First-line pembrolizumab, trastuzumab, and chemotherapy in advanced HER2-positive gastric cancer with sequential genomic proling

Hyun Cheol Chung (  [email protected] ) Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center, Yonsei University College of Medicine Choong-kun Lee Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center, Yonsei University College of Medicine https://orcid.org/0000-0001-5151-5096 Sun Young Rha Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center, Yonsei University Health System https://orcid.org/0000-0002-2512-4531 Hyo Song Kim Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center, Yonsei University Health System Minkyu Jung Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center, Yonsei University Health System Beodeul Kang Medical Oncology, Department of Internal Medicine, CHA Bundang Medical Center, CHA University Jingmin Che Sondang Institute for Cancer Research, Yonsei University College of Medicine Woo Sun Kwon Sondang Institute for Cancer Research, Yonsei University College of Medicine Woo Kyun Bae Division of Hematology-Oncology, Department of Internal Medicine, Chonnam National University Medical School and Hwasun Hospital Dong-Hoe Koo Kangbuk Samsung Hospital, Sungkyunkwan University School of Medicine Su-Jin Shin Department of Pathology, Yonsei University College of Medicine Hyunki Kim Yonsei University, College of Medicine Hei-Cheul Jeung Department of Internal Medicine, Gangnam Severance Hospital, Yonsei University College of Medicine Dae Young Zang Hallym University Medical Center, Hallym University College of Medicine Sang Kil Lee Division of Gastroenterology, Department of Internal Medicine, Yonsei University College of Medicine Chung Mo Nam Department of Biostatistics, Yonsei University College of Medicine

Article

Keywords: Programmed Cell Death, Human Epidermal Receptor, Next-generation Panel Sequencing, Quadruplet Regimen

Posted Date: February 23rd, 2021

DOI: https://doi.org/10.21203/rs.3.rs-254208/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

1 First-line pembrolizumab, trastuzumab, and chemotherapy in advanced HER2-

2 positive gastric cancer with sequential genomic profiling

3

4 Choong-kun Lee1,2†, Sun Young Rha1,2†, Hyo Song Kim1,2, Minkyu Jung1,2, Beodeul Kang3,

5 Jingmin Che2, Woo Sun Kwon2, Woo Kyun Bae4, Dong-Hoe Koo5, Su-Jin Shin6, Hyunki

6 Kim6, Hei-Cheul Jeung7, Dae Young Zang8, Sang Kil Lee9, Chung Mo Nam10, Hyun Cheol

7 Chung1,2*

8

9 1Division of Medical Oncology, Department of Internal Medicine, Yonsei Cancer Center,

10 Yonsei University College of Medicine, Seoul, South Korea

11 2Sondang Institute for Cancer Research, Yonsei University College of Medicine, Seoul,

12 South Korea

13 3CHA Bundang Medical Center, CHA University, Seongnam, South Korea

14 4Division of Hematology-Oncology, Department of Internal Medicine, Chonnam National

15 University Medical School and Hwasun Hospital, Jeollanam-do, South Korea

16 5Kangbuk Samsung Hospital, Sungkyunkwan University School of Medicine, Seoul, South

17 Korea

18 6Department of Pathology, Yonsei University College of Medicine, Seoul, South Korea

19 7Department of Internal Medicine, Gangnam Severance Hospital, Yonsei University College

20 of Medicine, Seoul, South Korea

21 8Hallym University Medical Center, Hallym University College of Medicine, Anyang, South

22 Korea

23 9Division of Gastroenterology, Department of Internal Medicine, Yonsei University College

24 of Medicine, Seoul, South Korea

25 10Department of Biostatistics, Yonsei University College of Medicine, Seoul, South Korea

1 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

26

27 †These authors contributed equally.

28 * Correspondence and requests for materials should be addressed to H.C.C.

29 ([email protected]).

30

31

2 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

32 ABSTRACT

33 Combining programmed cell death 1 (PD-1) and human epidermal receptor 2 (HER2)

34 targeting agents has shown synergy in HER2-positive preclinical cancer models. Here, we

35 describe a single-arm multi-institutional phase Ib/II trial combining pembrolizumab,

36 trastuzumab, and chemotherapy (capecitabine plus cisplatin) as first-line therapy for HER2-

37 positive advanced gastric cancer (AGC). With a median follow-up of 18.2 months, 3 out of

38 43 enrolled patients remained in the treatment and 7 finished 2-year treatment without

39 progression. Objective response rate was 76.7% (complete response 16.3%, partial response

40 60.5%, conversion surgery 4.6%), with 26 patients (56.6%) showing tumor reduction of more

41 than 50%. Median progression-free survival was 8.6 months (95% confidence interval [CI]

42 7.2–16.4) and median overall survival was 19.3 months (95% CI 16.5–not reached). There

43 was no patient with microsatellite instability-high or Epstein–Barr virus-positive tumor. No

44 patient discontinued pembrolizumab because of immune-related adverse events. Programmed

45 death ligand-1 (PD-L1) status, metastatic organ, or baseline tumor burden were not related to

46 survival. Molecular analyses of pre-treatment tumor samples using next-generation panel

47 sequencing (NGS) showed that HER2 amplification, RTK/RAS pathway alteration, and

48 neoantigen load corrected by HLA-B were related to survival. Comparison analyses of

49 sequentially biopsied samples identified sensitive and resistant sub-clones with spatial and

50 longitudinal tumor heterogeneity. This study provided insight into potentially relevant NGS-

51 based genomic changes to identify patients who may benefit from quadruplet regimen

52 (pembrolizumab, trastuzumab, capecitabine, and cisplatin), which was active and safe for

53 HER2-positive AGC.

54

3 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

55 INTRODUCTION

56 Gastric cancer is the third leading cause of cancer-related deaths and the fifth most common

57 cancer worldwide1,2. Human epidermal growth factor receptor 2 (HER2; erythroblastic

58 oncogene B2, ERBB2) is a transmembrane tyrosine kinase receptor and is overexpressed or

59 amplified in 10-20% of gastric cancer patients3. First-line anti-HER2 antibody trastuzumab in

60 combination with cytotoxic chemotherapy is the standard treatment for HER2-positive gastric

61 cancer patients after the Trastuzumab for Gastric Cancer (ToGA) study4. However, because

62 the majority of patients develop intrinsic or acquired resistance within the first year,

63 elucidating the molecular mechanisms for trastuzumab resistance is warranted to improve the

64 survival outcome in these patients. One strategy expected to overcome intrinsic or acquired

65 resistance to anti-HER2 treatment is by combining it with immune checkpoint inhibitors,

66 such as anti-PD-1 antibodies. Although the mechanism underlying anti-PD-1 and anti-HER2

67 treatment synergies has not been clearly identified, preclinical and clinical evidence support

68 combining anti-PD-1 antibody with anti-HER2 agents for HER2-positive cancers5-7.

69 In the present study, we report the results from a single-arm multi-institutional phase Ib/II

70 trial of a quadruplet regimen of pembrolizumab, trastuzumab, capecitabine, and cisplatin as

71 first-line therapy for HER2-positive advanced gastric and gastroesophageal junction

72 (AGC/GEJ) cancer (PANTHERA trial, NCT02901301). The primary endpoint was objective

73 response rate (ORR) per Response Evaluation Criteria in Solid Tumors (RECIST) version

74 1.1, and secondary endpoints included progression-free survival (PFS), overall survival (OS),

75 duration of response (DOR), disease control rate (DCR), safety, and biomarker exploration in

76 relation with anti-HER2- or immune checkpoint-related pathway (see Online Methods).

77

78 RESULTS

79

4 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

80 Clinical efficacy and safety of pembrolizumab, trastuzumab, and chemotherapy

81

82 Forty-three patients were enrolled in this study between February 2017 to March 2019 (Table

83 1). Full schedule and combination doses of pembrolizumab 200 mg on the day (D) 1,

84 trastuzumab 6 mg/kg (after 8 mg/kg load) D1, capecitabine 1,000 mg/m2 bid D1-14, and

85 cisplatin 80 mg/m2 D1 every 3 weeks were tested as a starting dose in phase Ib, and no dose-

86 limiting or unexpected toxicities were observed. The starting phase Ib doses were chosen as

87 recommended phase II dose, and three patients of the phase Ib cohort were included in the

88 final analyses of phase II part (Extended Data Fig. 1). At the time of data lock on Aug 31,

89 2020, the median follow-up duration was 18.2 months (95% confidence interval [CI] 16.5–

90 23.1). The median age of the patients was 63 years (range, 34–82), and the majority were

91 men (n=33, 76.7%). The lymph node (n=36, 83.7%) and liver (n=21, 48.8%) were

92 metastasized with highest frequencies. All the patients were HER2-positive, detected either

93 through immunohistochemistry (IHC) 3+ (n=30, 69.8%) or IHC 2+ with silver in situ

94 hybridization (SISH)-positive (n=13, 30.2%; median amplification index=2.74; range, 2.05–

95 4.07). Thirty-eight patients had pretreatment PD-L1 status evaluated with 22C3 antibody, and

96 21 (48.8%) were PD-L1 positive (combined positive score [CPS] 1 or higher) and 5 (11.6%)

97 were CPS 10 or higher. There was no patient with Epstein–Barr virus (EBV) positivity (based

98 on EBV-encoded small RNA in situ hybridization) or mismatch repair deficiency (based on

99 IHC for MLH1, MSH2, PMS2, and MSH6). At the time of the data lock, three patients

100 remained on the treatment, and seven finished the 2-year treatment without progression (six

101 patients still showing no evidence of disease progression). Two patients underwent

102 conversion surgery with curative intent after a remarkable tumor shrinkage (Fig. 1a). The

103 median PFS was 8.6 months (95% CI 7.2–16.4) with a 1-year PFS rate of 41.9%, and the

104 median OS was 19.3 months (95% CI 16.5–not reached [NR]) with a one-year OS rate of

5 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

105 80.1% (Fig. 1b). DCR was 97.7% and ORR was 76.7% (complete response [CR] 16.3% and

106 partial response [PR] 60.5%) (Fig. 1c), with 30% tumor burden reduction rate of 86.0% in 37

107 patients and 50% reduction rate of 56.6% in 26 patients. The median duration of response

108 was 10.8 months (95% CI 7.2–21.8), and the median time to response was 1.7 months (95%

109 CI 1.4–2.6) (Extended Data Table 1). Tumor burden percentage changes from baseline with

110 total tumor lesions of measurable target and non-target lesions over time showed an initial

111 decreasing pattern of tumor burden, except in one patient (Fig. 1d), suggesting tumor

112 shrinkage rate of 98%. The liver and peritoneum were the organs that showed the most tumor

113 shrinkage (mean, 62.5%; Extended Data Fig. 2a), and eventually 1/3rd of the liver lesion

114 progressed (n=11 out of 32, 34.4%; Extended Data Fig. 2b).

115 Among the various clinicopathologic parameters, we evaluated the potential predictive

116 markers in relation to survival. Patients were enrolled irrespective of PD-L1 status, and PD-

117 L1 expression was not correlated with survival or response with cut-off of CPS 1 (Fig. 1c-d,

118 Extended Data Fig. 3). Other laboratory test results that were reported to be associated with

119 immunotherapy response including lactate dehydrogenase (LDH), albumin, neutrophil-to-

120 lymphocyte ratio, and baseline tumor size were not related to survival (Extended Data Fig. 4).

121 The median number of treatment cycles was twelve (interquartile range [IQR], 8–24), and the

122 median number of cycles of capecitabine was eight (IQR, 6–12) and cisplatin was six (IQR,

123 4–7). Median number of maintenance cycles (beyond 6th cycle) was six (IQR, 2–12) with

124 median duration of 3.9 months (95% CI 2.13–11.57) (Extended Data Table 2). Treatment-

125 related adverse events occurred in 42 patients (97.7%) (Extended Data Table 3). Non-

126 hematologic adverse events were mostly graded 1 or 2, and the commonly reported grade 3

127 adverse events were mostly hematologic, including neutropenia (39.5%), anemia (16.3%),

128 febrile neutropenia (9.3%), and thrombocytopenia (7%). Grade 4 adverse events occurred in

129 two patients (neutropenia and thromboembolic event). One patient died from grade 5

6 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

130 pneumonia which is related to disease progression. Immune-related adverse events occurred

131 in 16 patients (37.2%), but none required pembrolizumab discontinuation. Grade 3 immune-

132 related adverse events included colitis (7%) and hypersensitivity (2.3%). Adding

133 pembrolizumab to trastuzumab and chemotherapy was tolerated and showed durable efficacy

134 in terms of tumor response or survival.

135

136 Genomic analyses of baseline tumor tissues

137

138 Among the 43 patients enrolled, either primary or metastatic tissues from 39 patients were

139 used for targeted next-generation sequencing (NGS) with an in-house DNA sequencing panel

140 (CancerMaster Panel V2, Extended Data Table 4). In the total 98 samples sequenced, 74

141 NGS cases from 35 patients were used for final analyses, with median sequencing depth of

142 1102x (range, 311–1,618). Baseline (pretreatment) tissues from 31 patients were available for

143 NGS (29 from primary tumor and two from metastatic liver), and their NGS results and

144 clinicopathologic features with response data are summarized in Fig. 2a. TP53/DNA repair,

145 receptor tyrosine kinase (RTK)/RAS, Wnt, epigenetic, Notch, and PI3K pathways were

146 commonly altered (selected genes altered in 6% or more of the patients are shown). No

147 patient had altered c-MET, and PD-L1 expression was not related to the genomic alteration

148 pattern. HER2 amplification was detected in 10 patients (23.3%) and 9 of them were HER2

149 IHC 3+ (Extended Data Table 5). When we compared the baseline HER2 overexpression

150 IHC status and survival, IHC 3+ (n=30) and IHC 2+ with SISH-positive (n=13) had similar

151 prognosis in terms of PFS and OS (Extended Data Fig. 5a-b). However, patients with HER2

152 amplification by NGS showed significantly longer PFS (22.0 vs. 7.7 months, hazard ratio

153 (HR) 0.35, 95% CI 0.13–0.93, P=0.0275) and OS (NR vs. 17.9 months, HR 0.21, 95% CI

154 0.05–0.92, P=0.0226), compared to those who did not (Fig. 2b, Extended Data Fig. 5c-d).

7 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

155 Interestingly, these patients with pre-treatment HER2 amplification assessed by NGS were

156 mostly durable responders (PFS over 12 months) (Extended Data Fig. 5e). De novo genetic

157 alterations in RTK/RAS pathway (excluding deletion of RTK and HER2 alterations) were

158 found in 67.7% of the patients (n=21, Extended Data Fig. 6a). Patients with these genetic

159 alterations showed significantly longer PFS (16.4 vs. 5.6 months, HR 0.12, 95% CI 0.04–

160 0.35, P<0.0001) and OS (NR vs. 10.4 months, HR 0.18, 95% CI 0.07–0.48, P=0.001),

161 compared to the patients without (Fig. 2c, Extended Data Fig. 6b). On the contrary, among

162 the first-line trastuzumab, capecitabine, and cisplatin-treated HER2-positive AGC patients in

163 our retrospective database cohort (n=18), patients with RTK/RAS pathway gene alterations

164 on pretreatment NGS (n=7) showed similar prognosis to that in patients without those

165 alterations (Extended Data Fig. 6c-d, Extended Data Table 6). This result suggests that

166 adding immunotherapy to the anti-HER2 targeting agent with cytotoxic chemotherapy might

167 overcome de novo resistance from, or even show synergy with, RTK/RAS pathway

168 alterations. One possible explanation could be a trend toward increased tumor mutation

169 burden (TMB) due to increased RTK/RAS pathway genes alterations (mean TMB, RTK/RAS

170 pathway altered vs. wild-type, 6.47 vs. 4.00 mutations/Mb, P=0.0846, Extended Data Fig. 6e-

171 f).

