Co-clinical Trial of Novel Bispecific Anti-HER2 Antibody Zanidatamab in Patient-Derived Xenografts

Therapies targeting HER2 have improved oncologic outcomes for patients with HER2-positive breast and gastric cancer, and are being studied across a variety of other tumor types with HER2 expression/amplification (1). As comprehensive genomic profiling has become commonplace in modern oncology practice, it is apparent that many tumors may be susceptible to HER2-targeted therapies, which has enabled their study in biomarker-driven, tumor-agnostic basket trials (2, 3). Although the monoclonal anti-HER2 antibody trastuzumab was the first agent to gain Federal Drug Administration approval for HER2-positive breast and gastric cancers, several novel anti-HER2 agents are currently being investigated, including small-molecule tyrosine kinase inhibitors, antibody–drug conjugates (ADC), and bispecific antibodies (4). Zanidatamab (ZW25, Jazz/Zymeworks) is a bispecific anti-HER2 antibody with biparatopic binding, which simultaneously binds to two extracellular domains (ECD) on the HER2 receptor, including the ECD4 (juxtamembrane) and ECD2 (dimerization) domains (5). Zanidatamab was recently tested in a first-in-human, multicenter, phase I, dose-escalation and expansion trial, to test the safety and tolerability of zanidatamab monotherapy and in combinations in HER2-expressing solid tumors (NCT02892123). The study demonstrated that zanidatamab was well tolerated and showed signals of activity in many tumor types. Notably, in the non-breast, non-gastric cancer expansion, 31 [37%; 95% combination index (CI), 27.0–48.7] of 83 evaluable patients had a confirmed objective response (6). Based on these results, zanidatamab progressed to pivotal clinical trials. Recently, the antitumor activity of zanidatamab was demonstrated in the HERIZONS-BTC-01 trial (NCT04466891) where confirmed objective responses were observed in 33 (41.3%; 95% CI, 30.4–52.8; ref. 7) of 80 patients with HER2-positive biliary tract cancer.

The paradigm of reverse translational research, or “bedside-to-benchtop” research, is a concept in which patient experiences and real-world data can inform benchtop research, in a cyclical process, to allow for rapid translation of research findings (8). Phase I clinical trials provide a unique opportunity for correlative studies which can facilitate a better understanding of novel anticancer therapies to guide clinical development. One method of investigation is through PDX models, which serve as one of the most patient-relevant tools for cancer research, as they accurately reflect the molecular profiles of patient tumors (9). Although PDX models are used extensively in translational oncology research, a remaining challenge is determining how responses observed from in vivo experiments correlate to clinical trial findings (10, 11). As many trials incorporate pre- and/or posttreatment biopsies for biomarker studies, this setting provides a valuable opportunity for PDX development (12). Likewise, for targeted therapies in clinical development, PDX models serve as a way to evaluate mechanisms of intrinsic and acquired resistance and facilitate the study of strategies to overcome these.

We developed PDX models from patients who were enrolled in or screened for the phase I study of the novel HER2 antibody zanidatamab (NCT02892123). We hypothesized that the generation of PDX models from the zanidatamab phase I study would enable “co-clinical trial” modeling and allow us to establish the association between patient and PDX responses, thus enabling clinically relevant treatment studies and broader biomarker analysis. We also hypothesized that biomarker-response assessments in PDXs and the development of PDXs from patients who develop acquired resistance can provide insights into mechanisms of resistance to zanidatamab.

We sought to establish PDX models from patients who were enrolled in or who underwent pretreatment biopsy for HER2 screening for the zanidatamab phase I trial (Fig. 1A). In total, samples from 36 tumors were implanted for PDX development, with a total of 19 models successfully established from 17 different patients (take rate of 52.7%). Most PDXs were developed from image-guided biopsies (18/19, 94.7%), and the other was developed from a surgically resected specimen (brain metastasis). There were 15 pretreatment PDXs developed and 4 posttreatment PDXs developed. The median time from implantation until the first passage was 111 days (range, 35–378 days). Patient and PDX characteristics are shown in Table 1.

We compared survival outcomes for patients who had pretreatment biopsies implanted for PDX development to assess if there was an association between PDX take rate and clinical outcomes. Median progression-free survival (PFS) was 58 days for patients who had biopsies that the successful “took” (i.e., grew) in mice compared with 112 days for patients whose biopsies did not take (log-rank test, P = 0.049; Fig. 1B). A similar trend was observed for overall survival (OS). Patients whose PDXs took had a median OS of 92 days versus 574 days for patients whose tumors did not take (log-rank test, P = 0.003; Fig. 1C). The cohort of PDXs which were successfully established represented a broad range of HER2-expressing cancers, including 4 colorectal, 4 esophageal, 3 gastric/gastroesophageal junction (GEJ), 3 breast, 3 gallbladder, 1 cholangiocarcinoma, and 1 small bowel cancer (Fig. 1D). There was a higher take rate observed for PDXs derived from gastrointestinal (GI) tumors (64.0%) compared with non-GI tumors (27.3%).

We then evaluated the clinical responses of the 17 patients who had PDXs that were successfully established from pretreatment biopsies. Of these 17 patients, 14 patients received zanidatamab and had a response assessment by Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 (13). Of this cohort, there were 5 (33.3%) patients with partial response, 4 (26.7%) with stable disease, and 5 (33.3%) with progressive disease, with best responses (change from baseline in sum diameter of target lesions) ranging from −52% to +26% (Fig. 1E). Most PDXs were from patients who enrolled in the zanidatamab monotherapy arms (either 10 mg/kg or 20 mg/kg), aside from a patient (corresponding to PDX.003.263) which enrolled in the zanidatamab + paclitaxel arm and a patient (corresponding to PDX.003.204) who enrolled in the zanidatamab + capecitabine arm.

