Clinicopathological and genomic features of dMMR/MSI-H GAC.
The clinicopathological and genomic features of 175 resected samples from patients with primary G/GEJ adenocarcinoma are described in Supplementary Table S5 and Figure 1A. IHC, PCR (5 loci), and NGS (100 loci) testing identified 115 concordant-dMMR/MSI-H samples, 46 concordant MMR-proficent (pMMR)/microsatellite stable (MSS) samples, and 14 discordant samples. The distribution of MSI score (by NGS testing), PCR score, and IHC result are displayed in Supplemental Figure S2 and Supplemental Table S6. In the 161 samples with concordant results of IHC and PCR testing, NGS was able to correctly detect MSI status with 100% sensitivity and 100% specificity. Furthermore, among the 14 samples where IHC result showed incomplete loss of MMR protein expression (e.g., loss of MSH6 expression in 50% of tumor cells and intact expression of MLH1/PMS2/MSH2) or was inconsistent with PCR results (e.g., IHC-dMMR but PCR-MSS), NGS identified 7 NGS-MSI-H cases with the highest variability of the tested microsatellites and the highest frameshift burden in the tested coding sequence (Supplemental Table S7). Taken together, NGS performed better in identifying MSI-H cases compared to IHC and PCR, especially when geographical heterogeneity of MMR protein expression is observed, or IHC and PCR results are inconsistent.
Among the concordant samples, dMMR/MSI-H status was associated with an early tumor stage (P=0.050), intestinal type of Lauren classification (P=0.058), lower HER2 expression (P=0.034), and EBV negativity (P=0.008, Supplementary Table S5). As for genetic aberration, mutations of TP53 (P=0.021) and CCNE1 (P=0.003) were enriched in pMMR/MSS samples, while the mutations in ARID1A, ACVR2A, KMT2C, TGFBR2, KMT2D, and RNF43 genes were dominant in dMMR/MSI-H samples (P<0.001) with mutational frequencies over 70% (Figure 1B). Significantly higher tumor mutational burden (TMB) and the numbers of frameshift mutation, missense mutation, non-frameshift insertion and deletion (indel), stopgain mutation, and splice site mutation were observed in concordant dMMR/MSI-H cases. On the contrary, copy number variation (CNV) and fusion were enriched in concordant pMMR/MSS cases (Supplemental Table S5 and Figure 1C).
Most dMMR/MSI-H samples carried mutations in the DNA damage response (99%), chromatin remodeling (99%), WNT (99%), TGFβ (98%), PI3K-AKT-mTOR (91%), Hippo (90%), and NOTCH (90%) pathways, while relatively lower mutational rates were discovered in the RTKs (73%), Hedgehog (68%), cell cycle (63%), Ras-Raf-MEK-ERK/JNK (62%), p53 (56%), and telomere maintenance (19%) pathways (Figure 1D). In the dMMR/MSI-H tumors with such high mutational burden and intratumoral heterogeneity, a singular mutation (e.g., PTEN mutation) can hardly reflect the functional alteration of the whole pathway (the PI3K-AKT-mTOR pathway) and the overall characteristics of the tumor. Therefore, we sought to define the number of mutated genes in the pre-specified pathway as a parameter representing the pathway’s potential changes and to explore its correlation with the density of DTICs and response to ICI treatment in the dMMR/MSI-H GAC.
Correlates of DTIC in dMMR/MSI-H GAC
Of the 115 concordant-dMMR/MSI-H cases, the evaluation of DTICs was missing in twelve cases (Supplementary Table S8), and another fourteen cases were excluded for prior neoadjuvant chemotherapy which could affect DTICs (Supplementary Table S9) [23-25]. Therefore, 89 samples were included for the following analysis of genomic correlates of DTICs.
A comprehensive correlation matrix was created to seek the correlates of DTICs, including CD3+, CD4+, CD8+, CD68+, and FOXP3+ cells in central-tumor and invasive-margin areas (Figure 2). Among the clinicopathological and genomic characteristics, the number of mutated members in the PI3K-AKT-mTOR pathway (NMP) exhibited the strongest negative correlation with DTICs, including CD3+ (P<0.001), CD4+ (P=0.065), CD8+ (P=0.004), and FOXP3+ (P=0.033) cells in central tumor area (marked by a red arrow, Figure 2). The correlations of NMP with central-tumor DTICs were markedly stronger than its correlations with invasive-margin DTICs, suggesting the potential difference of PI3K-AKT-mTOR function in central tumor and invasive margin. Sensitivity analysis further indicated the robustness of the above-mentioned results (Supplementary Table S10). The mutations of the members in the PI3K-AKT-mTOR pathway are illustrated in Supplementary Figure S3. The scatter diagrams of the above-mentioned results and representative images of immunohistochemical staining of tumor-infiltrating immune cells are displayed in Figure 3.
Of note, TMB was strongly correlated with nearly all the numbers of mutated members in the pre-specified pathways (marked by a purple arrow, Figure 2) including the PI3K-AKT-mTOR pathway, rather than DITCs (marked by a blue arrow, Figure 2), consistent with previous results in MSI-H CRC [14, 26].