172 Patients with high TMB (defined as ≥10 mutations/Mb) had tendency for better survival

173 compared to those with low TMB (median PFS, 28.6 vs. 8.2, P=0.1274; median OS, 33.2 vs.

174 18.4, P=0.2401; Extended Data Fig. 7). Besides, patients with high human leukocyte antigen-

175 B (HLA-B)-corrected neoantigen load (> median) predicted by convolutional neural network

176 (CNN) model8 (calculating binding of the MHC class I molecules and somatic mutated

177 peptides from NGS results as previously described) showed better prognosis in terms of PFS

178 (22.0 vs. 7.5 months, HR 0.41, 95% CI 0.17–0.99, P=0.0393) and OS (NR vs. 17.4 months,

179 HR 0.32, 95% CI 0.11–0.92, P=0.0263) compared to those with low neoantigen load (Fig.

8 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

180 2d, Extended Data Fig. 8). In addition, HLA homozygosity, subtle HLA supertypes or alleles

181 (HLA-B44, HLA-B*15:01) which predicted immunotherapy response in a previous report9,

182 was partly related to the survival (Extended Data Fig. 9, 10). DNA damage response pathway

183 gene alterations, however, were not related to survival (Extended Data Fig. 11). When we

184 performed multivariate analysis, RTK/RAS pathway alteration was correlated with survival

185 (PFS, HR 0.17, 95% CI 0.05–0.58, P=0.005; OS, HR 0.28, 95% CI 0.08–0.92, P=0.046)

186 independent of HLA-B corrected neoantigen load and HER2 amplification by NGS

187 (Extended Data Table 7).

188 Finally, we could group the patients who would get less benefit from quadruplet regimen

189 predicted by baseline NGS results (de novo resistant), using three predictive features we

190 found (patients with Her2 non-amplification by NGS, low HLA-B corrected neoantigen load,

191 and non-altered RTK/RAS pathway) (Extended Data Fig. 12a). Among 26 patients who had

192 one or more those predictive features, 16 patients (61.5%) had two or more of those features,

193 meaning possible shared mechanism for de novo response/resistance. Nine patients (29.0%

194 out of available 31 patients) who had all three features for de novo resistance showed poor

195 survival to quadruplet regimen (median PFS, 5.6 vs. 15.1 months, P<0.0001; median OS,

196 11.4 vs. 31.2 months, P=0.0009) compared to those had two or less features by baseline NGS

197 (Extended Data Fig. 12b-e).

198

199 Sequential genomic analyses of multiple biopsied tumor tissues

200

201 On-treatment biopsied samples (n=24 from 15 patients) and post-progression biopsied

202 samples (n=16 from 12 patients) were available for additional NGS analyses (Extended Data

203 Fig. 13). First, we looked for HER2 loss10,11 or HER2 mutations12, which might serve as a

204 resistance mechanism to anti-HER2 treatment. While only the tissues of four patients whose

9 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

205 baseline HER2 amplification was confirmed were available for on-treatment or post-

206 progression NGS, tissues from 21 patients were available for HER2 IHC/SISH at baseline

207 and on-treatment or post-progression for comparison. Of these 21 patients, eight showed

208 negative conversion of HER2 expression (HER2 loss). Patients with HER2-negative

209 conversion showed worse survival compared to those with maintained HER2 positivity

210 (mPFS, 11.5 vs. 5.9, P=0.0361; mOS, 22.8 vs. 10.5, P=0.0156; Extended Data Fig. 14).

211 There was no patient with significant HER2 mutations from baseline NGS. In four patients,

212 HER2 mutations were detected from on-treatment or post-progression NGS (Fig. 3a)

213 including D769H, D769Y, and L869R HER2 alterations. Detected HER2 mutations were

214 previously known to be resistant to anti-HER2 treatment. Among them, HER2 mutations

215 explained the acquired resistance mechanism from the increase in resistant sub-clones in

216 three patients (patients YCC010, YCC026, and YCC037), which advocated the necessity for

217 re-biopsy.

218 To analyze sub-clonal evolution within the primary tumor, only paired NGS data from

219 stomach tissues were compared from 14 patients, either comparing on-treatment samples

220 (n=9) or post-progression samples (n=10). We hypothesized that sub-clone frequency

221 changed over 2-fold at post-progression or on-treatment samples compared to baseline

222 samples were either sensitive (decrement) or resistant (increment) sub-clones. More than 20

223 genes were selected as hotspot mutations from possible sensitive or resistant sub-clones, with

224 no dominant gene alteration (Fig. 3b, Extended Data Fig. 15, and Extended Table 8).

225 Interestingly, pathway analysis of these genes indicated that the pathways related to the

226 PI3K/AKT signaling, IRS-mediated signaling or ERBB4 signaling were related to sensitive

227 sub-clones, whereas CD28-related pathways or homologous DNA pairing and strand-

228 exchange-related pathways might be related to the resistant sub-clones (Fig. 3b, Extended

229 Table 9).

10 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

230

231 Some patients had serial-biopsied tissues available from both the primary tumor (stomach)

232 and metastatic site (liver). Serial comparison analyses among these patients showed spatial

233 tumor heterogeneity between the primary and metastatic sites, along with longitudinal tumor

234 heterogeneity throughout the treatment. The first representative case (best response: PR, Fig.

235 3c) showed that major sub-clones were decreased in the metastatic liver in frequency and

236 minor sub-clones appeared as progression at the primary tumor site, which indicated acquired

237 resistance. The second representative case (best response: progressive disease [PD], Fig. 3d)

238 showed an increase in major sub-clone frequency in the metastatic liver as the disease

239 progressed, which indicated a rather primary resistance. Both cases showed clonal expansion

240 as the disease progressed but without any recurrent actionable hotspot mutation.

241

242

243 DISCUSSION

244

245 In this study, patients with HER2-positive gastric cancer were treated with first-line

246 quadruplet regimen (pembrolizumab, trastuzumab, capecitabine, and cisplatin). The study has

247 reached its primary endpoint, with an ORR of 76.7% (planned target 60%) (CR 16.3% and

248 PR 60.5%). Absolute survivals (median PFS 8.6 months and median OS 19.3 months) were

249 longer than those in ToGA trial4 (median PFS 6.7 months and median OS 13.8 months),

250 which could be explained by the effect of adding pembrolizumab. With high efficacy, the

251 quadruplet regimen was also tolerable, which is comparable to previously reported results of

252 first-line trastuzumab plus chemotherapy4 or pembrolizumab, trastuzumab, and

253 chemotherapy13. This result suggests that adding pembrolizumab to trastuzumab plus

11 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

254 chemotherapy, an existing first-line standard, to treat patients with advanced HER2-positive

255 gastric and gastroesophageal junction cancer is effective and safe.

256

257 A recently reported phase II trial on pembrolizumab, trastuzumab, and cytotoxic

258 chemotherapy for HER2-positive esophageal or gastric adenocarcinoma patients13 was

259 different in design from our study; they included esophageal adenocarcinoma patients (38%)

260 (rarely present in Asia) and HER2-negative patients (16%). Their regimen required an

261 induction phase (1 cycle) of pembrolizumab and trastuzumab alone without cytotoxic

262 chemotherapy, which is usually not feasible in a clinical setting. In addition, most of the

263 patients were treated with cytotoxic chemotherapy consisting of oxaliplatin (97%), whereas

264 our study only used the same chemotherapy backbone (capecitabine plus cisplatin) and dose

265 administration schedule as those used in the ToGA trial. Although controversial, meta-

266 analyses have suggested that oxaliplatin-based regimens were superior in terms of response

267 or survival compared to cisplatin-based regimens as first-line treatment for advanced gastric

268 cancer patients14,15. Moreover, oxaliplatin was regarded as capable of inducing more

269 immunogenic tumor cell death than cisplatin16,17. We hope that the durable efficacy from our

270 study will be confirmed in the currently ongoing phase III Keynote-811 study

271 (NCT03615326), which is using the same dose administration schedule as used in our study,

272 permitting a choice of the cytotoxic chemotherapy backbone between capecitabine plus

273 oxaliplatin or 5-fluorouracil plus cisplatin.

274

275 Subsets of patients treated with anti-PD-1 or anti-PD-L1 inhibitors often suffer from an

276 unexpected acceleration of tumor growth following initiation of immunotherapy, which is

277 termed as hyperprogressive disease18. Immune checkpoint blockade monotherapies are

278 generally recognized to have higher chances of showing such early progression compared to

12 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

279 cytotoxic chemotherapy19-22. Patients treated with quadruplet regimen in our study, however,

280 showed a low rate of early progression (only one patient with the best response as progressive

281 disease), which also could be confirmed by the low hazard rate at the initial phase of

282 treatment (Extended Data Fig. 16). As demonstrated in the ATTRACTION-0423 and

283 Checkmate-64924 studies in the chemotherapy plus immunotherapy arm in a first-line setting

284 for gastric cancer, chemotherapy and immunotherapy combination reduced the rate of

285 hyperprogression and early death.

286

287 Although only HER2-overexpressing patients were enrolled, not every patient demonstrated

288 HER2 amplification through NGS. Reports regarding concordances between HER2

289 amplification through NGS and HER2 expression through IHC/ISH are controversial25,26. In

290 this study, AGC patients with HER2 amplification detected through NGS responded better to

291 the quadruplet regimen in terms of response and survival, whereas HER2 IHC status (2+ or

292 3+) was not related to survival. Pretreatment HER2 SISH amplification index was correlated

293 with HER2 copy number by NGS (Extended Data Fig. 5d), implying that our results (survival

294 benefit for patients with HER2 amplified through NGS) indicated that the level of HER2 gene

295 amplification was important, and not the method to detect it, as previously reported27. In this

296 study, every patient with HER2 amplification through NGS was HER2 3+ through IHC;

297 among them, HER2 amplification index investigation by SISH was done in only three

298 patients. Although performing NGS to detect HER2 amplification for HER2 IHC-positive

299 patients or performing SISH or FISH for assessing patients with HER2 IHC 3+ is not

300 mandatory, accessing HER2 amplification in HER2-overexpressing AGC patients might

301 categorize a subset of patients with higher response and survival.

302

13 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

303 We utilized the NGS panel to detect genetic alterations in our patients. Patients with

304 RTK/RAS pathway gene alterations showed a favorable prognosis, which is contrary to the

305 previous report indicating that RTK/RAS pathway genes could serve as a bypass mechanism

306 or induce resistance to anti-HER2 treatments28. One technical difference is that our NGS

307 assessed mutations with VAF >1.0%, whereas targeted panel sequencing used in previous

308 studies13,28 usually assessed mutations with VAF > 5.0%. Our results showing higher HER2

309 amplification among patients with altered RTK/RAS pathway (six HER2 amplification by

310 NGS out of 21 altered RTK/RAS pathway patients [28.6%] vs. one HER2 amplification by

311 NGS out of 10 non-altered RTK/RAS pathway patients [10%]) suggest that anti-HER2

312 treatment combined with immunotherapy may play an active role in inhibiting HER2 and

313 RTK/RAS combined pathway activation. RTK/RAS pathway gene alterations showed a

314 tendency for higher TMB, with every TMB-high patient being in the RTK/RAS pathway

315 alteration group also (Extended Data Fig 6e-f). Indeed, the RTK/RAS pathway is related to

316 enhanced response to immunotherapy through stabilization or overexpression of PD-L129-31.

317

318 TMB has been regarded as an important marker to predict immunotherapy response32, but not

319 in this study, possibly due to the small number of TMB-high patients (n=3). HLA is

320 becoming important in predicting immunotherapy response8,9,33-35. In this study, we estimated

321 neoantigen load by predicting the binding of individual patient’s somatic mutation peptides to

322 the MHC class I molecules using deep learning algorithm (CNN). Interestingly, HLA-B-

323 corrected neoantigen load was significantly related to PFS or OS. Among HLA molecules,

324 HLA-B could be especially important for immune responses as HLA-B-restricted T-cell

325 responses are immunodominant compared to HLA-A or -C, in human immunodeficiency

326 virus36 or tuberculosis37 infection. Our study suggests that HLA-B-corrected neoantigen load,

327 compared to TMB, might be important for predicting response to immunotherapy in AGC

14 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

328 patients. Because HLA-B*15:01 allele was also related to favorable survival in this study

329 (Extended Data Fig. 10e-f), further exploration of the role of HLA subtle subtype/allele is

330 warranted in gastric cancer.

331

332 To the best of our knowledge, this study analyzed the largest number of paired biopsy

333 samples in AGC patients treated with homogenous immunotherapy-containing regimen.

334 Comprehensive analyses of serial biopsied samples confirmed HER2 loss or HER2 mutations

335 as acquired resistant mechanisms to anti-HER2 treatment. To get insight into different

336 responses among AGC patients treated with quadruplet regimen, paired biopsy samples from

337 primary (stomach) tumors were compared for identifying sensitive or resistant sub-clones;

338 sub-clonal evolution patterns were found to vary among patients (Extended Data Fig. 15). By

339 comparing cancer cell fractions between baseline and on-treatment or post-progression NGS,

340 sensitive and resistant genes were identified. Pathway analyses revealed mutations in the

341 signaling pathways related to PI3K, IRS, or ERBB4 as sensitive sub-clonal mutations. The

342 mutation sites from sensitive sub-clonal genes were mostly in non-kinase domains,

343 suggesting that these mutations might have acted by increasing antigenicity for

344 immunotherapy to be active rather than being directly resistant to HER2-targeted treatment.

345 In the resistant sub-clones, CD28 or DNA damage response were identified as enriched

346 pathways. Both CD28-mediated co-stimulation of T cell38 and DNA damage response

347 pathways39,40 are related to immunotherapy response, indicating that the main resistant

348 mechanism derived from sub-clonal evolution analyses are related to immunotherapy. When

349 we grouped the patients according to the mechanisms of acquired resistance (Her2 loss,

350 acquired HER2 mutation, and emergence of resistant sub-clones by clonal evolution), no

351 patients had all three features. This implies that acquired resistance to the quadruplet regimen

352 varies patient to patient (Extended Data Fig. 17).

15 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

353

354 This study has limitations, including the single-arm design, lack of immune cell profiling

355 from lack of specimen, and inadequate baseline tumors for NGS (12 patients). However, the

356 strength of this study is in including comprehensive NGS analyses of paired biopsy samples.

357 First-line quadruplet regimen (immune checkpoint inhibitor, anti-HER2 targeted agent,

358 doublet chemotherapy) led to significant tumor shrinkage in HER2-positive AGC. Tumor

359 tissue NGS is recommended for detecting co-pathway activation such as RTK-RAS, HER2

360 amplification, and HLA-B-adjusted neoantigen load. Re-biopsy after progression is

361 recommended for detecting such as HER2 loss or mutation, and sub-clonal changes with

362 resistance-associated mutations in immune pathway-related genes. Proposed correlative

363 biomarkers and molecular mechanisms that could be related to the response to quadruplet

364 regimen found from the NGS data of our study need to be validated through the on-going

365 phase III Keynote-811 multinational double-blind study, which is based on the protocol of

366 this study.

