D). Clinically, none of the tumors with compound mutations of L858R and E709A/G responded to EGFR TKIs (table S5). These findings suggest a mechanism of TKI resistance mediated by the combination of E709A/G and L858R. Although the compound mutants were less sensitive to gefitinib and erlotinib, their sensitivity to afatinib was not remarkably different (IC50 values of less than 0.1 nM).

Cetuximab sensitivity of EGFR mutations assessed using the MANO method

We further used the MANO method to evaluate the sensitivities of EGFR mutants to cetuximab, a monoclonal antibody to EGFR. All of the missense mutations in the EC domain except for S492R (L62R, R108K, A289D/T/V, H304Y, P596L, and G598V), E709 mutations, K714R, L718Q, V769 mutations, V774M, C797S, and L833V, were sensitive to cetuximab (Fig. 6A and table S9). An additional L858R mutation in the cis allele conferred varying degrees of resistance to cetuximab, according to the alamarBlue cell viability assay (Fig. 6B). The growth of 3T3 cells with EGFR mutations (R108K, L718Q, and R108K_L858R_Cis) was completely inhibited in vivo after cetuximab treatment, and no tumor growth was observed for more than 1 month, whereas the growth of EGFR(L718Q_L858R_Cis) cells was initially inhibited but rebounded within 3 weeks (Fig. 6C).

Fig. 6. Cetuximab sensitivity of EGFR mutations assessed by the MANO method.

(A) Ba/F3 cells expressing 158 EGFR mutants (indicated at top) were treated with either DMSO or cetuximab at 0.001, 0.01, 0.1, 1, 10, or 100 μg/ml, and the viability of cetuximab-treated cells relative to that of the corresponding DMSO-treated cells is color-coded according to the scheme indicated at the top right. (B) Ba/F3 cells expressing the indicated mutants of EGFR were treated with the indicated concentrations of cetuximab for 72 hours. Viable cells (%) were measured using the alamarBlue cell viability assay, and the results are shown as the means of five independent experiments. (C) Evaluation of cetuximab sensitivity in vivo. Mouse 3T3 cells (1 × 106) transfected with the indicated expression constructs were injected into the subcutaneous tissue of mice (n = 5 per group). Tumor volumes were calculated as described in Materials and Methods. Error bars, SD; *P < 0.01, the tumor volume of the cells expressing R108K_L858R_Cis (left) or L718Q_L858R_Cis (right) treated with vehicle is compared with that of the corresponding cells treated with cetuximab at day 20. P < 0.01, the tumor volume of the cells expressing R108K (left) or L718Q (right) treated with vehicle is compared with that of the corresponding cells treated with cetuximab at day 20 (left) or day 38 (right).

The estimated drug sensitivities are summarized in Fig. 7, and the recommended drug for each mutation is suggested in table S10. On the basis of this evaluation, among the L858R-compound mutations (n = 39), 25, 7, and 6 mutations showed sensitivities distinct from L858R alone toward gefitinib, erlotinib, and afatinib, respectively (table S6). Thus, in total, 12.8, 3.6, and 3.1% of the L858R mutations showed altered drug sensitivity for gefitinib, erlotinib, and afatinib, respectively. In the examined cohort, clinical data were available for seven patients with EGFR compound mutations. Among these, four cases with EGFR compound mutations predicted as sensitive to EGFR TKIs obtained a partial response, whereas the other three cases with mutations predicted as partially sensitive exhibited stable disease after EGFR TKI treatment.

Fig. 7. Assessment of drug sensitivity for EGFR mutants.

The drug sensitivities of the indicated EGFR mutants were evaluated using the MANO method in Ba/F3 cells. The drug sensitivity was categorized as sensitive, partially sensitive, or resistant based on the IC90 of each mutant with each drug.