Given the correlations of NMP with DTICs, we next sought to investigate the associations of NMP with other potential predictors of immunotherapy, including TMB, PD-L1 expression, and immune-related mRNA signatures.
Biological characteristics of the DTIC-enriched subtype with lower NMP
Since higher TMB and PD-L1 levels were commonly associated with more clinical benefit from ICI treatment in microsatellite stable (MSS) G/GEJ cancer [27-34] and two recent retrospective studies with small sample size suggested the association between TMB and immunotherapy efficacy in MSI-H GAC [35, 36], we set out to evaluate TMB and PD-L1 expression in the DTIC-enriched subtype of MSI-H GAC with lower NMP.
We retrieved the data of 32 MSI-H GACs (primary lesion) with assessments of mutations and PD-L1 from the 3DMed database. Among these, lower NMP was correlated with lower TMB (P<0.001, Figure 4A), but not the PD-L1 expression in tumoral and immune cells (tumor proportion score [TPS]: P=0.961; immune proportion score [IPS]: 0.484; combined positive score [CPS]: P=0.699; Figure 4B). Detailed characteristics and data are shown in Supplementary Table S11. The representative images of PD-L1 staining are shown in Figure 4C, and the images of corresponding hematoxylin-eosin (HE) staining and positive/negative controls are enclosed in Supplementary Figure S4.
Immune-related gene signatures were further explored in the TCGA cohort by GSEA (characteristics are shown in Supplementary Table S12). Multiple significant enrichments of immune-related signatures were revealed in the PI3K-AKT-mTORWT group (Figure 4D), including MHC-II-mediated antigen presentation (P=0.001), cross presentation (P=0.001), B cell receptor downstream signaling (P=0.017), T cell receptor downstream signaling (P<0.001), interferon signaling (P=0.045), and PD-1 signaling (P=0.006), while the activity of TGFβ signaling was decreased in the PI3K-AKT-mTORWT group (P<0.001). Of note, despite the enrichment of PD-1 signaling in PI3K-AKT-mTORWT group, the mRNA expression of CD274 (PD-L1) was not higher in PI3K-AKT-mTORWT group than in PI3K-AKT-mTORmut group, suggesting that PD-L1 expression is irrelevant to PI3K-AKT-mTOR mutations (P=0.341), consistent with the results in Figure 4B. Similar enrichments in the PI3K-AKT-mTORWT group were observed in interleukin (IL) pathways, including IL-1 (P=0.031), IL-12 (P=0.014), and IL-12 family (P=0.059, Figure 4E).
In addition to the immune-related gene signatures, we further assessed the signatures of NOTCH signaling. We previously reported that the downregulation of NOTCH pathway was associated with higher level of immune gene transcription and better immunotherapeutic efficacy in non-small cell lung cancer (NSCLC) . Here, in the MSI-H GACs, the DTIC-enriched PI3K-AKT-mTORWT group displayed trends towards downregulation of NOTCH-related signatures, especially the ones concerning transcriptional impact (Figure 4F).
PI3K-AKT-mTOR inhibitor efficacy of MSI-H GAC cell lines with different NMPs
Furthermore, we aimed to discover the association between PI3K-AKT-mTOR mutation and drug sensitivity. Among the five MSI-H GAC cell lines (NUGC-3, TGBC11TKB, SNU-1, IM95, and 23132/87), NUGC-3 and TGBC11TKB had two mutations of the members of the PI3K-AKT-mTOR pathway, and SNU-1, IM95, and 23132/87 had no less than four member mutations (Figure 4G). The PI3K-AKT-mTOR inhibitors (targeting AKT/mTORC/PDK1/PI3K/S6K1) exhibited lower IC50 in the SNU-1, IM95, and 23132/87 cell lines (P<0.001) than in NUGC-3 and TGBC11TKB cell lines, and the inhibitors targeting VEGFR (P=0.142) or EGFR (P=0.540) and chemotherapeutic drugs (P=0.378) showed no significant difference of sensitivity (Figure 4G). Taken together, these results indicate that the MSI-H STADs with high NMP might benefit more from PI3K/AKT/mTOR inhibitors, compared to the ones with low NMP.
Immunotherapy efficacy of the DTIC-enriched subtypes with lower NMP
The subtype with lower NMP was characterized by higher DTICs and immune-related gene transcription (potentially associated with better ICI response) [32, 38], and lower TMB (potentially associated with poorer ICI response) [27, 28, 39] in MSI-H GAC. Given these opposite predictive values, we sought to investigate the ICI efficacy of this subtype.