367

368

16 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

369 ONLINE METHODS

370

371 Trial design and procedure

372 The PANTHERA trial was an open-label, phase Ib/II trial performed at five academic cancer

373 centers in South Korea to evaluate the efficacy and safety of a quadruplet regimen

374 (pembrolizumab, trastuzumab, capecitabine, and cisplatin) as first-line therapy for HER2-

375 positive advanced gastric cancer. The trial was conducted in accordance with the Declaration

376 of Helsinki and the Guidelines for Good Clinical Practice (ClinicalTrials.gov identifier:

377 NCT#02901301). The trial protocol was approved by the Institutional Review Board of

378 Severance Hospital (Seoul, South Korea), Chonnam National University Hwasun Hospital

379 (Jeollanam-do, South Korea), Kangbuk Samsung Hospital (Seoul, South Korea), Gangnam

380 Severance Hospital (Seoul, South Korea), and Hallym University Medical Center (Anyang,

381 South Korea), and all patients have provided written informed consent before the enrollment.

382 The inclusion criteria were: (1) HER2-positive advanced gastroesophageal junction or gastric

383 adenocarcinoma. HER2-positive tumor was defined as either IHC 3+ or IHC 2+ in

384 combination with ISH +, as assessed by a local laboratory on primary or metastatic tumor, (2)

385 willing and able to provide written informed consent/assent for the trial, (3) at least 19 years

386 of age on the day of signing informed consent, (4) have measurable disease based on RECIST

387 (Response Evaluation Criteria In Solid Tumors) version 1.141, (5) have a performance status

388 of 0 or 1 on the Eastern Cooperative Oncology Group (ECOG) Performance Scale, (6)

389 demonstrate adequate organ functions including cardiac function, and (7) in case of female

390 subjects of childbearing potential, negative urine or serum pregnancy test result or the

391 willingness to use birth control. The exclusion criteria were: (1) currently participating and

392 receiving study therapy or having participated in a study of an investigational agent within 4

393 weeks of the first dose of treatment, (2) diagnosed with immunodeficiency or receiving

17 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

394 systemic steroid therapy or any other form of immunosuppressive therapy within 7 days prior

395 to the first dose of trial treatment (physiologic dose of systemic steroid was permitted), (3)

396 have a known history of active TBc (Bacillus Tuberculosis), (4) hypersensitivity to

397 pembrolizumab or any of its excipients, (5) has received a prior anti-cancer monoclonal

398 antibody (mAb) within 4 weeks prior to study Day 1 or who has not recovered (i.e., ≤ Grade

399 1 or at baseline) from adverse events due to agents administered more than 4 weeks earlier,

400 (6) has received prior chemotherapy, targeted small molecule therapy, or radiation therapy

401 within 2 weeks prior to study Day 1 or who has not recovered (i.e., ≤ Grade 1 or at baseline)

402 from adverse events due to a previously administered agent, (7) has a known additional

403 malignancy that is progressing or requires active treatment within 3 years, except basal cell

404 carcinoma of the skin, squamous cell carcinoma of the skin, in situ cervical cancer, and

405 thyroid cancer and who has undergone curative therapy, (8) has known active central nervous

406 system (CNS) metastases and/or carcinomatous meningitis, (9) has an active autoimmune

407 disease that required systemic treatment in the past 2 years, (10) has a known history of, or

408 any evidence of active, non-infectious pneumonitis, (11) has an active infection requiring

409 systemic therapy, (12) has known psychiatric or substance abuse disorders that would

410 interfere with cooperation with the requirements of the trial, (13) pregnant or breastfeeding,

411 or expecting to conceive or father children within the projected duration of the trial, starting

412 with the pre-screening or screening visit through 120 days after the last dose of trial

413 treatment, (14) received prior therapy with an anti-PD-1, anti-PD-L1, or anti-PD-L2 agent,

414 (15) has a known history of Human Immunodeficiency Virus (HIV) (HIV 1/2 antibodies),

415 (16) has known active Hepatitis B (HBsAg reactive and HBV DNA is detected) or Hepatitis

416 C (anti-HCV reactive and HCV RNA [qualitative] is detected), and (17) has received a live

417 vaccine within 30 days of the planned start of study therapy.

418 Phase Ib had a 3+3 design with dose level 1 and level -1. The starting dose (dose level 1) was

18 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

419 intravenous pembrolizumab 200 mg on day 1, intravenous trastuzumab (Trastuzumab-pkrb,

420 Herzuma®, Celltrion, Inc.) 6 mg/kg (after 8 mg/kg load) on day 1, oral capecitabine 1000

421 mg/m2 twice daily on days 1 to 14, and intravenous cisplatin 80 mg/m2 on day 1 every 3

422 weeks, and dose level -1 was reduced dose of cisplatin as 60 mg/m2. In phase II, the

423 recommended phase II dose (RP2D) from phase Ib was used. All the patients were treated

424 until documented disease progression, unacceptable toxicity, or up to 6 cycles for

425 capecitabine and cisplatin and up to 24 months for pembrolizumab and trastuzumab

426 (chemotherapies beyond 6th cycles were permitted per the investigator’s decision). Tumor

427 responses were evaluated every two cycles according to the RECIST version 1.1 criteria.

428 Toxicities were graded based on the National Cancer Institute Common Terminology Criteria

429 for Adverse Events 4.03. Left ventricular ejection fraction was measured by multigated

430 acquisition (MUGA) scan or echocardiography every 9 weeks during the first 8 cycles, and

431 every 12 weeks afterward.

432

433 Sample size

434 According to Simon’s two-stage minimax design42, a minimum sample size of 38 patients

435 was needed to accept the hypothesis that the true response rate was 60% with 80% power and

436 to reject the hypothesis that the response rate was less than 45%, with type I error of 0.2. At

437 the first stage, if there were fewer than 6 responses among the initial 13 patients, the study

438 would be stopped. The primary endpoint of the trial was the objective response rate (ORR)

439 according to RECIST version 1.1. The secondary endpoints included progression-free

440 survival (PFS), overall survival (OS), disease control rate (DCR), safety profile, and

441 exploratory biomarker analysis. PFS was defined as the time from the start of treatment until

442 the date of disease progression or death resulting from any cause. OS was measured from the

443 start of treatment to the date of death from any cause. PFS was measured until the last clinical

19 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

444 evaluation for those without a PFS event and at date ‘last seen’ for those still alive. The

445 response rate was calculated as the percentage of patients who achieved a confirmed CR or

446 PR, and DCR was calculated as response rate + stable disease, per the RECIST version 1.1

447 guidelines.

448

449 Clinicopathologic feature analyses

450 We collected data on the following variables: age at the beginning of treatment; sex; ECOG

451 performance status; pathology; plasma levels of albumin, lactate dehydrogenase (LDH),

452 carcinoembryonic antigen (CEA), and carbohydrate antigen (CA) 19-9; neutrophil and

453 lymphocyte counts; neutrophil-to-lymphocyte ratio (NLR); and metastatic organs. Parameters

454 that were previously reported to be related to immunotherapy response – LDH (LDH > upper

455 limit of normal), albumin (albumin <3.5 mg/dL), NLR (NLR >6), and baseline tumor size

456 (baseline tumor size > median) - were analyzed43,44.

457

458 Histological analyses

459 Tumor tissues were fixed in 10% formalin and embedded in paraffin and cut into 4 μm-thick

460 tissue sections for further analyses. Immunohistochemistry (IHC) staining was performed

461 using the Ventana Benchmark XT automated staining system (Ventana Medical Systems,

462 Tucson, AZ, USA) according to the manufacturer’s protocol.

463 For HER2 status, anti-HER2/neu antibody (Clone 4B5; Ventana Medical Systems) was used

464 for IHC, and HER2 expression scoring system was applied according to the guideline45. In

465 addition, HER2 DNA amplification was assessed using the silver in situ hybridization (SISH)

466 method. SISH was performed with INFORM® HER2 DNA and chromosome 17 (CEP17)

467 probes (Ventana Medical Systems) using a Ventana Benchmark XT automated staining

20 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

468 system, according to the manufacturer’s instructions. HER2 DNA amplification was defined

469 as HER2/CEP17 ratio of ≥ 2.0.

470 Epstein–Barr virus (EBV) status was assessed using EBV-encoded small RNA in situ

471 hybridization (EBER-ISH) using standard protocols. The INFORM® EBER Probe (Ventana

472 Medical Systems, Tucson, USA) was used to perform automated staining, according to the

473 manufacturer’s instructions. Positive staining was defined by diffuse staining of tumor cells.

474 Tumor tissue microsatellite instability status was determined using IHC for MutL homolog 1

475 (MLH1; Clone M1; Ventana Medical Systems), MutS protein homolog 2 (MSH2; clone

476 G219-1129; Cell Marque, Rocklin, CA, USA), postmeiotic segregation increased 2 (PMS2;

477 clone MRQ-28; 1:40; Cell Marque), and MutS homolog 6 (MSH6; Clone 44; 1:100; Cell

478 Marque) in FFPE tissue sections. Loss of staining was defined as complete loss of nuclear

479 staining in all the tumor nuclei with preserved staining of lymphocytes and/or non-neoplastic

480 gastric foveolar epithelium.

481 For PD-L1 status, IHC staining was carried out using Dako PD-L1 IHC 22C3 pharmDx kit

482 (Agilent, Santa Clara, CA, USA) with EnVision FLEX visualization system and

483 counterstained with hematoxylin according to the manufacturer’s instructions. PD-L1 protein

484 expression was determined using a combined positive score (CPS), which was defined as the

485 number of PD-L1 staining cells (tumor cells, lymphocytes, macrophages) divided by the total

486 number of viable tumor cells.

487 Antibodies recognizing EGFR (1:100, EP38Y, Abcam, Cambridge, UK), PTEN (1:100, clone

488 138G6, Cell Signaling Technology, Danvers, MA, USA) were also used for IHC, as

489 previously described46.

490

491 Tumor sample collection and in-house next generation sequencing (NGS)

21 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

492 Tumor tissues either from primary or metastatic tumors were obtained. Of quality-controlled

493 (QC) samples, 98 were sequenced with the in-house panel sequencing, CancerMaster Panel

494 V247. The CancerMaster Panel V2 covers 524 genes for single nucleotide variants (SNVs),

495 143 for copy-number variations (CNVs), and 18 for fusions (Extended Data Table 3). Tumor

496 DNA was extracted from freshly obtained tissues or formalin-fixed, paraffin-embed archival

497 tissues using a QIAamp DNA FFPE Tissue Kit (Qiagen, Germany) according to the

498 manufacturer’s instructions. We determine that DNA Integrity Number (DIN), measured

499 using 4200 Tapestation (Agilent) < 3.5 or sheared DNA ratio (ratio of total amount of DNA

500 after shearing, divided by initial sample DNA amount) < 0.4 as QC cutoff. Sequencing

501 libraries were prepared using the Celemics Library Preparation Kit (Celemics). To capture all

502 target regions, NGS libraries and capture probes were hybridized using Celemics Targeted

503 Sequencing Kit (Celemics). Pooled libraries containing captured DNA fragments were

504 subsequently sequenced on Illumina NextSeq 500 Sequencing System as 2x 100 bp paired-

505 end reads.

506 Sequencing reads were processed by our in-house bioinformatics pipeline following the

507 GATK best practices workflows for somatic short variant discovery or the recommended best

508 practices for the Illumina system. Paired-end reads were mapped to the reference genome

509 (GRCh37/hg19) using BWA (v0.7.10). Removal of duplicate reads and base quality

510 recalibration were processed using Picard (v1.115). SNVs and small insertion and deletions

511 (INDELs) were detected using VarScan 2.4.0 and GATK IndelRealigner 2.3.9 with default

512 parameters. The following quality criteria were further applied for the detected variant: total

513 depth ≥ 200, alter read depth ≥ 2 and strandRatio < 0.9, variant allele frequency (VAF) of

514 SNV ≥ 1%, VAF of InDel ≥ 10%, and no strand bias. Manual curations from the annotation

515 databases COSMIC v91, OncoKB, ClinVar, ANNOVAR, and literature analysis were also

516 performed by expert reviewers, and clinically relevant mutations were additionally included.

22 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

517 Copy number alterations were identified by comparing depth of coverages over targeted

518 regions in a tumor sample, relative to a reconstructed baseline. For internal normalization of

519 sequencing depth variation, regional depth was divided by the median depth of each sample.

520 For evaluating copy number alteration, we also used segmental copy number which was

521 calculated using a circular binary segmentation (CBS) algorithm with R package ‘DNAcopy’.

522 Copy gain or loss was defined genetic fold change over 4 or lower than 0.5, and also at least

523 more than 3 regions had the same events. Fusion genes were identified with TopHat-Fusion

524 2.0.13 and SOAP-HLA (v2.2) were used for typing of human leukocyte antigen (HLA).

525 For patients in whom in-house NGS was not performed but pre-treatment FoundationOne

526 CDx assay was done, HER2 amplification status from the reports was used (Extended Data

527 Table 4).

528

529 Tissue genomic analyses

530 The tumor mutation burden (TMB; mutations[mut]/megabase[Mb]) was estimated as the total

531 number of detected nonsynonymous mutations (SNP VAF > 5% and indel VAF > 10%;

532 putative germline mutations reported in population databases [Korean Variant Archive48,

533 Korean Reference Genome Database49, The Exome Aggregation Consortium50, 1000

534 Genomes Phase 351 were removed) divided by the length of the covered coding regions. TMB

535 value over 10 mut/Mb was considered as TMB-high.

536 HLA-corrected neoantigen load was calculated according to the previous report52. Briefly, a

537 pre-constructed convolutional neural network (CNN) model was used to predict the binding

538 of HLA-A or HLA-B molecules and mutated genes detected from in-house panel sequencing.

539 The interactions between 9-mer peptide sequences containing altered amino acids of mutated

540 genes and HLA sequence were projected into a 365 x 9 input matrix and the kernel of the

541 CNN model detected a specific binding pattern having high interaction preference53 in amino

23 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

542 acid level with positional effect (Extended Data Fig. 8A). Patients having HLA-corrected

543 neoantigen load more than the median value were predicted to have high neoantigen (13 for

544 HLA-A and 4 for HLA-B).

545 The genomic landscape plot (oncoplots, Fig. 2a) was generated using the R package

546 ‘maftools’54. Signaling pathways were annotated manually according to previously reported

547 curated oncogenic signaling pathways55. For the RTK/RAS pathway alteration group, HER2

548 gene alteration (amplification and mutations) and RTK gene deletions were excluded.

549 A DNA damage response (DDR)-related gene list was assembled56-58. Patients were classified

550 into DDR alteration group or wild-type group according to the results of targeted sequencing.

551 A patient was assigned to DDR alteration group if a homozygous deletion or deleterious

552 mutation were detected in DDR-related genes.

553 For the analyses of sub-clonal evolution in serial-biopsied samples, clonality inference in

554 tumors using phylogeny (CITUP) was used59. Mutations were removed as CITUP advised

555 (located on a copy number alteration region and X or Y chromosomes and VAF > 50%). R

556 packages ‘timescape’ and ‘fishplot’ were used for visualization. To determine resistant or

557 sensitive sub-clones, NGS results from baseline tissues were compared with those from on-

558 treatment or post-progression samples. Sub-clone frequencies that were changed over 2-fold

559 at post-progression or on-treatment samples compared to baseline samples were regarded as

560 sensitive (decrement) or resistant (increment) sub-clones (when both on-treatment and post-

561 progression samples were available, only the sub-clones with continuously decreasing or

562 increasing during treatment were selected). Biological pathways enriched by chosen genes as

563 sensitive or resistant sub-clones were annotated from Reactome database

564 (https://reactome.org) using R package ‘ReactomePA’60.

565

566 Statistical analyses

24 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

567 All statistical analyses were performed using statistical software package R 4.0.3 (R project;

568 The R Foundation for Statistical Computing, Vienna, Austria) or GraphPad Prism version 8

569 (GraphPad Software, San Diego, CA, USA). Survival curves were plotted using Kaplan–Meier

570 analysis and compared using the log-rank test. A Cox proportional hazards regression model

571 was used for univariate and multivariable analysis. Statistical differences between the means

572 were compared by the two-tailed, unpaired t-test for two groups, or determined using one-way

573 ANOVA followed by Tukey’s multiple comparison test for multiple groups unless otherwise

574 noted. P<0.05 was considered statistically significant.