EGFR mutants resistant to all EGFR TKIs tested (S768I, D770_N771insSVD, T790M_C797S_Cis, L718Q_L858R_Cis, and T790M_G719A_Cis) were further examined to detect sensitivities to other TKIs. As shown in fig. S11, the alamarBlue cell viability assay with dacomitinib (an irreversible pan-ErbB inhibitor) (35), nazartinib (an irreversible mutant-selective EGFR inhibitor) (36), or neratinib (a highly selective HER2 and EGFR inhibitor) (37) revealed that EGFR(S768I) was sensitive to dacomitinib, whereas the other mutants were all less sensitive or resistant to the three TKIs.


Recently, a number of high-throughput genetic perturbation assays have been developed to investigate the functional significance of mutant genes. Kim et al. (38), for example, assembled a collection of 1163 cDNAs (including 474 mutants reported in fewer than three tumors) for the in vivo assessment of transforming potential. Berger et al. (39) developed an expression-based variant impact phenotyping method that measures the ability to induce gene expression changes across a set of landmark transcripts, enabling the classification of VUS into gain-of-function, loss-of-function, or change-of-function mutations. In contrast with these technologies, the approach presented herein applies a high-throughput functional analysis using cytokine-dependent cells (Ba/F3) to investigate the susceptibility of the given variants to therapeutic drugs. The MANO method thus enables an assessment of not only the oncogenic potential but also the drug sensitivity of hundreds of VUS of genes of interest within a short period of time.

Using the MANO method, the EC domain mutants of EGFR were demonstrated to be relatively insensitive to gefitinib and erlotinib compared with L858R or exon 19 deletions, providing a possible explanation for the poor activity of erlotinib against gliomas. Because EC domain mutants exhibited high sensitivity to afatinib and osimertinib, individuals with gliomas harboring these mutations may be eligible for afatinib or osimertinib treatment. Similarly, E709 mutations may be suitable for treatment with afatinib, whereas G719 mutations showed varying drug sensitivities to EGFR TKIs, depending on the precise amino acid substitution: G719C is sensitive to any type of EGFR TKI, whereas G719A is sensitive to only afatinib. Consistent with the results of a previous study (13), the MANO method revealed that all exon 20 insertions, except for A763_Y764insFQEA, are resistant to gefitinib, erlotinib, and afatinib.

Our report also shows that the activating missense mutations within exon 19 are generally insensitive to gefitinib and erlotinib but sensitive to afatinib and osimertinib. Notably, in the COSMIC database, there are nine lung cancer patients with EGFR exon 19 missense mutations who were treated with gefitinib, and none of these individuals responded to TKIs.

Considering that the plasma concentration of osimertinib at 24 hours after 80-mg oral administration is about 200 nM (40), we set the sensitivity threshold of osimertinib to an IC90 of 100 nM. According to this threshold, we observed distinct sensitivities to osimertinib among various EGFR exon 20 insertions. Compared to T790M, A763_Y764insFQEA was more sensitive, whereas other exon 20 insertions, including D770_N771insSVD and V769_D770insASV (the two most common exon 20 insertions), exhibited resistance. These results suggest that the sensitivity of each exon 20 insertion should be carefully interpreted. Similarly, a different mutation within exon 21 showed divergent sensitivities to EGFR TKIs. Our findings demonstrate that L833V, A839T, V851I, A871T, and G873E are resistant to gefitinib.

These data reveal that compound mutations of EGFR (L858R and E709A/G) are a relatively frequent mechanism for gefitinib resistance among tumors without the T790M mutation. Provided that mutations for E709A/G and L858R exist in cis and that only specific combinations emerge in TKI-resistant tumors, such double changes likely affect the protein structure of EGFR. Considering that 12.8% of L858R-positive tumors potentially have compound mutations that limit gefitinib efficacy, not only the hotspot mutation but also sequencing analysis of the entire EGFR gene/cDNA is clinically important to select suitable TKIs for each patient. Because the transforming potential of compound mutations is stronger than that of minor mutations alone, compound mutations likely result from multistep mutagenesis within one gene. All compound mutations observed in our study were present in cis, although the underlying biological mechanism remains elusive.