As HER2 expression/amplification were inclusion criteria for the zanidatamab phase I study, we sought to investigate the concordance of HER2 expression in patients with corresponding PDX samples (Supplementary Fig. S1). Both IHC and fluorescence in situ hybridization (FISH) were performed on separate patient biopsies taken at the same time as the biopsy for PDX development. Patient next-generation sequencing (NGS) results obtained on prior local testing was obtained from a retrospective data review and shown in Supplementary Table S1 and Supplementary Fig. S1. Of 15 patients who had pretreatment biopsies for PDX development, there were 10 who had exact concordance on HER2 IHC. Of the 5 patients who were discordant, there were 3 who were IHC 3+ on patient biopsy with PDXs that were 2+, one that was IHC 3+ whose PDX was 1+, and one that was 2+ whose PDX was 3+. Interestingly, the PDXs generated from the 7 patients who demonstrated response or stable disease with zanidatamab were all HER2 IHC 3+ and HER2 amplified by NGS, whereas there was more variability in HER2 status among the 8 PDXs arising from pretreatment biopsies from patients who progressed (clinically or radiographically): two IHC 3+, four 2+ and two 1+.

To further investigate the molecular profiles of PDX models generated on the zanidatamab phase I study, we performed whole-exome sequencing (WES) on early passage PDXs. Corresponding patient clinical data, including prior lines of therapy, primary tumor type, and best response on the zanidatamab phase I trial, were reported along with WES results assessing for mutations, indels, and copy-number variations in cancer-related genes (Fig. 2). Of the 19 PDXs which were assessed, 16 had ERBB2 amplification by WES (4 also had ERBB2 mutations). We also identified frequent coalterations in cancer-related genes, including CDK12, RARA, and TOP2A genes on the same chromosome as ERBB2 (Supplementary Table S1).

In eight of the patients, frozen fine needle aspirate (FNA) samples from the PDX biopsies were available. We performed WES on these 8 samples and compared genomic alterations in the first passage PDX to the biopsy (Supplementary Fig. S2). Although it took a while to obtain PDXs from tumor biopsies, raising concern about the emergence of genomic drift, most therapeutically actionable single nucleotide variants (SNV) and significant copy number variants (CNV) were concordant between the matched samples. Most discordances were seen in CNVs, suggesting the increase tumor cellularity in the PDX versus FNA may contribute to discordances. Most discordances in SNV were seen in cases when patient FNA was found to have a lower number of alterations pointing to the possibility discordance was due to lower tumor cellularity in the FNA as well. Interestingly, TP53 mutations were observed in WES of PDX.003.256 and PDX.003.309 and not in matched biopsies (Supplementary Fig. S2); however, TP53 mutations had been reported in prior Clinical Laboratory Improvement Amendments (CLIA) NGS testing on earlier tumor samples on both of these patients (Supplementary Fig. S1), suggesting that this potential discordance is not an acquired mutation in the PDX but rather represents analytic differences or differences due to tumor heterogeneity.

We then proceeded with in vivo testing of zanidatamab at three different doses, using this cohort of PDX models. We selected models developed at both pretreatment/screening and postprogression. We first tested PDXs generated from biopsies obtained from biopsy at the time of HER2 screening: one PDX was generated from a patient that did not enter treatment, three from patients who had a partial response and two from those who had best response of progressive disease (Fig. 3A–F; Supplementary Table S2). We first tested PDX.003.085, a model developed from a patient with extrahepatic cholangiocarcinoma. This patient ultimately did not enroll in the trial due to declining performance status. This model was HER2 3+ on IHC and had an ERBB2 S310Y mutation and ERBB2 amplification. This model demonstrated dose-dependent antitumor activity of zanidatamab. Following treatment with zanidatamab at a lower dose of 4 mg/kg, PDX.003.085 exhibited some tumor growth inhibition but continued to grow (event-free survival [EFS-2] 28.5 vs. 16.5 days, P = 0.034). In contrast, the higher dose of 8 mg/kg and 16 mg/kg zanidatamab led to durable tumor regression [EFS-2 not estimable (>42 days) vs. 16.5 days, P = 0.007; Fig. 3A).

We next tested PDXs developed from pretreatment biopsies of patients with known responses to zanidatamab. This included three models from patients who had a partial response as the best response, and two patients who had progressive disease. We selected a GEJ adenocarcinoma model, PDX.003.256, which was developed from a patient who had a confirmed partial response (best response of −33%) to zanidatamab. PDX.003.256 was HER2 3+ by IHC and ERBB2 amplified by NGS. This patient had previously been treated with several lines of prior therapy including 5-FU + oxaliplatin + trastuzumab, paclitaxel + ramucirumab, pembrolizumab, and radiation. Zanidatamab, at all dose levels (4, 8, and 16 mg/kg), demonstrated antitumor activity in this model, as demonstrated by no tumors in any zanidatamab treatment arm doubling in size by day 82 [EFS-2 not estimable (>81 days) vs. 16.5 days, P = 0.017; Fig. 3B]. Similarly, we tested PDX.003.230, which was developed from a gastric tumor that had a partial response to zanidatamab (best response = −31%). PDX.003.230 was HER2 3+ and ERBB2 amplified by NGS. The PDX showed some growth inhibition with both 8 mg/kg and 16 mg/kg but survival was only significantly extended with 16 mg/kg [EFS-2 not estimable (>38 days) vs. 11 days, P = 0.006; Fig. 3C]. We next tested a model developed from a patient with colorectal cancer, PDX.003.019, with HER2 IHC 3+ expression and ERBB2 amplification on NGS, who had a partial response. This model had growth inhibition with all three doses of zanidatamab but had disease stabilization with only 16 mg/kg [4 mg/kg; EFS-2 10 days vs. 35 days, P = 0.004; 8 mg/kg: EFS-2 10 days vs. 25 days, P = 0.002; 16 mg/kg: 10 days vs. EFS-2 not estimable (>42 days), P = 0.002; Fig. 3D]. As the patient had a partial response while the PDX only had disease stabilization, we reviewed the clinical records for more details. It was noted that this patient had stable disease for the first two restaging scans, with a delayed partial response noted.

