Published papers

We consider tensor factorizations based on sparse measurements of the components of relatively high rank tensors. The measurements are designed in a way that the underlying graph of interactions is a random graph. The setup will be useful in cases where a substantial amount of data is missing, as in completion of relatively high rank matrices for recommendation systems heavily used in social network services. In order to obtain theoretical insights on the setup, we consider statistical inference of the tensor factorization in a high dimensional limit, which we call as dense limit, where the graphs are large and dense but not fully connected. We build message-passing algorithms and test them in a Bayes optimal teacher-student setting in some specific cases. We also develop a replica theory to examine the performance of statistical inference in the dense limit based on a cumulant expansion. The latter approach allows one to avoid blind usage of Gaussian ansatz which fails in some fully connected systems.

With the widespread use of programmed death receptor-1 (PD-1) blockade therapy, sensitive and specific predictive biomarkers that guide patient selection are urgently needed. T-cell receptor (TCR) repertoire, which reflects antitumor T-cell responses based on antigen specificity, is expected as a novel biomarker for PD-1 blockade therapy. In the present study, the TCR repertoire of eight patients with gastrointestinal cancer treated with anti-PD-1 antibody (nivolumab) was analyzed. To analyze the tumor-associated T-cell clones in the blood and their mobilization into the tumor, we focused on T-cell clones that presented in both blood and tumor (blood-tumor overlapping clones). Responders to PD-1 blockade tended to exhibit a higher number of overlapping clones in the tumor and a higher total frequency in the blood. Moreover, a higher total frequency of overlapping clones in blood CD8+ T cells before treatment was associated with a favorable clinical response. Collectively, these results suggest the possibility of blood-tumor TCR repertoire overlap to predict clinical response to PD-1 blockade and guide patient selection before the treatment.

OBJECTIVE: To investigate an association between the gut microbiome and efficacy of poly(ADP-ribose) polymerase inhibitors (PARPi) in ovarian cancer. METHODS: This study conducted fecal microbiome analysis (16S rRNA gene sequencing) and circulating tumor DNA (ctDNA) profiling for ovarian cancer patients who underwent PARPi maintenance therapy. Fecal and blood samples were collected at the baseline and the progressive disease (PD) or last follow-up. The relative abundance of gut microbes and progression-free survival (PFS) were analyzed using linear discriminant analysis of effect size and the Cox proportional hazard model according to BRCA1/2 mutation (BRCA1/2mut) status detected by ctDNA sequencing. RESULTS: Baseline samples were available from 23 BRCA1/2mut-positive patients and 33 BRCA1/2mut-negative patients. The microbes enriched in the baseline samples with long PFS were Bifidobacterium, Roseburia, Dialister, Butyricicoccus, and Bilophila for BRCA1/2mut-positive patients and Phascolarctobacterium for BRCA1/2mut-negative patients. In multivariate analyses dividing patients by the median values of relative abundances, no bacteria were associated with PFS in BRCA1/2mut-positive patients, whereas high Phascolarctobacterium abundances (≥1.11%) was significantly associated with longer PFS in BRCA1/2mut-negative patients (median 14.0 vs. 5.9 months, hazard ratio=0.28; 95% confidence interval=0.11-0.69; p=0.014). In the last samples, the relative abundances of Phascolarctobacterium were significantly higher in patients without PD (n=5) than those with PD (n=15) (median 1.25% vs. 0.06%; p=0.016). CONCLUSION: High fecal composition of Phascolarctobacterium was associated with prolonged PFS in patients with BRCA1/2mut-negative ovarian cancer receiving PARPi therapy. Our results would provide new insights for future research.