The MANO method also showed the utility of cetuximab for tumors with several minor mutants of EGFR, including L718Q, which is resistant to all types of EGFR TKIs (41), providing a treatment option for patients with these mutations. Mutants sensitive to cetuximab likely depend on the EGF ligand signal and subsequent receptor dimerization for transforming activities.

In the DNA-mutation Inventory to Refine and Enhance Cancer Treatment (DIRECT) database, 42 EGFR mutations were associated with disease progression after gefitinib or erlotinib therapy (42), and 10 of these mutations were also evaluated in the present study. Eight such mutations (80%) were predicted as partially sensitive or resistant to gefitinib or erlotinib using the MANO method, confirming that our preclinical data are consistent with the findings of other studies.

The limitation of our study is the lack of sufficient clinical data for validating the prediction by the MANO method. Especially, the results regarding cetuximab remain to be verified by clinical evidence because little has been done to examine the sensitivity of EGFR minor mutants to cetuximab in human. Although suitable TKIs for a given EGFR mutant can be readily predicted using the MANO method (and such combinations have been retrospectively verified in a small cohort), the clinical application of the output of the MANO method should be carefully translated. For example, patients with different types of minor EGFR mutations predicted as sensitive using the MANO method may be enrolled in open basket–type clinical trials, because the incidence of each mutation is too rare to conduct clinical trials for each subtype.

The annotation of VUS in terms of drug sensitivity and transforming potential is urgently needed in the catalog of the cancer genome atlas. Thus, we propose that the MANO method could accelerate the evaluation of the VUS of TKs, enabling the determination of the best drug for each mutation.


Study design and patient specimens

To evaluate the tumorigenicity and drug sensitivity of EGFR mutants, EGFR mutants reported more than four times in COSMIC database were selected, and their cDNAs were constructed by site-directed mutagenesis from wild-type EGFR cDNA in a retroviral vector. 3T3 cells and Ba/F3 cells overexpressing EGFR mutants were subjected to the MANO method. All in vitro experiments for the assessment of EGFR mutants by the MANO method (Figs. 4, A and C; 5C; 6A; and fig. S6) were performed in triplicate. To evaluate the tumorigenicity and drug sensitivity of oncogenic mutants, an in vivo MANO method was conducted in 10 mice per each group. Mice were sacrificed to obtain DNA from the tumors at the indicated day in each experiment (Figs. 2C and 3C and fig. S5). To investigate the involvement of infrequent mutations within EGFR in the tolerance against TKIs in clinical settings, we performed target amplicon sequencing of EGFR of 52 primary biopsy and rebiopsy specimens from 21 cases that recurred after gefitinib treatment, without acquiring the T790M mutation (obtained from the Department of Respiratory Medicine, Juntendo University, Graduate School of Medicine, Tokyo, Japan). We excluded the data of 10 cases, because we could not detect the EGFR mutations that were reported in the primary biopsy specimens of those cases. The samples included endobronchial biopsies, computed tomography–guided core needle biopsies, and small surgical biopsies. To determine the prevalence of EGFR compound mutations, 420 surgically resected NSCLC specimens positive for EGFR(L858R), EGFR exon 19 deletion, G719 mutations, or L861Q were obtained from the Department of General Thoracic Surgery, Juntendo University, Graduate School of Medicine, Tokyo, Japan. We excluded the data of 30 cases, because we could not detect the EGFR mutations which the specimens of those cases were reported to have. To investigate whether the compound mutations exist in the cis or trans allele, we performed gDNA- or cDNA-based amplicon sequencing in 12 and 27 specimens, respectively. If both mutations were on the same exon and closer than 200 nucleotides, then gDNA-based amplicons were used for sequencing, and otherwise, cDNA-based amplicons were used for sequencing because gDNA amplicons covering two different EGFR exons are too big for the library preparation. We could not evaluate the allele status of the other 23 specimens harboring compound mutations because we could not obtain reliable cDNA libraries and therefore stopped the sequencing analysis. Tumor tissue specimens were collected and analyzed under a protocol approved by the institutional review boards of The University of Tokyo (no. G3546) and Juntendo University (no. 2014176). Informed consent was obtained from all patients involved in the present study.