575

25 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

576 Acknowledgements: This study was supported by Merck Sharp & Dohme Corp. (USA),

577 Celltrion (South Korea), Celemics (South Korea), Daewoong Pharmaceutical (South Korea),

578 and grants from the National R&D Program for Cancer Control, Ministry of Health and

579 Welfare, Republic of Korea (HA15C0005, HA16C0018). We thank the patients and their

580 families and caregivers for their participation in the study, all primary investigators and site

581 personnel, Hye Jin Choi (Yonsei University College of Medicine) for CancerMaster NGS

582 program, Sejung Park (Yonsei University College of Medicine) for statistical support, and

583 Seulkee Lee (Samsung Medical Center) for HLA-corrected neoantigen load prediction

584 modeling.

585

586 Author Contributions: S.Y.R., H.S.K., M.J., C.M.N., and H.C.C. designed the trial. C.-k.L.,

587 S.Y.R., and H.C.C. analyzed and interpreted the clinical data. C.-k.L., S.Y.R., J.C., W.S.K.,

588 and H.C.C. performed genomic analyses. S.-J.S. and H.K. performed the

589 immunopathological analyses. C.-k.L., J.C., W.S.K., and C.M.N. performed statistical

590 analyses. S.K.L. performed endoscopies and tissue samplings. S.Y.R., H.S.K., M.J., B.K.,

591 W.K.B., D.-H.K., H.-C.J., D.Y.Z., and H.C.C. conducted clinical trial and contributed to the

592 collection and assembly of data. The manuscript was written by C.-k.L., S.Y.R. and H.C.C. in

593 collaboration with co-authors, who vouch for the accuracy of the data reported and adherence

594 to the protocol.

595 Competing Interests: S.Y.R. reported grant/research support from MSD, Celltrion,

596 Boehringer-Ingelheim, Eli Lilly, Taiho, Bristol-Myers Squibb, ASLAN, Incyte, consultation

597 for Daiichi Sankyo, MSD, Eli Lilly, Bristol-Myers Squibb, Eisai, and speaker’s bureau for Eli

598 Lilly, Bristol-Myers Squibb, MSD, outside of the submitted work. H.C.C. reported

599 grants/research support from Eli Lilly, GlaxoSmithKline, MSD, Merck-Serono, Bristol-

600 Myers Squibb, Taiho, Amgen, Beigene, Incyte, Zymework, honoraria from Merck-Serono,

26 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

601 Eli Lilly, and consultation for Taiho, Celltrion, MSD, Eli Lilly, Bristol-Myers Squibb,

602 Merck-Serono, Gloria, Beigene, Amgen, Zymework, outside of the submitted work. The

603 other authors declare no conflicts of interest.

604 Author Information: Reprints and permissions information is available at

605 www.nature.com/reprints. Correspondence and requests for materials should be addressed to

606 H.C.C. ([email protected]).

27 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

607 REFERENCE

608 609 1. Bray, F., et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and 610 mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 68, 394-424 611 (2018). 612 2. Hong, S., et al. Cancer Statistics in Korea: Incidence, Mortality, Survival, and 613 Prevalence in 2017. Cancer Res Treat 52, 335-350 (2020). 614 3. Van Cutsem, E., et al. HER2 screening data from ToGA: targeting HER2 in gastric and 615 gastroesophageal junction cancer. Gastric Cancer 18, 476-484 (2015). 616 4. Bang, Y.J., et al. Trastuzumab in combination with chemotherapy versus chemotherapy 617 alone for treatment of HER2-positive advanced gastric or gastro-oesophageal junction 618 cancer (ToGA): a phase 3, open-label, randomised controlled trial. Lancet 376, 687- 619 697 (2010). 620 5. Chaganty, B.K.R., et al. Trastuzumab upregulates PD-L1 as a potential mechanism of 621 trastuzumab resistance through engagement of immune effector cells and stimulation 622 of IFNgamma secretion. Cancer Lett 430, 47-56 (2018). 623 6. Stagg, J., et al. Anti-ErbB-2 mAb therapy requires type I and II interferons and 624 synergizes with anti-PD-1 or anti-CD137 mAb therapy. Proc Natl Acad Sci U S A 108, 625 7142-7147 (2011). 626 7. Catenacci, D.V.T., et al. Margetuximab plus pembrolizumab in patients with previously 627 treated, HER2-positive gastro-oesophageal adenocarcinoma (CP-MGAH22-05): a 628 single-arm, phase 1b-2 trial. Lancet Oncol 21, 1066-1076 (2020). 629 8. Kim, K., et al. Predicting clinical benefit of immunotherapy by antigenic or functional 630 mutations affecting tumour immunogenicity. Nat Commun 11, 951 (2020). 631 9. Chowell, D., et al. Patient HLA class I genotype influences cancer response to 632 checkpoint blockade immunotherapy. Science 359, 582-587 (2018). 633 10. Pietrantonio, F., et al. HER2 loss in HER2-positive gastric or gastroesophageal cancer 634 after trastuzumab therapy: Implication for further clinical research. Int J Cancer 139, 635 2859-2864 (2016). 636 11. Seo, S., et al. Loss of HER2 positivity after anti-HER2 chemotherapy in HER2-positive 637 gastric cancer patients: results of the GASTric cancer HER2 reassessment study 3 638 (GASTHER3). Gastric Cancer 22, 527-535 (2019). 639 12. Wen, W., et al. Mutations in the Kinase Domain of the HER2/ERBB2 Gene Identified 640 in a Wide Variety of Human Cancers. J Mol Diagn 17, 487-495 (2015). 641 13. Janjigian, Y.Y., et al. First-line pembrolizumab and trastuzumab in HER2-positive 642 oesophageal, gastric, or gastro-oesophageal junction cancer: an open-label, single-arm, 643 phase 2 trial. Lancet Oncol 21, 821-831 (2020). 644 14. Montagnani, F., Turrisi, G., Marinozzi, C., Aliberti, C. & Fiorentini, G. Effectiveness 645 and safety of oxaliplatin compared to cisplatin for advanced, unresectable gastric cancer: 646 a systematic review and meta-analysis. Gastric Cancer 14, 50-55 (2011). 647 15. Zhang, F., Zhang, Y., Jia, Z., Wu, H. & Gu, K. Oxaliplatin-Based Regimen is Superior 648 to Cisplatin-Based Regimen in Tumour Remission as First-line Chemotherapy for 649 Advanced Gastric Cancer: A Meta-Analysis. J Cancer 10, 1923-1929 (2019). 650 16. Tesniere, A., et al. Immunogenic death of colon cancer cells treated with oxaliplatin. 651 Oncogene 29, 482-491 (2010). 652 17. Galluzzi, L., Humeau, J., Buque, A., Zitvogel, L. & Kroemer, G. Immunostimulation 653 with chemotherapy in the era of immune checkpoint inhibitors. Nat Rev Clin Oncol 17, 654 725-741 (2020).

28 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

655 18. Champiat, S., et al. Hyperprogressive Disease Is a New Pattern of Progression in 656 Cancer Patients Treated by Anti-PD-1/PD-L1. Clin Cancer Res 23, 1920-1928 (2017). 657 19. Shitara, K., et al. Efficacy and Safety of Pembrolizumab or Pembrolizumab Plus 658 Chemotherapy vs Chemotherapy Alone for Patients With First-line, Advanced Gastric 659 Cancer: The KEYNOTE-062 Phase 3 Randomized Clinical Trial. JAMA Oncol (2020). 660 20. Rizvi, N.A., et al. Durvalumab With or Without Tremelimumab vs Standard 661 Chemotherapy in First-line Treatment of Metastatic Non-Small Cell Lung Cancer: The 662 MYSTIC Phase 3 Randomized Clinical Trial. JAMA Oncol 6, 661-674 (2020). 663 21. Carbone, D.P., et al. First-Line Nivolumab in Stage IV or Recurrent Non-Small-Cell 664 Lung Cancer. N Engl J Med 376, 2415-2426 (2017). 665 22. Andre, T., et al. Pembrolizumab in Microsatellite-Instability-High Advanced 666 Colorectal Cancer. N Engl J Med 383, 2207-2218 (2020). 667 23. Boku, N., et al. Nivolumab plus chemotherapy versus chemotherapy alone in patients 668 with previously untreated advanced or recurrent gastric/gastroesophageal junction 669 (G/GEJ) cancer: ATTRACTION-4 (ONO-4538-37) study. Annals of Oncology 31, 670 S1192-S1192 (2020). 671 24. Moehler, M., et al. Nivolumab (nivo) plus chemotherapy (chemo) versus chemo as 672 first-line (1L) treatment for advanced gastric cancer/gastroesophageal junction cancer 673 (GC/GEJC)/esophageal adenocarcinoma (EAC): First results of the CheckMate 649 674 study. Annals of Oncology 31, S1191-S1191 (2020). 675 25. Ross, D.S., et al. Next-Generation Assessment of Human Epidermal Growth Factor 676 Receptor 2 (ERBB2) Amplification Status: Clinical Validation in the Context of a 677 Hybrid Capture-Based, Comprehensive Solid Tumor Genomic Profiling Assay. J Mol 678 Diagn 19, 244-254 (2017). 679 26. Niu, D., et al. Evaluation of Next Generation Sequencing for Detecting HER2 Copy 680 Number in Breast and Gastric Cancers. Pathol Oncol Res 26, 2577-2585 (2020). 681 27. Gomez-Martin, C., et al. Level of HER2 gene amplification predicts response and 682 overall survival in HER2-positive advanced gastric cancer treated with trastuzumab. J 683 Clin Oncol 31, 4445-4452 (2013). 684 28. Janjigian, Y.Y., et al. Genetic Predictors of Response to Systemic Therapy in 685 Esophagogastric Cancer. Cancer Discov 8, 49-58 (2018). 686 29. Jiang, J., et al. Integrated genomic analysis identifies a genetic mutation model 687 predicting response to immune checkpoint inhibitors in melanoma. Cancer Med 9, 688 8498-8518 (2020). 689 30. Stutvoet, T.S., et al. MAPK pathway activity plays a key role in PD-L1 expression of 690 lung adenocarcinoma cells. J Pathol 249, 52-64 (2019). 691 31. Garcia-Aranda, M. & Redondo, M. Targeting Protein Kinases to Enhance the Response 692 to anti-PD-1/PD-L1 Immunotherapy. Int J Mol Sci 20(2019). 693 32. Yarchoan, M., Hopkins, A. & Jaffee, E.M. Tumor Mutational Burden and Response 694 Rate to PD-1 Inhibition. N Engl J Med 377, 2500-2501 (2017). 695 33. Brown, S.D. & Holt, R.A. Neoantigen characteristics in the context of the complete 696 predicted MHC class I self-immunopeptidome. Oncoimmunology 8, 1556080 (2019). 697 34. Chowell, D., et al. Evolutionary divergence of HLA class I genotype impacts efficacy 698 of cancer immunotherapy. Nat Med 25, 1715-1720 (2019). 699 35. Shim, J.H., et al. HLA-corrected tumor mutation burden and homologous 700 recombination deficiency for the prediction of response to PD-(L)1 blockade in 701 advanced non-small-cell lung cancer patients. Ann Oncol 31, 902-911 (2020). 702 36. Kiepiela, P., et al. Dominant influence of HLA-B in mediating the potential co- 703 evolution of HIV and HLA. Nature 432, 769-775 (2004).

29 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

704 37. Lewinsohn, D.A., et al. Immunodominant tuberculosis CD8 antigens preferentially 705 restricted by HLA-B. PLoS Pathog 3, 1240-1249 (2007). 706 38. Hui, E., et al. T cell costimulatory receptor CD28 is a primary target for PD-1-mediated 707 inhibition. Science 355, 1428-1433 (2017). 708 39. Chung, J.H., et al. Prospective Comprehensive Genomic Profiling of Primary and 709 Metastatic Prostate Tumors. JCO Precis Oncol 3(2019). 710 40. Robinson, D., et al. Integrative Clinical Genomics of Advanced Prostate Cancer. Cell 711 162, 454 (2015). 712 713

30 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

714 Reference for Online Methods

715

716 41. Eisenhauer, E.A., et al. New response evaluation criteria in solid tumours: revised 717 RECIST guideline (version 1.1). Eur J Cancer 45, 228-247 (2009). 718 42. Simon, R. Optimal two-stage designs for phase II clinical trials. Control Clin Trials 10, 719 1-10 (1989). 720 43. Joseph, R.W., et al. Baseline Tumor Size Is an Independent Prognostic Factor for 721 Overall Survival in Patients with Melanoma Treated with Pembrolizumab. Clin Cancer 722 Res 24, 4960-4967 (2018). 723 44. Bigot, F., et al. Prospective validation of a prognostic score for patients in 724 immunotherapy phase I trials: The Gustave Roussy Immune Score (GRIm-Score). Eur 725 J Cancer 84, 212-218 (2017). 726 45. Bartley, A.N., et al. HER2 Testing and Clinical Decision Making in Gastroesophageal 727 Adenocarcinoma: Guideline From the College of American Pathologists, American 728 Society for Clinical Pathology, and the American Society of Clinical Oncology. J Clin 729 Oncol 35, 446-464 (2017). 730 46. Kim, H.S., et al. Comprehensive expression profiles of gastric cancer molecular 731 subtypes by immunohistochemistry: implications for individualized therapy. 732 Oncotarget 7, 44608-44620 (2016). 733 47. Kwon, W.S., et al. Development and validation of a targeted sequencing panel for 734 application to treatment-refractory solid tumor. Journal of Clinical Oncology 39, 245- 735 245 (2021). 736 48. Lee, S., et al. Korean Variant Archive (KOVA): a reference database of genetic 737 variations in the Korean population. Sci Rep 7, 4287 (2017). 738 49. Jung, K.S., et al. KRGDB: the large-scale variant database of 1722 Koreans based on 739 whole genome sequencing. Database (Oxford) 2020(2020). 740 50. Lek, M., et al. Analysis of protein-coding genetic variation in 60,706 humans. Nature 741 536, 285-291 (2016). 742 51. Genomes Project, C., et al. A global reference for human genetic variation. Nature 526, 743 68-74 (2015). 744 52. Kim, K., et al. Predicting clinical benefit of immunotherapy by antigenic or functional 745 mutations affecting tumour immunogenicity. Nat Commun 11, 951 (2020). 746 53. Jha, A.N., Vishveshwara, S. & Banavar, J.R. Amino acid interaction preferences in 747 proteins. Protein Sci 19, 603-616 (2010). 748 54. Mayakonda, A., Lin, D.C., Assenov, Y., Plass, C. & Koeffler, H.P. Maftools: efficient 749 and comprehensive analysis of somatic variants in cancer. Genome Res 28, 1747-1756 750 (2018). 751 55. Sanchez-Vega, F., et al. Oncogenic Signaling Pathways in The Cancer Genome Atlas. 752 Cell 173, 321-337 e310 (2018). 753 56. Lin, J., et al. Alterations in DNA Damage Repair Genes in Primary Liver Cancer. Clin 754 Cancer Res 25, 4701-4711 (2019). 755 57. de Bono, J., et al. Olaparib for Metastatic Castration-Resistant Prostate Cancer. N Engl 756 J Med 382, 2091-2102 (2020). 757 58. Abida, W., et al. Non-BRCA DNA Damage Repair Gene Alterations and Response to 758 the PARP Inhibitor Rucaparib in Metastatic Castration-Resistant Prostate Cancer: 759 Analysis From the Phase II TRITON2 Study. Clin Cancer Res 26, 2487-2496 (2020).