We next tested models from pretreatment biopsies from patient tumors who had the best response of progressive disease on the trial. We first tested PDX.003.045, which was developed from a patient with breast carcinoma who had progressive disease (best response +26%) when treated with zanidatamab. PDX.003.045 was HER2 2+ by IHC and ERBB2 amplified. Of note, this patient had received several prior lines of HER2-targeted therapy, including docetaxel + carboplatin + trastuzumab + pertuzumab, trastuzumab emtansine (T-DM1), and trastuzumab deruxtecan (T-DXd). This model showed a trend toward growth inhibition with 8 and 16 mg/kg zanidatamab, but this model demonstrated progressive disease, without a statistically significant increase in time to tumor doubling with zanidatamab at any dose range (Fig. 3E).

We next tested PDX.003.326 developed from a cholangiocarcinoma patient who had progressive disease (best response +12%, with unequivocal nontarget progression). PDX.003.326 was HER2 2+ by IHC. Interestingly, the PDX responded to all three doses of zanidatamab [EFS-2 not estimable (>78 days) vs. 7 days, P = 0.018; Fig. 3F]. This was the most discordant of the models tested; interestingly, this model was the only model tested with loss of ERRB2 amplification compared with patient tumor in the NSG analysis, with limited concordance of genomic alterations on WES except for a preserved IDH1_R132 L mutation (Supplementary Fig. S2), and further, the patient CLIA genomic testing on prior archival sample had several additional alterations such as CDKN2A and CDKN2B deletion, and TSC2 truncation that were not detected from on NGS of the pretreatment biopsy and the PDX that was tested. Also of note, reviewing the clinical data, this patient has stable disease in their target lesions overall, but with heterogeneity in antitumor activity, with two target lesions growing, two decreasing in size, but with unequivocal progression in non-target lesions (lung lesions).

Next, we sought to investigate the efficacy of zanidatamab in PDX models developed from biopsies following the progression on zanidatamab (Fig. 3G–I). Moreover, these “acquired resistance” models were generated after the patients initially had a RECIST partial response on zanidatamab before progressing, after which a postprogression biopsy was collected for PDX generation. We first tested PDX.003.285, a HER2 IHC 3+ model with ERBB2 amplification on NGS, developed from a patient with gallbladder adenocarcinoma who achieved a prolonged partial response on zanidatamab (best response: −44%). Notably, prior to enrollment on the zanidatamab trial, this patient had previously received 8 lines of therapy, all of which included trastuzumab as a component. This postprogression PDX model did not exhibit sensitivity to zanidatamab at any dose tested (Fig. 3G). Additionally, we tested PDX.003.300, a model also developed from a patient with gallbladder adenocarcinoma at the time of progression, following a prolonged partial response with zanidatamab (best response: −52%). PDX.003.300 was HER2 1–2+ by IHC with ERBB2 amplification on NGS. The model did not exhibit sensitivity to zanidatamab at any dose level (Fig. 3H). We also tested PDX.003.094, a model developed from a patient with colorectal cancer with HER2 IHC 3+ expression and ERBB2 amplification on NGS. PDX.003.094 was developed from a resected brain metastasis from a patient who had prolonged treatment with zanidatamab with an initial partial response, but subsequent progression due to the development of a brain metastasis. This model was not sensitive to zanidatamab at any dose level (Fig. 3I).

In summary, we demonstrated zanidatamab antitumor activity in all three models that were tested from patients who were zanidatamab sensitive. We demonstrated zanidatamab resistance in all acquired resistance models. Of the two patients with PD as the best response, the PDX of one patient progressed but another responded to zanidatamab; but notable that this latter patient had a mixed response clinically. A summary of the statistical analysis of the PDX experimental data can be found in Supplementary Table S2.

A challenge with HER2-targeted therapies is that while often a promising signal of activity is observed in clinical trials, essentially all patients ultimately progress on therapy, often due to the development of acquired resistance (14, 15). We therefore developed four PDX models from postprogression biopsies; all four were still HER2 positive, all were HER2 amplified and two had MET amplifications.

PDX.003.300 was developed from a patient with gallbladder adenocarcinoma who had a confirmed clinical response (PR, best response -52%) with zanidatamab, but ultimately the tumor progressed after 280 days on trial. The PDX model was developed 3 months following the demonstration of progression on zanidatamab but prior to initiating other lines of therapy (Fig. 4). We also had a pretreatment PDX model from this patient, PDX.003.148, and compared the molecular profiles between the pretreatment and postprogression PDX models. PDX.003.148, which was developed from the pretreatment biopsy, was HER2 IHC 3+, whereas PDX.003.300, which was developed from the postprogression biopsy, was HER2 IHC 1–2+. Both pretreatment and postprogression PDXs had ERBB2 amplification on NGS as well as JAK2 S1032F and TP53 R248Q mutations. WES of the postprogression model (PDX.003.300) revealed new amplifications of both MET and MYC (estimated copy numbers of 34 and 85, respectively). Notably, MET and MYC amplification was not detected in the clinical NGS testing the patient had performed prior to study enrollment (Supplementary Figs. S1 and S3). MET and MYC amplifications were confirmed on NGS performed on a postprogression patient biopsy samples obtained at the same time as the sample for PDX generation (Fig. 4; Supplementary Fig. S2).

To confirm these findings, we performed biomarker assessment of c-MET and c-MYC by IHC and FISH (Fig. 4). When comparing MYC expression and copy number between PDX.003.148 (pretreatment) and PDX.003.300 (postprogression), we found that the percentage of c-MYC staining was 28% (H-score 32) pretreatment versus 48% (H-score 76) postprogression and that MYC copy number was normal pretreatment versus amplified postprogression by FISH. More strikingly, when we reviewed the MET expression and copy number, we found that the percentage of c-MET staining was 26% (H-score 36) pretreatment versus 98% (H-score 213) posttreatment and that MET copy number was normal pretreatment versus amplified posttreatment by FISH.

