Determinants of Survival with Combined HER2 and PD-1 Blockade in Metastatic Esophagogastric Cancer

Approximately 20% of esophagogastric cancer (EGC) patients’ tumors carry amplifications of ERBB2 or overexpress its product, human epidermal growth factor receptor 2 (HER2). For these patients, the anti-HER2 antibody trastuzumab and chemotherapy (TC) have been the standard treatment for over a decade (1). The combination of these agents with pembrolizumab, an anti-programmed death 1 (PD-1) antibody, gained fast-track approval in first-line HER2-positive esophagogastric cancer by the United States Food and Drug Administration based on a 74.4% objective response rate in the Phase III KEYNOTE 811 study with a 22.7% improvement in objective responses versus trastuzumab and chemotherapy alone (2).

In addition to providing updated survival results, which build upon concordant short follow-up results seen in INTEGA and PANTHERA, we present expanded biomarker analysis from our antecedent phase II trial (3–5). In the phase II cohort, more than half of the patients had developed new non-target “escape” lesions upon progression, either due to loss of HER2 expression or via genomically mediated resistance mechanisms. We hypothesized that escape lesions reflect preexisting tumor heterogeneity between disease sites (intrapatient) and within a disease site (intratumor) that eventually conferred resistance. Therefore, we sought to understand how we can better predict durable responders by using biomarkers derived from integrated clinical characteristics, tumor and plasma genomics, and molecular imaging features generated from patients treated in a phase II trial or in clinically annotated institutional cohorts of HER2-positive EGC patients treated with trastuzumab and chemotherapy with or without PD-1 inhibition (Supplementary Table S1) to identify putative biomarkers of intrapatient tumor heterogeneity and their prognostic significance. Finally, additional paired pre- and on-treatment TC single-cell RNA sequencing (scRNA-seq) was used to identify a transcriptional resistance program that allowed certain tumor clones to persist despite therapy and may serve as future therapeutic targets.

Patients with HER2⁺ [defined as IHC 3+ or IHC 2+ and HER2:CEP17 fluorescence in situ hybridization (FISH) ratio ≥ 2.0] metastatic esophageal, gastroesophageal junction, or gastric adenocarcinoma (n = 37) were recruited to an open-label, nonrandomized, single-arm, investigator-initiated, single-institution phase II trial of trastuzumab, pembrolizumab, capecitabine, and oxaliplatin. The study protocol and all amendments were approved by the Memorial Sloan Kettering (MSK) Institutional Review Board (IRB 16–937) and the trial was registered with ClinicalTrials.gov (NCT02954536). Treatment and procedure details have been published (3). In addition to standard response assessment (RECIST 1.1), up to 5 lesions from each affected disease site (i.e., up to 5 lymph nodes, 5 liver metastases, 5 lung nodules, etc.) were measured for “expanded assessment” to better account for response heterogeneity, including lesions smaller than 1 cm at baseline. The phase II protocol was performed in accordance with the protocol and its amendments and Good Clinical Practice guidelines and was overseen by the MSK Data and Safety Monitoring Committee. All patients provided written informed consent before enrollment.

An additional retrospective cohort of patients in the MSK esophagogastric database who received platinum chemotherapy with HER2-directed therapy (trastuzumab or zanidatamab) from 2010 to 2022 were manually abstracted for demographic, pathologic, radiographic, and genomic features under a separate IRB-approved retrospective protocol.

HER2 clinical testing was performed by IHC using the PATHWAY anti-HER2/neu (4B5) assay (Ventana) and programmed death-ligand 1 (PD-L1) testing using the E1L3N clone (Cell Signaling Technology). MET IHC was performed, when relevant, using Ventana clone SP44. HER2 IHC was considered uniformly 3+ if all pretreatment testing at MSK or outside institutions was interpreted as 3+. Conversely, if pretreatment HER2 IHC was ever interpreted as ≤2+, the patient's tumor was considered 2+/heterogeneous.

HER2 FISH was performed in cases with equivocal (2+ intensity) HER2 immunohistochemistry staining using HER2 IQFISH pharmDx (Dako) with scoring per clinical guidelines (6). FISH for MET was performed on paraffin section using a 2-color MET/CEN7 probe consisting of BAC clones spanning the MET gene (RP11–39K12 and RP11–163L9 labeled with red dUTP) and a centromeric repeat plasmid for Chr 7 (P7t1 labeled with green dUTP) as control. MET amplification was defined as ≥ 2 MET/CEN7 ratio or ≥ 6 MET copies (discrete signal) or the presence of at least one MET cluster (≥ 4 copies; low-level amplicon resulting from tandem duplications) in >10% of cells.