31 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

760 59. Malikic, S., McPherson, A.W., Donmez, N. & Sahinalp, C.S. Clonality inference in 761 multiple tumor samples using phylogeny. Bioinformatics 31, 1349-1356 (2015). 762 60. Yu, G. & He, Q.Y. ReactomePA: an R/Bioconductor package for reactome pathway 763 analysis and visualization. Mol Biosyst 12, 477-479 (2016). 764 765 766

32 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

767 FIGURE LEGENDS 768 769 Fig. 1 Clinical efficacy of the first-line quadruplet regimen in patients with HER2-positive

770 gastric cancer and their baseline genomic landscape.

771 a, A swimmer plot showing outcomes in all patients from the start of treatment to either disease

772 progression or the last follow-up. Note that seven patients finished 2-year treatment. b, Kaplan–

773 Meier survival curves with progression-free survival and overall survival in all patients. c,

774 Maximum percentage change from baseline in size of total tumor lesions with corresponding

775 best responses by RECIST 1.1 and PD-L1 combined positive score (CPS) from baseline tissue.

776 Lower dotted line represents tumor reduction of 30%. d, Spider plot showing percentage

777 change from baseline in target lesions over time during treatment, with corresponding PD-L1

778 CPS.

779

780 Fig. 2 Baseline genomic landscape in the enrolled HER2-positive gastric cancer patients

781 treated with the first-line quadruplet regimen.

782 a, Baseline (pretreatment) tumor tissue-targeted DNA sequencing results grouped by best

783 response and related clinicopathologic features (n=31). Curated pathways and selected genes

784 altered in 6% or more of the patients are shown. Other pathways include cell cycle, Hippo,

785 MYC, and NRF2 pathways. Vertical dashed lines indicate groups by best response. See Online

786 Methods for details. b-d, Kaplan–Meier survival curves with progression-free survival (PFS)

787 stratified by pretreatment HER2 amplification by NGS (b), RTK-RAS pathway genes

788 alteration (c), and HLA-B-corrected neoantigen load (d).

789

33 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

790 Fig. 3 Genomic analyses from serial biopsy samples of the HER2-positive gastric cancer

791 patients treated with the first-line quadruplet regimen.

792 a, Spider plot showing patients with HER2 mutation found in serial NGS analyses of primary

793 tumor. Only HER2 mutations and variant allele frequencies (VAFs) with the corresponding

794 patient IDs for detected cases are shown. b, Sensitive or resistant sub-clones in which the sub-

795 clone frequency changed over 2-fold in post-progression (Post-PD) samples (n=10) or on-

796 treatment (On-Tx) samples (n=9) compared to paired baseline samples (n=18) are selected per

797 patient. Selected enriched Reactome pathways from sensitive or resistant sub-clone genes are

798 also shown. All paired tissue samples are from primary tumor (stomach). c-d, Representative

799 cases showing sub-clonal evolution by fish plot and corresponding clinicopathologic features

800 and computed tomography or endoscopic images from paired tissue NGS from both primary

801 tumor (stomach) and metastatic liver. Selected hotspot mutations are labeled. Representative

802 case from good responder (c) and poor responder (d).

803

34 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

804 TABLES

805 Table 1. Baseline patients’ characteristics (N=43)

n (%) Median Age (year, range) 63 (34–82) Sex Male 33 (76.7%) Female 10 (23.3%) Previous gastrectomy performed with curative intent Yes 12 (27.9%) No 31 (72.1%) ECOG performance status 0 26 (60.5%) 1 17 (39.5%) Pathology AWD 2 (4.7%) AMD 27 (62.8%) APD 13 (30.2%) SRC 1 (2.3%) Metastatic organs Lymph node 36 (83.7%) Liver 21 (48.8%) Peritoneum 11 (25.6%) Lung 7 (16.3%) Bone 3 (7.0%) Adrenal gland 1 (2.3%) HER2 positivity IHC 3+ 30 (69.8%) IHC 2+ and SISH+ 13 (30.2%) Pretreatment PD-L1 status CPS < 1 (negative) 17 (39.5%) CPS ≥ 1 (positive) 21 (48.8%) CPS ≥ 10 5 (11.6%) Not done 5 (11.6%) Baseline CEAa < 5 ng/mL 15 (34.9%) ≥ 5 ng/mL 28 (65.1%) Baseline CA 19-9a,b < 34 U/mL 19 (46.3%)

35 1st-Line anti-PD-1 and anti-HER2 with chemotherapy for HER2-positive AGC Lee et al.

≥ 34 U/mL 22 (53.7%) EBV-positivityc 0 (0%) Deficient MMRd 0 (0%)

ECOG, Eastern Cooperative Oncology Group; AWD, adenocarcinoma well differentiated; AMD, adenocarcinoma moderately differentiated; APD, adenocarcinoma poorly differentiated; SRC, signet ring cell; IHC, immunohistochemistry; SISH, silver in situ hybridization; CPS, combined positive score; CEA, carcinoembryonic antigen; CA, carbohydrate antigen; EBV, Epstein-Barr virus. aTumor markers are grouped by upper limit of normal range. b41 out of 43 patients available cEBV-positivity by EBV-encoded small RNA in situ hybridization ddeficient MMR by immunohistochemistry for MLH1, MSH2, PMS2 and MSH6

806

36 Fig. 1 | Clinical efficacy of the first-line quadruplet regimen in patients with HER2-positive gastric cancer and their baseline genomic landscape. a, A swimmer plot showing outcomes in all patients from the start of treatment to either disease progression or the last follow-up. Note that seven patients finished 2-year treatment. b, Kaplan–Meier survival curves with progression-free survival and overall survival in all patients. c, Maximum percentage change from baseline in size of total tumor lesions with corresponding best responses by RECIST 1.1 and PD-L1 combined positive score (CPS) from baseline tissue. Lower dotted line represents tumor reduction of 30%. d, Spider plot showing percentage change from baseline in target lesions over time during treatment, with corresponding PD-L1 CPS.

a Complete response start b 100100 Partial response start Progression-free survival Two-year treatment end Overall survival Progression Conversion surgery Ongoing treatment 50 Follow up without Treatment 50 Subjects Percent survival

0 0 12 24 36 48 0 12 24 36 0 12 24 36 48 Months Time on treatment (months)

c 140 d PD-L1 < 1% (pretreatment) 120 60 PD-L1 ≥ 1% (pretreatment) PD-L1 ≥ 1%* Progressive disease 100 40 Stable disease Not available at diagnosis PD-L1 < 1%* 80 Confirmed partial response 2-year treatment end 20 Complete response 60 New lesion progression 40 0 20 -20 0 -20 -40 -40 -60 -60

-80 -80 Tumor burden change from baseline (%) -100 Maximumtumor reduction (Total lesions, %) -100 10 20 30 40 50 60 70 80 90 100 Time since start of treatment (weeks) Fig. 2 | Baseline genomic landscape in the enrolled HER2-positive gastric cancer patients treated with the first-line quadruplet regimen. a, Baseline (pretreatment) tumor tissue-targeted DNA sequencing results grouped by best response and related clinicopathologic features (n=31). Curated pathways and selected genes altered in 6% or more of the patients are shown. Other pathways include cell cycle, Hippo, MYC, and NRF2 pathways. Vertical dashed lines indicate groups by best response. See Online Methods for details. b-d, Kaplan–Meier survival curves with progression-free survival (PFS) stratified by pretreatment HER2 amplification by NGS (b), RTK-RAS pathway genes alteration (c), and HLA-B corrected neoantigen load (d).

a b 100 HER2 amplified by NGS

HER2 non-amplified by NGS freesurvival - 50 TP53/DNA repair

Percent progression mPFS: 22.0 vs 7.7, P=0.0275 0 0 12 24 36 48 Months RTK/RAS c 100 Selected RTK/RAS pathway genes

Altered

Wild-type freesurvival - WNT 50 Epigenetic

Percent progression mPFS: 16.4 vs 5.6, P<0.0001 0

NOTCH 0 12 24 36 48 Months PI3K

d 100 High Neoantigen (HLA-B corrected) Others Low Neoantigen (HLA-B corrected) freesurvival - 50

Percent progression mPFS: 22.0 vs 7.5 P=0.0393 0 0 12 24 36 48 Months Fig. 3 | Genomic analyses from serial biopsy samples of the HER2-positive gastric cancer patients treated with the first-line quadruplet regimen. a, Spider plot showing patients with HER2 mutation found in serial NGS analyses of primary tumor. Only cases with HER2 mutations and variant allele frequencies (VAFs) with the corresponding patient IDs are shown. b, Sensitive or resistant sub-clones in which the sub-clone frequency changed over 2-fold in post- progression (Post-PD) samples (n=10) or on-treatment (On-Tx) samples (n=9) compared to paired baseline samples (n=18) are selected per patient. Selected enriched Reactome pathways from sensitive or resistant sub-clone genes are also shown. All paired tissue samples are from primary tumor (stomach). c-d, Representative cases from good responder (c) and poor responder (d) showing sub-clonal evolution by fish plot and corresponding clinicopathologic features and computed tomography or endoscopic images from paired tissue NGS from both primary tumor (stomach) and metastatic liver. Selected hotspot mutations are labeled.

a b

TP53 Enriched Reactome Pathways 120 On-treatment NGS P=0.05 YCC026 0.6 BAP1

clones KMT2B Diseases of signal transduction 100 D769H 6.34% Post-progression NGS - 0.4 80 PI3K/AKT Signaling in Cancer YCC026 Progression with new lesion ELMO1 60 D769H 7.57% 0.2 PIK3CA IRS-mediated signaling

clone frequency PIK3CB 40 - 0.0 Signaling by ERBB4 Sensitive sub YCC037 Sub -log Adjusted P-value 20 Baseline On-Tx Post-PD 10 1 2 3 D769H 3.16% 0 Enriched Reactome Pathways -20 0.20 ERG Regulation of gene expression in P=0.05 YCC023 -40 clones beta cells - 0.15 D769Y 1.11% GABRA6 -60 APOB CD28 dependent PI3K/Akt 0.10 ROBO1 SEMA3E signaling -80 EPHB1 0.05 Homologous DNA pairing and YCC010 YCC010 clone frequency FGF12 Tumor burden change from baseline (%) -100 - strand exchange L869R 1.58% L869R 15.49% 0.00 GATA6 -log Adjusted P-value

Resistant sub 10 0 20 40 60 80 100 Sub Baseline On-Tx Post-PD 1 2 3 Time since start of treatment (weeks)

c Baseline On-treatment Progression d Baseline Progression ID: YCC006 ID: YCC033 PFS: 8.1 m PFS: 1.4 m OS: 22.8 m OS: 7.0 m TMB: 8.0 TMB: 5.1 May-2017 Nov-2017 Feb-2018 Aug-2018 Oct-2018

Stomach PD-L1 CPS 1 PD-L1 CPS 3 PD-L1 CPS 1 Stomach PD-L1 CPS 10 PD-L1 CPS 0 HER2 2+ AI 2.19 HER2 2+ AI 2.1 HER2 1+ AI 2.4 HER2 3+ HER2 3+ AI 4.45 ATM R2443X

SDHA L639V TP53 S149fs FLCN A264V PTPRD R141H SF3B1 R387W CRBN T403M Clonal Prevalence Clonal Prevalence

8.1 months on Pembrolizumab+Trastuzumab+XP 1.4 months on Pembrolizumab+Trastuzumab+XP

PD-L1 CPS 0 PD-L1 CPS 0 Liver HER2 2+ PD-L1 CPS 1 Liver HER2 3+ AI 2.9 HER2 1+

PTPRD R141H SDHA L639V ATM R2443X Clonal Prevalence Clonal Prevalence DAXX S138X MAP3K1 S1358L Baseline On-treatment Post-progression Baseline Baseline Post-progression May-2017 Dec-2017 Feb-2018 Aug-2018 Sep-2018 Oct-2018 a

Assessed for eligibility (n=45)

Screening failure (n=2) - Not met inclusion criteria (n=2)

Enrolled and evaluated for efficacy and safety (n=43) - Phase 1b (n=3) - Phase 2 (n=40)

On treatment (n=3) Discontinued treatment (n=40) - Disease progression (n=33) - 2-year treatment end (n=7)

Analyzed for NGS (n=35) - Pretreatment tissues (n=31) - On-treatment tissues (n=15) - Post-progression tissues (n=12)

b Phase Ib: RP2D Phase II: Efficacy of RP2D

HER2-positive Level (1) Treatment naïve Pembrolizumab (Pem, 200mg) Quadruplet Maintenance Recurrent Trastuzumab (T, 8/6mg) Combination Pem+T and/or + Pem+T+XP (Until PD metastatic Capecitabine (X, 100mg/m2) (6 cycles) or up to 2 yrs) gastric cancer Cisplatin (P, 80mg/m2)

Level (-1) Pembrolizumab (Pem, 200mg) Trastuzumab (T, 8/6mg) + Capecitabine (X, 100mg/m2) Cisplatin (P, 60mg/m2)

Extended Data Fig. 1 | Consort diagram and clinical trial design. Progression-free survival (PFS) 8.6 months Median PFS (95% CI 7.2-16.4) 6-month PFS 79.1% 12-month PFS 41.9% Overall survival (OS) 19.3 months Median OS (95% CI 16.5-NR) 6-month OS 95.3% 12-month OS 80.1% Best response* Complete response 7 (16.3%) Partial response 26 (60.5%) Stable disease 9 (20.9%) Progressive disease 1 (2.3%) Objective response rate (ORR) 33 (76.7%) Disease control rate 42 (97.7%) ORR at 12 week 31 (72.1%) ORR at 24 week 29 (67.4%) Duration of response (DOR) 1.7 months Median time to response (range, 1.2-5.8) 10.8 months Median DOR (95% CI 7.2-21.8) Median DOR for CR as best response 29.5 months Median DOR for PR as best response 8.0 months *confirmed response per RECIST v1.1 Extended Data Table 1 | Efficacy a

20

-30 Maximal tumor size reduction (%) reduction size tumor Maximal

Mean -47.5 -62.5 -62.5 -40.6 +40.9 -41.1 b 15

10 Patients 5

0 Liver Peritoneum LN Lung Others (n=11) (n=9) (n=8) (n=4) (n=3)

Extended Data Fig. 2 | Responses per metastatic organs a, Maximal metastatic tumor size reduction per organ in evaluable 41 patients. Others included adrenal gland (n=1) and esophageal (n=1) metastatic lesions. b, Bar graph showing metastatic organs of progression in evaluable 32 patients. Others included adrenal gland (n=2) and brain (n=1) metastatic lesions. a b

100100 100100 PD-L1 < 1% PD-L1 < 1% PD-L1 ≥ 1% PD-L1 ≥ 1% freesurvival - 5050 5050 Percent overall survival

Percent progression mPFS: 7.1 vs 8.4, P=0.8007 mOS: 17.9 vs 22.8, P=0.3662 0 0 00 1212 24 24 36 36 48 48 00 1212 24 24 36 36 48 48 Months Months c d

4040 100100 PD-L1 < 1% PD-L1 ≥ 1% 3030

2020 5050 L1 combinedL1 positive score - 1010 PD Percent duration ofresponse mDOR: 20.5 vs 7.2, P=0.2470 0 0 00 1212 24 24 36 36 48 48 CR PR SD PD Months

Extended Data Fig. 3 | Baseline PD-L1 status and survival a-c, Kaplan Meier survival curves with progression-free survival (PFS, a), overall survival (OS, b), and duration of response (DOR, c) stratified by pretreatment PD-L1 combined positive score (n=38). d, Association between PD-L1 combined positive score from baseline tumor tissue and best response by RECIST. Bars and error bars, mean±SD. a b

