Purpose:

Intrapatient heterogeneity of programmed death ligand 1 (PD-L1) expression and tumor mutational burden (TMB) in gastroesophageal adenocarcinoma (GEA) could influence their roles as predictive biomarkers for response to immune checkpoint inhibitors (ICI). In this retrospective analysis, we evaluated the spatiotemporal heterogeneity and prognostic relevance of PD-L1 expression and TMB in GEA.

Experimental Design:

A cohort of 211 patients with stage II–IV GEA was retrospectively reviewed for a total of 407 tumor samples with PD-L1 expression data and 319 tumor samples with TMB data. PD-L1 status was defined as positive if combined positive score (CPS) ≥1 using the 22C3 pharmDx assay. TMB levels were categorized as low, intermediate, or high (≤5, 5–15, or >15 mutations/Mb), or using a single threshold (<10 or ≥10 mutation/Mb), determined by next-generation sequencing using a targeted gene panel.

Results:

Of 407 tumors, 56% were PD-L1 negative and 44% PD-L1 positive. Of 319 tumors, 50% were TMB-low, 45% TMB-intermediate, and 5% TMB-high; 86% had <10 and 14% ≥10 mutations/Mb. TMB level was significantly associated with MSI-status. PD-L1 expression and TMB exhibited marked spatial heterogeneity between baseline primary and metastatic tumors (61% and 69% concordance), and temporal heterogeneity between tumors before and after chemotherapy (57%–63% and 73%–75% concordance). PD-L1 expression and TMB were not significantly associated with overall survival.

Conclusions:

PD-L1 expression and TMB exhibit marked spatial and temporal heterogeneity in GEA. This heterogeneity should be considered when obtaining tumor samples for molecular testing and when deciding whether ICI therapy is appropriate.

See related commentary by Klempner et al., p. 6401

Translational Relevance

Programmed death ligand 1 (PD-L1) expression and tumor mutational burden (TMB) are potential biomarkers for response to ICIs. Many biomarkers exhibit heterogeneity in gastroesophageal adenocarcinoma (GEA). Intrapatient heterogeneity of PD-L1 expression and TMB in GEA could influence their roles as predictive biomarkers. In this analysis, the largest of its kind for GEA, PD-L1 expression and TMB exhibited marked spatial and temporal heterogeneity, and did not affect prognosis in advanced disease. Spatial heterogeneity was particularly noted in baseline primary tumors, which were PD-L1 positive and TMB-high but frequently PD-L1 negative and TMB-low in paired metastases. These results may provide explanation for lack of benefit of ICI therapy in most patients despite positive PD-L1 scoring in primary tumors. Evaluation of intrapatient heterogeneity of PD-L1 expression and/or TMB may better predict which patients with GEA are most likely to benefit from ICI therapy.

Gastroesophageal adenocarcinoma (GEA) commonly presents with advanced disease and is associated with poor prognosis (1). Despite advances in targeted therapies, the treatment of advanced GEA still relies primarily on cytotoxic chemotherapy (2). In recent years, immune checkpoint inhibitors (ICI) have emerged as promising treatments for locally advanced or metastatic GEA, and the anti-programmed cell death protein 1 (anti-PD-1) antibody pembrolizumab was conditionally approved by the FDA for use in a subset of patients with GEA after progression on initial systemic treatments (1, 3, 4). Biomarkers including programmed death ligand 1 (PD-L1) expression and tumor mutational burden (TMB) might help define which patients are most likely to benefit from ICI therapy (5). Tumor cell PD-L1 expression (‘TPS’) is associated with response to anti-PD-1/anti-PD-L1 therapy in some cancers (6, 7). In GEA, a PD-L1 combined positive score (CPS) of ≥1 helps predict benefit from pembrolizumab therapy with high negative predictive value but less-than-desirable positive predictive value (1, 3, 8). TMB has also emerged as a putative biomarker associated with ICI response in multiple tumor types (9–11). In patients with advanced GEA treated with the anti-PD-1 antibody toripalimab, TMB ≥12 mutations per megabase (mut/Mb) was associated with improved response and overall survival (OS; ref. 12). Recently, the FDA approved the use of pembrolizumab for advanced solid tumors with TMB ≥10 mut/Mb that progressed on prior therapy, marking the second tumor-agnostic approval of ICI therapy. This approval was based on a retrospective analysis of the nonrandomized KEYNOTE-158 trial. In this analysis, tumors with TMB ≥10 mut/Mb and microsatellite stability (MSS), found in 102 patients (13% of all successfully evaluated), was associated with an overall response rate of 29% to pembrolizumab monotherapy in various chemorefractory cancers, notably not including any GEA tumors (13). However, further work is needed to more precisely define the clinical utility of TMB as a predictive biomarker in various tumor types, including GEA (5).

Molecular biomarkers in GEA exhibit significant heterogeneity (14, 15). Heterogeneity of PD-L1 expression and TMB could influence their roles as predictive biomarkers for ICI response. Heterogeneity of PD-L1 expression within tumors and between tumor sites has been described for multiple cancer types (16–18). In surgically resected GEAs, significant spatial discordance of PD-L1 expression was observed between primary tumors and lymph node metastases (19). In non–small cell lung cancer (NSCLC), TMB exhibited both intratumoral heterogeneity and discordance between primary and metastatic tumors, but it is unclear whether this observation applies to other tumor types (20). PD-L1 expression and TMB may also demonstrate temporal heterogeneity from before to after chemotherapy. However, conflicting studies suggest that chemotherapy might increase, decrease, or have no effect on these biomarkers (21–25). Further characterization of the spatial and temporal heterogeneity of PD-L1 expression and TMB could define optimal sites and timing for their measurement, and aid in their interpretation as predictive biomarkers for ICI therapeutic benefit.