A subset of phase II trial patients were additionally consented to investigational PET imaging on MSK IRB protocol 13–165. Whole-body PET/CT scans from mid skull to proximal thigh were performed on days 5 to 7 following intravenous injection of 5 mCi of ⁸⁹Zr-trastuzumab. All scans were obtained on a dedicated PET/CT scanner (Discovery STE GE Healthcare) using low-dose CT scans for attenuation correction (10–80 mA current) and PET acquisition for 7 to 8 minutes per bed position in 3-dimensional mode and attenuation, scatter, and other standard corrections were applied to generate images using iterative reconstruction. Localization in the tumor was defined as positive if focal accumulation of activity was greater than adjacent background activity and distinct from areas of expected physiologic activity. Volumes of interest were drawn to include lesions on PET images using dedicated software (Hermes Medical Solutions; ref. 7), and images were independently assessed by a nuclear radiologist. Only lesions computed tomography (CT) measurements per “expanded assessment” as described above were evaluated. Lesion uptake was classified based upon uptake values (SUV) categorized as low (3 ≤ SUVmax < 5), moderate (5 ≤ SUVmax < 10), intense (10 ≤ SUVmax < 15), or very intense (15 ≤ SUVmax; ref. 8).

Collection, processing, and analysis of tissue and plasma NGS samples are described in the Supplementary Methods section.

Fresh tissue was collected from a patient receiving capecitabine and oxaliplatin (CAPOX) and trastuzumab ± pembrolizumab at MSK and underwent processing and analysis as described in the Supplementary Methods section. Additionally, paired scRNA-seq data collected before cycles 1 and 3 from patients who received CAPOX with (n = 3) or without trastuzumab (n = 4) was provided by Samsung Medical Center and processed as previously described (9).

Patient demographic characteristics were summarized using descriptive summary statistics and compared between groups of interest using Fisher exact test for categorical data or Wilcoxon rank-sum test for continuous variables. Progression-free survival (PFS) and overall survival (OS) were calculated from the date of treatment until date of first progression or death, whichever occurred first for PFS or date of death for OS. Patients alive and without progression of disease (POD) were censored at their last available scan. PFS and OS were estimated using Kaplan–Meier methods and compared between subgroups using the log-rank test. Among patients treated with pembrolizumab, trastuzumab, and chemotherapy on-protocol, a landmark time of 9 weeks was used to examine the association between response status and clearance of all tumor-matched ctDNA at 9 weeks with OS and PFS. Patients without the 9-week follow-up were excluded. Univariate PFS analysis was done using Cox proportional hazards model and included all patients who received trastuzumab and chemotherapy with or without PD-1 inhibitor (n = 217). The final multivariable PFS model was created by including the main covariate of interest (trastuzumab and chemotherapy with or without PD-1 inhibitor) and adjusting for factors associated with PFS from univariate analysis at P < 0.05. TP53 alteration was not included in the final multivariable analysis since it was highly correlated with ERBB2 amplification status. All statistical analyses were done using R version 4.0.4. All P values were two-sided and P < 0.05 was considered statistically significant.

Data were analyzed using publicly available computational packages as described in the NGS and scRNA-seq sections above.

Clinical data and key genomic features for each patient are provided in the Supplementary Data. Targeted sequencing data generated for this study are available at the European Variation Archive (EVA) accession #PRJEB63072 (https://www.ebi.ac.uk/eva/?eva-study=PRJEB63072) and at https://www.cbioportal.org/study/summary?id=egc_trap_ccr_msk_2023.

Clinical data for the phase II study was updated on April 1, 2022, providing nearly three additional years of follow-up since the prior publication. Six of 37 patients have not yet progressed, and 5 of these patients are still being treated on-study with a median follow-up time of 44 (range: 41–63) months. Median PFS was 13 [95% confidence interval (CI): 8.5–20] months, median OS was 27 (95% CI: 21–44) months, and 31 of 35 evaluable patients (89%; 95% CI: 73%–97%) achieved a RECIST response. Since study enrollment was completed, an additional 29 patients received PD-1 inhibition (nivolumab or pembrolizumab) in combination with TC in standard practice, and achieved a median 14.7-month [95% CI, 11.2–not reached (NR)] PFS, median OS was not reached, 80% (23/29; 95% CI: 60%–92%) objective response rate (ORR) with a median of 10.4 months of follow-up (Supplementary Figs. S1 and S2; Supplementary Tables S1 and S2; Supplementary Data S1).

