Diverse defects in HR DNA repair genes, such as germline mutations in BRCA1, BRCA2, and PALB2, somatic mutations in BRCA1 and BRCA2, and promoter methylation of BRCA1, have been reported in breast and ovarian cancers (35, 36). BRCA mutations are also associated with an increased risk for pancreatic cancer, and 4% to 7% of patients with pancreatic cancer have a germline BRCA mutation (8). BRCA genes encode for proteins involved in the HR repair of DNA double-stranded breaks. Cells with deficient HR repair are sensitive to PARP inhibition. PARP enzymes are key components in the repair of DNA single-stranded breaks and replication fork damage (37). PARP inhibition causes accumulation of such lesions through catalytic inhibition and trapping of PARP on DNA at the sites of single-stranded breaks. These processes eventually result in double-stranded breaks, which cannot be accurately repaired in tumors with HRD. Thus, PARP inhibitors cause accumulation of DNA damage and tumor-cell death. Accordingly, PARP inhibitors are selectively effective for cells with HRD due to BRCA1 or BRCA2 mutations (38, 39).

Recent investigations of genomic profiling in large cohorts of PDAC have reported the significance of HRD in predicting sensitivity to platinum-based therapy and PARP inhibitors (3, 4, 6, 7). According to ASCO guidelines, treatment with platinum-based chemotherapy or the PARP inhibitor olaparib is recommended for patients who have a germline BRCA1 or BRCA2 mutation. A recent randomized phase III trial (POLO) demonstrated the efficacy of olaparib, a PARP inhibitor, in germline BRCA-mutated metastatic PDAC (8). Among the 154 enrolled patients, progression-free survival (PFS) was significantly longer in the olaparib group than the placebo group (7.4 vs. 3.8 months). Furthermore, another recent randomized phase II trial showed that patients with germline BRCA1/2- or PALB2-mutated PDAC benefit from first-line platinum chemotherapy, with median OS and PFS of 15.5 (14.6–19) and 7 (6.1–8.1) months, respectively. Patients with HRD had improved PFS compared with no HRD when treated with first-line platinum therapy but not with first-line non-platinum therapy (9). These results suggest that HRD can effectively be targeted in PDAC.

In addition to mutations in canonical HR genes, a comprehensive evaluation of HR gene mutations is needed beyond germline BRCA mutations to understand their sensitivity to DDR-targeted therapies, including platinum-based therapy. Multiple groups have identified a broader group of patients with HRD sensitive to DDR-targeted therapies (3, 4, 7). The concept of “BRCAness” was introduced to describe the clinical and biological features in some sporadic tumors shared with tumors harboring germline BRCA1/2 mutations. Polak et al. (40) explored signature 3, a mutational signature prevalent in tumors with BRCAness, and found altered expression of PALB2 and RAD51, which are genes that are important in the HRR pathway. In addition, signature 3 has been found in tumors with both germline and somatic BRCA1/2 mutations (40). Thus, signature 3 can be considered a potential biomarker that could lead to BRCAness-targeting therapies. O’Reilly and colleagues (7) evaluated the mutational status of HR genes and HRD genetic signatures to determine their benefit to platinum therapy. They observed that patients with HRD had significantly improved PFS when treated with first-line platinum-based therapy compared with those who received first-line non-platinum-based therapy. Subgroup analyses suggest that patients with either pathogenic somatic or germline BRCA1, BRCA2, or PALB2 mutations, as well as biallelic loss of other rarer HR genes, such as ATM and CHEK2, could be recommended for platinum-based therapy. Aguirre and colleagues (4) observed four samples that did not have clear DNA changes or mRNA downregulation of BRCA1, BRCA2, PALB2, or RAD51C but nevertheless had enrichment of HRD/signature 3 compared with samples with HRD. These data suggest that signature 3 can be recommended for platinum-based therapy and PARP inhibitors. Furthermore, patients with biallelic HRD show higher tumor mutation burden (TMB), indicating the potential benefit from immunotherapy as shown in other types of cancer (41, 42).