PD-L1 expression and TMB might also have intrinsic prognostic value. PD-L1 expression correlated with poor prognosis in some tumor types, but has no correlation or a correlation with favorable prognosis in others (26, 27). The prognostic significance of PD-L1 expression in GEA also has conflicting results (28–30). TMB-high has been linked to poor prognosis in NSCLC and esophageal cancer, and to favorable prognosis in colorectal cancer (31, 32). In addition to potentially real prognostic differences between histologies, discrepancies could also be due to differences in methodology and scoring thresholds for PD-L1 expression and TMB across studies. Moreover, underlying confounding factors such as Epstein–Barr virus (EBV) or microsatellite instability (MSI) status may have contributed to the observed discordance. Overall, the prognostic significance of PD-L1 expression and TMB in GEA remains unclear.

In this retrospective analysis, the largest and most comprehensive of its kind, we examined the heterogeneity and prognostic significance of PD-L1 expression and TMB in 211 patients with GEA (33, 34). To evaluate spatial and temporal heterogeneity, we compared baseline PD-L1 expression and TMB in paired primary and metastatic tumors from newly diagnosed stage IVB GEA, and in paired pre- and posttreatment tumors from stage II–IV GEA. To assess prognostic significance, we evaluated OS of patients with stage IVB GEA stratified by PD-L1 expression or TMB.

Patients

The studies were conducted in accordance with the Declaration of Helsinki. We retrospectively reviewed a cohort of 211 patients diagnosed with stage II–IV GEA between 2011 and 2020 with at least one tumor sample with PD-L1 expression or TMB data. Informed written consent was obtained from each patient or each patient's guardian. Patient and tumor characteristics were obtained from the clinically annotated University of Chicago GI database using institutional review board (IRB)-approved tissue banking protocols. No assurance was filed as this was not necessary. Sample size was determined by available data. Baseline samples were defined as obtained prior to or up to 10 days after initiation of first-line chemotherapy. Samples obtained >10 days after initiation of first-line chemotherapy were considered posttreatment samples. OS was defined as time from diagnosis to death. Median follow-up time was 15.5 months for analysis of PD-L1 expression, and 14.5 months for analysis of TMB.

Biomarkers

All samples were processed through routine Clinical Laboratory Improvement Amendments (CLIA)-certified labs for clinical care and stored per standards set by the College of American Pathologists (CAP). Assays were conducted by independent commercial companies with CLIA labs, and blinded to all patient clinical/pathologic characteristics, as well as treatment and outcomes. PD-L1 CPS (8) and HER2 IHC (with HER2 FISH if necessary; ref. 35) were evaluated per routine clinical standards. PD-L1-status was defined as positive (PD-L1+) if CPS ≥1, and negative (PDL1-) if CPS <1. Samples were submitted for TMB and MSI assessment to Foundation Medicine (310 samples; ref. 36), or to Tempus Labs with normal blood (9 samples; ref. 37). TMB level was defined as low (TMB-L, ≤5 mut/Mb), intermediate (TMB-I, >5 mut/Mb, ≤15 mut/Mb), or high (TMB-H, >15 mut/Mb), based on the thresholds used in Foundation Medicine reports. A post hoc analysis was performed using a binary cutoff at 10 mut/Mb given the recent FDA approval (13). Each patient had all their tumor samples assessed at either Foundation Medicine or Tempus Labs, with no intrapatient sample crossover between vendors.

Statistical analysis

Relationships between PD-L1 status, MSI status, and TMB level were evaluated by Fisher exact test, using one tumor sample per patient to respect the assumption of independence. PD-L1 status or TMB levels of paired baseline primary and metastatic tumors, and of paired pre- and posttreatment tumors, were compared by McNemar test using one pair of tumors per patient to respect the assumption of independence. OS was analyzed using Kaplan–Meier curves, log-rank tests, and Cox proportional-hazards models, using one tumor or pair of tumors per patient to respect the assumption of independence. Single tumors or pairs of tumors were selected by prioritizing the earliest primary tumor sample, then earliest metastatic tumor sample. Candidate variables considered for inclusion in Cox proportional-hazards models were: age, sex, race, ethnicity, ECOG status at diagnosis, tumor location, Lauren histologic classification, differentiation, presence of signet cells, HER2 status, PD-L1 status, MSI status, exposure to anti-HER2 therapy, and exposure to ICI therapy. In these models, missing data were treated as a separate category (“unknown”). For all analyses, we considered P < 0.05 as significant. Statistical analysis was performed using R-v3.6.2.

In a cohort of 211 patients with stage II–IV GEA, 189 patients had 407 tumor samples with PD-L1 expression data, and 162 patients had 319 tumor samples with TMB data (Table 1).

Relationship between PD-L1 status, MSI status, and TMB level

Among 407 tumors, PD-L1 CPS ranged 0–100 with median 0 and interquartile range 0–3.8. PD-L1-status was positive for 44% of tumors, and CPS ≥10 for 15% (Fig. 1A and Supplementary Fig. S1A). Among 319 tumors, TMB ranged 0–54.7 mut/Mb with median 6 mut/Mb and interquartile range 4–8. Half of the tumors were TMB-L, 45% TMB-I, and 5% TMB-H (Fig. 1B), while 14% of the tumors had TMB ≥10 mut/Mb (Fig. 1C). After exclusion of MSI-H tumors, 2% of MSS tumors were TMB-H (>15 mut/Mb), and 12% had TMB ≥10 mut/Mb (Fig. 1DE). Of note, none of the MSI-H tumors were germline. TMB level was significantly associated with MSI-status (P = 1 × 10−7; Fig. 1D; Supplementary Fig. S1B). Notably, 89% (8/9) of MSI-high (MSI-H) tumors were TMB-H, while only 53% (8/15) of TMB-H tumors were MSI-H. TMB was also significantly associated with MSI-status using a threshold of 10 mut/Mb (P = 9.4 × 10−6; Fig. 1E; Supplementary Fig. S1C). PD-L1 status was not significantly associated with MSI status or TMB level, although more PD-L1 tumors were TMB-L, and more PD-L1+ tumors were TMB-I and TMB-H (Fig. 1F and G; Supplementary Fig. S1D and S1E).