Building upon prior findings that high ERBB2 gene copy number and homogeneity predict improved PFS (3, 10), we evaluated the correlation between intensity and homogeneity of pretreatment HER2 IHC overexpression and PFS in 193 patients with HER2-positive EGC treated with first-line chemotherapy with trastuzumab with or without PD-1 inhibitor. Patients with uniformly high pretreatment tumor HER2 overexpression [IHC 3+; n = 129/193 (67%)] who received either PD-1 inhibition and TC or TC alone had longer median PFS [15 (95% CI: 13–20) vs. 8.5 (95% CI: 6.4–13) months; CI, 0.43–0.86; Fig. 1A] compared with patients with indeterminate (HER2 IHC 2+) or heterogeneous (HER2 IHC 3+ and HER2 IHC 0–1+) HER2 expression within a disease site or between sites (n = 64/193) pretreatment. This correlation may result from greater and/or more uniform HER2 overexpression in ERBB2-amplified tumors, as ERBB2 was amplified in the tumors of 110 of 125 patients (88%) with HER2 3+ overexpression, versus 21 of 58 (36%) evaluable patients with HER2 2+/heterogeneous expression (P = 1.8e−12; Supplementary Data S1).

As pretreatment biopsies of multiple metastatic disease sites are not feasible in routine care, we examined whether noninvasive detection of HER2 expression heterogeneity via ⁸⁹Zr-trastuzumab imaging could serve as a biomarker of treatment response. Among the 25 on-protocol patients who received a dose of pembrolizumab and trastuzumab “induction” 3 weeks before initiation of concurrent chemotherapy, 8 patients (32%) underwent ⁸⁹Zr-trastuzumab PET prior to initiating therapy; 7 of 8 patients (88%) had at least 1 PET-avid lesion (Fig. 1B; Supplementary Table S3; Supplementary Data S2 and S3), including all evaluable patients with high HER2 expression. We observed significant variability in PET avidity between disease sites within patients pre-treatment. Of 47 assessable pretreatment lesions, 31 (66%) were PET-avid (SUV ≥ 3); of these, 48% (15/31) demonstrated intense ⁸⁹Zr-trastuzumab avidity (SUV ≥ 10; Supplementary Data S2). All 15 intensely ⁸⁹Zr-trastuzumab avid lesions decreased in size by CT after a single dose of pembrolizumab and trastuzumab, whereas only 9 of 32 non-intensely avid lesions (8%) showed reduction (P = 2.4 ×10⁻⁶ by Fisher exact test; Supplementary Fig. S3). All 4 patients with ≥ 1 intensively avid (SUV ≥ 10) lesion at baseline achieved 6-month PFS (Supplementary Data Table S2; example shown in Fig. 1C), versus only 2 of 4 patients without intensely avid lesions (SUV < 10; example in Fig. 1D). Three of these patients also had plasma ctDNA ERBB2 amplification, and exceeded a 6-month PFS, while the two patients lacking both ⁸⁹Zr-trastuzumab intense avidity and plasma ctDNA ERBB2 amplification failed to achieve a 6-month PFS (Supplementary Data Table S2). However, 1 patient with low-volume lymph node–only disease lacked PET-avid lesions at baseline yet achieved an 8.7-month PFS. These results, though limited by the small sample size, suggest that ⁸⁹Zr-trastuzumab PET can noninvasively reveal the intensity and uniformity of HER2 expression, which may correlate with clinical benefit from pembrolizumab and trastuzumab.

We previously demonstrated that pretreatment plasma ERBB2 amplification predicts durable PFS (3, 11), and so we next explored the utility of early ctDNA decline as a biomarker of ≥6-month PFS on pembrolizumab with chemotherapy and trastuzumab. Of patients with tumor-matched (see Materials and Methods) mutations at baseline, 12 of 16 (75%) had a decline in maximum variant allelic frequency (maxVAF) by week 3, and 9 of these patients (75%) achieved 6-month PFS, versus 0 of 4 patients (0%) with rising maxVAF at week 3 (Supplementary Fig. S4; Supplementary Data S3). Plasma ctDNA dynamics were then compared with lesion-level CT scan dynamics. At 3-week CT, no responding lesions were identified in patients lacking ERBB2-amplified ctDNA (Supplementary Fig. S5A). In contrast, lesions from 5 of 9 patients (56%) lacking ERBB2 tissue amplification responded to pembrolizumab and trastuzumab by 3-week CT, suggesting intrapatient HER2 heterogeneity (Supplementary Fig. S5B; Supplementary Data S3). Only 3 of 10 patients (30%) who achieved both 6-month PFS and decline in ctDNA demonstrated a uniform CT response in all disease sites (P = 0.58), highlighting the discordance of early ctDNA with CT imaging.