Cell lines

Human embryonic kidney–293T (HEK293T) cells and mouse 3T3 fibroblasts were obtained from the American Type Culture Collection and maintained in Dulbecco’s modified Eagle’s medium–F12 (DMEM-F12) supplemented with 10% FBS (both from Thermo Fisher Scientific). Ba/F3 cells were cultured in RPMI 1640 (Thermo Fisher Scientific) supplemented with 10% FBS and mouse IL-3 (20 U/ml; Sigma-Aldrich).

Construction of retroviral vector with random bar codes

The pcx5bleo vector was developed by modifying the pcx4bleo vector (43) to harbor the 6-bp DNA bar code sequence upstream of the start codon of the genes of interest. Plasmids encoding wild-type human EGFR cDNA were isolated by PCR and ligated into pcx5bleo. The cDNAs encoding the EGFR mutants were generated using the QuikChange II Site-Directed Mutagenesis kit (Agilent Technologies) and ligated into pcx5bleo.

Preparation of retrovirus and transduction of cell lines

The recombinant plasmids were introduced together with packaging plasmids (Takara Bio) into HEK293T cells to obtain recombinant retroviral particles. For the focus-formation assay, 3T3 cells were infected with ecotropic recombinant retroviruses using polybrene (4 μg/ml) (Sigma-Aldrich) for 24 hours and further cultured in DMEM-F12 supplemented with 5% calf serum for up to 2 weeks. Cell transformation was assessed through either phase-contrast microscopy or staining with Giemsa solution.

MANO detection

gDNA from the cell lysates was PCR-amplified using the primers 5′-TGGAAAGGACCTTACACAGTCCTG-3′ and 5′-GACTCGTTGAAGGGTAGACTAGTC-3′. The PCR products were purified using AMPure beads (Beckman Coulter). The sequencing libraries were generated using the NEBNext Ultra DNA Library Prep Kit (New England Biolabs) according to the manufacturer’s instructions, and index bar codes were added. The library quality was assessed using a Qubit 2.0 fluorometer (Thermo Fisher Scientific) and the Agilent 2200 TapeStation system. The library was sequenced on an Illumina MiSeq using Reagent Kit V2 (300 cycle) with the 150-bp paired-end option. These sequence reads included the bar code sequence 5′-CTAGACTGCCXXXXXXGGATCACTCT-3′ (where X denotes any nucleotide) and their complementary sequences. We detected the amount of each cDNA by counting these bar code sequences. DMSO-treated cell mixtures were used as the reference control for scaling of each cell clone signal. Thus, the signal from each treated cell line was calculated as 100 × (median read number across replicates)/(median read number of the DMSO control).

In vivo MANO method

Individually transduced cell clones were mixed in equal numbers, and 2.5 × 106 cells of this mixture (1 × 105 cells from each of 25 cell clones) were subcutaneously injected into 20 6-week-old female nude mice according to the animal use protocol reviewed and approved by the University of Tokyo Animal Care and Use Committee. The mice were treated once daily for 16 days by gavage with the EGFR TKI erlotinib (50 mg/kg body weight), afatinib (20 mg/kg body weight), or vehicle control (1% sodium carboxymethyl cellulose), beginning 5 days after injection of the cell lines. The tumors were resected and homogenized with gentleMACS Octo Dissociator with Heaters (Miltenyi Biotec) to obtain gDNA from each component of the tumor uniformly. The bar code number for each cell line was normalized to the total bar code numbers of the tumor, and the calculated number was subsequently used to determine the percentage contribution of each cell line to the tumors, treated either with EGFR inhibitor (n = 10) or with vehicle alone (n = 10).

