Patient-Derived Cells to Guide Targeted Therapy for Advanced Lung Adenocarcinoma

Adequate preclinical model and model establishment procedure are required to accelerate translational research in lung cancer. We streamlined a protocol for establishing patient-derived cells (PDC) and identified effective targeted therapies and novel resistance mechanisms using PDCs. We generated 23 PDCs from 96 malignant effusions of 77 patients with advanced lung adenocarcinoma. Clinical and experimental factors were reviewed to identify determinants for PDC establishment. PDCs were characterized by driver mutations and in vitro sensitivity to targeted therapies. Seven PDCs were analyzed by whole-exome sequencing. PDCs were established at a success rate of 24.0%. Utilizing cytological diagnosis and tumor colony formation can improve the success rate upto 48.8%. In vitro response to a tyrosine kinase inhibitor (TKI) in PDC reflected patient treatment response and contributed to identifying effective therapies. Combination of dabrafenib and trametinib was potent against a rare BRAF K601E mutation. Afatinib was the most potent EGFR-TKI against uncommon EGFR mutations including L861Q, G719C/S768I, and D770_N771insG. Aurora kinase A (AURKA) was identified as a novel resistance mechanism to olmutinib, a mutant-selective, third-generation EGFR-TKI, and inhibition of AURKA overcame the resistance. We presented an efficient protocol for establishing PDCs. PDCs empowered precision medicine with promising translational values.

Statistical analysis. In univariate analysis, the Fisher's exact test and Mann-Whitney U test were applied to investigate association between PDC establishment and variables. In multivariate analysis, multivariate logistic regression model was used.

Results
positive cytological diagnosis of malignancy and tumor colony formation impact pDc establishment. A total of 23 PDCs were established from malignant effusions of advanced lung adenocarcinoma at a success rate of 24.0%. Established PDCs were free of stromal cells by light microscopy, strongly positive for EpCAM (an epithelial cell marker), could be frozen/thawed, and propagated at least 10 times (Supplementary Table 3 and Supplementary Fig. 3A) 7,21,22 .
Previous studies have shown that several factors including genetic alteration impact success rate of patient-derived xenograft model establishment, whereas little is known about establishing PDC from advanced lung adenocarcinoma [23][24][25] . To address this question, we reviewed association of factors to PDC establishment. Univariate analysis revealed that positive cytological diagnosis of malignancy (M+) and tumor colony formation in the initial primary culture (TCF+) were strongly correlated with PDC establishment (OR = 8.3654, P < 0.001; OR = 22.0772, P < 0.001) as well as multivariate analysis (OR = 4.8336, P = 0.0239; OR = 14.1733, P = 0.0131) ( Table 1 and Supplementary Fig. 1). As expected, high concordance was observed between M+ and TCF+ group in malignant effusions (Fig. 1A). The success rate was high in M+/TCF+subgroup (20/41, 48.8%), implying that these factors may be a powerful indicator of successful model establishment (Fig. 1B). A major reason for failure of model establishment was a paucity of tumor cells in samples (62/73; 84.9%) followed by tumor cell senescence (11/73; 15.1%). characteristics of pDcs. We characterized PDCs by direct sequencing (n = 23) and whole-exome sequencing (WES) (n = 7) ( Table 2, Supplementary Table 4, and Supplementary Fig. 4). Fourteen EGFR-mutant cell lines were generated from EGFR-mutant tumors progressing to first-(n = 8), second-(n = 1), or third-generation EGFR-TKIs (n = 5). Routine genetic testing of rebiopsy samples at recurrence were available in 9 patients with EGFR-mutant NSCLC. Notably, EGFR genotypes detected in the rebiopsy samples were concordant to those in corresponding PDCs (Table 2). Three PDCs which were originated from ALK-positive NSCLC maintained EML4-ALK fusion genes. Five ROS1-fusion PDCs which were generated from ROS1-positive NSCLC maintained various ROS1 fusion genes (SLC34 A2-ROS1, CD74-ROS1, and TPM3-ROS1) ( Table 2). WES identified BRAF K601E as a driver mutation in YU-1070 cells that were derived from NSCLC without druggable genomic alterations (Supplementary Table 4 and Supplementary Fig. 4A). These results demonstrate that PDCs largely maintain known patient driver mutations.