Aim: Esophagectomy may alter gut microbiota, a key regulator of antitumor immunity and immune checkpoint inhibitor (ICI) efficacy; however, its impact in patients with esophageal squamous cell carcinoma (ESCC) remains unclear. This study evaluated the association between the tumor immune microenvironment (TIME) and gut microbiota and also investigated gut microbiota composition in relation to nivolumab response in patients with ESCC. Methods: Two cohorts were analyzed. In Cohort 1 (n = 18), fecal samples collected before and three months after esophagectomy were analyzed alongside TIME profiling, and patients were classified as hot or cold tumors based on CD8+ T-cell infiltration and PD-L1 expression. In Cohort 2 (n = 35), the relationship between baseline gut microbiota and nivolumab response was assessed using a retrospective analysis of a prospectively collected multicenter cohort. Results: Patients with hot tumors exhibited higher microbial alpha diversity than those with cold tumors. Before esophagectomy, cold tumors showed enrichment of Streptococcus, Veillonella, Haemophilus, and Gemella, whereas hot tumors were enriched in Flavonifractor and Phascolarctobacterium; these differences disappeared postoperatively, indicating surgery-associated compositional shifts. In Cohort 2, higher alpha diversity was associated with favorable nivolumab response, while patients with Streptococcus- or Veillonella-enriched microbiota more frequently had residual tumors and poor outcomes. Conclusion: Gut microbiota diversity was associated with the tumor immune microenvironment and nivolumab response in patients with esophageal cancer. Esophagectomy was also associated with alterations in gut microbial composition.

67Background: The global phase 3 FRESCO-2 study, (NCT04322539) demonstrated that fruquintinib (F) vs placebo (P) significantly improved OS (HR=0.66 [95% CI: 0.55-0.80]; p<0.001) and PFS (HR=0.32 [95% CI: 0.27-0.39]; p<0.001) in heavily pre-treated patients (pts) with refractory metastatic colorectal cancer (mCRC). F safety profile was consistent with the established monotherapy profile. Here we report the HRQoL and tolerability results. Methods: Pts were randomized 2:1 to F + BSC or P + BSC. EORTC QLQ-C30 and EQ-5D-5L were assessed at baseline and on Day (D) 1 of each Cycle (C) until treatment discontinuation, and ECOG PS was assessed at baseline, D1 of each C, and D21 of C1 to C3. Least-squares mean (LSM) change from baseline to post-baseline visits and the difference between F and P in QLQ-C30 scale scores (e.g. global health status [GHS]/QoL) and EQ-5D-5L scale scores (e.g. visual analog scale [VAS]) were calculated using mixed model repeated measures approach. For each scale, the appropriate minimally important difference (MID) thresholds were determined to evaluate the improvement or deterioration. Time to deterioration (TTD), defined as worsening from baseline in scale-specific MID or death, was analyzed using Kaplan-Meier method, adjusted log-rank test, and stratified Cox PH model. QLQ-C30 and EQ-5D-5L analyses were conducted on the ITT population, and ECOG PS was on the safety population. Results: 691 pts were randomized (F: 461 vs P: 230) and 686 pts received study drug (F: 456 vs P: 230). Median treatment Cycles received (range) were 3 (1, 20) for F vs 2 (1, 13) for P. More than 79% of pts on both arms had a baseline and ≥1 post-baseline assessment for QLQ-C30, EQ-5D-5L, and ECOG PS. GHS/QoL was similar between F and P at baseline; the LSM differences between F and P were 1.7 (95% CI: -1.7, 5.0) for C2 and 1.6 (95% CI: -3.2, 6.4) for C3. At C4, <30 P patients were available. The % of patients who remained stable (MID -6.38 to <8.43) or improved (≥8.43) was numerically higher for F vs P (C2: 61.5% vs 57.1%; C3: 56.4% vs 50.9%). Median TTD was 2.1 months in F and 1.8 months in P (HR=0.9; 95% CI: 0.7-1.0; P=0.098). EQ-5D VAS was similar between F and P at baseline; the LSM differences between F and P were 0.6 (95% CI: -2.3, 3.5) for C2 and 1.4 (95% CI: -2.8, 5.6) for C3. The % of patients who remained stable (MID -7 to <7) or improved (≥7) was similar for F and P (C2: 64.6% vs 58.3%; C3: 64.2% vs 64.8%). Median TTD was 2.6 months in F and 1.9 months in P (HR=0.8; 95% CI: 0.6-0.9; P=0.001). The % of patients with ≥1-point increase from baseline in ECOG PS was 52.1% in F vs 54.0% in P. Conclusions: HRQoL is not negatively impacted by treatment with F. TTD in health utility instrument EQ-5D is improved for patients receiving F. These results, along with improved OS and PFS and favorable toxicity profile, further support F as a potential new treatment option for patients with refractory mCRC. Clinical trial information: NCT04322539.