Based on these findings suggesting that amplification/overexpression of MET may be a mechanism of acquired resistance to zanidatamab, we sought to assess combinatorial strategies with MET inhibitors (capmatinib and crizotinib). To assess if the expression of MET affected sensitivity to zanidatamab, we first tested three HER2-positive breast cancer cell lines with different levels of MET expression. We found that a cell line with high MET expression (HCC-1954) did not respond to zanidatamab, whereas cell lines with low MET levels (BT-474 and SK-BR-3) displayed much smaller IC₅₀ values at 2.35 and 0.7 μg/mL, respectively (Fig. 5A). We also found that BT-474 and SK-BR-3 cell lines with low MET levels were less sensitive to MET inhibitor crizotinib than HCC-1954 cells with high MET expression (Supplementary Fig. S3). Next, we evaluated the combinatorial efficacy in these cell lines. Cell viability assay showed that combinations of zanidatamab with MET inhibitors capmatinib or crizotinib demonstrated synergistic antitumor cell efficacy in all three HER2⁺ cell lines, with CI ranging from 0.1 to 0.5 (Fig. 5B). The enhanced therapeutics was also seen in cell colony formation assay (Fig. 5C). Moreover, flow cytometry results in HCC-1954 cells with Annexin V staining showed that the apoptotic effect induced by single drug treatment with zanidatamab and MET inhibitor (capmatinib or crizotinib) was significantly enhanced by their combination (Fig. 5D and E).

We then assessed the efficacy of these combinations on PDX.003.300, the model developed from the patient whose tumor developed amplification/overexpression of MET and MYC following progression on zanidatamab. We tested zanidatamab at 16 mg/kg in combination with two MET inhibitors (capmatinib or crizotinib). The MET inhibitor crizotinib had single-agent activity at the dose tested (EFS-2 not estimable vs. 11 days, P = 0.001) without an apparent enhancement by the addition of zanidatamab (Fig. 5F). Similarly at the high dose of capmatinib, both as a single agent and in combination with zanidatamab led to prolonged tumor regression (EFS-2 not estimable vs. 11 days, P = 0.002; Fig. 5G). To dissect the combinatorial effect of zanidatamab and MET inhibition, we then repeated the experiment with lower doses of capmatinib. At 6 and 9 mg/kg, capmatinib + zanidatamab showed improved efficacy over either single-agent EFS-2 (6 mg/kg: not estimable vs. 8 days, P = 0.002, 9 mg/kg: not estimable vs. 16 days, P = 0.08; Fig. 5H).

Preclinical studies have demonstrated that MYC transcription is regulated by BET bromodomains and that BET inhibitors (such as JQ1) can suppress MYC transcription and deplete c-Myc protein expression (16). Considering the new amplification of MYC, we thus tested the efficacy of JQ1 alone and in combination with zanidatamab in PDX.003.300. JQ1 had limited activity in monotherapy and in combination with zanidatamab (Supplementary Fig. S4).

Although xenograft models are used extensively in precision oncology research to better understand predictors of therapeutic response and identify mechanisms of intrinsic/acquired resistance, a major gap in knowledge is how PDX response correlates to observed clinical responses. In the present study, we prospectively developed PDX models from pretreatment biopsies of patients with HER2-expressing/amplified solid tumors enrolled in the zanidatamab phase I study. We successfully established 19 PDX models from 17 individual patients for an overall take rate of >50%, representing a diverse cohort of 7 different HER2-expressing solid tumors. We found that both PFS and OS durations were shorter in patients with biopsies from which PDXs were established versus those who did not take, suggesting that the PDXs developed represent more aggressive tumors. This finding is aligned with a prior study from our group, where we observed that patients with breast cancer whose tumors developed PDXs upon implantation of surgical samples had lower recurrence-free survival and OS (17). However, in the current study, although the growing PDXs may represent more aggressive tumors, we were able to develop PDXs from patients with differing antitumor responses in the trial. Notably, zanidatamab had in vivo antitumor activity in many of the PDXs generated in baseline biopsies, the postprogression models tested were not sensitive to zanidatamab at any dose level. We additionally identified MET and MYC overexpression/amplification as potential mechanisms of acquired resistance to zanidatamab and provide evidence for rational combinations of zanidatamab with MET inhibitors as a potential strategy to overcome this. Taken together, these findings suggest that PDXs can be prospectively established in clinical trials and provide an opportunity to better understand investigational agents and their potential mechanisms of resistance.

An ongoing challenge in PDX translational studies is the question of how PDX experimental results correlate to patient responses regarding an investigational agent. Several prior studies have proposed the concept of “co-clinical trial” or “mouse hospital” as a means for therapeutic discovery in cancer research, considering how well PDX molecular profiles recapitulate patient tumor profiles (9–11, 18). We prospectively established PDX models from pretreatment biopsies on a phase I study with the purpose of correlating clinical responses to in vivo experimental results. We found that zanidatamab demonstrated robust regression in PDX developed from screening biopsy in a patient who unfortunately had a decline of their performance status before they could be treated on trial. Zanidatamab demonstrated significant growth inhibition in models developed from patients who had a partial response on trial and zanidatamab had more limited activity in models developed from patients with progressive disease or in models from postprogression biopsies (Supplementary Table S3). PDX.003.019 had prolonged EFS with treatment, but the activity in the PDX seemed less robust than in the patient. However, reviewing this case, the patient also initially had stable disease, raising the possibility that zanidatamab may have additional mechanisms of action such as immune effects, that are not captured by our PDX work in immune-deficient mice. In contrast, in one model, PDX.003.326, we observed significant antitumor activity with zanidatamab at all doses, but the patient demonstrated progressive disease. Notably, this patient had stable disease in their target lesions, with an increase in some and a decrease in some target lesions, but unequivocal progression in the nontarget lesions and clinical deterioration. Thus, this case, may in part represent a discordance attributable to tumor heterogeneity.