Extensive passaging may result in a genetic drift of cell lines 26,27 . To investigate this issue, we analyzed 5 PDCs at early and later passages using direct sequencing (YU-1092, YU-1096, YU-1152, and YU-1097) or WES (YU-1094). A mutation allele frequency (MAF) of EGFR mutations were preserved upto approximately 30 passages ( Supplementary Fig. 4B). Furthermore, somatic mutations and copy number variations were stably maintained between passages ( Supplementary Fig. 4A,C, and D). These results may suggest that driver mutations and tumor-related genes are stably maintained at least in tested PDCs.
Next, we compared in vitro sensitivity to TKI in PDC with response in the clinic. Ten patients in our study received subsequent TKI therapy after PDC establishment (5 osimertinib, 3 first-generation EGFR-TKI, 2 entrectinib). Twelve PDCs established from these patients were screened with TKI which the patients were treated with ( Fig. 2A). Two out of five patients with EGFR-mutant NSCLC were positive for EGFR T790M mutation, a marker of sensitivity to osimertinib, and received clinical benefits from osimertinib therapy, achieving a partial response (PR) and relatively long progression-free survival (PFS) 3 . Two corresponding PDCs (YU-1090 and YU-1073) exhibited in vitro sensitivity to osimertinib. Three PDCs (YU-1093, YU-1152, and YU-1094) generated from patients who were treated with osimertinib and had progressive disease as a best response were resistant to the drug (Fig. 2B). Three patients who received first-generation EGFR-TKI treatment did not achieve a partial response and had short PFS (n = 3). Accordingly, 4 corresponding PDCs (YU-1088, YU-1099, YU-1095, and YU-1091) were not responsive to gefitinib (Fig. 2C). Two patients with ROS1-positive NSCLC received entrectinib. One patient experienced a partial response with PFS of 6.5 months and corresponding PDC (YU-1080) was sensitive to entrectinib ( Supplementary Fig. 3B). The other patient displayed cardiac toxicity to entrectinib therapy [not evaluable according to RECIST (Response Evaluation Criteria In Solid Tumors)] and switched to crizotinib. PFS on crizotinib was 4.2 months, indicating intrinsic resistance to the therapy (Fig. 2D). YU-1082 and YU-1083 cells were established from the patient before the start of crizotinib therapy and were resistant to the drug in vitro. A similar pattern was observed for YU-1085 cells that were established from the patient after crizotinib therapy ( Fig. 2D and E). Together, these data suggest that PDCs may reflect patient treatment response to TKI.
PDCs can guide the selection of potentially effective therapy in oncogene-driven lung adenocarcinoma. BRAF mutations are found in 1-3% of lung adenocarcinoma 2 . The two main types of BRAF mutations, www.nature.com/scientificreports www.nature.com/scientificreports/ V600E and non-V600E, are associated with different clinicopathological features of lung adenocarcinoma and exhibit different therapeutic response to BRAF-targeted targeted agents 1,28 . While dabrafenib alone or in combination with trametinib demonstrated promising efficacy in BRAF V600E mutant NSCLC, appropriate treatment paradigms are still under investigation for non-V600E mutations 29,30 .
To identify an effective therapy for treatment of non-V600E BRAF mutant NSCLC, we tested efficacy of the single-agent and combination targeted therapy in YU-1070 cells harboring a BRAF K601E mutation. YU-1070 cells were highly resistant to vemurafenib, dabrafenib, and trametinib ( Supplementary Fig. 5A). On the other hand, treatment with trametinib sensitized YU-1070 cells to dabrafenib (Fig. 3A). The combination of dabrafenib with trametinib induced c-Raf phosphorylation and completely blocked ERK phosphorylation (Fig. 3B). These data demonstrate that the BRAF K601E mutation may respond to the dabrafenib/trametinib combination therapy.
Most NSCLC patients harboring common EGFR mutations, such as deletions in exon 19 or the L858R mutation in exon 21, respond dramatically to EGFR-TKIs. However, there is a paucity of data regarding the activity of www.nature.com/scientificreports www.nature.com/scientificreports/ EGFR-TKIs in NSCLC harbor uncommon EGFR mutations, such as G719X, L861Q, S768I alone or in combination with each other, which occur in approximately 10% of EGFR-mutant NSCLC 31 .