Spatial heterogeneity of PD-L1 expression and TMB

We compared PD-L1 expression in paired baseline primary and baseline metastatic tumors from 62 patients (Fig. 2AB). Thirty-six primary tumors were PD-L1+, compared with 18 metastatic tumors (Fig. 2A). Baseline paired primary and metastatic tumor PD-L1-status were 61% concordant (38/62), indicating marked spatial heterogeneity. Of 26 PD-L1 primary tumors, 23 (88%) remained PD-L1 in the metastatic tumor. In contrast, of 36 PD-L1+ primary tumors, only 15 (42%) remained PD-L1+ in the metastatic tumor (P = 2.4 × 10−4). PD-L1 status and spatial concordance of PD-L1 status did not show a clear dependence on metastatic site (Supplementary Table S1). With a threshold CPS of 10 (Fig. 2B), concordance was 84% (52/62). Of 53 primary tumors with CPS <10, 48 (91%) maintained CPS <10 in the metastatic tumor. In contrast, of 9 primary tumors with CPS ≥10, only 4 (44%) maintained CPS ≥10 (P = 1).

We compared TMB in paired baseline primary and baseline metastatic tumors from 45 patients (Fig. 2C and D). Baseline primary and metastatic tumor TMB levels were 69% concordant (31/45), indicating marked spatial heterogeneity (Fig. 2C). Of 41 primary tumors with low-to-intermediate TMB (TMB-L/I), 41 (100%) remained TMB-L/I in the metastatic tumor. In contrast, of four TMB-H primary tumors, only two (50%) remained TMB-H in the metastatic tumor (P = 0.16). When TMB was dichotomized by low versus intermediate–high, McNemar test also did not show significance (P = 0.41). Of note, both pairs of tumors that maintained TMB-H were MSI-H, while all other tumors were not. Using a single TMB threshold of 10 mut/Mb (Fig. 2D), concordance was 89% (40/45). Of 38 primary tumors with TMB <10 mut/Mb, 47 (97%) maintained TMB <10 mut/Mb in the metastatic tumor. In contrast, of 7 primary tumors with TMB ≥10 mut/Mb, only 3 (43%) maintained TMB ≥10 mut/Mb (P = 0.18) in the metastasis. Similar results were obtained when the two pairs of MSI-H tumors were excluded from the analysis (Supplementary Fig. S2). TMB and spatial concordance of TMB did not show a clear dependence on metastatic site (Supplementary Table S2).

Paired baseline primary and baseline metastatic tumor samples were obtained 0–100 days apart, with a median of 14.5 days between samples in the analysis of spatial heterogeneity of PD-L1 expression, and a median of 17 days between samples in the analysis of TMB. Paired samples with a longer interval between the primary and metastatic tumor samples could potentially be influenced by temporal changes in PD-L1 expression or TMB. However, we did not find a significant temporal effect. Concordance of PD-L1 status was 70% (21/30) for paired samples obtained ≤14 days apart, and 55% (17/31) for paired samples obtained >14 days apart. Concordance of TMB levels was 63% (12/19) for paired samples obtained ≤14 days apart, and 73% (19/26) for paired samples obtained >14 days apart. Concordance of TMB with a threshold of 10 mut/Mb was 95% (18/19) for paired samples obtained ≤14 days apart, and 85% (22/26) for paired samples obtained >14 days apart.

Given the marked spatial heterogeneity between baseline primary and metastatic tumor samples, we considered the possibility that concordant PD-L1+ status or TMB-H levels might predict better response to ICI therapy. To test this hypothesis, we compared progression-free survival in patients with concordant or discordant PD-L1 expression or TMB (Supplementary Table S3). However, only a small number of patients received ICI therapy, and with this small sample size we were unable to reach any conclusions regarding the relationship between spatial concordance of PD-L1 expression or TMB and response to ICI therapy.

Temporal heterogeneity of PD-L1 expression and TMB

We compared PD-L1 expression in paired pretreatment primary tumors and posttreatment primary or metastatic tumors from 83 patients (Fig. 3A). Pretreatment primary and posttreatment tumor PD-L1 status were 63% concordant (52/83), indicating marked temporal heterogeneity. Similar proportions of PD-L1 and PD-L1+ pretreatment tumors maintained their initial PD-L1 status after treatment (P = 0.86). With a threshold CPS of 10, concordance was 75% (62/83; Fig. 3B). Similar results were obtained when the single pair of pre- and posttreatment MSI-H tumors was excluded from the analysis (Supplementary Fig. S3A and S3B). Moreover, similar concordance of PD-L1 status was observed between pre- and posttreatment primary tumors (63% concordance; Supplementary Fig. S3C and S3D), or pre- and posttreatment primary or metastatic tumors (57% concordance; Supplementary Fig. S3E and S3F). Higher concordance (79%) was noted between pre- and posttreatment metastatic tumors at matching sites, but the sample size was small (Supplementary Fig. S3G and S3H). Three PD-L1 pretreatment metastatic tumors converted to PD-L1+ posttreatment, whereas no PD-L1+ pretreatment metastatic tumors converted to PD-L1 posttreatment (P = 0.083). Of note, all the patients received at least first-line platinum-based therapy between pretreatment and posttreatment tumor samples. However, some patients also received other treatments such as radiotherapy, targeted therapy, or ICIs. These additional therapies did not have a clear effect on temporal concordance of PD-L1 status (Supplementary Table S4).

We compared TMB in paired pretreatment primary tumors and posttreatment primary or metastatic tumors from 63 patients (Fig. 3C). Pretreatment primary and posttreatment tumor TMB levels were 73% concordant (46/63). Of 61 pretreatment TMB-L/I primary tumors, 61 (100%) remained TMB-L/I after treatment, whereas of two TMB-H pretreatment primary tumors, one (50%) remained TMB-H after treatment (P = 0.32). The difference between conversion from low to intermediate–high TMB and from intermediate–high to low TMB was also not significant (P = 0.32). Notably, the pair of tumors that maintained TMB-H was MSI-H, while all other tumors were not. Using a single TMB threshold of 10 mut/Mb (Fig. 3D), concordance was 90% (57/63). Of 56 pretreatment primary tumors with TMB <10 mut/Mb, 53 (95%) maintained TMB <10 mut/Mb after treatment, whereas of 7 pretreatment primary tumors with TMB ≥10 mut/Mb, 4 (57%) maintained TMB ≥10 mut/Mb after treatment (P = 1). Similar results were obtained when the pair of MSI-H tumors was excluded from the analysis (Supplementary Fig. S4A and S4B). Similar concordance of TMB levels was observed between pre- and posttreatment primary tumors (74% concordance; Supplementary Fig. S4C and S4D), or pre- and posttreatment primary or metastatic tumors (75% concordance; Supplementary Fig. S4E and S4F). Higher concordance (80%) was noted between pre- and posttreatment metastatic tumors at matching sites, but the sample size was small (Supplementary Fig. S4G and S4H).