INTRODUCTION: Maintaining or improving health-related quality of life (HRQoL) is as important as extending survival in metastatic colorectal cancer. We report an HRQoL analysis from FRESCO-2 (NCT04322539). METHODS: Patients were randomized to fruquintinib +best supportive care (BSC; n = 461) or placebo +BSC (n = 230). Instruments of EORTC QLQ-C30 and 5-level EQ-5D, and ECOG performance status (PS) were assessed. Changes from baseline scores for QLQ-C30 and EQ-5D were evaluated and minimally important difference thresholds were used to define stable, improved, or deteriorated QoL. Time to deterioration (TTD) was assessed. RESULTS: With fruquintinib versus placebo, baseline QLQ-C30 global health status (GHS) and EQ-5D visual analog scale (VAS) scores were 65.2 versus 64.6 and 67.0 versus 66.6, respectively. Least-squares mean changes from baseline fluctuated throughout treatment. At end of treatment (EOT), mean scores with fruquintinib versus placebo were 53.8 versus 52.3 (QLQ-C30 GHS) and 58.9 versus 58.5 (EQ-5D VAS). For QLQ-C30 GHS, 38.3 % versus 36.5 % of patients receiving fruquintinib versus placebo had stable or improved scores at EOT; median TTD was 2.1 versus 1.8 months (HR, 0.9; 95 % CI, 0.7-1.0). For EQ-5D VAS, 47.9 % versus 42.7 % had stable or improved scores at EOT; median TTD was 2.6 versus 1.9 months (HR, 0.8; 95 % CI, 0.6-0.9). Median TTD to ECOG PS ≥ 2 or death within 30+ /7 days after EOT was 6.6 versus 2.9 months with fruquintinib versus placebo (HR, 0.6; 95 % CI, 0.4-0.7). CONCLUSIONS: Fruquintinib delayed TTD of ECOG PS and did not negatively impact HRQoL versus placebo.

BACKGROUND: In the KEYNOTE-177 study, pembrolizumab monotherapy provided statistically significant and clinically meaningful improvements in progression-free survival versus chemotherapy as first-line treatment in patients with microsatellite instability-high or mismatch repair-deficient metastatic colorectal cancer. To further support the efficacy and safety findings of the KEYNOTE-177 study, results of the health-related quality of life (HRQOL) analyses are reported here. METHODS: KEYNOTE-177 is an open-label, randomised, phase 3 trial being done at 192 cancer centres in 23 countries, in patients aged 18 years and older with microsatellite instability-high or mismatch repair-deficient metastatic colorectal cancer, with an Eastern Cooperative Oncology Group performance status of 0 or 1, and who had not received previous systemic therapy for metastatic disease. Eligible patients were randomly assigned (1:1) centrally by use of interactive voice response or integrated web response technology to receive pembrolizumab 200 mg intravenously every 3 weeks or investigator's choice chemotherapy (mFOLFOX6 [leucovorin, fluorouracil, and oxaliplatin] or FOLFIRI [leucovorin, fluorouracil, and irinotecan] intravenously every 2 weeks with or without intravenous bevacizumab or cetuximab). Patients and investigators were not masked to treatment assignment. The primary endpoints were progression-free survival (previously reported) and overall survival (data to be reported at the time of the final analysis). HRQOL outcomes were evaluated as prespecified exploratory endpoints. The analysis population comprised all randomly assigned patients who received at least one dose of study treatment and completed at least one HRQOL assessment. HRQOL outcomes were mean change from baseline to prespecified week 18 in European Organisation for Research and Treatment of Cancer Quality of Life Questionnaire Core 30 (EORTC QLQ-C30) and EORTC Quality of Life Questionnaire-Colorectal 29 (EORTC QLQ-CR29) scale and item scores, and in the EuroQoL 5 Dimensions 3 Levels (EQ-5D-3L) visual analogue scale and health utility scores; the proportion of patients with improved, stable, or deteriorated scores from baseline to prespecified week 18 in EORTC QLQ-C30 scales and items; and time to deterioration in EORTC QLQ-C30 global health status/quality of life (GHS/QOL), physical functioning, social functioning, and fatigue scores and EORTC QLQ-CR29 urinary incontinence scores. The threshold for a small and clinically meaningful mean difference in EORTC QLQ-C30 score was 5-8 points. This study is registered with ClinicalTrials.gov, NCT02563002 and is ongoing; recruitment is closed. FINDINGS: Between Feb 11, 2016, and Feb 19, 2018, 307 patients were enrolled and randomly assigned to receive pembrolizumab (n=153) or chemotherapy (n=154). The HRQOL analysis population comprised 294 patients (152 receiving pembrolizumab and 142 receiving chemotherapy). As of Feb 19, 2020, median time from randomisation to data cutoff was 32·4 months (IQR 27·7-37·8). Least squares mean (LSM) change from baseline to prespecified week 18 showed a clinically meaningful improvement in EORTC QLQ-C30 GHS/QOL scores with pembrolizumab versus chemotherapy (between-group LSM difference 8·96 [95% CI 4·24-13·69]; two-sided nominal p=0·0002). Median time to deterioration was longer with pembrolizumab versus chemotherapy for GHS/QOL (hazard ratio 0·61 [95% CI 0·38-0·98]; one-sided nominal p=0·019), physical functioning (0·50 [95% CI 0·32-0·81]; one-sided nominal p=0·0016), social functioning (0·53 [95% CI 0·32-0·87]; one-sided nominal p=0·0050), and fatigue scores (0·48 [95% CI 0·33-0·69]; one-sided nominal p<0·0001). INTERPRETATION: Pembrolizumab monotherapy led to clinically meaningful improvements in HRQOL compared with chemotherapy in patients with previously untreated microsatellite instability-high or mismatch repair-deficient metastatic colorectal cancer. These data, along with the previously reported clinical benefits, support pembrolizumab as a first-line treatment option for this population. FUNDING: Merck Sharp & Dohme, a subsidiary of Merck & Co, Kenilworth, NJ, USA.