Other studies have suggested that PDX responses to chemotherapy/targeted therapies may be correlated with patient responses, and have proposed their preclinical use as a screening tool for therapy (10, 19, 20). Although the approach of using PDXs as avatars for treatment selection is complex and difficult to scale, our finding that antitumor efficacy in PDXs and patients is similar gives strength to the idea of using PDXs in basic and translational research as a clinically relevant model. As zanidatamab is an agent currently in clinical development, these findings add to the body of knowledge of this agent and may permit additional modeling of rational combinations to overcome intrinsic/acquired resistance. Furthermore, this study allowed for the establishment of a broad cohort of HER2-expressing PDXs of a wide variety of tumor types, enabling further study with other HER2-targeted agents to perform a tumor agnostic PDX “basket trial.”

HER2-targeted therapies have dramatically improved the oncologic outcomes for patients with HER2-expressing and/or amplified tumors; however, resistance remains a therapeutic challenge (15). HER2 expression is emerging as an important determinant of response for antibody-based therapies (21). Notably in the recent HERIZON-BTC-01 trial of patients with HER2-amplified BTC treated with zanidatamab, the objective response rate among patients who had HER2 IHC 3+ tumors was 51.6%, whereas it was 5.6% in patients who were 2+, and there were no responses among the seven patients who were HER2 amplified but IHC 1+ or 0. In this study, we found that the PDXs generated from the 7 patients who demonstrated response or stable disease with zanidatamab were all HER2 IHC 3+ and amplified by NGS, whereas there was more variability in HER2 status among the 8 PDXs arising from pretreatment biopsies from patients who progressed (two 3+, four 2+, and two 1+), suggesting that tumor heterogeneity in HER2 expression may be a contributor to zanidatamab resistance. Interestingly, the postprogression PDX, PDX.003.300, also had lower HER2 expression (1/2+) in comparison with the pretreatment model; this may also be a contributing factor in the acquired resistance to zanidatamab.

As the landscape of HER2-targeted therapies is vast and continues to evolve, unique mechanisms of resistance (both intrinsic and extrinsic) have been identified for several different HER2-directed monoclonal antibodies, including impaired binding to HER2, altered intracellular signaling (through downstream/bypass pathways), cell-cycle regulation, and metabolic factors, among others (15). In the present study, we confirmed the amplification of MET and MYC following zanidatamab treatment in a patient who had PDX developed from both a pretreatment and postprogression biopsy, and that the postprogression PDX was not sensitive to zanidatamab. Interestingly, coamplification of MET in ERBB2-amplified esophagogastric cancer has been identified as a potential mechanism of acquired resistance to the irreversible pan-HER kinase inhibitor afatinib in both preclinical and clinical studies (22, 23). Another study identified MET amplification in 2 of 20 patients with HER2-positive metastatic gastric cancer who were resistant to trastuzumab (24). In contrast, Saeki and colleagues primarily noted loss of HER2 expression with treatment and did not observe an increase in MET amplification among 33 patients who progressed after trastuzumab-containing therapy for HER2-positive gastric cancer (25). However, they reported that 17.9% of patients had MET amplification at baseline; this demonstrates the need to understand the interplay between MET amplification and HER2-inhibitor sensitivity. Our in vitro studies identified improved efficacy in terms of inhibiting cell growth with the combination of crizotinib/capmatinib and zanidatamab. We found robust in vivo activity of capmatinib and crizotinib, and demonstrated that lower doses of capmatinib and crizotinib are able to enhance the activity of zanidatamab in vivo. Taken together, these data suggest targeting MET inhibition alone or in combination with zanidatamab may represent a potential opportunity to overcome acquired resistance to zanidatamab.

There are several limitations to the present study. First, although our take rate was over 50%, PDXs were not able to be generated from all patients evaluated. With the standardization of PDX practices, take rates are often above 20%, with many series developing PDXs from surgical specimens where larger samples are available for implantation, the take rate here compares favorably with what has been observed in prior studies (26). We found that PDXs that were taken had worse OS/PFS durations, indicating that PDX development does have a selection bias and our cohort of models represents more aggressive disease, consistent with our prior observations in breast cancer PDX models (17). This is also consistent with our finding that when we developed PDXs from longitudinal biopsies, the latter PDX had a shorter time to passage, suggesting they were more aggressive (PDX.003.148 vs. PDX.003.300: 238 vs. 148 days, respectively, and PDX.003.019 vs. PDX.003.094 197 vs. 65 days, respectively). However, one can argue that this is an advantage as the more aggressive cases need more intensive modeling to identify mechanisms of intrinsic and acquired resistance. We also found that our PDX take rate was higher in those derived from GI tumors, which limits our ability to assess response correlations in other HER2-expressing tumor types. Second, although patients had HER2 IHC and FISH performed on biopsies obtained synchronously as PDX development, we compared WES performed on the PDX with clinical NGS data available from prior testing; this limited our ability to directly compare patient PDX genomics. Third, this study was carried out in a relatively heavily treated patient population, many of which had been treated with other targeted therapies, some with HER2-targeted therapy and all treated at a single comprehensive cancer center. Further study is needed to determine the full range of mechanisms of intrinsic and acquired resistance.

These data have several implications moving forward. This study provides compelling evidence for the prospective establishment of PDXs in phase I clinical trials, and that this is feasible using samples from image-guided biopsies. Additionally, these data support the convention that the molecular profiles of PDXs reflect the patient's tumor characteristics, and in the cohort tested here, the in vivo responses to zanidatamab which were observed were similar to patients’ responses on trial. Lastly, utilizing genomic data from pre- and posttreatment PDXs, we were able to identify a novel mechanism of acquired resistance to zanidatamab and develop therapeutic strategies to overcome this, which further supports the use of PDX models to study intrinsic and acquired resistance mechanisms in precision oncology and early drug development.