We considered the possibility that temporally concordant PDL1+ status or TMB-H levels might predict better response to ICI therapy. To test this hypothesis, we compared progression-free survival in patients with temporally concordant or discordant PD-L1 expression or TMB (Supplementary Table S3). However, due to small sample sizes, we were unable to reach any conclusions regarding the relationship between temporal concordance of PD-L1 expression or TMB and response to ICI therapy.

Prognostic significance of PD-L1 expression and TMB

To determine the prognostic significance of PD-L1 expression and TMB, we analyzed OS among patients with stage IVB GEA. We found no significant difference in OS among 125 patients stratified by baseline primary tumor PD-L1 status (P = 0.65; Fig. 4A). Thirty-four (27%) patients received ICI therapy, including 20 (59%) with PD-L1+ baseline primary tumors (Supplementary Fig. S5A). After adjustment for multiple covariates including exposure to ICI therapy, the impact of baseline primary tumor PD-L1-status on OS remained insignificant (Padj = 0.56). We also found no significant difference in OS with stratification by baseline metastatic tumor PD-L1 status (P = 0.58; Supplementary Fig. S5B) or baseline primary or metastatic tumor PD-L1 status (P = 0.75; Supplementary Fig. S5C). Similar results were obtained with a threshold CPS of 10 (Supplementary Fig. S5D–S5F). OS also did not depend on change in PD-L1 expression after chemotherapy (Supplementary Fig. S5G–H).

TMB-level did not have a clear effect on prognosis among patients with stage IVB GEA. OS did not significantly differ among 115 patients stratified by baseline primary tumor TMB level (P = 0.74; Fig. 4B). Thirty-eight (33%) patients received ICI therapy, including 20 (53%) with TMB-I and 5 (13%) with TMB-H baseline primary tumors (Supplementary Fig. S6A). After adjustment for multiple covariates including exposure to ICI therapy, the impact of baseline primary tumor TMB level on OS remained insignificant (Padj = 0.12). OS also did not significantly differ with stratification by baseline metastatic tumor TMB level (P = 0.61; Supplementary Fig. S6B) or baseline primary or metastatic tumor TMB level (P = 0.96; Supplementary Fig. S6C). Likewise, OS did not significantly differ with stratification using a threshold TMB of 10 mut/Mb (Fig. 4C; Supplementary Fig. S6D and S6E). OS also did not depend on change in TMB level or TMB after chemotherapy (Supplementary Fig. S6F and S6G).

In this study, we investigated the heterogeneity and prognostic significance of PD-L1 expression and TMB in GEA. Both PD-L1 expression and TMB exhibited marked spatial heterogeneity, particularly between PD-L1+ primary tumors and PD-L1 metastatic tumors. PD-L1 expression and TMB also demonstrated notable temporal heterogeneity. This spatial and temporal heterogeneity of PD-L1 expression and TMB may impact their roles as predictive biomarkers for ICIs. Neither PD-L1 expression nor TMB significantly impacted OS among patients with stage IVB GEA.

Consistent with previous studies across tumor types (36, 38), we found that TMB-H was significantly associated with MSI-H in GEA. Of note, a substantial proportion of TMB-H tumors were MSS. These MSS, TMB-H tumors might be associated with genetic alterations in other DNA damage repair pathways, which may have additional prognostic and therapeutic implications (39). Conversely, nearly all MSI-H tumors were TMB-H, and all had ≥10 mut/Mb. A different TMB threshold may be more predictive of ICI response among the subset of tumors that are MSI-H, as has been previously explored in metastatic colorectal cancer (40). Previous studies found that PD-L1 expression correlated with high TMB in gastric but not in esophageal cancers (41, 42), while PD-L1+ correlated with MSI-H in surgical resections of GEA (41, 43). We found no significant association between PD-L1 status and either TMB level or MSI status in GEA, possibly due to inclusion of both gastric and esophageal adenocarcinomas at advanced stages, differences in measurement of TMB or MSI among studies, or lack of power to detect a correlation of PD-L1 status with either TMB or MSI given small numbers. Notably, in the KEYNOTE-059, -061, and -062 pooled analysis, 92.5% (62/67) of MSI-H tumors had PD-L1 CPS ≥1 (44).

We found that PD-L1 expression and TMB in GEA demonstrated substantial spatial heterogeneity, with under 70% concordance of PD-L1 status and TMB levels between baseline primary and metastatic tumors. In previous studies of surgically resected GEA, regional lymph node metastases had higher PD-L1+ rates than matched primary tumors (19, 43). In contrast, we found that baseline distant metastatic tumors had lower PD-L1+ rates than matched baseline primary tumors. This discrepancy could be due to differences in biology of regional lymph node metastases and distant metastases, or in measurement of PD-L1 between studies. Together, the results from our study and previous reports suggest that PD-L1 expression differs at baseline between primary tumors, lymph node metastases, and distant metastases in a large fraction of patients with GEA. Spatial heterogeneity of TMB was previously reported in NSCLC (20). We observed that TMB also exhibits spatial heterogeneity between primary and metastatic tumors in GEA. Higher spatial concordance was observed using the now clinically relevant threshold TMB of 10 mut/Mb, as expected when using one rather than two thresholds. Of note, spatial heterogeneity using this threshold was highly directional, favoring pairing of baseline primary tumors with TMB ≥10 mut/Mb to baseline metastatic tumors with TMB <10 mut/Mb, although this directionality was not statistically significant. Such spatial heterogeneity could influence treatment decisions, especially given the recent approval of pembrolizumab for solid tumors with TMB ≥10 mut/Mb. Further study is needed to determine the impact of discordant baseline metastatic tumor TMB on response to ICIs.