The TITANIA (mulTi omIcs daTa cANcer dIagnostics therApies) study is an international collaboration to generate high-quality multi-omics cancer profiles. We conducted an exploratory investigation within this framework to understand the molecular basis of lymphovascular invasion (LVI), a critical determinant of metastatic potential in clear cell renal cell carcinoma (ccRCC). We analyzed 31 ccRCC specimens (11 LVI+, 20 LVI-) from the National Cancer Center Hospital East using whole-genome sequencing, RNA sequencing, and proteomic profiling. Our findings were integrated with public single-cell RNA sequencing (GSE159115) and spatial transcriptomics (GSE175540) datasets to provide a broader biological context. LVI+ tumors consistently showed a distinctive hepatic-lineage gene expression signature, with significant upregulation of aldolase B (ALDOB) and other liver-specific metabolism genes at both the RNA and protein levels. Single-cell analysis identified a previously unrecognized hepatic-differentiated tumor subpopulation expressing master transcription factors HNF1A and HNF4A, which was positioned at the terminal stages of tumor evolution. Comparison with established hepatocyte gene signatures from three independent databases confirmed enrichment of hepatic metabolic programs in this subpopulation. Spatial transcriptomics revealed preferential localization within hypoxic tumor regions. A metabolic program resembling hepatic lineage differentiation, associated with aggressive disease features including LVI and hypoxic microenvironments, offers preliminary insights into renal cancer progression and potential biomarker development. This is a discovery-based, hypothesis-generating study; all findings require independent functional validation before clinical application.