As decline in VAF fails to account for copy number alteration events, we also evaluated clearance of all tumor-matched ctDNA alterations as a biomarker of survival. Of the 23 patients who had tumor-matched ctDNA alterations, the 17 in whom all were cleared by 9 weeks exhibited longer PFS than in the 6 patients who did not (HR 0.18; 95% CI, 0.06–0.53; Fig. 1D). While 13 of 17 patients (76%) who cleared ctDNA achieved 6-month PFS, only 1 of 6 (17%) who failed to clear ctDNA did. This performed similarly to follow-up imaging where phase II pembrolizumab, trastuzumab, and chemotherapy-treated patients with uniform CT responses (i.e., all lesions radiographically improving) at 9 weeks (21/35; 60%) had a decreased risk of progression or death (HR 0.38; 95% CI, 0.17–0.81; Fig. 1E) compared with patients with mixed responses (14/35; 40%). This suggests that ctDNA clearance may have a role in reducing the need for frequent cross-sectional imaging.

Of the 31 pembrolizumab, trastuzumab, and chemotherapy-treated trial patients whose disease progressed, 19 (61%) developed escape lesions that were not present on baseline CT, including 3 (10%) who developed brain metastases. While all 37 protocol patients had HER2-positive tumors at baseline, only 11 of 21 patients (52%) remained HER2-positive upon progression, corresponding to 9 of 18 (50%) with ERBB2 amplification in tissue and 8 of 21 (38%) in plasma (variation in denominators reflects data availability; Supplementary Fig. S6). There was no significant change seen in tumor mutation burden post-treatment. As site-matched pre- and posttreatment biopsies were not available for these lesions, ctDNA evaluation improved sensitivity to detect baseline and acquired resistance mechanisms. Eighteen of 21 (86%) evaluable patients had tumor-matched ctDNA mutations upon progression (Supplementary Data S1). However, only 49% of plasma NGS alterations could be tumor-matched, as the nonmatched mutations presumably arose outside the sequenced tumor site or via clonal hematopoiesis (Supplementary Table S4). Tissue and/or ctDNA testing identified putative mechanisms of resistance in nearly all patients, including loss of HER2 expression and alterations in PIK3CA, KRAS, MET, and EGFR (Supplementary Fig. S6). One patient had pretreatment heterogeneity with a HER2−, PD-L1 combined positive score (CPS) 100 retroperitoneal lymph node (Fig. 2A; Supplementary Fig. S7) which resolved on-treatment, while their esophageal primary tumor was HER2 3+, PD-L1 CPS <1. Upon progression, their metastases demonstrated MET IHC expression and concurrent MET ctDNA amplification. We also demonstrated that PIK3CD S367L, currently a variant of undetermined significance, increases phosphorylation of AKT, suggesting functional relevance (Supplementary Fig. S8). These findings indicate that the predominant mechanisms of resistance were either loss of HER2 expression or selection for resistant subclones, similar to previous findings with trastuzumab and chemotherapy (10).

To explore the contribution of intratumor heterogeneity to pre-existing subclonal resistance, we performed scRNA-seq on tumor tissue collected from a metastatic EGC patient treated with chemotherapy and trastuzumab ± pembrolizumab (Fig. 2B). At the single-cell level, we identified extensive ERBB2 expression heterogeneity and co-occuring high expression of FGFR3, a putative biomarker of trastuzumab resistance, in a subpopulation of highly ERBB2-expressing cells. Furthermore, pretreatment plasma NGS identified a subclonal FGFR3-TACC3 fusion event (not detected by targeted sequencing of a pretreatment tumor biopsy) that supports scRNA-seq findings and also identified EGFR, ERBB2, and PIK3CA amplifications (Fig. 2B; Supplementary Fig. S9).