In this study, marked temporal heterogeneity of PD-L1 expression and TMB was observed, with under 65% concordance of PD-L1 status and under 75% concordance of TMB levels between pre- and posttreatment tumors. This temporal heterogeneity was not explained by spatial heterogeneity between primary and metastatic tumors, because similar levels of concordance were observed for matched pre- and posttreatment primary tumors. However, we cannot exclude intratumoral heterogeneity as a cause of the observed temporal heterogeneity. Previous studies found that tumor cell PD-L1 expression decreased after chemotherapy for esophageal adenocarcinoma (23) or metastatic gastric cancer (24), but we found no significant directional change in PD-L1 status after chemotherapy for GEA. Previous studies on the effect of treatment on TMB yielded conflicting results in multiple tumor types, possibly due, in part, to differences in radiotherapy and chemotherapy regimens (22, 25, 45). We found marked temporal heterogeneity of TMB in GEA, with no statistically significant directional change in TMB. As with spatial concordance, higher temporal concordance was observed using a TMB threshold of 10 mut/Mb, with nonstatistically significant directionality favoring conversion from pretreatment tumors with TMB ≥10 mut/Mb to posttreatment tumors with TMB <10 mut/Mb. Given the notable temporal heterogeneity of both PD-L1 expression and TMB, repeat biopsy and testing of these biomarkers should be considered to make more informed real-time decisions regarding the use of ICI therapy in clinical practice.

Studies have yielded conflicting results on the prognostic impact of PD-L1 expression in GEA (28–30). One study found that CPS ≥1 was associated with longer OS in stage I–II, but not stage III–IV, GEA (41). We also found no significant effect of PD-L1-status on OS in stage IVB GEA. Another study found that metastatic GEA patients who experienced a decrease in PD-L1 expression after chemotherapy had prolonged progression-free survival, but no difference in OS (24). Consistent with this result, we found that temporal changes in PD-L1 expression did not predict OS in stage IVB GEA. While TMB has been found to be associated with either poor or favorable prognosis in other tumor types (31, 32), the prognostic value of TMB in GEA is largely unknown. In patients with esophageal cancer, high TMB was associated with decreased OS among patients who did not receive radiotherapy, but not in the whole patient population (46). We also did not find a significant association of TMB with OS in patients with stage IVB GEA.

Our study has limitations. First, our results might be affected by the choice of biomarker thresholds. We used two PD-L1 CPS thresholds that are currently either approved or thought to enrich for further benefit from ICI therapy for GEA in clinical practice and in trials (1, 4, 8). However, measurement and thresholds for TMB were not yet standardized at the time we conducted this study (5, 12). We stratified TMB using an assay and thresholds that are readily available in clinical practice, but this stratification led to very few samples being classified as TMB-H, which limited statistical power. In the interim, pembrolizumab was approved for use in tumors with TMB ≥10 mut/Mb, so we repeated our analysis using this threshold, which yielded overall higher spatial and temporal concordance, but also more pronounced directional discordance. Second, although we found significant heterogeneity between baseline primary and metastatic tumors, we did not investigate intratumoral heterogeneity. Intratumoral heterogeneity has been demonstrated for other molecular biomarkers in GEA (14, 15), and for PD-L1 expression and TMB in other tumor types (16, 18, 20). Of note, a recent study found significant intratumoral heterogeneity of PD-L1 expression in advanced gastric cancer and concluded that at least five biopsies are likely necessary for accurate assessment of tumor PD-L1-status (47). In our study, the vast majority (>80%) of tumor samples were biopsies, so some of the observed discordance of PD-L1 status could be due to intratumoral heterogeneity. It could also be valuable to further characterize intratumoral heterogeneity of TMB in GEA in future studies. Third, in our analysis of temporal heterogeneity, we did not account for varying time on therapy, chemotherapy regimens (although all were pre/post first-line platinum-based therapy), or other treatments such as radiotherapy, targeted therapy, or ICIs. Moreover, we could not distinguish between the effects of chemotherapy-induced changes versus intrinsic natural tumor evolution. A subgroup analysis of our small sample did not show any obvious effects of radiotherapy, targeted therapies, or ICIs on temporal concordance of PD-L1 status. Notably, tyrosine kinase inhibitors have been previously reported to substantially increase PD-L1 expression in a subset of patients with metastatic gastric cancer (48). However, none of the patients in our analysis received tyrosine kinase inhibitors between the pre- and posttreatment samples. Large prospective studies are needed to determine the effect of chemotherapy and other therapies on PD-L1 expression and TMB in GEA, and mutational signatures could help distinguish variants caused by chemotherapy or tumor evolution (25, 49). Fourth, while we did not find associations of PD-L1 or TMB with OS in stage IVB GEA, our analysis may have been affected by residual confounding, and we cannot exclude associations with progression-free survival or in early-stage GEA. Finally, given the retrospective nature and small sample sizes in this study, larger independent studies are needed to confirm our conclusions regarding spatial and temporal heterogeneity of PD-L1 expression and TMB in GEA.

In conclusion, this study evaluating heterogeneity of PD-L1 expression and TMB in GEA demonstrated marked spatial and temporal discordance, which may impact their use as predictive biomarkers for ICIs. Further studies are needed to confirm our results, and to determine the optimal location and timing for measurement of these biomarkers. Multisite sampling, repeat testing after treatment, and blood-based testing each warrant further investigation as possible approaches to account for heterogeneity of PD-L1 expression and TMB. It may be that homogeneous disease, with concordant PD-L1+ and/or TMB-H profiles throughout, or particularly the metastatic disease profile if discordant, could be significant positive predictors of outcome. While our dataset was relatively small and heterogeneous to address this important question, it is being addressed within some studies (50). Understanding the heterogeneity and prognostic value of these biomarkers could ultimately help select GEA patients who are most likely to benefit from ICIs.