EGFR exon 20 insertions are among the rarer EGFR mutations (approximately 9% of EGFR-mutant NSCLC patients) and treatment for these mutations remain elusive without an approved inhibitor 32,33 . To identify optimal EGFR-TKIs, we investigated YU-1074 cells harboring the EGFR D770_N771insG mutation (Fig. 3C). Afatinib   www.nature.com/scientificreports www.nature.com/scientificreports/ potently inhibited growth of YU-1074 cells, whereas osimertinib was less effective than afatinib (Fig. 3C). Together, these data suggest that afatinib among all EGFR-TKIs tested may be the most effective treatment for the uncommon EGFR mutations.
EGFR C797S mutation is one of the most commonly reported mechanisms of acquired resistance to third-generation EGFR-TKIs 5 . EGFR T790M mutation in cis to C797S mutation confers resistance to third-generation EGFR-TKIs as well as first-generation EGFR-TKIs 34 . A combination of brigatinib and cetuximab has been introduced to overcome the C797S-mediated resistance 35 . We aimed to evaluate EGFR-TKI efficacies in YU-1097 cells harboring an EGFR exon 19 del/T790M/C797S mutation (T790M in cis to C797S). YU-1097 cells were resistant to single-agent gefitinib, afatinib, osimertinib, and brigatinib ( Supplementary Fig. 5B). Notably, YU-1097 cells were highly sensitive to the combination of brigatinib and cetuximab (Fig. 3E). The drug combination synergistically suppressed phosphorylation of AKT and ERK (Fig. 3F). These results show that the triple mutation may respond to the brigatinib/cetuximab combination therapy.
AURKA as a potential therapeutic target in EGFR-mutant nScLc resistant to third-generation eGfR-tKi. Mechanisms    www.nature.com/scientificreports www.nature.com/scientificreports/ progressing to third-generation EGFR-TKIs, posing a challenge to clinical decision making for these patients 36,37 . Using our clinically-relevant cell lines, we aimed to provide therapeutic strategies in this setting. In our panel of PDCs resistant to third-generation EGFR-TKIs, WES revealed genetic alterations (EGFR C797S, MET amplification, PIK3CA amplification, and PTEN loss) associated with osimertinib resistance [36][37][38] . However, known genetic alteration associated with drug response was not observed in YU-1089 cells (Fig. 4A, Supplementary Fig. 4A and E). First-, second-, and third-generation EGFR-TKIs failed to inhibit growth of YU-1089 cells (Fig. 4B). The EGFR-TKIs suppressed phosphorylation of EGFR and ERK but had no effect on phosphorylation of AKT (Fig. 4C).
To overcome the EGFR-independent mechanism of olmutinib resistance using YU-1089 cells, we comprised a panel of 79 investigational or FDA-approved drugs which target a wide range of kinases (Supplementary Table 5). Then, we performed drug combination screening on YU-1089 cells with olmutinib and each drug in the panel to nominate potent drug combinations. The screening identified 41 drugs with synergistic effects (CI < 1). The most strong synergy was observed with tozasertib, which targets Aurora kinases (Fig. 4D) 39 .
We next characterized the synergistic effect of combined EGFR and aurora kinase inhibition. The combination of olmutinib with tozasertib potently inhibited colony formation of YU-1089 cells compared to either agent alone (Fig. 4E). The robust synergism was confirmed in a 5 × 5 dose response matrix by using the Chou-Talalay method, resulting in a combination index (CI) value of 0.029 at 50% growth inhibition. Furthermore, the combination of olmutinib with tozasertib synergistically decreased phosphorylation of AKT and ERK and increased expression of apoptotic markers in YU-1089 cells. The comparable antitumor synergy was also shown by a combination of olmutinib with alisertib, a highly selective Aurora A kinase inhibitor under clinical development (CI = 0.196 at 44% growth inhibition), and a combination of osimertinib with tozasertib (CI = 0.189 at 52% growth inhibition) ( Supplementary Fig. 7A) 40 . Recently, Shah et al. has shown that AURKA confers resistance to third-generation EGFR-TKIs in NSCLC and inhibition of AURKA can resensitize the tumor to EGFR-TKIs 41 . Thus, we tested if this drug combination strategy is applicable to other osimertinib-resistant PDCs. However, the osimertinib/alisertib combination was less potent in YU-1095, YU-1096, and YU-1097 cells than in YU-1089 cells (Supplementary Fig. 7B). These differential responses to combined EGFR and AUKRA inhibition may be due to difference in AURKA expression 41,42 . Supporting this hypothesis, AURKA expression was lower in PDCs that were not responsive to the drug combination ( Supplementary Fig. 7C) 41,42 . Together, these results suggest that Aurora kinase A may be an actionable therapeutic target to overcome acquired resistance to third-generation EGFR-TKIs in EGFR-mutant NSCLC.