Immune checkpoint inhibitors have been an effective therapy for MMR-D/MSI-H cancers regardless of tumor type, although activity may vary by tumor type. MMR-D occurs as a consequence of loss-of-function changes in MMR genes (MLH1, PMS2, MSH2, MSH6) because of the inherited germline mutations known as Lynch syndrome, or because of the biallelic somatic inactivation of MMR genes. In PDAC, approximately 1% of patients have MMR-D or MSI-H due to Lynch syndrome or somatic MMR gene mutations (43, 44). The normal MMR system can correct the process of DNA replication errors but MMR-D results in an inappropriate response to DNA mismatches, increasing the possibility of gene mutation. MMR-D causes MSI-H by missing or inserting one or more of the repeating units in the inappropriate process of DNA replication and repair. MMR-D and MSI-H are generally associated with high TMB. A high TMB increases the potential number of neoantigens, and these neoantigens can be presented by the tumor cell and recognized by host immune cells, which are also known as tumor-infiltrating lymphocytes (TILs) that migrate into TME (45, 46). TILs, particularly CD8+ cytotoxic T cells, orchestrate a significant antitumor response to eliminate tumor cells (45, 46). Detection of MMR-D and/or MSI-H was proposed as a biomarker of an immunogenic tumor and response to ICB therapy, such as anti-programmed cell death protein 1 (PD-1) inhibition. The immune checkpoint inhibitor pembrolizumab is currently approved for treatment of MMR-D/MSI-H cancer regardless of the histology. In ASCO guidelines, pembrolizumab is recommended as a second-line therapy for PDAC patients with MMR-D or MSI-H (34).

The recently published KEYNOTE 158 study (10) and NCI-MATCH study (5), which investigated the efficacy of ICB therapy in non-colorectal cancers, clearly suggested that MMR-D in solid tumors is a predictor of the response to ICB therapy. However, ICB therapy has low efficacy against pancreatic cancer, suggesting that cancer type-specific responses show variable clinical outcomes, and that disease-specific biological factors may have an independent impact on ICB response, regardless of MMR-D status. Regarding the underlying mechanisms, the degree of T-cell infiltration is critical for predicting the efficacy of ICB therapy in other types of cancers (47–52), and a small subset of patients with MSI-H tumors exhibit T-cell infiltration and sensitivity to immunotherapy (53). In pancreatic cancer, dense stroma with desmoplastic reaction may function as a physical barrier and affect the infiltration of myeloid-derived suppressor cells and T cells in tumor stroma (54, 55). In addition, PDAC exhibits substantial immunological heterogeneity, with tumors influencing T-cell infiltration (45, 56–58). For example, Stanger et al. (46) observed heterogeneity in the degree of T-cell infiltration in their cohort of 12 PDAC patients (none of whom were MSI-H), in line with prior reports for PDAC (45). The abundance of PD-1+CD8+ T cells was more predictive of immunotherapy response than was total CD8+ T cell infiltration alone. Furthermore, Leach and colleagues (59) identified mucin 16 neoantigens as T-cell targets in PDAC and as potential biomarkers of immunogenic tumors that may guide the application of immunotherapies (59). These results suggest that both the quality and quantity of tumor-infiltrating CD8+ T cells are critical for predicting the immunotherapy response, and novel biomarkers are needed to predict the status of tumor-infiltrating CD8+ T cells.

Furthermore, the abovementioned results suggest that MMR-D status is not a perfect predictor of immunotherapy response. Loss-of-function changes in MMR genes can sometimes be a passenger mutation/change and responses to immune checkpoint inhibitor therapy could be affected by founder mutations that determine the molecular behavior of cancer (60). MSI-H and high TMB may be a better predictor of the immune checkpoint inhibitor response, as these markers are highly associated with MMR-D-driven carcinogenesis (61). The basis for these differences in T-cell infiltration is poorly understood in PDAC, where most tumors share the same oncogenic mutations. Further prospective studies are needed to evaluate the predictor role of characterization of T-cell infiltration on ICB response/resistance in cancer patients with MMR-D tumors.