AlamarBlue cell viability assay

After incubating the cells in 96-well plates (with 100 μl of culture medium per well), 10 μl of alamarBlue (Thermo Fisher Scientific) was added, and the fluorescence was measured by a microplate reader (2030 ARVO X3, PerkinElmer) (excitation, 530 nm; emission, 590 nm) at the indicated times. Wells without cells were assayed as negative controls. Adjustment for fluorescence gain for every well was performed against the well with the maximum fluorescence intensity.

EGFR sequencing and allele quantification

gDNA was prepared from frozen or formalin-fixed paraffin-embedded samples using the DNeasy kit from Qiagen. EGFR exons were PCR-amplified using the following 13 primer sets: set 1: 5′-ACGAGTAACAAGCTCACGCA-3′, 5′-ATTCTGCCCAGGCCTTTCTC-3′; set 2: 5′-TTGCCCTCAACACAGTGGAG-3′, 5′-TTATGAACCCCCAGCCTTGG-3′; set 3: 5′-CTGCGACATCCCTGGGAAAT-3′, 5′-CATCTTACCAGGCAGTCGCT-3′; set 4: 5′-ACTTACCTCACTTGCCCAGC-3′, 5′-GACAAGGATGCCTGACCAGT-3′; set 5: 5′-AAGCCAAAGGAGGATGGAGC-3′, 5′-AGGCCCTTCGCACTTCTTAC-3′; set 6: 5′-TTCTCTTGCAGTCGTCAGCC-3′, 5′-GGACCCATTAGAACCAACTCCA-3′; set 7: 5′-TGTGCCCACTACATTGACGG-3′, 5′-TTGCCGGAAAACTTGGGAGA-3′; set 8: 5′-TCCACCTCATTCCAGGCCTA-3′, 5′-ACTGCTAATGGCCCGTTCTC-3′; set 9: 5′-AGCCTCTTACACCCAGTGGA-3′, 5′-ACAGCTTGCAAGGACTCTGG-3′; set 10: 5′-GGCACCATCTCACAATTGCC-3′, 5′-AAAAGGTGGGCCTGAGGTTC-3′; set 11: 5′-TCATGCGTCTTCACCTGGAA-3′, 5′-AGGTACTGGGAGCCAATATTGT-3′; set 12: 5′-CACAGCAGGGTCTTCTCTGT-3′, 5′-GGTGTCAGGAAAATGCTGGC-3′; and set 13: 5′-GGCTCTGTGCAGAATCCTGT-3′, 5′-CAGGCTAATTTGGTGGCTGC-3′. Sequencing libraries were generated from the PCR products and sequenced on the MiSeq platform. The nucleotides in the sequence reads with a Q value of <20 were masked, and we further extracted unique reads that were subsequently mapped to the reference human genome (hg38) using BWA (, Bowtie2 (, and NovoAlign ( Mutations were called by an in-house pipeline based on the following detection rules: (i) Mutations are detected at a position of total read depth of ≥1000, (ii) the mutation allelic frequency for the tumor is ≥0.01, and (iii) the mutations were supported by both strands of the genome.