Discussion
In this study, we established 23 PDCs that represent molecularly heterogeneous subsets of advanced lung adenocarcinoma. Among them, cell lines of ROS1 fusions with various fusion partners, uncommon EGFR mutations, a resistant C797S mutation, and a rare BRAF mutation were included ( Table 2) 5,43 . To the best of our knowledge, there are no commercially-available NSCLC cell lines endogenously harboring these mutations. Using novel cell lines, we presented effective therapeutic strategies which may inform future clinical decision making.
Selection of appropriate tumor specimens is important for successfully establishing patient-derived models 23,24 . Previous studies have shown that tumor cellularity in malignant effusions of NSCLC is highly variable, ranging from 0.1% to 90% 44 . Furthermore, cytological diagnosis of malignant effusions can be misleading because of potential mimics such as reactive mesothelial cells 45 . To use the malignant effusion as starting material, there is an urgent need to optimize establishment procedures. Our findings provided the evidence that both positive cytological diagnosis of malignancy (M+) and tumor colony formation (TCF+) were crucial to establishing PDCs from malignant effusions. Indeed, using M+/TCF+ malignant effusions can increase the success rate of PDC by approximately 2-fold (24.0% vs 48.8%). Additionally, M−/TCF−, M−/TCF+, and M+/TCF− malignant effusions (55/96, 57.3%) that have a low potential for establishing PDCs (3/55; 5.5%) can be excluded in a stepwise manner, thereby substantially reducing time and effort needed for sample processing and subsequent long-term culture. Carter et al. has shown that tumor cellularity in malignant effusions of advanced NSCLC is not correlated to sample volume 44 . Accordingly, we observed sample volume did not impact cytological diagnosis (P = 0.42372) or PDC establishment (P = 0.58232) ( Table 1 and Supplementary Fig. 9). Although the difference was not statistically significant (OR = 0.5904, P = 0.3064) ( Table 1), we observed a higher success rate in EGFR wild-type cases (31.0%) than EGFR mutant cases (20.9%). Similarly, John et al. and our group reported the negative correlation between EGFR mutations and NSCLC PDX model establishment from surgical resection, which may reflect a favorable prognostic value of EGFR mutations [46][47][48] . Interestingly, we noted tumor cell senescence in some M+/ TCF+ primary cultures (11/41; 26.8%) between 4 to 7 passages. Despite high tumor purity, 5 PDCs became senescent between 10 to 23 passages, whereas other PDCs stably propagated over serial passage (Supplementary  Table 3). These results show that some advanced lung adenocarcinoma (16/41; 39.0%) may depend on niche factors, which are not provided by R10 medium or autocrine signaling, for optimal growth. Notedly, recent study has utilized Wnt, FGF7, and FGF10 to establish NSCLC organoid models 49 . The success rate for organoids was higher than the success rate for PDX or PDC, implying that these specific factors may be associated with niche factor dependency observed in the subset of advanced lung adenocarcinoma 48,49 . Direct comparison between these control, # p < 0.05 vs the value at the indicated comparison, n = 3). (F) YU-1097 cells were treated with the indicated concentrations of brigatinib alone or in combination with cetuximab for 6 hours. Cell lysates were immunoblotted with the indicated antibodies. (A and C) Cell viability was measured by CellTiter-Glo. Data are presented as the mean ± SEM (n = 3). (B, D, and F) Immunoblots are representative of 3 independent experiments. The full-length blots can be found in Supplementary Fig. 6  www.nature.com/scientificreports www.nature.com/scientificreports/ patient-derived models may provide insight into tumorigenesis of NSCLC and therapeutic potential for targeting these niche factors and related signaling pathways.
To demonstrate clinical relevance, we tested efficacy of single-agent or combination targeted therapies in our PDCs harboring a BRAF K601E mutation and uncommon EGFR mutations (L861Q, G719C/S768I, D770_ N771GinsG). Our data suggest that the BRAF K601E mutation may respond to a combination of dabrafenib and trametinib in a similar manner to a BRAF V600E mutation 29 . Indeed, the drug combination has demonstrated efficacy in a PDX model of BRAF K601E mutated melanoma 50 . Generally, NSCLC with uncommon EGFR mutation has been known to be less sensitive to first-generation EGFR-TKIs 31,51,52 . Similar to our findings in a PDC harboring the L861Q mutation, afatinib has shown lower IC 50 values than first-or third-generation EGFR-TKIs in genetically-engineered Ba/F3 cells 31,53,54 . To our knowledge, we first reported in vitro efficacy of EGFR-TKIs against the EGFR G719C/S768I mutation and demonstrated that afatinib was the most potent among other EGFR-TKIs against the EGFR G719C/S768I mutation. These previous findings and ours corroborate the clinical activity of afatinib in patients with the uncommon mutations with an overall response rate (ORR) of 71.1% and median PFS of 10.7 months 32 .