We postulated that durable clinical benefit is inversely proportional to intratumor clonal diversity and evaluated on-treatment clonal selection using recently published paired scRNA-seq data collected before cycles 1 and 3 from patients who received CAPOX with (TC; n = 3) or without trastuzumab (C; n = 4; ref. 9). Using inferCNV (Supplementary Methods), we reconstructed the clonal composition of the pre- and on-treatment tumor biopsies (Supplementary Fig. S10A) and were successfully able to demonstrate rapid clearance of most highly ERBB2 expressing clones, which was consistent with our prior ctDNA findings. However, we also identified many clones that expanded despite treatment, suggesting resistance to both trastuzumab and chemotherapy (Supplementary Fig. S10B). To identify a trastuzumab resistance program, gene sets associated with clonal expansion versus contraction were devised using GeneVector on pretreatment expression from patients who received either TC or C (Supplementary Methods). Therefore, the equation [(CAPOX+T_Expanded_T0 − CAPOX+T_Contracted_T0) − (CAPOX_Expanded_T0 − CAPOX_Contracted_T0)], allowed us to identify that MT1H, MT1E, MT2A, and MSMB expression were associated with CAPOX+trastuzumab resistance (Supplementary Fig. S10C and S10D). Though exploratory and requiring orthogonal validation, these genes may serve as future therapeutic targets in overcoming resistance.

The combined PD-1 inhibitor with trastuzumab and chemotherapy cohort (n = 66) was evaluated in conjunction with patients who received chemotherapy and HER2 inhibition without PD-1 inhibitor (n = 151). Another 9 patients were included, where able, whose concurrent PD-1 inhibitor receipt was unknown. We sought to identify additional prognostic features including sites of metastatic involvement and molecular alterations (Supplementary Tables S5 and S6; Supplementary Data S4), in patients with EGC treated with trastuzumab and chemotherapy with or without anti-PD-1 therapy. Univariate analysis identified that patients with fewer disease sites involved and with ERBB2 amplification had superior PFS, while those with oncogenic alterations in KRAS, MYC, CCND1, and CDKN2A/B had inferior PFS (Supplementary Table S7). Though no specific disease sites were associated with prognosis, the multivariable model identified ERBB2 amplification as a positive prognostic feature, and alterations in MYC, CDKN2A/B as negative prognostic features for PFS (Supplementary Table S8).

These results demonstrate that pretreatment tumor heterogeneity, in particular HER2 expression heterogeneity, is associated with shorter PFS among patients with HER2+ EGC treated with trastuzumab, pembrolizumab, and chemotherapy. Since our initial clinical publication (3), INTEGA and PANTHERA similarly demonstrated PD-L1 agnostic improved response and survival with the addition of immune checkpoint blockade to chemotherapy and trastuzumab, albeit with shorter follow-up (4, 5). Although combined PD-1 and HER2 blockade augments trastuzumab efficacy across all 3 studies, our long-term follow-up confirms that preexistent lesion-to-lesion and even intratumoral HER2 heterogeneity remains a major barrier to HER2-targeted therapy efficacy and a negative predictor of survival (12, 13).

While biopsy of multiple disease sites to assess for HER2 and other molecular heterogeneity is not feasible in clinical practice, we found that noninvasive ctDNA analysis and ⁸⁹Zr-trastuzumab molecular imaging can identify patients most likely to have durable benefit on first-line pembrolizumab and trastuzumab therapy. However, these modalities have limitations and require relatively high disease burden for accurate assessment.

We found that declining tumor-matched ctDNA maxVAF at 3 weeks, after a single dose of pembrolizumab and trastuzumab, predicted 6-month PFS, which demonstrates ctDNA's ability to predict treatment response in EGC, mirroring results in other cancers (14–17). While uniformity of 3-week CT response predicted ≥ 6-month PFS, changes in individual lesions were often too small to reliably predict future response. However, several patients with progressive lesions after induction therapy responded to the addition of chemotherapy, highlighting the role for combination approaches in overcoming heterogeneity-driven drug resistance. This may explain why previous studies of targeted monotherapies failed to associate early ctDNA decline with improved PFS (18). The observation that ctDNA is an early prognostic biomarker of treatment response may inform future targeted and immunotherapy trials in which therapy escalation or alternate molecularly guided agents are used as salvage in patients with mixed responses resulting from tumor heterogeneity.