Investigation of compound mutation allelic status

RNA was isolated from the frozen samples using the RNeasy kit from Qiagen, and 1 μl of total RNA (1 μg) and 4 μl of SuperScript IV VILO Master Mix (Invitrogen) were mixed with 15 μl of nuclease-free water, followed by incubation at 25°C for 10 min, 50°C for 10 min, and 85°C for 5 min. The resulting cDNA was amplified by reverse transcription (RT)–PCR using PrimeSTAR HS DNA polymerase (Takara Bio) and specific primer sets to obtain a fragment containing both compound mutations. The following primer sets were used: set 1: 5′-GTCTTGAAGGCTGTCCAACGAATG-3′ and 5′-TCCAATGCCATCCACTTGATAGGC-3′ for detecting E709 or L718Q and L858R; set 2: 5′-TCTGGATCCCAGAAGGTGAGAAAG-3′ and 5′-TCCAATGCCATCCACTTGATAGGC-3′ for detecting I744M or S768I and L858R; set 3: 5′-ATCTGCCTCACCTCCACCGTG-3′ and 5′-TCCAATGCCATCCACTTGATAGGC-3′ for detecting T790M and L858R; set 4: 5′-GTCTTGAAGGCTGTCCAACGAATG-3′ and 5′-AGGTGAGGCAGATGCCCAGCA-3′ for detecting G719 and S768I; set 5: 5′-GTCTTGAAGGCTGTCCAACGAATG-3′ and 5′-CTTTGCGATCTGCACACACCAGTTG-3′ for detecting G719 and T790M; and set 6: 5′-TCTGGATCCCAGAAGGTGAGAAAG-3′ and 5′-CTTTGCGATCTGCACACACCAGTTG-3′ for detecting E746_A750del and T790M. The PCR amplification was conducted for 40 cycles at 98°C for 10 s, 55°C for 15 s, and 72°C for 1 min. The PCR products were purified using AMPure beads and subjected to library construction with NEBNext Ultra DNA Library Prep Kit. The library was sequenced on an Illumina MiSeq using Reagent Kit V2 (300 cycles) with 150-bp paired-end option.

Droplet digital PCR

One microliter of total RNA (=1 μg) and 1 μl of 2 μM gene-specific reverse primer (5′-GTCCTGGTAGTGTGGGTCTC-3′) were mixed with 1 μl of 10 mM deoxynucleotide triphosphate mix and 10 μl of nuclease-free water and heated at 65°C for 5 min, followed by cooling on ice for 1 min. Subsequently, 1 μl of dithiothreitol (100 mM), 4 μl of 5× SuperScript IV buffer, 1 μl of RNaseOUT recombinant RNase inhibitor, and 1 μl of SuperScript IV reverse transcriptase (Invitrogen) were added. The combined reaction mixture was incubated at 23°C for 10 min, 50°C for 10 min, and 80°C for 10 min. The resulting cDNAs were subsequently amplified by RT-PCR using the same procedure as described above. Digital PCR was performed using the QX100 Droplet Digital PCR system (Bio-Rad Laboratories) with EGFR-E709K primers (5′-CCAACCAAGCTCTCT-3′ and 5′-GCCCAGCACTTTGAT-3′) and the EGFR-E709K BHQ1 probe (HEX-GAGGATCTTGAAGAAAACTGA) at a final concentration of 1 μM primers and 250 nM probe, and EGFR-L858R primers (5′-GTATTCTTTCTCTTCCGCA-3′ and 5′-CAGCATGTCAAGATCACA-3′) and EGFR-L858R BHQ1 probe (FAM-TTGGGCGGGCCAAAC) at a final concentration of 1 μM primers and 250 nM probe. The 20-μl aqueous-volume digital PCR contained final concentrations of 1× Droplet Digital PCR supermix for probes (Bio-Rad Laboratories), each primer and probe (EGFR-L858R and EGFR-E709K), and templates. We used 600-pg to 60-ng cDNA, 100 to 10,000 copies of pcx5-plasmid, or 100 to 10,000 copies of PCR products as templates per sample. The reaction mixture was loaded onto a plastic cartridge with 70 μl of droplet generation oil and placed in the droplet generator. The droplets generated from each sample were transferred to a 96-well PCR plate, and the PCR amplification was conducted using a T100 Thermal Cycler with the following conditions: 95°C for 10 min followed by 45 cycles at 94°C for 30 s and 50°C for 90 s, a 10-min incubation at 98°C, and a final hold at 4°C. After amplification, the digital PCR data were collected and analyzed using a Bio-Rad QX100 droplet reader and QuantaSoft v1.3.2.0 software. Crosshair gating was used to split the data into four quadrants in a procedure analogous to that applied in flow cytometry analysis (44). About 15,000 droplets were analyzed per well.