However, we noted that IC 50 values of osimertinib in YU-1092 cells and YU-1074 cells were comparable to the reported mean plasma concentration in patients receiving osimertinib (≈120 nmol/L), suggesting a potential of osimertinib against these mutations 55 . Consistent with our preclinical findings, osimertinib was shown to achieve an ORR of 60% in 5 patients with NSCLC harboring uncommon EGFR mutations (G719X, G719X/ S768I, and L861Q) 56 . Additionally, nazartinib, a third-generation EGFR-TKI, also demonstrated preclinical activity against major variants of EGFR exon 20 insertions (D770_N771insSVD, V769_D770insASV, and H773_ V774insNPH) 57 . More recently, a patient with lung adenocarcinoma harboring EGFR exon 20 insertion (S768_ D770dup) was shown to respond to osimertinib 58 . Together, patients with NSCLC harboring an EGFR L861Q or D770_N771insG mutation may respond to osimertinib.
To date, heterogeneous mechanisms of osimertinib resistance have been reported 5 . Our data suggest that EGFR C797S-mediated resistance can be overcome by a combination of brigatinib and cetuximab, consistent with the previous finding 35 . Recent study has shown that overexpression of AURKA and its upstream TPX2 confers resistance to osimertinib and rociletinib 41 . Indeed, we found that combined inhibition of EGFR and AURKA is efficacious in YU-1089 cells that were established from patient tumor progressing to olmutinib (Fig. 4). It is plausible that YU-1089 cells responded to tozasertib and alisertib due to elevated expression of AURKA 41,42 .
We observed that EGFR-TKI treatment in EGFR-mutant PDCs increases total EGFR protein (Fig. 3D,F and 4C). Previous studies and ours imply that this phenomenon may be common among EGFR-TKI resistant cell lines, although a molecular mechanism behind the phenomenon remains unclear 34,35 . It is well established that inhibition of receptor tyrosine kinase (RTK) signaling pathway causes temporary relief of RTK-dependent negative feedback mechanisms, resulting in a rebound in RTK expression or downstream signaling activation 59,60 . EGFR signaling is regulated by various EGFR inducible negative regulators such as LRIG1, MIG6, SOCS4, and SOC5 61 . Furthermore, LRIG1 and MIG6 are overexpressed in EGFR-mutant NSCLC cell lines and function as a negative regulator of EGFR signaling [61][62][63] . These findings may suggest a possible involvement of EGFR inducible negative regulators in EGFR upregulation after EGFR-TKI treatment. Further studies are required to investigate mechanisms of the EGFR rebound and its relationship to EGFR-TKI resistance 59,60 .
This study had several limitations. Previous studies have demonstrated that long-term culture of patient-derived models results in accumulation of somatic mutations and subclonal selection. Occasionally, these genetic drifts may functionally impact drug sensitivity 26,27 . PDCs in our study varied in their growth rates and time they take to achieve high tumor purity (Supplementary Table 3). We observed that majority of PDCs at early passages (1 to 8, median passage number of 3) were contaminated with fibroblasts (0%-51.9%, median value of 3.94%) in line with previous findings (Supplementary Table 1) 8 . Because of differential trypsinization, fibroblasts did not overgrow tumor cells, although fibroblast contamination generally resulted in additional cell passaging and a delay in functional tests (Supplementary Tables 1 and 3). Particularly, drug testing in 1 PDC was only available after 20 passages, which may not well represent patient tumor. Therefore, improved culture conditions should be tested in M+/TCF+ malignant effusions to accelerate tumor growth and turn-around time for functional assays. We also acknowledge that the presented therapeutic strategies should be validated in prospective clinical studies.
In summary, we streamlined a protocol for establishing PDCs and showed that these PDCs can be valuable preclinical platforms for designing therapeutic strategies.

Data availability
Materials and data are available upon reasonable request to corresponding authors.