100 100 LDH ≤ ULN LDH ≤ ULN LDH > ULN LDH > ULN freesurvival -

50 50 Percent overall survival

Percent progression mPFS: 8.4 vs 10.7, P=0.9378 mOS: 21.1 vs 19.3, P=0.7880 0 0 0 12 24 36 48 0 12 24 36 48 Months Months c d 100 100 Albumin < 3.5g/dL Albumin < 3.5g/dL Albumin ≥ 3.5g/dL Albumin ≥ 3.5g/dL freesurvival -

50 50 Percent overall survival

Percent progression mPFS: 11.3 vs 8.4, P=0.8882 mOS: 18.4 vs NR, P=0.2545 0 0 0 12 24 36 48 0 12 24 36 48 Months Months

e f 100 100 NLR ≤ 6 NLR ≤ 6 NLR > 6 NLR > 6 freesurvival -

50 50 Percent overall survival

Percent progression mPFS: 9.0 vs 8.5, P=0.9560 mOS: 19.3 vs 19.3, P=0.9298 0 0 0 12 24 36 48 0 12 24 36 48 Months Months g h

100 100 Baseline tumor size ≤ median Baseline tumor size ≤ median Baseline tumor size > median Baseline tumor size > median freesurvival -

50 50 Percent overall survival

Percent progression mPFS: 11.9 vs 8.1, P=0.0806 mOS: 18.9 vs 21.1, P=0.9007 0 0 0 12 24 36 48 0 12 24 36 48 Months Months

Extended Data Fig. 4 | Baseline clinical characteristics and survival a-f, Kaplan Meier survival curves with progression-free survival (PFS, a, c, e, g) and overall survival (OS, b, d, e, h) stratified by pretreatment lactate dehydrogenase level (LDH, a, b), albumin level (c, d), neutrophil- to-lymophocyte ratio (NLR, e, f) and baseline tumor size (median baseline tumor size=100.74mm, g, h). LDH, lactate dehydrogenase; ULN, upper limit of normal (247 U/L for LDH). Median (Interqaurtile range) Total treatment cycles 12 (8-24) Pembrolizumab+Trastuzumab (Pem+T) 12 (8-24) Capecitabine (X) 8 (6-12) Cisplatin (P) 6 (4-7) Quadruplet cycles (Pem+T+XP) 6 (4-7) Chemotherapy-free maintenance Cisplatin-free cycles (Pem+T±X) 5 (1-20) Capecitabine-free cycles (Pem+T±P) 1 (0-13) XP-free cycles (Pem+T) 1 (0-13) Maintenance cycles (beyond 6th cycle)* 6 (2-18) *Number of cycles before 6th cycles w as also included w hen XP w ere omitted earlier.

Extended Data Table 2 | Treatment cycles Any Grade Grade 3 Grade 4 Event n % n % n % Hematologic Neutrophil count decreased 20 46.5 17 39.5 1 2.3 Anemia 14 32.6 7 16.3 0 0 Platelet count decreased 7 16.3 3 7 0 0 Febrile Neutropenia 5 11.6 4 9.3 0 0 Non-hematologic Anorexia 17 39.5 2 4.7 0 0 Nausea 14 32.6 0 0 0 0 Creatinine increased 13 30.2 2 4.7 0 0 Diarrhea 13 30.2 1 2.3 0 0 Hand-foot syndrome 10 23.3 0 0 0 0 Oral mucositis 10 23.3 1 2.3 0 0 General weakness 8 18.6 1 2.3 0 0 Peripheral neuropathy 8 18.6 0 0 0 0 Abdominal pain 6 14 0 0 0 0 Fever 6 14 0 0 0 0 Hypoalbuminemia 2 4.7 1 2.3 0 0 Hyperkalemia 2 4.7 2 4.7 0 0 Tinnitus 2 4.7 0 0 0 0 Thromboembolic event 2 4.7 0 0 1 2.3 Immune-related Hypothyroidism 5 11.6 0 0 0 0 Hypersensitivity 4 9.3 1 2.3 0 0 Pruritus 4 9.3 0 0 0 0 Colitis 3 7 3 7 0 0 Adrenal insufficiency 2 4.7 0 0 0 0 Hyperglycemia 2 4.7 0 0 0 0 Skin rash 2 4.7 0 0 0 0 Total 42 97.7 33 76.7 2 4.7 One patient died from G5 pneumonia related to disease progression.

Extended Data Table 3 | Treatment-related adverse events ABL1 ABL2 ACVR1 ACVR1B ADAM29 ADGRA2 AKT1 AKT2 AKT3 ALK ALOX12B ALOX15B AMER1 APC APCDD1 APEX1 APOB APOBEC1 APOBEC3A APOBEC3B AR ARAF ARFRP1 ARID1A ARID1B ARID2 ASXL1 ATM ATP11B ATR ATRX AURKA AURKB AXIN1 AXL B2M B3GA T1 BACH1 BAP1 BARD1 BCL2 BCL2A1 BCL2L1 BCL2L2 BCL6 BCL9 BCOR BCORL1 BCR BIRC2 BIRC3 BLM BRAF BRCA1 BRCA2 BRD2 BRD3 BRD4 BRIP1 BTG1 BTK BTLA CARD11 CASP5 CASP8 CBFB CBL CCDC150 CCDC168 CCDC43 CCL2 CCL4 CCND1 CCND2 CCND3 CCNE1 CD27 CD274 CD276 CD28 CD3D CD3E CD3G CD4 CD40 CD44 CD79A CD79B CD8A CDC42 CDC73 CDH1 CDH2 CDH20 CDH5 CDK12 CDK4 CDK6 CDK8 CDKN1A CDKN1B CDKN2A CDKN2B CDKN2C CDX2 CEBPA CHD1 CHD2 CHD4 CHEK1 CHEK2 CHUK CIC CRBN CREBBP CRKL CRLF2 CSF1R CSF2 CSF2RA CSF2RB CSNK2A1 CTCF CTLA4 CTNNA1 CTNNB1 CUL3 CUL4A CUL4B CXCL10 CXCL11 CXCL9 CXCR3 CYLD CYP17A1 DAXX DCUN1D1 DDR2 DICER1 DIS 3 DNMT1 DNMT3A DOCK2 DOT1L EGFR ELMO1 EML4 EMSY EP300 EPHA3 EPHA5 EPHA6 EPHA7 EPHB1 EPHB4 EPHB6 ERBB2 ERBB3 ERBB4 ERCC1 ERCC2 ERG ERRFI1 ESR1 ETV1 ETV4 ETV5 ETV6 EWSR1 EYA 2 EZH2 FAM46C FANCA FANCC FANCD2 FANCE FANCF FANCG FANCI FANCL FANCM FAS FAT1 FAT3 FBXW7 FGF1 FGF10 FGF12 FGF14 FGF19 FGF2 FGF23 FGF3 FGF4 FGF6 FGF7 FGFR1 FGFR2 FGFR3 FGFR4 FH FLCN FLT1 FLT3 FLT4 FOXA1 FOXL2 FOXO3 FOXP1 FOXP3 FRS2 FUBP1 GABRA6 GAS 6 GATA1 GATA2 GATA3 GATA4 GATA6 GID4 GLI1 GNA 11 GNA 13 GNA Q GNA S GRIN2A GRM 3 GSK3B GUCY1A 2 GZM A GZM B GZM H H3F3A HGF HIS T1H3B HLA-A HLA-B HLA-C HLA-DRA HLA-E HLA-F HLA-G HNF1A HOXA3 HRAS HSD3B1 HSP90AA1 IDH1 IDH2 IDO1 IDO2 IFITM1 IFITM3 IFNA1 IFNB1 IFNG IGF1 IGF1R IGF2 IGF2R IKBKE IKZF1 IL12A IL12B IL2 IL23A IL6 IL7R INHBA INPP4B INSR IRF2 IRF4 IRS2 ITGA E ITK JAK1 JAK2 JAK3 JUN KAT6A KDM5A KDM5C KDM6A KDR KEAP1 KEL KIT KLF4 KLHL6 KMT2A KMT2B KMT2C KNSTRN KRAS LA G3 LMO1 LRP1B LRP6 LTK LYN LZTR1 M A GI2 M A GOH MAML1 MAP2K1 MAP2K2 MAP2 K4 MAP3K1 MAP3K13 MAPK1 MAX MCL1 MDM2 MDM4 MED12 MEF2B MEN1 MET MITF MLH1 MPL MRE1 1 MS H2 MSH6 MTOR M UTYH M YB MYC MYCL MYCN M YD88 MYO18A NCOA3 NCOR1 NF1 NF2 NFE2L2 NFKBIA NKX2-1 NKX2-8 NKX3-1 NOTCH1 NOTCH2 NOTCH3 NOTCH4 NPM1 NRAS NSD1 NSD3 NTRK1 NTRK2 NTRK3 NUP93 NUTM1 PAK3 PAK5 PALB2 PARP1 PARP2 PARP3 PARP4 PAX5 PBRM1 PDCD1 PDCD1LG2 PDGFRA PDGFRB PDK1 PGR PHF6 PHLPP2 PIK3C2B PIK3C3 PIK3CA PIK3CB PIK3CG PIK3R1 PIK3R2 PKHD1 PLCG1 PLCG2 PMS2 PNP PNRC1 POLD1 POLE PPARG PPP2R1A PRDM1 PREX2 PRF1 PRKAR1A PRKCI PRKDC PARK2 PRPF40B PRSS8 PTCH1 PTCH2 P TEN PTK2 PTPN11 PTPRC PTPRD QKI RAB35 RAC1 RAC2 RAD17 RAD50 RAD51 RAD51B RAD51C RAD51D RAD52 RAD54L RAF1 RANBP2 RARA RB1 RBM10 REL RET RHEB RHOA RHOB RICTOR RNF43 ROBO1 ROBO2 ROS1 RPA1 RPS 6KB1 RPTOR RUNX1 RUNX1T1 RUNX3 SDHA SDHB SDHC SDHD S EMA3 A SEMA3E SET S ETB P 1 SETD2 SF3A1 SF3B1 S H2B3 S KP2 SLIT2 SMAD2 SMAD3 SMAD4 SMARCA1 SMARCA4 SMARCB1 SMARCD1 SMO SNCAIP SOCS1 SOX10 SOX2 SOX9 SPEN SPOP SPTA1 SRC SRSF1 SRSF2 SRSF7 STA G2 STAT3 STAT4 S TK11 SUFU SYK TACC3 TAF1 TBX22 TBX3 TERC TERT TET2 TGFBR2 TIAF1 TIGIT TIPARP TMPRSS2 TNF TNFAIP3 TNFRSF14 TNFRSF18 TNFRSF4 TNFSF13B TNKS TNKS 2 TOP1 TOP2A TP53 TRAF7 TRRAP TSC1 TSC2 TSHR U2AF1 U2AF2 USP9X VEGFA VHL VSIR VTCN1 WISP3 WNT1 WT1 WWP1 XBP1 XPO1 XRCC3 ZBTB2 ZNF217 ZNF703 ZRSR2 Bold : CNV (143 genes), italic : translocation (18 genes)

Extended Data Table 4 | In-house NGS sequencing panel Pretreatment HER2 Pretreatment HER-2 HER2 Best Patient ID NGS HER2 copy NGS tissue IHC SISH AI# Response amplification number YCC030 Stomach Amplified 8.78 3+ Not done CR YCC001 Stomach Amplified 6.13 3+ 20 CR YCC021 Stomach Non-amplified 1.66 2+ 2.74 CR YCC015 Stomach Non-amplified 1.59 2+ 3.7 CR YCC042 Stomach Non-amplified 1.18 2+ 2.01 CR YCC003 Not done 2+ 2.9 CR YCC037 Not done Amplified* 3+ Not done CR YCC016 Stomach Amplified 28.21 2+ >2.0 PR YCC040 Stomach Amplified 20.10 3+ Not done PR KBSMC-01 Stomach Amplified 13.28 3+ 14.16 PR YCC010 Stomach Amplified 13.07 3+ Not done PR YCC036 Stomach Amplified 9.95 3+ Not done PR YCC012 Stomach Non-amplified 3.07 3+ 6.6 PR YCC023 Stomach Non-amplified 3.03 3+ Not done PR YCC004 Stomach Non-amplified 2.94 3+ Not done PR YCC011 Stomach Non-amplified 2.88 3+ Not done PR YCC005 Liver Non-amplified 2.13 3+ Not done PR YCC026 Stomach Non-amplified 1.99 3+ Not done PR YCC018 Stomach Non-amplified 1.92 3+ Not done PR YCC024 Stomach Non-amplified 1.79 2+ 4.07 PR YCC006 Stomach Non-amplified 1.52 2+ 2.19 PR YCC027 Stomach Non-amplified 1.36 3+ 2.9 PR YCC025 Stomach Non-amplified 1.18 3+ 1.921 PR YCC008 Stomach Non-amplified 1.10 3+ Not done PR YCC031 Liver Non-amplified 1.07 2+ 3.12 PR YCC017 Stomach Non-amplified 1.00 2+ 2.05 PR YCC009 Stomach Non-amplified 0.98 2+ 2.05 PR CNUH-01 Not done 2+ 2.38 PR CNUH-02 Not done 3+ Not done PR YCC019 Not done 3+ Not done PR YCC028 Not done 3+ Not done PR YCC035 Not done 3+ Not done PR YCC041 Not done Amplified* 3+ Not done PR YCC022 Stomach Amplified 5.10 3+ 7.452 SD YCC034 Stomach Non-amplified 1.83 3+ Not done SD YCC039 Stomach Non-amplified 1.41 2+ 2.89 SD YCC014 Stomach Non-amplified 1.41 3+ 11.67 SD YCC029 Stomach Non-amplified 1.05 3+ 1.643 SD YCC002 Not done 3+ Not done SD YCC020 Not done Non-amplified* 2+ 2.08 SD YCC032 Not done 3+ Not done SD YCC038 Not done Non-amplified* 3+ Not done SD YCC033 Stomach Non-amplified 2.52 3+ Not done PD *FoundationOne® panel sequencing results #Amplification index from silver in situ hybridazion Extended Data Table 5 | HER2 status of each patient measured by different technologies and clinical responses a b

100 100 HER2 IHC 2+/SISH+ HER2 IHC 2+/SISH+

HER2 IHC 3+ HER2 IHC 3+ freesurvival -

50 50 Percent overall survival

Percent progression mPFS: 9.0 vs 8.5 P=0.926 mOS: 21.1 vs 19.3 P=0.726 0 0 0 12 24 36 48 0 12 24 36 48 Months Months

c HER2 amplified by NGS d 100 15 HER2 non-amplified by NGS 15

Y = 0.4325X + 0.3147 1010 (P=0.0010 deviation from zero)

50

5 Percent overall survival HER2 Copy Number by NGS by Number Copy HER2

mOS: Not reached vs 17.9, P=0.0226 0 00 0 12 24 36 48 0 51 102 153 204 Months HER2 SISH Amplification Index

e 140 120 HER2 amplified by NGS

100 HER2 non-amplified by NGS 80 60 40 20 0 -20 -40 -60 -80 Tumor burden change from baseline (%) -100

10 20 30 40 50 60 70 80 90 100 Time since start of treatment (weeks)

Extended Data Fig. 5 | Correlation between baseline HER2 status and survival a-b, Kaplan Meier survival curves with progression-free survival (PFS) and overall survival (OS) in months stratified by baseline HER2 IHC status (n=43). c, Kaplan Meier survival curves with OS in months according to baseline HER2 amplification by NGS (n=35). d, Correlation between HER2 copy number by NGS and HER2 SISH amplification index (n=17). e, Percentage change from baseline with total tumor lesions of measurable target and non-target lesions over time, according to baseline HER2 amplification by NGS (n=31). All the patients were HER2-positive by IHC or SISH. a b