Xenograft tumor assays

For xenograft generation, 1.0 × 106 cells were subcutaneously injected into 6-week-old female nude mice. The mice were treated twice a week with an intraperitoneal injection of either cetuximab (10 mg/kg body weight) (Merck Serono) or vehicle control, beginning 5 days after the injection of the cell lines. The average tumor volume in each group was expressed in cubic millimeters and calculated using the formula π/6 × (large diameter) × (small diameter)2. Tumor injections and volume measurements were performed blinded to the constructs expressed by the cells used for injection. All procedures in mice were performed according to the protocols reviewed and approved by the University of Tokyo Animal Care and Use Committee.

Statistical analysis

Data are means ± SD or means only, as stated in the figure legends. Differences between two experimental groups were determined by two-tailed Student’s t test. P < 0.05 was considered statistically significant.


Fig. S1. Quantification of focus-formation assay.

Fig. S2. Temporal changes in the proportion of 3T3 cells expressing 25 different genes.

Fig. S3. Transforming activity evaluation using the MANO method in vitro.

Fig. S4. AlamarBlue cell viability assay in Ba/F3 cells expressing EGFR mutants.

Fig. S5. In vivo evaluation of sensitivity to afatinib using the MANO method.

Fig. S6. In vitro transforming activity of 101 EGFR mutants evaluated by the MANO method.

Fig. S7. Immunoblot analysis of Ba/F3 cells expressing EGFR mutants.

Fig. S8. The droplet digital PCR assay for the detection of EGFR E709A and L858R.

Fig. S9. Focus-formation assay of 3T3 cells with EGFR L858R compound mutations.

Fig. S10. Focus-formation assay of 3T3 cells expressing EGFR G719 compound mutations or exon 19 deletion compound mutations.

Fig. S11. The sensitivity of EGFR mutants to dacomitinib, nazartinib, and neratinib.

Table S1. The raw data of Fig. 3A (provided as an Excel file).

Table S2. One hundred one recurrent EGFR mutations in COSMIC database analyzed with the MANO method (provided as an Excel file).

Table S3. Barcode sequences used in the MANO method for EGFR compound mutations (provided as an Excel file).

Table S4. The raw data of Fig. 4C (provided as an Excel file).

Table S5. Recurrent NSCLC cases after gefitinib treatment without the EGFR T790M mutation.

Table S6. Clinical and molecular characteristics of surgically resected lung adenocarcinoma with EGFR compound mutations (provided as an Excel file).

Table S7. The frequency and pattern of E709 compound mutations.

Table S8. The raw data of Fig. 5C (provided as an Excel file).

Table S9. The raw data of Fig. 6A (provided as an Excel file).

Table S10. Candidate drug for each EGFR mutation (provided as an Excel file).


  1. Acknowledgments: We thank A. Maruyama and H. Tomita for technical assistance. We also thank T. Akagi and K. Sasai for providing the pcx4bleo plasmid. Funding: This study was financially supported in part through grants from the Leading Advanced Projects for Medical Innovation (LEAP) and the Practical Research for Innovative Cancer Control from the Japan Agency for Medical Research and Development, AMED. Author contributions: S.K. and H.M. conceived the project and wrote the paper. S.K. designed and performed the experiments and analyses. M.N. performed the experiments. T.U. performed the bioinformatic analyses. Y.S., T.H., N.S., K. Takahashi, K.S., K. Takamochi, and F.T. provided clinical specimens and performed the analyses. Competing interests: The authors declare that they have no competing interests.

Stay Connected to Science Translational Medicine

Navigate This Article