KRAS mutation is a major driver mutation in pancreatic cancer and more than 90% of pancreatic cancer patients harbor KRAS mutation. Recently drugs targeting KRAS^G12C are available and clinical trials suggested promising results (62–64). However, KRAS^G12D and KRAS^G12V mutations are more common in pancreatic cancer, and these mutations are still undruggable. In KRAS^WT PDAC patients, NTRK fusions, ALK rearrangements, ROS, NRG1 rearrangements, BRAF, PIK3CA, and a number of cancer-associated genes representing potential drivers have been identified (e.g., ERBB2, STK11, GNAS, CHEK2, and RB1), as potential targets.

Gene fusions involving NTRK1, NTRK2, or NTRK3 (TRK fusions) are found in many pediatric and adult malignancies (65). NTRK fusions are rare in PDAC and are identified in less than 1% of tumors (66). Pishvaian et al. (67) reported a partial response to entrectinib, a potent TRK and ROS1 inhibitor in a subgroup of advanced PDAC patients. Larotrectinib, a highly selective small-molecule inhibitor of the TRK kinases, has shown efficacy in preclinical models and in patients with tumors harboring TRK fusions. In ASCO guidelines, in patients with tumors harboring NTRK fusions, treatment with larotrectinib or entrectinib is recommended as treatment options after first-line therapy such as FOLFIRINOX and gemcitabine plus nab-paclitaxel (GnP) (34). NRG1 rearrangement contributes to susceptibility to ERBB inhibitors and anti-EGFR antibodies and clinical trials are ongoing (68, 69). ALK gene translocations have been reported in 0.16% of PDAC, and crizotinib is reportedly effective for a PDAC patient with ALK gene translocation (70).

BRAF^V600E mutations occurred at a frequency of 3% and were mutually exclusive with KRAS mutations. BRAF^V600E could be a driver event based on mouse models (71). Analyses of PDAC cells revealed that BRAF^V600E cells are sensitive to the FDA-approved BRAF inhibitor PLX-4032, while cells with KRAS mutations are resistant (33). These data suggest that a subset of patients may benefit from targeted therapy along the KRAS/BRAF axis. Aguirre et al. (4) reported the first therapeutic experience with mitogen-activated protein kinase (MAPK) inhibition in a patient harboring a BRAF in-frame deletion. The patient had a partial response to the MEK inhibitor trametinib. A second patient with rapidly progressive disease harboring a BRAF mutation was also treated with trametinib but failed to show a response. This heterogeneity in resistance mechanisms will require effective combination treatments with MAPK inhibitors. The Cancer Genome Atlas Research Network reported that the KRAS^WT tumors had significantly elevated tuberous sclerosis complex/mammalian target of rapamycin (TSC/mTOR) signaling pathway activity compared with KRAS mutant tumors, indicating that functional activation of the mTOR signaling pathway may be an alternative oncogenic driver in KRAS^WT pancreatic cancer (15).

These data suggest that larger multicenter clinical trials are needed to fully investigate the therapeutic efficacy of the inhibition of upstream and downstream signaling of RAS in KRAS^WT patients with other oncogenic mutations.