D.V.T. Catenacci reports personal fees from Merck, BMS, Five Prime, Gritstone, Astellas, Pieris, Daiichi Sankyo, Lilly, Genentech Roche, Tempus, Foundation Medicine, Guardant 360, Seattle Genetics, Taiho, Zymeworks, QED, Archer, and Natera during the conduct of the study. No potential conflicts of interest were disclosed by the other authors.

K.I. Zhou: Conceptualization, formal analysis, investigation, writing-original draft, writing-review and editing. B. Peterson: Data curation, writing-review and editing. A. Serritella: Writing-review and editing. J. Thomas: Writing-review and editing. N. Reizine: Writing-review and editing. S. Moya: Writing-review and editing. Y.-H. Tan: Writing-review and editing. Y. Wang: Conceptualization, resources, supervision, funding acquisition, project administration, writing-review and editing. D.V.T. Catenacci: Conceptualization, resources, supervision, funding acquisition, project administration, writing-review and editing.

This work was supported by the NIH Medical Scientist Training Program (T32GM007281) and the University of Chicago Biological Sciences Division and Frank Family Endowment (to K.I. Zhou); and NIH K23 award (CA178203–01A1), UCCCC (University of Chicago Comprehensive Cancer Center) Award in Precision Oncology—CCSG (Cancer Center Support Grant; P30CA014599), Castle Foundation, LLK (Live Like Katie) Foundation Award, Ullman Scholar Award and the Sal Ferrara II Fund for PANGEA (to D.V.T. Catenacci).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Zayac
A
,
Almhanna
K
. 
Esophageal, gastric cancer and immunotherapy: small steps in the right direction
?
Transl Gastroenterol Hepatol
2020
;
5
:
9
.
2.
Wagner
AD
,
Syn
NL
,
Moehler
M
,
Grothe
W
,
Yong
WP
,
Tai
B-C
, et al
Chemotherapy for advanced gastric cancer
.
Cochrane Database Syst Rev
2017
;
8
:
CD004064
.
3.
Fuchs
CS
,
Doi
T
,
Jang
RW
,
Muro
K
,
Satoh
T
,
Machado
M
, et al
Safety and efficacy of pembrolizumab monotherapy in patients with previously treated advanced gastric and gastroesophageal junction cancer: phase 2 clinical KEYNOTE-059 trial
.
JAMA Oncol
2018
;
4
:
e180013
.
4.
Joshi
SS
,
Maron
SB
,
Catenacci
DV
. 
Pembrolizumab for treatment of advanced gastric and gastroesophageal junction adenocarcinoma
.
Future Oncol
2018
;
14
:
417
30
.
5.
Walk
EE
,
Yohe
SL
,
Beckman
A
,
Schade
A
,
Zutter
MM
,
Pfeifer
J
, et al
The cancer immunotherapy biomarker testing landscape
.
Arch Pathol Lab Med
2020
;
144
:
706
24
.
6.
Gandhi
L
,
Rodríguez-Abreu
D
,
Gadgeel
S
,
Esteban
E
,
Felip
E
,
De Angelis
F
, et al
Pembrolizumab plus chemotherapy in metastatic non-small-cell lung cancer
.
N Engl J Med
2018
;
378
:
2078
92
.
7.
Weber
JS
,
D'Angelo
SP
,
Minor
D
,
Hodi
FS
,
Gutzmer
R
,
Neyns
B
, et al
Nivolumab versus chemotherapy in patients with advanced melanoma who progressed after anti-CTLA-4 treatment (CheckMate 037): a randomised, controlled, open-label, phase 3 trial
.
Lancet Oncol
2015
;
16
:
375
84
.
8.
Kulangara
K
,
Zhang
N
,
Corigliano
E
,
Guerrero
L
,
Waldroup
S
,
Jaiswal
D
, et al
Clinical utility of the combined positive score for programmed death ligand-1 expression and the approval of pembrolizumab for treatment of gastric cancer
.
Arch Pathol Lab Med
2019
;
143
:
330
7
.
9.
Goodman
AM
,
Kato
S
,
Bazhenova
L
,
Patel
SP
,
Frampton
GM
,
Miller
V
, et al
Tumor mutational burden as an independent predictor of response to immunotherapy in diverse cancers
.
Mol Cancer Ther
2017
;
16
:
2598
608
.
10.
Snyder
A
,
Makarov
V
,
Merghoub
T
,
Yuan
J
,
Zaretsky
JM
,
Desrichard
A
, et al
Genetic basis for clinical response to CTLA-4 blockade in melanoma
.
N Engl J Med
2014
;
371
:
2189
99
.
11.
Rizvi
NA
,
Hellmann
MD
,
Snyder
A
,
Kvistborg
P
,
Makarov
V
,
Havel
JJ
, et al
Cancer immunology. mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer
.
Science
2015
;
348
:
124
8
.
12.
Wang
F
,
Wei
XL
,
Wang
FH
,
Xu
N
,
Shen
L
,
Dai
GH
, et al
Safety, efficacy and tumor mutational burden as a biomarker of overall survival benefit in chemo-refractory gastric cancer treated with toripalimab, a PD-1 antibody in phase Ib/II clinical trial NCT02915432
.
Ann Oncol
2019
;
30
:
1479
86
.
13.
Marabelle
A
,
Fakih
MG
,
Lopez
J
,
Shah
M
,
Shapira-Frommer
R
,
Nakagawa
K
, et al
Association of tumour mutational burden with outcomes in patients with select advanced solid tumours treated with pembrolizumab in KEYNOTE-158
.
Annals of Oncology
2019
;
30
:
v477
8
.
14.
Murugaesu
N
,
Wilson
GA
,
Birkbak