PURPOSE: The tumor-agnostic approach has been increasingly adopted in precision oncology. Immunohistochemistry (IHC) is the standard biomarker testing for human epidermal growth factor receptor 2 (HER2)-targeted therapies, whereas next-generation sequencing (NGS) has been widely incorporated in routine clinical practices. Here, we investigated the concordance between NGS-based assays and IHC in HER2 testing. Interfering factors leading to discordant results between the assays were also studied. MATERIALS AND METHODS: Over 78,000 solid tumors across various types from two independent cohorts in the United States and Japan were investigated for HER2 DNA copy number, mRNA expression, and protein overexpression by whole-exome sequencing (WES), whole-transcriptome sequencing (WTS), and IHC, respectively. RESULTS: In the US cohort (n = 77,267), HER2 DNA amplification, mRNA overexpression, and IHC-positive (IHC-P) were detected in 4.9%, 10.1%, and 4.7% of the tumors, respectively. Positive results in at least one of the three assays were observed in 10.7% of tumors, while 3.9% were positive for all three assays. Using IHC as a comparator, WTS showed better sensitivity than WES but a lower positive predictive value. These results were consistent in the Japanese cohort (n = 1,225). Although the overall HER2 RNA expression level correlated well with IHC score and DNA copy number, the degree of correlation varied among tumor types. Heterogeneous distribution of IHC-P tumor cells was associated with discordant results between NGS-based assays and IHC. CONCLUSION: HER2-positive status in protein, mRNA, and DNA showed concordance in general but varied among tumor types. NGS-based assays, especially WTS, could be a useful predictive tool for HER2 testing in tumor-agnostic settings. Intratumor heterogeneity in HER2 protein expression should be considered when bringing bulk sequencing tests into clinical settings.

DESTINY-CRC01 (NCT03384940) was a multicentre, open-label, phase 2 study that investigated the safety and efficacy of trastuzumab deruxtecan (T-DXd) in patients with human epidermal growth factor receptor 2 (HER2)-expressing metastatic colorectal cancer (CRC). The present exploratory biomarker analysis aims to investigate relationships between biomarkers and clinical outcomes in patients with HER2-positive (immunohistochemistry [IHC] 3+ or IHC 2+ and in situ hybridization [ISH] positive) Cohort A (N = 53) of DESTINY-CRC01. Higher levels of HER2 biomarkers in baseline tissue and liquid biopsies, including HER2 status (IHC/ISH), HER2/CEP17 ratio, HER2 ISH signals, HER2 H-score, plasma HER2 (ERBB2) amplification status, HER2 adjusted plasma copy number, and HER2 extracellular domain correlate with antitumor activity (indicated by objective response rate, progression-free survival, and overall survival) of T-DXd. Baseline circulating tumor DNA (ctDNA) analysis suggests antitumor activity of T-DXd in patients who had baseline activating RAS, PIK3CA, or HER2 mutations detected in ctDNA.

Although surgical resection is the standard therapy for stage II/III colorectal cancer, recurrence rates exceed 30%. Circulating tumor DNA (ctNDA) detects molecular residual disease (MRD), but lacks spatial and tumor microenvironment information. Here, we develop a deep learning (DL) model to predict disease-free survival from hematoxylin & eosin stained whole slide images in stage II-IV colorectal cancer. The model is trained on the DACHS cohort (n = 1766) and validated on the GALAXY cohort (n = 1404). In GALAXY, the DL model categorizes 304 patients as DL high-risk and 1100 as low-risk (HR 2.31; p < 0.005). Combining DL scores with MRD status improves prognostic stratification in both MRD-positive (HR 1.58; p < 0.005) and MRD-negative groups (HR 2.1; p < 0.005). Notably, MRD-negative patients predicted as DL high-risk benefit from adjuvant chemotherapy (HR 0.49; p = 0.01) vs. DL low-risk (HR = 0.92; p = 0.64). Combining ctDNA with DL-based histology analysis significantly improves risk stratification, with the potential to improve follow-up and personalized adjuvant therapy decisions.

We propose a method, HotSpot, for optimizing neural signed distance functions. Existing losses, such as the eikonal loss, act as necessary but insufficient constraints and cannot guarantee that the recovered implicit function represents a true distance function, even if the output minimizes these losses almost everywhere. Furthermore, the eikonal loss suffers from stability issues in optimization. Finally, in conventional methods, regularization losses that penalize surface area distort the reconstructed signed distance function. We address these challenges by designing a loss function using the solution of a screened Poisson equation. Our loss, when minimized, provides an asymptotically sufficient condition to ensure the output converges to a true distance function. Our loss also leads to stable optimization and naturally penalizes large surface areas. We present theoretical analysis and experiments on both challenging 2D and 3D datasets and show that our method provides better surface reconstruction and a more accurate distance approximation.

No abstract available