Patients with advanced/metastatic HER2-expressing tumors who were evaluated for the zanidatamab phase I study consented to an image-guided biopsy as a component of pretrial screening to have central evaluation of HER2 status; this treatment and pretreatment screening protocol was approved by the Institutional Review Board at the University of Texas MD Anderson Cancer. In addition, patients prospectively provided written informed consent for an additional biopsy for PDX development, under a separate investigator-initiated protocol approved by the Institutional Review Board at MD Anderson. Patients underwent a biopsy and, simultaneously, patients had a second image biopsy for the purposes of PDX development. Studies were conducted in accordance with the Declaration of Helsinki and the U.S. Common Rule. Demographic, clinical, and pathologic characteristics were retrospectively abstracted from the electronic medical record and a prospectively maintained clinical genomic database.

PFS was determined in patients who had image-guided biopsies for pretrial screening and was calculated as time (days) from cycle 1 (C1) day 1 (D1) to radiologic progression or death from any cause. OS was determined in the same cohort as time (days) from C1D1 to death from any cause. Data were censored at the time of last contact if the endpoints of progression or death had not yet been met. The Kaplan–Meier method was used to estimate median survival durations and the log-rank test was used for statistical comparisons, with P < 0.05 considered statistically significant. Radiologic responses for patients were reported as percent change from baseline in the sum diameter of target lesions. Clinical responses were reported using RECIST 1.1 criteria (13).

Cells were seeded in 96-well plates at densities of 0.15–0.6 × 10⁴ cells/100 μL per well in triplicates for each treatment dose. After adhering overnight, 100 μL of serially diluted drug solutions (single agent or combination) were added. Cells were incubated at 37°C for 72 hours. Cells were then fixed with 50% trichloroacetic acid (TCA) followed by staining with 0.4% sulforhodamine B (SRB) solution. OD values were read at 490 nm by plate reader Synergy 4 (BioTek). The half maximal inhibitory concentration (IC₅₀) was determined using CalcuSyn software (Biosoft). To evaluate combination efficacy, CI was determined based on Chou–Talalay IC₅₀ model using CalcuSyn. CI < 1.0 (curve left-shift): synergistic; CI = 1.0, additive; CI > 1.0 (curve right-shift): antagonistic).

Cells were seeded in 6-well plates at a density of 1,000 cells per well in triplicates for each treatment group. The next day, cells were treated with single drugs or a combination at different concentrations. The culture medium was changed with fresh drugs twice a week. Cells were cultured for 3 weeks. Cell colonies were then fixed in 10% formalin and stained with 0.05% crystal violet in 25% methanol. The stained colonies were imaged, and the total colony area was quantitated using NIH ImageJ v.1.48 software.

Cells were seeded in 6-cm plates at a density of 3 × 10⁵ cells per well in triplicates for each treatment group. The following day, cells were treated with a single agent or combination. After 72 hours, floating and attached cells were collected. Using the Annexin V–FLUOS Staining Kit (Roche), cells were stained with Annexin V fluorescence and propidium iodide. Samples were analyzed by flow cytometry at The Flow Cytometry and Cellular Imaging Core Facility at MD Anderson Cancer Center. The percentage of Annexin V–positive apoptotic cells was calculated.

Cells were washed with cold PBS and lysed in 1× Laemmli buffer. The protein concentrations in the cell lysates were measured using a Pierce BCA protein assay Kit (Thermo Fisher). The protein samples were loaded into SDS-PAGE gel, followed by transferring proteins to a 0.2-μm nitrocellulose membrane (Bio-Rad Laboratories). Membranes were blocked with blocking buffer Blocker Casein in PBS (Thermo Fisher) at room temperature for 1 hour, followed by immunoblotting with the primary antibodies in 5% BSA in TBST buffer at room temperature overnight. After washing with TBST buffer, the immunoblotting membrane was then probed with the secondary antibodies with fluorescence conjugation. The immunoblots were visualized and the immunoblotting signal intensity was quantitated using the Odyssey IR imaging system (Li-Cor Biosciences).

Animal experiments were approved by the Institutional Animal Care and Use Committee at the University of Texas MD Anderson Cancer Center. Tumor fragments from image-guided or surgical biopsies were implanted into the flanks of highly immunodeficient NSG mice followed by passaging into athymic nu/nu mice for PDX development. Confirmation of tumor origin was performed using short-tandem repeat (STR) DNA fingerprinting as described previously (17). PDX tumors were screened with lymphocyte common antigen CD45 IHC to confirm the absence of human lymphocytic infiltration (27). Once early passage PDX tumors were of adequate size, they were passaged into athymic nu/nu mice for experimental testing. Treatments were initiated when tumors were approximately 200 to 400 mm³ in size. Mice were euthanized when the experimental endpoint was reached or when tumor volume reached 2,000 mm³.

Formalin-fixed paraffin-embedded (FFPE) cell lines/tissue of 4-μm thickness slides were used to perform all IHCs using Leica Bond III Auto Stainer (Leica Biosystems) except HER2 that was performed using Ventana Benchmark ULTRA IHC/ISH platform (Roche). In brief, after deparaffinization with bond dewax (cat no. AR9222; Leica Biosystems) solution followed by heat-mediated epitope retrieval [Bond TM Epitope Retrieval Solution 1&2 (Leica Biosystem) cat nos. AR9961 and AR9640, respectively] were blocked by peroxidase blocking (3% H₂O₂) for 20 minutes. After blocking, slides were incubated for 15 minutes with the following primary antibodies. All the antibodies were detected using a Bond polymer Detection kit (cat no. DS9800, Leica Biosystem) with diaminobenzidine (DAB) as a chromogen. Supplementary Table S4 provides detailed antibody information. Subsequent transient washing, postprimary, and polymer incubation, followed by hematoxylin staining, were performed. Finally, slides were air dried, and cover slipped with mounting media (Cytoseal XYL, Thermo Scientific). and digitalized in an Aperio AT2 scanner (Leica Biosystems; Vista) under a 20× objective magnification.