NJ
,
Watkins
T
,
McGranahan
N
,
Kumar
S
, et al
Tracking the genomic evolution of esophageal adenocarcinoma through neoadjuvant chemotherapy
.
Cancer Discov
2015
;
5
:
821
31
.
15.
Pectasides
E
,
Stachler
MD
,
Derks
S
,
Liu
Y
,
Maron
S
,
Islam
M
, et al
Genomic heterogeneity as a barrier to precision medicine in gastroesophageal adenocarcinoma
.
Cancer Discov
2018
;
8
:
37
48
.
16.
Stovgaard
ES
,
Bokharaey
M
,
List-Jensen
K
,
Roslind
A
,
Kümler
I
,
Høgdall
E
, et al
PD-L1 diagnostics in the neoadjuvant setting: implications of intratumoral heterogeneity of PD-L1 expression in triple negative breast cancer for assessment in small biopsies
.
Breast Cancer Res Treat
2020
;
181
:
553
60
.
17.
Manson
QF
,
Schrijver
W
,
Ter Hoeve
ND
,
Moelans
CB
,
van Diest
PJ
. 
Frequent discordance in PD-1 and PD-L1 expression between primary breast tumors and their matched distant metastases
.
Clin Exp Metastasis
2019
;
36
:
29
37
.
18.
Haragan
A
,
Field
JK
,
Davies
MPA
,
Escriu
C
,
Gruver
A
,
Gosney
JR
. 
Heterogeneity of PD-L1 expression in non-small cell lung cancer: Implications for specimen sampling in predicting treatment response
.
Lung Cancer
2019
;
134
:
79
84
.
19.
Gao
Y
,
Li
S
,
Xu
D
,
Chen
S
,
Cai
Y
,
Jiang
W
, et al
Prognostic value of programmed death-1, programmed death-ligand 1, programmed death-ligand 2 expression, and CD8(+) T cell density in primary tumors and metastatic lymph nodes from patients with stage T1–4N+M0 gastric adenocarcinoma
.
Chin J Cancer
2017
;
36
:
61
.
20.
Kazdal
D
,
Endris
V
,
Allgäuer
M
,
Kriegsmann
M
,
Leichsenring
J
,
Volckmar
A-L
, et al
Spatial and temporal heterogeneity of panel-based tumor mutational burden in pulmonary adenocarcinoma: separating biology from technical artifacts
.
J Thorac Oncol
2019
;
14
:
1935
47
.
21.
Fournel
L
,
Wu
Z
,
Stadler
N
,
Damotte
D
,
Lococo
F
,
Boulle
G
, et al
Cisplatin increases PD-L1 expression and optimizes immune check-point blockade in non-small cell lung cancer
.
Cancer Lett
2019
;
464
:
5
14
.
22.
Park
S
,
Joung
J-G
,
Min
YW
,
Nam
J-Y
,
Ryu
D
,
Oh
D
, et al
Paired whole exome and transcriptome analyses for the Immunogenomic changes during concurrent chemoradiotherapy in esophageal squamous cell carcinoma
.
J Immunother Cancer
2019
;
7
:
128
.
23.
Van Der Kraak
L
,
Goel
G
,
Ramanan
K
,
Kaltenmeier
C
,
Zhang
L
,
Normolle
DP
, et al
5-Fluorouracil upregulates cell surface B7-H1 (PD-L1) expression in gastrointestinal cancers
.
J Immunother Cancer
2016
;
4
:
65
.
24.
Yang
JH
,
Kim
H
,
Roh
SY
,
Lee
MA
,
Park
JM
,
Lee
HH
, et al
Discordancy and changes in the pattern of programmed death ligand 1 expression before and after platinum-based chemotherapy in metastatic gastric cancer
.
Gastric Cancer
2019
;
22
:
147
54
.
25.
O'Donnell
T
,
Christie
EL
,
Ahuja
A
,
Buros
J
,
Aksoy
BA
,
Bowtell
DDL
, et al
Chemotherapy weakly contributes to predicted neoantigen expression in ovarian cancer
.
BMC Cancer
2018
;
18
:
87
.
26.
Zhang
M
,
Sun
H
,
Zhao
S
,
Wang
Y
,
Pu
H
,
Wang
Y
, et al
Expression of PD-L1 and prognosis in breast cancer: a meta-analysis
.
Oncotarget
2017
;
8
:
31347
54
.
27.
Pyo
J-S
,
Kang
G
,
Kim
JY
. 
Prognostic role of PD-L1 in malignant solid tumors: a meta-analysis
.
Int J Biol Markers
2017
;
32
:
e68
74
.
28.
Zhang
M
,
Dong
Y
,
Liu
H
,
Wang
Y
,
Zhao
S
,
Xuan
Q
, et al
The clinicopathological and prognostic significance of PD-L1 expression in gastric cancer: a meta-analysis of 10 studies with 1,901 patients
.
Sci Rep
2016
;
6
:
37933
.
29.
Kawazoe
A
,
Kuwata
T
,
Kuboki
Y
,
Shitara
K
,
Nagatsuma
AK
,
Aizawa
M
, et al
Clinicopathological features of programmed death ligand 1 expression with tumor-infiltrating lymphocyte, mismatch repair, and Epstein-Barr virus status in a large cohort of gastric cancer patients
.
Gastric Cancer
2017
;
20
:
407
15
.
30.
Wang
Y
,
Zhu
C
,
Song
W
,
Li
J
,
Zhao
G
,
Cao
H
. 
PD-L1 expression and CD8+ T cell infiltration predict a favorable prognosis in advanced gastric cancer
.
J Immunol Res
2018
;
2018
:
4180517
.
31.
Owada-Ozaki
Y
,
Muto
S
,
Takagi
H
,
Inoue
T
,
Watanabe
Y
,
Fukuhara
M
, et al
Prognostic impact of tumor mutation burden in patients with completely resected non-small cell lung cancer: brief report
.
J Thorac Oncol
2018
;
13
:
1217
21
.
32.
Innocenti
F
,
Ou
F-S
,
Qu
X
,
Zemla
TJ
,
Niedzwiecki
D
,
Tam
R
, et al
Mutational analysis of patients with colorectal cancer in CALGB/SWOG 80405 identifies new roles of microsatellite instability and tumor mutational burden for patient outcome