Upon progression, IHC and DNA sequencing of tumor and/or ctDNA identified low HER2 expression or lack of ERBB2 amplification in approximately half of tested patients. This was higher than in prior studies, although limited by sample size (19, 20). In patients with persistent HER2 expression, escape lesions had co-occurring alterations in KRAS, MET, EGFR, or PI3K-associated genes or were in reservoir sites with poor trastuzumab penetration, such as the central nervous system. scRNA-seq in conjunction with serial ctDNA sequencing illustrated de novo pretreatment subclonal coexpression of a FGFR3-TACC3 fusion and ERBB2, reiterating the correlation between genomic heterogeneity and therapeutic resistance, and the role of ctDNA testing in identifying tumor heterogeneity.

Furthermore, scRNA-seq suggested that MT1H, MT1E, MT2A, and MSMB expression were associated with trastuzumab resistance. MT1H, MT1E, and MT2A encode isoforms of metallothioneins, which are found in all tissues and are associated with trace element homeostasis and protection against oxidative stress. While MT2A has been associated with chemotherapy resistance (21, 22), MT1G expression contributes to P53 activation and improved oxaliplatin cytotoxicity, though this may not be relevant in HER2-expressing EGC as nearly all of these patients have TP53 mutations (23). Similarly, MSMB encodes microseminoprotein-beta. Microseminoprotein expression is stimulated by hypoxia, which leads to MAPK pathway activation and has been associated with vascular endothelial growth factor inhibitor resistance in ovarian cancer models (24). Further preclinical validation of metallothionein inhibition in overcoming chemotherapy resistance is merited.

Multivariable analysis of our larger institutional cohort of HER2-targeted patients did not reveal a difference in PFS with or without concurrent immune checkpoint blockade despite impressive results with adding PD-1 inhibitors in multiple trials. Survival outcomes from the phase 3 KEYNOTE-811 trial are expected shortly. Pre-clinically, combining trastuzumab and PD-1 blockade augments HER2-specific T-cell response, promotes T-cell and dendritic cell trafficking, and induces peripheral memory T-cell expansion (25–29). Further clinical validation of these mechanisms is needed. While we identified oncogenic MYC and CDKN2A/B alterations as prognostic of inferior PFS, PANTHERA suggested that pretreatment RTK/RAS pathway alterations were associated with improved PFS. This is contrary to our prior findings (3, 10) and likely reflects their liberal definition of oncogenic RTK-RAS alterations (61% vs. 28% of patients), and would benefit from validation in a larger cohort. Both studies identified that ERBB2 amplification was associated with improved PFS.

Promising initial phase III KEYNOTE-811 study findings suggest that the combination of pembrolizumab and trastuzumab is a transformative treatment for patients with EGC, and mechanistic evaluation of this combination remains ongoing. Here, we identified prognostic pre- and early-treatment biomarkers of durable response. We also confirmed that tumor heterogeneity remains the greatest barrier to precision oncology in EGC, and that preexistent molecular heterogeneity can be identified noninvasively through molecular imaging and ctDNA analysis. Future trials should assess whether noninvasive assessment of mixed response and molecular heterogeneity can dynamically inform adaptive treatment strategies prior to disease progression.