For the assessment of HER2 status in patients, both HER2 IHC and FISH were performed on pretreatment biopsies taken in the same setting as the biopsy for PDX development. ERBB2 amplification status was obtained from archival NGS data in a prospectively maintained clinical genomic database. For the assessment of HER2 status in corresponding PDX models, tumors from early passage PDXs were collected, rapidly fixed in 10% neutral buffered formalin for 24 hours, and then washed with 70% ethanol. Tissues were FFPE by the MDACC Research Histology Core Facility. A tissue microarray (TMA) of HER2-amplified/expressing PDXs was made for the purpose of biomarker assessment. HER2 IHC was performed on TMA blocks (three sections per tumor) or individual tumor sections using the MDACC CLIA-certified lab and reviewed by a clinically trained pathologist. HER2 interpretation was performed with brightfield microscopy using the American Society of Clinical Oncology (ASCO)/College of American Pathologists guidelines for IHC and FISH as described previously (28). For all other markers, H-Scores (0–300) based on intensity and percentage were digitally obtained using HALO (Indica Labs) image analysis software under the supervision of a pathologist.

Fragments of flash-frozen PDX tissues were lysed in a buffer containing protease K and homogenized. DNA was then extracted using the Qiagen DNA Mini Kit per the manufacturer's protocol. Normal DNA was also purified from whole blood samples using Qiagen Blood Mini Kit per the manufacturer's protocol. WES of samples was performed using the MD Anderson Cancer Center Cancer Genomics Laboratory, NIH/NCI, or the Translational Research to AdvanCe Therapeutics and Innovation in ONcology (TRACTION) platforms at MD Anderson Cancer Center.

Zanidatamab was diluted in sterile water and delivered intravenous (i.v.) tail-vein injection twice per week in the following concentrations for monotherapy studies: 4, 8, and 16 mg/kg. For combination studies, zanidatamab was delivered at the maximal concentration used (16 mg/kg, i.v., twice per week) in combination with one of the following agents: JQ1 [50 mg/kg, intraperitoneal (i.p.), weekly (qw)], capmatinib [17.5 mg/kg, oral gavage (p.o.), twice per day (b.i.d.)], or crizotinib (100 mg/kg, p.o., qd).

Zanidatamab was provided in frozen vials as a gift from Zymeworks Inc for in vivo studies. Capmatinib (#CT-CAPM), crizotinib (#CT-CRIZS), and JQ1 (#CT-JQ1) were purchased from ChemieTek. The following antibodies were purchased for IHC staining of PDX tissues: CD45 (Agilent, no. M0701), HER2 (Ventana, no. 790-2991), c-MET (Ventana, no. 790-4430), and c-MYC (Ventana, no. 790-4628). Probes for MET and MYC FISH were purchased from Abbott Molecular/Vysis Products.

Statistical comparisons and figures were performed using Prism version 8 (GraphPad) and SPSS Statistics Version 24 (IBM). Tumor volume (Vt) was calculated as Vt (mm³) = [(width)² × length]/2 and percent change in Vt from baseline was calculated as (Vt, day 0 – Vt, day X)/Vt, day 0 × 100%. Relative treatment-to-control (T/C) ratio was calculated as (Vt, day 21/Vt, day 0)/(Vc, day 21/Vc, day 0), where t = treatment and c = control. Event-free survival (EFS-2) was defined as the day on which tumor volume doubled in size from baseline. Log-rank test was used to compare EFS-2 curves.

Actionable alteration data and patient characteristic data are included in the Supplementary Material. Detailed PDX whole exome and patient panel testing sequencing data will be made available from the investigator upon reasonable request.