.
J Clin Oncol
2019
;
37
:
1217
27
.
33.
Zhou
KI
,
Peterson
B
,
Serritella
A
,
Reizine
N
,
Wang
Y
,
Catenacci
DVT
. 
Evaluation of spatiotemporal heterogeneity of PDL1 expression in gastroesophageal adenocarcinoma (GEA) at baseline diagnosis and after chemotherapy
.
J Clin Oncol
2020
;
38
:
4528
.
34.
Zhou
KI
,
Peterson
B
,
Serritella
A
,
Reizine
N
,
Wang
Y
,
Catenacci
DVT
. 
Evaluation of spatiotemporal heterogeneity of tumor mutational burden (TMB) in gastroesophageal adenocarcinoma (GEA) at baseline diagnosis and after chemotherapy
.
J Clin Oncol
2020
;
38
:
4546
.
35.
Bartley
AN
,
Washington
MK
,
Colasacco
C
,
Ventura
CB
,
Ismaila
N
,
Benson
AB
, et al
HER2 Testing and clinical decision making in gastroesophageal adenocarcinoma: guideline from the college of American pathologists, American society for clinical pathology, and the American society of clinical oncology
.
J Clin Oncol
2017
;
35
:
446
64
.
36.
Chalmers
ZR
,
Connelly
CF
,
Fabrizio
D
,
Gay
L
,
Ali
SM
,
Ennis
R
, et al
Analysis of 100,000 human cancer genomes reveals the landscape of tumor mutational burden
.
Genome Med
2017
;
9
:
34
.
37.
Beaubier
N
,
Tell
R
,
Lau
D
,
Parsons
JR
,
Bush
S
,
Perera
J
, et al
Clinical validation of the tempus xT next-generation targeted oncology sequencing assay
.
Oncotarget
2019
;
10
:
2384
96
.
38.
Vanderwalde
A
,
Spetzler
D
,
Xiao
N
,
Gatalica
Z
,
Marshall
J
. 
Microsatellite instability status determined by next-generation sequencing and compared with PD-L1 and tumor mutational burden in 11,348 patients
.
Cancer Med
2018
;
7
:
746
56
.
39.
Parikh
AR
,
He
Y
,
Hong
TS
,
Corcoran
RB
,
Clark
JW
,
Ryan
DP
, et al
Analysis of DNA damage response gene alterations and tumor mutational burden across 17,486 tubular gastrointestinal carcinomas: implications for therapy
.
Oncologist
2019
;
24
:
1340
7
.
40.
Schrock
AB
,
Ouyang
C
,
Sandhu
J
,
Sokol
E
,
Jin
D
,
Ross
JS
, et al
Tumor mutational burden is predictive of response to immune checkpoint inhibitors in MSI-high metastatic colorectal cancer
.
Ann Oncol
2019
;
30
:
1096
103
.
41.
Liu
X
,
Choi
MG
,
Kim
K
,
Kim
K-M
,
Kim
ST
,
Park
SH
, et al
High PD-L1 expression in gastric cancer (GC) patients and correlation with molecular features
.
Pathol Res Pract
2020
;
216
:
152881
.
42.
Yarchoan
M
,
Albacker
LA
,
Hopkins
AC
,
Montesion
M
,
Murugesan
K
,
Vithayathil
TT
, et al
PD-L1 expression and tumor mutational burden are independent biomarkers in most cancers
.
JCI Insight
2019
;
4
:
e126908
.
43.
Svensson
MC
,
Borg
D
,
Zhang
C
,
Hedner
C
,
Nodin
B
,
Uhlén
M
, et al
Expression of PD-L1 and PD-1 in chemoradiotherapy-naïve esophageal and gastric adenocarcinoma: relationship with mismatch repair status and survival
.
Front Oncol
2019
;
9
:
136
.
44.
Chao
J
,
Fuchs
CS
,
Shitara
K
,
Tabernero
J
,
Muro
K
,
Cutsem
EV
, et al
Pembrolizumab (pembro) in microsatellite instability-high (MSI-H) advanced gastric/gastroesophageal junction (G/GEJ) cancer by line of therapy
.
Journal of Clinical Oncology
2020
;
38
:
430
.
45.
Sakai
H
,
Takeda
M
,
Sakai
K
,
Nakamura
Y
,
Ito
A
,
Hayashi
H
, et al
Impact of cytotoxic chemotherapy on PD-L1 expression in patients with non-small cell lung cancer negative for EGFR mutation and ALK fusion
.
Lung Cancer
2019
;
127
:
59
65
.
46.
Yuan
C
,
Xiang
L
,
Cao
K
,
Zhang
J
,
Luo
Y
,
Sun
W
, et al
The prognostic value of tumor mutational burden and immune cell infiltration in esophageal cancer patients with or without radiotherapy
.
Aging (Albany NY)
2020
;
12
:
4603
16
.
47.
Ye
M
,
Huang
D
,
Zhang
Q
,
Weng
W
,
Tan
C
,
Qin
G
, et al
Heterogeneous programmed death-ligand 1 expression in gastric cancer: comparison of tissue microarrays and whole sections
.
Cancer Cell Int
2020
;
20
:
186
.
48.
Lee
J
,
Kim
ST
,
Kim
K
,
Lee
H
,
Kozarewa
I
,
Mortimer
PGS
, et al
Tumor genomic profiling guides patients with metastatic gastric cancer to targeted treatment: the VIKTORY umbrella trial
.
Cancer Discov
2019
;
9
:
1388
405
.
49.
Secrier
M
,
Li
X
,
de Silva
N
,
Eldridge
MD
,
Contino
G
,
Bornschein
J
, et al
Mutational signatures in esophageal adenocarcinoma define etiologically distinct subgroups with therapeutic relevance
.
Nat Genet
. 
2016
;
48
:
1131
41
.
50.
Catenacci
DVT
,
Lomnicki
S
,
Chase
L
,
Peterson
B
,
Moore
K
,
Markevicius
U
, et al
Personalized ANtibodies for gastroesophageal adenocarcinoma (PANGEA): primary efficacy analysis of the phase II platform trial (NCT02213289)
.
J Clin Oncol
2020
;
38
:
356
.