In total, 36 patients with locally advanced or metastatic concordant-dMMR/MSI-H G/GEJ adenocarcinoma were included. The key baseline characteristics, individual response to ICI treatment, and mutational events of the members of the PI3K-AKT-mTOR pathway are illustrated in Figure 5A and Supplementary Table S13. The best responses are shown in Figure 5B. To explore the optimal cut-off of the NMP for predicting immunotherapy efficacy in dMMR/MSI-H G/GEJ adenocarcinoma, a ROC curve was plotted based on the objective response in patients with evaluable target lesion (n=31). The AUC of NMP was significantly higher than 0.5 (AUC=0.792, 95% CI 0.628-0.956, P=0.006, Figure 5B), suggesting the feasibility of using NMP to predict response to ICIs. The optimal cut-off was set as 1 when the largest Youden’s index was achieved. In the NMP-high patients (NMP≥2), ORR was 29.4% and 4-month PFS rate was 35.3%, while in the NMP-low patients (NMP=0/1), ORR and 4-month PFS rate were significantly higher as 85.7% (P=0.002) and 93.3% (P=0.001), respectively (Table 1).
Consistent with our previous findings, the NMP-high group exhibited higher TMB compared to the NMP-low group (Figure 5C), and NMP was not associated with PD-L1 CPS (Figure 5D). In addition, the ROC curves based on the objective response demonstrate that TMB (AUC=0.582, 95% CI 0.377-0.787, P=0.44) and PD-L1 CPS (AUC=0.529, 95% CI 0.291-0.767, P=0.82) were not associated with the response to ICI treatment in the dMMR/MSI-H G/GEJ adenocarcinomas (Supplementary Figure S5).
The clinical characteristics were comparable between the NMP-high and NMP-low groups (Supplementary Table S13). The median time to progression or death was 3.4 months in the NMP-high patients versus not reached in the NMP-low patients (HR=3.40, 95% CI 1.16-10.00, Log-rank P=0.019, Figure 5E). Similarly, shorter median OS was observed in the NMP-high patients as 15.0 months, as compared with not reached in the NMP-low patients (HR=3.59, 95% CI 0.94-13.78, Log-rank P=0.048, Figure 5F). The maturity of OS was 30.6%, contributing to the slight difference between the results of Cox regression and Log-rank statistics (P=0.063 and 0.048, respectively), and therefore we set PFS as the major outcome in the following analyses.
Despite that higher TMB was observed in the NMP-high group, we found that the predictive effect of NMP on PFS was similar in the TMB-high (≥40 mutations/Mb) and TMB-low (<40 mutations/Mb) subgroups (Figure 5G). This cut-off of TMB (40) was selected based on the first study pointing out the predictive effect of TMB in MSI-H gastrointestinal (GI) tumors . Further subgroup analysis suggests the consistent predictive effects of NMP in patients with different characteristics, including treatment lines, ECOG, differentiation grade, PD-L1 CPS, and NGS testing technique (Figure 5G). Univariable analyses of PFS and OS did not reveal potential predictive biomarker other than NMP and the multivariable analyses identified NMP as a predictor independent of TMB and PD-L1 for PFS (multivariable HR=5.99, 95% CI 1.21-29.61, P=0.028) and OS (multivariable HR=11.88, 95% CI 1.30-108.4, P=0.028) (Supplemental Table S14). Consistent with the results in Figure 3, compared to the NMP-high subgroup, higher DTICs, including CD3+ (P=0.022) CD4+ (P=0.013), CD8+ (P=0.10), and FOXP3+ cells (P=0.073) were observed in the NMP-low subgroup by multiplex immunofluorescence, while CD68+ macrophage (P=0.37), M1 macrophage (CD68+/CD163-, P=0.29), and M2 macrophage (CD68+/CD163+, P=0.59) were not significantly differed (Figure 5H-I).
A previous retrospective study involving 45 dMMR/MSI-H GI tumors (18 gastric tumors) suggests that the mutation of PTEN, a key member of the PI3K-AKT-mTOR pathway, was associated with poor response to ICIs . Given this, we first tried to validate this result in our 36 cases and only observed non-significant trends in PFS (P=0.34, Supplemental Figure S6A) and OS (P=0.29, Supplemental Figure S6B). Moreover, in the 28 PTENWT cases, NMP-high remained as an indicator of poorer PFS (HR=5.98, 95% CI 1.47-24.22, P=0.005, Supplemental Figure S6C) and OS (HR=5.95, 95% CI 1.09-32.47, P=0.022, Supplemental Figure S6D). Taken together, NMP, as a predictor concerning multiple key members of the PI3K-AKT-mTOR pathway, performed more robustly and powerfully than a singular PTEN mutation.
In addition, to exclude the possibility that the association between NMP and ICI efficacy was impacted by the pre-treatment immune cell concentration in peripheral blood (reflecting systemic immunity), we detected the correlation between NMP and the pre-treatment status of peripheral blood lymphocyte subsets and observed no significant correlations (Supplementary Figure S7). Moreover, PI3K-AKT-mTORmut was not associated with the DFS (P=0.37) and OS (P=0.37) in the MSI-H GAC cases of the TCGA database (Supplementary Figure S8). Collectively, NMP-high was identified as a predictive, rather than prognostic biomarker, associated with inferior clinical benefit from ICI treatment in patients with dMMR/MSI-H G/GEJ adenocarcinoma.