100 Selected RTK/RAS pathway genes

Altered

Wild-type

50

mOS: Not reached vs 10.4, P=0.0001 Percent overall survival

0 0 12 24 36 48 Months

c d 100 100 Selected RTK/RAS pathway genes Selected RTK/RAS pathway genes

Altered Altered

Wild-type Wild-type freesurvival -

50 50 Percent overall survival

Percent progression mPFS: 12.1 vs 12.9, P=0.5067 mOS: 25.4 vs 21.1, P=0.3532 0 0 0 12 24 36 48 0 12 24 36 48 Months Months

e f 20 Selected RTK/RAS pathway genes Altered 100 RTK/RAS pathway altered group 15 Wild-type TMB-high

TMB-low freesurvival 10 - 50 Burden (Mutations/Mb)

5 Mutation mPFS: 12.6 vs 28.6, P=0.2892 Percent progression P=0.0846 0 Tumor 0 0 12 24 36 48 Months Altered Wild-type

Extended Data Fig. 6 | Selected RTK/RAS pathway genes and patient survival from baseline tissue NGS, n=31 a, The detected molecular alterations on RTK/RAS pathway genes, except ERBB2. b, Kaplan Meier survival curves with overall survival (OS, b) of this study in months by alteration of selected RTK/RAS pathway genes. c,d, Kaplan Meier survival curves with progression-free survival (PFS, c) and overall survival (OS, d) in months by alteration of selected RTK/RAS pathway genes from historical retrospective cohort (n=18) gastric cancer patients who were treated with Trastuzumab+Capeciatbine+Cisplatin as palliative 1st line treatment. e, Association between tumor mutation burden from baseline tumor tissue NGS and group by RTK/RAS pathway genes alteration. f, Kaplan Meier survival curves with progression-free survival in months by tumor mutation burden stratified by pre-defined cutoff (≥10 mutations/Mb indicates TMB-high, n=3) among RTK/RAS pathway altered group (n=23). Bars and error bars, mean±SD. Pretreatment Pretreatment Pretreatment NGS HER-2 HER2 Patient ID NGS HER2 PFS OS NGS tissue RTK/RAS pathway IHC SISH AI amplification Pt001 Esophagus Non-amplified Altered 3+ Not done 8.2 37.5 Pt002 Stomach Non-amplified Wild-type 2+ 4.42 15.5 30.3 Pt003 Stomach Non-amplified Wild-type 3+ Not done 21.0 27.4 Pt004 Bronchus Non-amplified Altered 3+ Not done 12.1 25.8 Pt005 Lymph Node Amplified Altered 3+ Not done 12.1 24.9 Pt006 Stomach Non-amplified Wild-type 2+ 2.5 11.5 21.1 Pt007 Stomach Non-amplified Wild-type 3+ Not done 8.1 17.5 Pt008 Stomach Amplified Wild-type 2+ 5.29 8.3 17.5 Pt009 Stomach Amplified Altered 3+ Not done 12.3 17.4 Pt010 Stomach Amplified Wild-type 3+ Not done 12.9 16.7 Pt011 Stomach Amplified Wild-type 3+ Not done 7.4 15.3 Pt012 Stomach Amplified Wild-type 3+ Not done 14.3 14.3 Pt013 Stomach Amplified Altered 3+ Not done 5.4 11.4 Pt014 Stomach Amplified Altered 3+ Not done 10.1 11.0 Pt015 Stomach Non-amplified Wild-type 3+ Not done 7.2 8.4 Pt016 Stomach Non-amplified Altered 3+ Not done 7.3 7.3 Pt017 Stomach Non-amplified Wild-type 2+ 2.14 1.3 7.3 Pt018 Stomach Non-amplified Wild-type 3+ Not done 5.4 5.4

Extended Data Table 6 | HER2 status, RTK/RAS pathway changes and survival in each patient from historical HER2-positive AGC cohort with different technologies. a b

100 100 100 100 High TMB High TMB Low TMB Low TMB freesurvival - 5050 5050 Percent overall survival

Percent progression mPFS: 28.6 vs 8.2 P=0.1274 mOS: 33.2 vs 18.4 P=0.2401 0 0 0 12 24 36 48 0 12 24 36 48 Months Months

c 20

15

10

55 Tumor Mutation Burden (Mutations/Mb)

P=0.9203 00 ResponderResponder Non-responderNon-responder

Extended Data Fig. 7 | Tumor mutation burden and survival from baseline tissue NGS, n=31 a, b, Kaplan Meier survival curves with progression-free survival (PFS, a) and overall survival (OS, b) in months by tumor mutation burden stratified by pre-defined cutoff (≥10 mutations/Mb indicates high TMB, n=3, and <10 mutations/Mb indicates low TMB). c, Association between tumor mutation burden from baseline tumor tissue NGS and objective response to the pembrolizumab, trastuzumab, and chemotherapy. Bars and error bars, mean±SD. a Peptide sequence (9-mer) Convolutional neural network Binding prediction

HLA sequence 1 Y YYL A RN......

B Kernel

Kernel X1 D mer) - Kernel detects binding pattern E …. 1 = Binding …. HLA Sequence . 0 = No binding .

. Peptide sequence(9 n candidates from hotspot mutant variants)

F Y Kernel X G n Kernel neoantigen (

b c

100 High Neoantigen (HLA-A corrected) 100 High Neoantigen (HLA-A corrected) Low Neoantigen (HLA-A corrected) Low Neoantigen (HLA-A corrected) free survival - 50 50 Percent overall survival mPFS: 13.8 vs 8.2 P=0.5628 mOS: 31.2 vs 19.8 P=0.3480 0 0 Percent progression 0 12 24 36 48 0 12 24 36 48 Months Months

d

100 High Neoantigen (HLA-B corrected) Low Neoantigen (HLA-B corrected)

50 Percent overall survival mOS: Not reached vs 17.4 P=0.0263 0 0 12 24 36 48 Months

Extended Data Fig. 8 | Neoantigen load predicted by the CNN model and survival a, Schematic 3D model for peptide-MHC class I binding (left). Schematic diagram showing HLA molecule and diagram showing estimation of neoantigen load by the convolutional neural network (CNN) according to individual patient’s somatic mutations and HLA-A (b, c) or HLA-B (d) alleles (right). Modified from Kim et al., Nature Communications 2020. See online methods for the details. b-d, Kaplan Meier survival curves with progression-free survival (PFS, b) and overall survival (OS, c, d) in months by neoantigen load predicted, stratified by pre-defined cutoff (> median indicates high neoantigen load and ≤ median indicates low neoantigen load). a b

100 100 HLA homozygous in at least one locus HLA homozygous in at least one locus HLA heterozygous at all loci HLA heterozygous at all loci free survival - 50 50 Percent overall survival mPFS: 16.4 vs 8.3 P=0.6025 mOS: 22.8 vs 19.8 P=0.8751

Percent progression 0 00 0 12 24 36 48 0 1212 24 24 36 36 48 Months Months

c d

100 100 HLA-A homozygous HLA-A homozygous HLA-A heterozygous HLA-A heterozygous free survival - 50 50 Percent overall survival mPFS: 22.0 vs 8.3 P=0.2800 mOS: 31.2 vs 18.4 P=0.2558

Percent progression 0 00 0 12 24 36 48 0 1212 24 24 36 36 48 Months Months

e f

100 100 HLA-B homozygous HLA-B homozygous HLA-B heterozygous HLA-B heterozygous free survival - 50 50 Percent overall survival mPFS: 16.4 vs 8.3 P=0.7195 mOS: 17.9 vs 22.0 P=0.8023

Percent progression 0 00 0 12 24 36 48 0 1212 24 24 36 36 48 Months Months

Extended Data Fig. 9 | HLA homozygosity and treatment outcome a-f, Association between homozygosity in at least one human leukocyte antigen (HLA)-A or –B locus and survival. Kaplan Meier survival curves with progression-free survival (PFS, a, c, d) and overall survival (OS, b, e, f) in months by HLA-A or –B homozygosity. a b

100 100 HLA-B44 (+) HLA-B44 (+) HLA-B44 (-) HLA-B44 (-) free survival - 50 50 Percent overall survival mPFS: 7.7 vs 16.4 P=0.0861 mOS: 19.8 vs Not reached P=0.1544

Percent progression 0 00 0 12 24 36 48 0 1212 24 24 36 48 Months Months

c d

100 100 HLA-B62 (+) HLA-B62 (+) HLA-B62 (-) HLA-B62 (-) free survival - 50 50 Percent overall survival mPFS: 8.1 vs 9.1 P=0.7789 mOS: Not reached vs 18.2 P=0.2527

Percent progression 0 00 0 12 24 36 48 0 1212 24 24 36 48 Months Months

e f

100 100 HLA-B*15:01 (+) HLA-B*15:01 (-) HLA-B*15:01 (+) free survival - HLA-B*15:01 (-) 50 50 Percent overall survival mPFS: Not reached vs 8.1 P=0.0682 mOS: Not reached vs 18.4 P=0.0493

Percent progression 0 00 0 12 24 36 48 0 1212 24 24 36 48 Months Months

Extended Data Fig. 10 | HLA supertypes or allele and treatment outcome a-f, Association between HLA supertypes or allele and survival. Kaplan Meier survival curves with progression-free survival (PFS, a, c, d) and overall survival (OS, b, e, f) in months by indicated HLA supertypes or allele. a

Gene AA change YCC031 YCC006 YCC008 KBSMC-01 YCC012 YCC001 YCC027 YCC010

PMS2 K706X 46

ATM R2443X 29

ATM R2993X 46

PALB2 R1086X 1

RAD51B Q371X 2 3

BAP1 Q392X 1

FANCA R853X 1

FANCM E454X 6

PARP1 R340X 3

PARP2 W74X 1

PARP4 R1512X 2

ARID1A E1895X 64

b c

100 DDR pathway genes 100 DDR pathway genes

Altered Altered

Wild-type Wild-type freesurvival -

50 50 Percent overall survival Percent progression mPFS: 8.4 vs 8.4, P=0.6577 mOS: 22.8 vs 18.4, P=0.7160 0 0 0 12 24 36 48 0 12 24 36 48 Months Months

Extended Data Fig. 11 | DNA damage response pathway genes and patient survival from baseline tissue NGS, n=31 a, Molecular alterations on DNA damage response (DDR) pathway genes among patients. Numbers indicate variant allele frequencies (VAF, %). b,c, Kaplan Meier survival curves with progression-free survival (PFS, b) and overall survival (OS, c) in months by alterations of DDR pathway genes. PFS OS HR 95% CI P HR 95% CI P HLA-B corrected neoAg load 0.76 0.25-2.33 0.636 0.69 0.18-2.61 0.582 RTK/RAS pathway alterationa 0.17 0.05-0.58 0.005 0.28 0.08-0.92 0.046 HER2 NGS amplification 0.39 0.11-1.44 0.159 0.47 0.10-2.29 0.352 aHER2 alterations excluded.

Extended Data Table 7 | Multivariate analysis with HLA-B corrected neo-antigen load, RTK/RAS pathway alterations, and HER2 amplification a De novo resistant (n=31)

Low HLA-B HER2 non- corrected amp by NGS Neoantigen load (n=23) 7 7 2 (n=18)

9 0 0

1

RTK/RAS pathway 5 non-altered (n=10)

b c

100 100 None None One One Two Two Three Three freesurvival survival - 50 50

mPFS: NR vs 19.8 vs 8.3 vs 5.6 P=0.0002 Percent overall mOS: 37 vs NR vs 18.4 vs 11.4 P=0.0034 Percent progression 0 0 0 12 24 36 48 0 12 24 36 48 Months Months

d e 100 100 None or 1,2 None or 1,2

three three freesurvival - 50 50 overallsurvival

Percent mOS: 31.2 vs 11.4 P=0.0009 mPFS: 15.1 vs 5.6 P<0.0001 Percent progression 0 0 0 12 24 36 48 0 12 24 36 48 Months Months

Extended Data Fig. 12 | Grouping of patients in according to de novo resistant mechanism a, Venn diagram showing grouping of patients who would get less benefit (de novo resistant) from quadruplet regimen predicted by baseline NGS results (n=31), using three predictive features we found (patients with Her2 non-amplification by NGS, low HLA-B corrected neoantigen load, and non-altered RTK/RAS pathway). b-e, Kaplan Meier survival curves with progression-free survival (PFS, b,d) and overall survival (OS, c,e) in months by number of predictive features for de novo resistance. HER2-positive treatment naïve Quadruplet regimen recurrent and/or (Pembrolizumab/Trastuzumab/Capecitabine/Cisplatin) Progression metastatic treatment gastric cancer

Baseline NGS On-treatment NGS Post-progression NGS

• 34 tissues from 31 patients • 24 tissues from 15 patients • 16 tissues from 12 patients - 30 primary tumor (stomach) - All from primary tumor - 14 primary tumor (stomach) - 4 metastatic liver (stomach) - 2 metastatic liver

Extended Data Fig. 13 | Schematic diagram showing sequential genomic analysis with available patients, n=35 a Pretreatment On-treatment Post-progression HER2 Best NGS HER2 HER-2 HER2 HER-2 HER2 HER-2 HER2 Overall Patient ID copy Response amplification IHC SISH AI IHC SISH AI IHC SISH AI HER2 Change number YCC003 CR 2+ 2.9 3+ Not done Not done Not done HER2+ preserved YCC015 CR Non-amplified 1.59 2+ 3.7 3+ 4.6 Not done Not done HER2+ preserved KBSMC-01 PR Amplified 13.28 3+ 14.16 Not done Not done 3+ 5.95 HER2+ preserved YCC005 PR Non-amplified 2.13 3+ Not done 3+ Not done Not done Not done HER2+ preserved YCC006 PR Non-amplified 1.52 2+ 2.19 2+ 2.1 3+ 2.9 HER2+ preserved YCC008 PR Non-amplified 1.10 3+ Not done Not done Not done 2+ 1.5 HER2 loss YCC009 PR Non-amplified 0.98 2+ 2.05 0 1.8 Not done Not done HER2 loss YCC010 PR Amplified 13.07 3+ Not done Not done Not done 0 1.4 HER2 loss YCC018 PR Non-amplified 1.92 3+ Not done Not done Not done 3+ 7.5 HER2+ preserved YCC019 PR 3+ Not done 3+ 12 2+ 9.1 HER2+ preserved YCC023 PR Non-amplified 3.03 3+ Not done 3+ 15 Not done Not done HER2+ preserved YCC026 PR Non-amplified 1.99 3+ Not done 3+ 4 3+ Not done HER2+ preserved YCC027 PR Non-amplified 1.36 3+ 2.9 Not done Not done 2+ 2.9 HER2+ preserved YCC028 PR 3+ Not done 3+ Not done 2+ 7.1 HER2+ preserved YCC031 PR Non-amplified 1.07 2+ 3.12 Not done Not done 1+ 1.5 HER2 loss YCC035 PR 3+ Not done 2+ 1.7 Not done Not done HER2 loss YCC020 SD Non-amplified* 2+ 2.08 Not done Not done 0 1.47 HER2 loss YCC032 SD 3+ Not done Not done Not done 3+ 8.03 HER2+ preserved YCC034 SD Non-amplified 1.83 3+ Not done 0 1.19 0 Not done HER2 loss YCC039 SD Non-amplified 1.41 2+ 2.89 Not done Not done 1+ 1.6 HER2 loss YCC033 PD Non-amplified 2.52 3+ Not done Not done Not done 3+ 4.45 HER2+ preserved *FoundationOne® panel sequencing results

b c 100 100 HER2+ preserved HER2+ preserved

HER2 loss HER2 loss

mPFS: 11.5 vs 5.9 P=0.0361 freesurvival survival -

50 50 Percent overall Percent progression mOS: 22.8 vs 10.5 P=0.0156 0 0 0 12 24 36 48 0 12 24 36 48 Months Months