Although genome-based precision medicine, such as platinum-based therapy and PARP inhibition, in PDAC patients with HRD is promising, tissue-based genomic sequencing for first-line treatment decision making in PDAC remains challenging due to the turn-around time of obtaining sequencing results, which is 3 to 6 weeks. To enroll patients in genome-based precision medicine, recent reports have suggested that diagnosis by liquid biopsy with a short turn-around time has emerged. Liquid biopsy includes analyses of tumor materials obtained in a minimally invasive or noninvasive manner by collecting blood or other body fluids. Liquid biopsy samples are obtained from saliva, stool, or urine. They include circulating tumor cells (CTCs), circulating tumor DNA (ctDNA), extracellular vesicles, cell-free DNA, and microRNA. More recently, next-generation sequencing-based methods have enabled ctDNA profiling as an alternative to tumor tissue sequencing (72, 73). For example, the TARGET study recently reported the screening of 100 patients using ctDNA sequencing for trial enrollment (74). Most recently, Yoshino and colleagues (11) reported that ctDNA genotyping significantly shortened the screening duration (11 vs. 33 days; P < 0.0001) and improved the trial enrollment rate (9.5 vs. 4.1%; P < 0.0001) compared with tumor tissue sequencing. Overall, ctDNA was detected in 91.4% (1,438/1,573) of patients; however, the ctDNA detection rate of PDAC was the lowest (83.4% (304/363) compared with other types of cancers, such as esophageal squamous-cell carcinoma cancer (99.1% (107/108) and CRC (96.0%, 628/654). Overall, they detected multiple biomarkers relevant to the selection of treatment, including KRAS, NRAS, BRAF, and PIK3CA mutations; ERBB2, FGFR1–2, and MET amplifications; FGFR2–3, ALK, NTRK1, and RET fusions; and MSI. Liquid biopsy also enables the collection of repeated samples during the course of the treatment of patients and the collection of clones resistant to ongoing therapy. Thus, this technology has the potential to promote innovation in precision medicine.

Targeting drugs, according to tumor subtypes, have improved treatment outcomes in other cancers. Identification of therapeutic molecular subtypes in PDAC has been challenging. In addition to genomic subtypes, transcriptomic subtypes have been evaluated to understand the biology of pancreatic cancer. The recent identification of PDAC transcriptional subtypes has provided a new perspective relevant to the development of therapies. These include basal-like/squamous and classical/progenitor (hereafter referred to as basal-like and classical, respectively) (12–14, 75–77). Basal-like tumors are associated with poor outcomes and treatment resistance (12–14, 16–19, 75–77). Two independent clinical trials revealed that basal-like tumors are resistant to FOLFIRINOX-based regimens (19, 78). In those studies, RNA in situ hybridization or immunohistochemical analysis of GATA-binding protein 6 (GATA6) was used to differentiate basal-like and classical tumors. The resistance of basal-like tumors to FOLFIRINOX is supported by a recent report by Tiriac et al. (24), who showed that patient-derived organoid (PDO) chemotherapy signatures may predict treatment response. The signatures representing individual cytotoxic agents were applied to the COMPASS cohort, suggesting that basal-like tumors are most likely to have a non–oxaliplatin-sensitive signature (24). To apply molecular subtyping in treatment decision-making for PDAC patients, Rashid et al. (79) revealed that the tumor-intrinsic two-subtype schema of Moffitt et al. is the most replicable, and they developed the Purity Independent Subtyping of Tumors, a clinically usable single-sample classifier based on gene expression data obtained using multiple platforms, including microarrays, RNA sequencing, and NanoString.

The development of subtype-based therapies remains challenging because the genetic and epigenetic aberrations that promote the stable or dynamic regulation of subtypes are unknown. The basal-like subtype consists of small subgroups that are regulated by different mechanisms (14). The master regulators of the basal-like subtype have been identified, and the basal-like subtype is associated with the activation of genes involved in the epithelial–mesenchymal transition, activation of transcription factors such as MYC and TP63, and downregulation of markers for endoderm such as HNF4A and GATA6 (12–14). In addition, expression of the ΔN isoform of TP63 (ΔNp63) and GLI2 promotes the basal-like identity in PDAC (80, 81). Several epigenetic regulator genes, including KDM6A, KMT2C, and KMT2D, are associated with the basal-like subtype (12, 14, 82). Mueller et al. (83) divided the basal-like subtype into the TP63-related transcriptional program with squamous differentiation and the RAS/epithelial–mesenchymal transition-related transcriptional program with undifferentiated cancers. Further characterization of the master regulators of molecular subtypes may lead to the identification of biomarkers and targets for tailored therapies.