T.P. DiPeri reports other support from the American Association for Cancer Research outside the submitted work. K.W. Evans reports grants from NCI PDXNET during the conduct of the study. A. Korkut reports grants from BostonGene outside the submitted work. E.E. Dumbrava reports grants from Bolt Therapeutics, Sanofi, Belicum Pharmaceuticals, grants and other support from Triumvira Immunologics, grants and personal fees from Mersana Therapeutics and Fate Therapeutics, and personal fees from Orum therapeutics outside the submitted work. S. Pant reports other support from Mirati Therapeutics, Lilly, Xencor, Novartis, Bristol-Myers Squibb, Astellas, Framewave, 4D Pharma, Boehringer Ingelheim, NGM Pharmaceuticals, Janssen, Arcus, Elicio, Biontech, Ipsen, Zymeworks, Pfizer, ImmunoMET, Immuneering, Amal Therapeutics, and other support from Zymeworks, Ipsen, Novartis, Janssen, AskGene Pharma, BPGBio, Jazz, AstraZeneca, Boehringer Ingelheim, USWorldmeds, Nihon Medi-Physics Co, Ltd, Alligator Bioscience, Theriva Biosciences outside the submitted work. J.A. Ajani reports personal fees from Jazz during the conduct of the study. P.R. Pohlmann reports personal fees from Frontiers, Pfizer, grants from Pfizer, Carisma Therapeutics, and Orum, other support from Seagen outside the submitted work; in addition, P.R. Pohlmann has a patent for United States Patent no. 9,745,377 issued and licensed to Immunonet Biosciences and a patent for United States Patent no. 8,501,417 issued and licensed to Immunonet Biosciences. M. Javle reports grants and personal fees from Astra Zeneca, Jazz, grants from El Lilly, grants from BMS, Novartis, QED, and Transthera outside the submitted work. J. Rodon reports nonfinancial support and reasonable reimbursement for travel from European Society for Medical Oncology and Loxo Oncology; receiving consulting and travel fees from Ellipses Pharma, Molecular Partners, IONCTURA, Sardona, Mekanistic, Amgen, Merus, MonteRosa, Aadi and Bridgebio (including serving on the scientific advisory board); consulting fees from Vall d'Hebron Institute of Oncology/Ministero De Empleo Y Seguridad Social, Chinese University of Hong Kong, Boxer Capital, LLC, Tang Advisors, LLC and Guidepoint, receiving research funding from Blueprint Medicines, Merck Sharp & Dohme, Hummingbird, AstraZenneca, Yingli, Vall d'Hebron Institute of Oncology/Cancer Core Europe; and serving as investigator in clinical trials with Cancer Core Europe, Symphogen, BioAlta, Pfizer, Kelun-Biotech, GlaxoSmithKline, Taiho, Roche Pharmaceuticals, Hummingbird, Yingli, Bicycle Therapeutics, Merus, AadiBioscience, ForeBio, Loxo Oncology, Hutchinson MediPharma, Ideaya, Amgen, Tango Therapeutics, Mirati, Linnaeus Therapeutics, MonteRosa, Kinnate, Yingli, Debio, BioTheryX, Storm Therapeutics, Beigene, MapKure, Relay, Novartis, FusionPharma, C4 Therapeutics, Scorpion Therapeutics, Incyte, Fog Pharmaceuticals, Tyra, Nuvectis Pharma. F. Meric-Bernstam reports personal fees from AbbVie, Aduro BioTech Inc., Alkermes, AstraZeneca, Daiichi Sankyo Co. Ltd., Calibr (a division of Scripps Research), DebioPharm, Ecor1 Capital, eFFECTOR Therapeutics, Exelixis, F. Hoffman-La Roche Ltd., GT Apeiron, Genentech Inc., Harbinger Health, IBM Watson, Incyte, Infinity Pharmaceuticals, Jackson Laboratory, Kolon Life Science, LegoChem Bio, Lengo Therapeutics, Menarini Group, OrigiMed, PACT Pharma, Parexel International, Pfizer Inc., Protai Bio Ltd, Samsung Bioepis, Seattle Genetics Inc., Tallac Therapeutics, Tyra Biosciences, Xencor, Zymeworks, personal fees from Black Diamond, Biovica, Eisai, FogPharma, Immunomedics, Inflection Biosciences, Karyopharm Therapeutics, Loxo Oncology, Mersana Therapeutics, OnCusp Therapeutics, Puma Biotechnology Inc., Seattle Genetics, Sanofi, Silverback Therapeutics, Spectrum Pharmaceuticals, Theratechnologies, Zentalis, grants from Jazz Pharmaceuticals, Zymeworks, Aileron Therapeutics, Inc. AstraZeneca, Bayer Healthcare Pharmaceutical, Calithera Biosciences Inc., Curis Inc., CytomX Therapeutics Inc., Daiichi Sankyo Co. Ltd., Debiopharm International, eFFECTOR Therapeutics, Genentech Inc., Guardant Health Inc., Klus Pharma, Takeda Pharmaceutical, Novartis, Puma Biotechnology Inc., Taiho Pharmaceutical Co., personal fees from Dava Oncology, and other support from European Organisation for Research and Treatment of Cancer (EORTC), European Society for Medical Oncology (ESMO), Cholangiocarcinoma Foundation, Dava Oncology, Artidis outside the submitted work. No disclosures were reported by the other authors.

T.P. DiPeri: Data curation, formal analysis, investigation, visualization, writing–original draft, writing–review and editing. K.W. Evans: Data curation, formal analysis, supervision, validation, investigation, visualization, writing–original draft, writing–review and editing. B. Wang: Data curation, formal analysis, investigation, writing–review and editing. M. Zhao: Formal analysis, investigation, visualization, writing–review and editing. A. Akcakanat: Formal analysis, investigation, visualization, writing–review and editing. M. Raso: supervision, investigation, visualization, writing–review and editing. Y.Q. Rizvi: Investigation, visualization, writing–review and editing. X. Zheng: Software, investigation, visualization, methodology, writing–review and editing. A. Korkut: Software, supervision, investigation, visualization, methodology, writing–review and editing. K. Varadarajan: Writing–review and editing. B. Uzunparmak: Data curation, investigation, writing–review and editing. E.E. Dumbrava: Investigation, writing–review and editing. S. Pant: Investigation, writing–review and editing. J.A. Ajani: Investigation, writing–review and editing. P.R. Pohlmann: Investigation, writing–review and editing. V. Jensen: Resources, investigation, project administration, writing–review and editing. M. Javle: Resources, investigation, writing–review and editing. J. Rodon: Supervision, investigation, writing–review and editing. F. Meric-Bernstam: Conceptualization, resources, formal analysis, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

We would like to sincerely thank the following personnel for their contributions to this manuscript: Susanna E. Brisendine for manuscript formatting and preparation; Kathleen Kong and Bryce Kirby for assistance with study enrollment; the University of Texas MD Anderson Cancer Center Department of Veterinary Medicine and Surgery for excellent care of animals through the duration of these studies; the University of MD Anderson Cancer Center Research Histology Core Facility for tissue processing and embedding and the Department of Veterinary Medicine and Surgery for their veterinary support (both supported by P30CA016672). The Mayo Cytogenetics Core, including Dr. Patricia T. Greipp, D.O. and Ryan A. Knudson, who performed FISH analysis for this study and is supported, in part, by the Mayo Clinic Comprehensive Cancer Center Grant, funded by National Cancer Institute (P30CA15083). This work was supported by the following: NIH Training of Academic Surgical Oncologists (Award #: 5T32CA009599-32), University of Texas PDX Development and Trial Center (Award #: 5U54CA224065-04), NIH/NCI PDX Supplement in support of the Texas Experimental Cancer Therapeutics Network (UM1CA186688), Center for Clinical and Translational Science (Award #: 5UL1TR003167-03), and the Sheikh Khalifa Bin Zayed Al Nahyan Institute for Personalized Cancer Therapy, MD Anderson Cancer Center support grant (Award #: P30 CA016672).