Extended Data Fig. 14 | HER2 loss on on-treatment or post-progression samples and patient survival from available patients, n=21 a, HER2 IHC and SISH amplification index in pretreatment, on-treatment, and post-progression tissues from available 21 patients and their response and baseline HER2 amplification by NGS. b,c, Kaplan Meier survival curves with progression-free survival (PFS, a) and overall survival (OS, b) in months by HER2 loss by IHC/SISH on on-treatment or post-progression samples. tissues. Daystissues. betweendate of tissue acquisition for baseline NGS and last (on Extended Data Fig. Fish Fish progression) NGS are also NGS are also shown.progression) plots time- in plots scale for sensitive orfor resistant sensitive scale Patient ID Patient

15 YCC010 YCC024 YCC023 YCC026 YCC006 YCC027 YCC034 YCC015 YCC018 YCC039 YCC008 YCC009 YCC016 YCC033 | Sequentially Baseline NGS Days 60 Days Days 88 Days Days 146 Days

On-Tx 174 Days On-Tx Days 178 Days Post-PD On-Tx On-Tx 208 Days Days 222 Days

Days 228 Days Post-PD biopsied stomachtissues and On-Tx On-Tx On-Tx Days 270 Days Days 278 Days Post-PD On-Tx Post-PD On-Tx Post-PD On-Tx Post-PD

Days 392 Days Post-PD Post-PD sub-clones sub-clones On-Tx Days 447 Days

Post-PD

Days 558 Days Post-PD On-Tx determination from serial biopsied stomach stomach biopsied serial from determination Post Post-PD Days 703 Days - progression NGS

Post-PD Post-PD sub -clonal - treatment or post treatment evolution - a b Gene Amino Acid Change Baseline On-Treatment Post-progression Gene Amino Acid Change Baseline On-Treatment Post-progression FLCN A264V 0.024541 0.000007 0.000008 FUBP1 A43E 0.00001601 0.04700201 N/A SF3B1 R387W 0.024541 0.000007 0.000008 GATA2 T176P 0.00001601 0.04700201 N/A ADAM29 W735R 0.02776 0.00001 0.000009 PREX2 S1140L 0.00001601 0.04700201 N/A DNMT3A V296G 0.02776 0.00001 0.000009 ADAM29 S768R 0.000005 0.022207 N/A LZTR1 A744V 0.032508 0.000008 N/A AKT1 T443M 0.000005 0.022207 N/A PTCH2 V471I 0.032508 0.000008 N/A PIK3CA D1029N 0.000005 0.022207 N/A ERBB4 K1223T 0.029078 0.000005 N/A BRCA2 G2837X 0.000006 N/A 0.026433 MYC Q48H 0.029078 0.000005 N/A ETV4 R387W 0 0.02499 N/A TP53 R342X 0.712591 0.00003902 N/A MYC Q48H 0 0.02499 N/A BAP1 R60Q 0.641324 0.00002001 N/A RANBP2 W1669C 0 0.02499 N/A KMT2B A1649T 0.641324 0.00002001 N/A WT1 H465N 0 0.02499 N/A MYC S77A 0.030552 0.000003 N/A APOB Q2352H 0 0.028192 N/A ADAM29 S768R 0.070098 N/A 0.000037 ATR S89R 0 0.028192 N/A ADAM29 T746M 0.022845 N/A 0.005042 FAT3 S3761F 0 0.204389 N/A CDH2 A610S 0.022845 N/A 0.005042 IKZF1 D290N 0 0.118886 N/A BIRC3 V102I 0.02859 0 N/A SETBP1 R498W 0 0.118886 N/A CSF1R R676X 0.02859 0 N/A APOB M1189X 0.005665 N/A 0.027377 DICER1 R745X 0.02859 0 N/A HNF1A A161T 0.005665 N/A 0.027377 DOCK2 G365D 0.02859 0 N/A MYC Q48H 0.005665 N/A 0.027377 GUCY1A2 M410I 0.02859 0 N/A APOB A43V 0 N/A 0.109623 MITF R272H 0.02859 0 N/A ERG A439V 0 N/A 0.174283 NCOR1 E1109X 0.02859 0 N/A EPHB1 A36T 0 N/A 0.055657 PIK3CA E39K 0.02859 0 N/A GABRA6 F27V 0.000009 0.000009 0.081739 SETBP1 R1146W 0.02859 0 N/A ROBO1 R1170G 0.000009 0.000009 0.081739 SMO R772H 0.02859 0 N/A SEMA3E E726D 0.000009 0.000009 0.081739 TP53 R213Q 0.515513 N/A 0.09520102 MYC Q48H 0.00412 N/A 0.026511 ADAM29 Q731K 0.410767 N/A 0.08407901 CRBN T403M 0.000019 N/A 0.025076 BCL2L2 V178M 0.027287 N/A 0 FGF12 G82V 0 0 0.024234 FGFR2 R210Q 0.027287 N/A 0 GATA6 T537A 0 0 0.024234 GRM3 R237C 0.027287 N/A 0 KMT2C R444W 0.027287 N/A 0 NF1 V288M 0.027287 N/A 0 ELMO1 A5T 0.103883 N/A 0.02277 PIK3CA E542K 0.103883 N/A 0.02277 PIK3CB R149Q 0.103883 N/A 0.02277 TP53 T81fs 0.371165 N/A 0.10677001 MAP2K4 E141K 0.031932 N/A 0.000028 NCOR1 Q1993X 0.031932 N/A 0.000028 MYC Q48H 0.035795 N/A 0.000022

Extended Data Table 8 | Genes with hotspot mutation from sensitive (a) or resistant (b) sub-clones and their sub-clonal fraction. a

ID Description GeneRatio BgRatio pvalue p.adjust qvalue geneID R-HSA-5663202 Diseases of signal transduction 7/26 377/10654 2.41E-05 0.004728 0.002758 ERBB4/MYC/NCOR1/PIK3CA/FGFR2/NF1/PIK3CB R-HSA-2219530 Constitutive Signaling by Aberrant PI3K in Cancer 4/26 75/10654 3.01E-05 0.004728 0.002758 ERBB4/PIK3CA/FGFR2/PIK3CB R-HSA-2219528 PI3K/AKT Signaling in Cancer 4/26 101/10654 9.69E-05 0.007711 0.004498 ERBB4/PIK3CA/FGFR2/PIK3CB R-HSA-6811558 PI5P, PP2A and IER3 Regulate PI3K/AKT Signaling 4/26 103/10654 0.000105 0.007711 0.004498 ERBB4/PIK3CA/FGFR2/PIK3CB R-HSA-199418 Negative regulation of the PI3K/AKT network 4/26 110/10654 0.000135 0.007711 0.004498 ERBB4/PIK3CA/FGFR2/PIK3CB R-HSA-109704 PI3K Cascade 3/26 44/10654 0.00016 0.007711 0.004498 PIK3CA/FGFR2/PIK3CB R-HSA-112399 IRS-mediated signalling 3/26 48/10654 0.000207 0.007711 0.004498 PIK3CA/FGFR2/PIK3CB R-HSA-1250342 PI3K events in ERBB4 signaling 2/26 10/10654 0.000255 0.007711 0.004498 ERBB4/PIK3CA R-HSA-2428928 IRS-related events triggered by IGF1R 3/26 52/10654 0.000263 0.007711 0.004498 PIK3CA/FGFR2/PIK3CB R-HSA-2428924 IGF1R signaling cascade 3/26 53/10654 0.000279 0.007711 0.004498 PIK3CA/FGFR2/PIK3CB R-HSA-2404192 Signaling by Type 1 Insulin-like Growth Factor 1 Receptor (IGF1R) 3/26 54/10654 0.000295 0.007711 0.004498 PIK3CA/FGFR2/PIK3CB R-HSA-74751 Insulin receptor signalling cascade 3/26 54/10654 0.000295 0.007711 0.004498 PIK3CA/FGFR2/PIK3CB R-HSA-1236394 Signaling by ERBB4 3/26 58/10654 0.000364 0.007981 0.004655 ERBB4/NCOR1/PIK3CA R-HSA-9027276 Erythropoietin activates Phosphoinositide-3-kinase (PI3K) 2/26 12/10654 0.000372 0.007981 0.004655 PIK3CA/PIK3CB R-HSA-1257604 PIP3 activates AKT signaling 5/26 264/10654 0.000386 0.007981 0.004655 ERBB4/TP53/PIK3CA/FGFR2/PIK3CB R-HSA-9607240 FLT3 Signaling 5/26 267/10654 0.000407 0.007981 0.004655 ERBB4/PIK3CA/FGFR2/NF1/PIK3CB R-HSA-1963642 PI3K events in ERBB2 signaling 2/26 16/10654 0.000673 0.011538 0.00673 ERBB4/PIK3CA R-HSA-2730905 Role of LAT2/NTAL/LAB on calcium mobilization 2/26 16/10654 0.000673 0.011538 0.00673 PIK3CA/PIK3CB R-HSA-449147 Signaling by Interleukins 6/26 461/10654 0.000698 0.011538 0.00673 MYC/TP53/CSF1R/PIK3CA/PIK3CB/MAP2K4 R-HSA-9006925 Intracellular signaling by second messengers 5/26 305/10654 0.000746 0.011715 0.006833 ERBB4/TP53/PIK3CA/FGFR2/PIK3CB

b ID Description GeneRatio BgRatio pvalue p.adjust qvalue geneID R-HSA-210745 Regulation of gene expression in beta cells 2/18 21/10654 0.000556 0.093094 0.06233 AKT1/HNF1A R-HSA-389357 CD28 dependent PI3K/Akt signaling 2/18 22/10654 0.00061 0.093094 0.06233 AKT1/PIK3CA R-HSA-389356 CD28 co-stimulation 2/18 33/10654 0.00138 0.113371 0.075907 AKT1/PIK3CA R-HSA-5693616 Presynaptic phase of homologous DNA pairing and strand exchange 2/18 39/10654 0.001925 0.113371 0.075907 BRCA2/ATR R-HSA-186712 Regulation of beta-cell development 2/18 42/10654 0.00223 0.113371 0.075907 AKT1/HNF1A R-HSA-5693579 Homologous DNA Pairing and Strand Exchange 2/18 42/10654 0.00223 0.113371 0.075907 BRCA2/ATR R-HSA-1227986 Signaling by ERBB2 2/18 50/10654 0.003148 0.128628 0.086122 AKT1/PIK3CA R-HSA-5685942 HDR through Homologous Recombination (HRR) 2/18 67/10654 0.005586 0.128628 0.086122 BRCA2/ATR R-HSA-8939211 ESR-mediated signaling 3/18 223/10654 0.005852 0.128628 0.086122 AKT1/PIK3CA/MYC R-HSA-3371453 Regulation of HSF1-mediated heat shock response 2/18 69/10654 0.005915 0.128628 0.086122 RANBP2/ATR R-HSA-8948751 Regulation of PTEN stability and activity 2/18 69/10654 0.005915 0.128628 0.086122 PREX2/AKT1 R-HSA-388841 Costimulation by the CD28 family 2/18 70/10654 0.006083 0.128628 0.086122 AKT1/PIK3CA R-HSA-157118 Signaling by NOTCH 3/18 235/10654 0.006767 0.128628 0.086122 AKT1/MYC/IKZF1 R-HSA-9009391 Extra-nuclear estrogen signaling 2/18 77/10654 0.007319 0.128628 0.086122 AKT1/PIK3CA R-HSA-69202 Cyclin E associated events during G1/S transition 2/18 83/10654 0.008461 0.128628 0.086122 AKT1/MYC R-HSA-69656 Cyclin A:Cdk2-associated events at S phase entry 2/18 85/10654 0.008859 0.128628 0.086122 AKT1/MYC R-HSA-1257604 PIP3 activates AKT signaling 3/18 264/10654 0.009318 0.128628 0.086122 PREX2/AKT1/PIK3CA R-HSA-3371556 Cellular response to heat stress 2/18 89/10654 0.009679 0.128628 0.086122 RANBP2/ATR R-HSA-5687128 MAPK6/MAPK4 signaling 2/18 89/10654 0.009679 0.128628 0.086122 ETV4/MYC R-HSA-9616222 Transcriptional regulation of granulopoiesis 2/18 90/10654 0.009889 0.128628 0.086122 GATA2/MYC

Extended Data Table 9 | Top 20 enriched Reactomes from selected sensitive (a) or resistant (b) sub- clone genes a 0.12 Median PFS (8.6 months)

0.08 Hazard Rate Hazard 0.04

0.00

0 5 10 15 20

Follow-up time (Months)

b Median OS (19.3 months) 0.06

0.04 Hazard Rate Hazard 0.02

0.00

0 5 10 15 20 25

Follow-up time (Months)

Extended Data Fig. 16 | Survival hazard rate function a,b Hazard rate function of progression-free survival (a) and overall survival (b). Note that hazard rate is relatively low at initial 6 months compared to the peak near median PFS or OS (gray dash line). a Acquired resistant mechanism (n=39)

Acquired HER2 HER2 loss mutation (n=8) (n=3) 4 1 1

0 3 1

5 Emergence of resistant sub-clones by 27 clonal evolution(n=9)

Extended Data Fig. 17 | Grouping of patients in according to acquired resistant mechanism a, Venn diagram showing grouping of patients according to acquired resistant mechanisms (patients with Her2 loss, acquired HER2 mutation, and emergence of resistant sub-clones by clonal evolution). Figures

Figure 1

Clinical ecacy of the rst-line quadruplet regimen in patients with HER2-positive gastric cancer and their baseline genomic landscape. a, A swimmer plot showing outcomes in all patients from the start of treatment to either disease progression or the last follow-up. Note that seven patients nished 2-year treatment. b, Kaplan–Meier survival curves with progression-free survival and overall survival in all patients. c, Maximum percentage change from baseline in size of total tumor lesions with corresponding best responses by RECIST 1.1 and PD-L1 combined positive score (CPS) from baseline tissue. Lower dotted line represents tumor reduction of 30%. d, Spider plot showing percentage change from baseline in target lesions over time during treatment, with corresponding PD-L1 CPS. Figure 2

Baseline genomic landscape in the enrolled HER2-positive gastric cancer patients treated with the rst- line quadruplet regimen. a, Baseline (pretreatment) tumor tissue-targeted DNA sequencing results grouped by best response and related clinicopathologic features (n=31). Curated pathways and selected genes altered in 6% or more of the patients are shown. Other pathways include cell cycle, Hippo, MYC, and NRF2 pathways. Vertical dashed lines indicate groups by best response. See Online Methods for details. b-d, Kaplan–Meier survival curves with progression-free survival (PFS) stratied by pretreatment HER2 amplication by NGS (b), RTK-RAS pathway genes alteration (c), and HLA-B-corrected neoantigen load (d). Figure 3

Genomic analyses from serial biopsy samples of the HER2-positive gastric cancer patients treated with the rst-line quadruplet regimen. a, Spider plot showing patients with HER2 mutation found in serial NGS analyses of primary tumor. Only HER2 mutations and variant allele frequencies (VAFs) with the corresponding patient IDs for detected cases are shown. b, Sensitive or resistant sub-clones in which the sub-clone frequency changed over 2-fold in post-progression (Post-PD) samples (n=10) or on-treatment (On-Tx) samples (n=9) compared to paired baseline samples (n=18) are selected per patient. Selected enriched Reactome pathways from sensitive or resistant sub-clone genes are also shown. All paired tissue samples are from primary tumor (stomach). c-d, Representative cases showing sub-clonal evolution by sh plot and corresponding clinicopathologic features and computed tomography or endoscopic images from paired tissue NGS from both primary tumor (stomach) and metastatic liver. Selected hotspot mutations are labeled. Representative case from good responder (c) and poor responder (d).