S.B. Maron reports grants from Society of MSK, Cycle for Survival, NIH, and non-financial support from Guardant Health during the conduct of the study, as well as personal fees from Natera, Bayer, Basilea, Daiichi Sankyo, Bicara, Novartis, Amgen, Elevation Oncology, and Purple Oncology outside the submitted work; S.B. Maron also previously owned stock in Calithera. N. Pandit-Taskar reports other support from Bayer, Fusion Pharma, Clarity Pharma, and Janssen, as well as personal fees and other support from Imaginab, Illumina-innervate, and Actinium Pharma outside the submitted work. J.S. Lewis reports grants from NIH during the conduct of the study. T. Biachi De Castria reports personal fees from Merck Sharp & Dohme Corp, Bristol Myers Squibb, Eli Lilly, AstraZeneca, and A2Bio during the conduct of the study. R.J. Nagy reports other support from Guardant Health outside the submitted work. G.Y. Ku reports grants and personal fees from AstraZeneca, Daiichi Sankyo, Merck, Pieris, and Zymeworks during the conduct of the study, as well as grants from Adaptimmune, CARsgen, and Oncolys outside the submitted work. T. Merghoub reports personal fees from Daichii Sankyo and Normunity during the conduct of the study, as well as grants from Bristol Myers Squibb, Surface Oncology, Kyn Therapeutics, Infinity Pharmaceuticals, Peregrine Pharmaceuticals, Adaptive Biotechnologies, Leap Therapeutics, and Aprea. T. Merghoub is a consultant for Daiichi Sankyo, Leap Therapeutics, Immunos Therapeutics, and Pfizer, and is co-founder of Imvaq Therapeutics with equity. In addition, T. Merghoub is an inventor on patent applications related to work on oncolytic viral therapy, alpha virus-based vaccine, neoantigen modeling, CD40, GITR, OX40, PD-1, and CTLA-4. S. Shah reports other support from Canexia Health Inc, as well as personal fees from AstraZeneca Inc outside the submitted work. D.B. Solit reports personal fees from Pfizer, Scorpion Therapeutics, FORE Therapeutics, Function Oncology, Fog Pharma, Elsie Biotechnologies, Rain Oncology, and BridgeBio outside the submitted work. Y.Y. Janjigian reports personal fees from Merck Serono, Micheal J. Hennessy Associates, Paradigm Medical Communications, PeerView Institute, Pfizer, Research to Practice, Geneos Therapeutics, GlaxoSmithKline, Imedex, Imugene, Lynx Health, Daiichi Sankyo, AbbVie, Amerisource Bergen, Ask-Gene Pharma, Inc., Arcus Biosciences, AstraZeneca, and Basilea Pharmaceutica; personal fees and other support from Rgenix, Seagen, Silverback Therapeutics, Zymeworks Inc., Merck, Eli Lilly, Bayer, and Bristol Myers Squibb; and other support from Cycle for Survival, Department of Defense, Fred's Team, Genentech/Roche, and NCI outside the submitted work. Y.Y. Janjigian also reports stock options from Rgenix. No disclosures were reported by the other authors.

S.B. Maron: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. W. Chatila: Formal analysis, visualization, methodology. H. Walch: Formal analysis, visualization, methodology. J.F. Chou: Conceptualization, formal analysis, writing–original draft. N. Ceglia: Investigation, visualization, methodology. R. Ptashkin: Resources, data curation, methodology. R.K.G. Do: Data curation. V. Paroder: Data curation. N. Pandit-Taskar: Investigation, methodology, writing–review and editing. J.S. Lewis: Conceptualization, resources. T. Biachi De Castria: Data curation, validation. S. Sabwa: Data curation. F. Socolow: Data curation. L. Feder: Data curation, project administration. J. Thomas: Investigation. I. Schulze: Investigation. K. Kim: Investigation, methodology. A. Elzein: Investigation. V. Bojilova: Investigation, visualization. M. Zatzman: Validation, investigation. U. Bhanot: Supervision, investigation. R.J. Nagy: Resources, funding acquisition. J. Lee: Data curation, validation. M. Simmons: Data curation, supervision. M. Segal: Project administration. G.Y. Ku: Investigation, writing–review and editing. D.H. Ilson: Supervision, writing–review and editing. M. Capanu: Formal analysis, writing–original draft. J.F. Hechtman: Conceptualization, data curation, supervision, investigation. T. Merghoub: Conceptualization, writing–review and editing. S. Shah: Conceptualization, resources. N. Schultz: Conceptualization, resources, formal analysis, visualization, methodology, writing–review and editing. D.B. Solit: Conceptualization, validation, investigation, visualization, writing–original draft, writing–review and editing. Y.Y. Janjigian: Conceptualization, formal analysis, supervision, validation, investigation, writing–original draft, writing–review and editing.

The original trial was funded by Merck & Co, which provided pembrolizumab (ClinicalTrials.gov trial registration ID: NCT01522768). We acknowledge the use of the Integrated Genomics Operation Core (iGO) for sequencing and the Radiochemistry and Molecular Imaging Probe Core for the production of ⁸⁹Zr-trastuzumab, both funded by the NCI Cancer Center Support Grant (CCSG, P30 CA08748). iGO is also supported by Cycle for Survival, and the Marie-Josée and Henry R. Kravis Center for Molecular Oncology. JSL was supported in part by NCI R35 CA232130. Other funding sources include the Robertson Foundation (grant to NS), the Department of Defense Congressionally Directed Medical Research Program (CA 150646 to YYJ), the MSK Paul Calabresi Career Development Award for Clinical Oncology (K12 CA184746 to SBM), Cycle for Survival (to YYJ) and the Society of Memorial Sloan Kettering (to SBM).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.