Because KRAS is the most commonly mutated gene in PDAC, the association between KRAS addiction (KRAS dependency) and a molecular subtype has been debated. KRAS-addicted cells have been previously observed as more classical and epithelial in monolayer cell cultures (12). Collisson et al. (12) showed that classical PDAC cells are relatively more dependent on KRAS and more sensitive to erlotinib than basal-like PDAC cells. Conversely, basal-like PDAC lines are more sensitive to gemcitabine than classical PDAC (12). KRAS ablation induces a basal-like phenotype in surviving cells in vivo (80). A study of the inducible KrasG12D;Trp53-/- PDAC mouse model (84) revealed cancer cell-intrinsic mechanisms enabling bypass of KRAS dependency and tumor recurrence (85). Specifically, Yap1 amplification and overexpression enabled escape in approximately one-third of KRAS-negative recurrent PDAC tumors (85) and serves a similar role in lung cancer (86). However, allelic imbalance and elevated expression of mutant KRAS have been associated with aggressive and undifferentiated histological phenotypes in PDAC (32, 83). Furthermore, increased dosage of mutant KRAS is sufficient to induce basal-like features (32, 87). These results suggest that mutant KRAS plays an important role in oncogenesis in PDAC, but other epigenetic or microenvironmental factors are critical in regulating molecular phenotypes.

Pancreatic cancer is characterized histologically by a dense stromal reaction with desmoplasia, which creates a physical barrier around the tumor cells and prevents appropriate vascularization and delivery of chemotherapeutic agents (88). The surrounding desmoplasia was formerly considered to promote cancer, and a number of clinical trials targeting the stroma have been conducted to prove this. However, most of those trials failed, and the current understanding is that the stroma is multi-faceted (89–91). To reveal the heterogeneity of stromal components, studies based on single-cell RNA sequencing have been conducted (92–95). Cancer-associated fibroblasts (CAFs) play an important role in the TME, and cancer-derived IL-1 or TGF-β can stimulate the differentiation of surrounding fibroblasts into inflammatory and myofibroblastic CAFs, respectively (93). IL-6 secreted by inflammatory CAFs promote proliferation of the tumor, whereas myofibroblastic CAFs produce the surrounding stroma. Because cancer cells create an environment favorable to themselves, these stromal subtypes are linked to the cancer cell subtypes mentioned above. Mauer et al. (77) reported CAF subtypes using laser capture microdissection and RNA sequencing of pathologically verified PDAC epithelia and their adjacent stroma. The authors detected two subtypes reflecting ECM deposition and remodeling (ECM-rich) versus immune-related processes (immune-rich). There was a strong association between ECM-rich stroma and basal-like tumors, whereas immune-rich stroma occurred more often in association with classical tumors (77, 96). As such, the epithelial and stromal subtypes were partially linked, suggesting potential biomarkers for stroma-targeted therapies in PDAC.

As mentioned previously, TILs are associated with the response to immune checkpoint inhibitors. Therapeutic strategies targeting immune modulators have emerged. Bailey et al. (14) identified an immunogenic cancer subtype, which shares many of the characteristics of classical tumors but is uniquely associated with significant immune-cell infiltration. This cancer cell subtype, as well as Mauer et al.’s immune-rich subtype, has potential as a biomarker for immune therapy. Furthermore, studies using mouse models have revealed potential targets such as colony-stimulating factor 1 receptor, cytotoxic T-lymphocyte-associated protein 4 (97, 98), and CXC chemokine receptor 2 (99, 100), which led to clinical trials. However, due to their complexity, TME-targeted therapies have largely failed (101, 102). Clinical trials of such therapies have been reviewed recently (103). Further investigations are warranted to discover effective TME-targeted therapies.