Introduction
Cancers of the digestive system represent a major fraction of the global tumor burden [
1,
2]. Despite their high incidence, the therapeutic options for most advanced and metastatic cancers of the digestive system are still limited. Molecular profiling approaches, in particular next-generation tumor genome sequencing, hold the promise of identifying alterations that can be exploited for cancer therapy. This concept of personalized oncology is supported by several large prospective clinical trials showing that actionable mutations can be detected in in 36.7–58.2% of patients with solid cancers [
3,
4]. Furthermore, results of these studies indicate that genomics-driven cancer therapy can improve overall survival and reduce toxicity-related mortality [
5]. Hence, precision oncology approaches have been integrated into routine oncological care in many academic medical centers. Retrospective real-world data from these centers suggest that actionable mutations can be detected in 20–40% of cancer patients [
6,
7], matching findings from prospective studies [
3,
4]. For cancers of the digestive systems, studies have reported that actionable genetic alterations can be found in 5.8 to 27.8% of cases, with highest rates in cholangiocellular and gastroesophageal cancers, and lowest rates in pancreatic cancer [
8]. Different methods for high-throughput DNA profiling have been applied for precision oncology, ranging from whole genome/exome sequencing to more focused approaches such as panel gene sequencing. While mutational profiling by whole genome/exome sequencing is comprehensive, it is also associated with higher costs and requires more extensive bioinformatic resources for analysis compared to more focused sequencing approaches [
9]. In contrast, gene panel sequencing is versatile and less costly, but allows only the assessment of hotspot mutational sites in predefined sets of genes [
9]. In clinical practice, translation of cancer genomics into novel therapeutic options for patients is frequently hindered by a lack of access to suitable clinical trials, delayed coverage of therapy cost by insurance companies or rapid disease progression [
10]. Thus, many studies indicate that only a minor fraction of patients with actionable genetic alteration benefit from alteration-specific therapies [
6,
11]. While the real-life clinical utility of panel gene sequencing has been evaluated across a broad spectrum of solid tumors [
12,
13], few studies focused on cancers of the digestive system.
In this retrospective study, we determined the results of panel gene sequencing in 155 patients with locally advanced or metastatic cancers of the digestive system who were treated at a tertiary academic medical center. We describe the spectrum of identified mutations, subsequent mutation-specific therapeutic approaches, and clinical courses of the patients receiving tailored therapies.
Discussion
This retrospective, real-world analysis of 155 patients with advanced cancers of the digestive system shows that gene panel sequencing can uncover actionable genetic alterations that result in individualized therapeutic options.
In our patient cohort, actionable molecular alterations were observed in 13.5–56.8% of cases, using the ESMO treatment guidelines or OncoKB database, respectively. When using OncoKB as reference, our number of detected actionable alterations is comparable to published, prospective studies that used panel gene sequencing and reported frequencies between 36.7 and 58.2% [
3,
4]. However, significantly fewer actionable alterations were detected when the ESMO treatment guidelines were applied.
In general, our study shows that the frequency of identified actionable alterations was dependent on two main factors: the genomic coverage of the assay type and the classifier used. In 29 out of 32 cases in which no mutations could be detected, a focused panel gene sequencing assay was used. Similarly, the percentage of tumors with actionable mutations, as classified by the OncoKB database, was significantly lower when the Oncomine Focused assay was used (Fig.
2B). These findings indicate that to maximize the clinical utility of gene panel sequencing, comprehensive gene panels are needed.
Secondly, we observed that the number of actionable alterations significantly differed between the OncoKB database and ESMO treatment guidelines. We found a major overlap between the two resources for actionable alterations with high levels of clinical evidence, such dMMR or
BRAF V600E. However, the OncoKB database also includes many actionable genetic alterations with low levels of clinical evidence which were not evaluated by the ESMO guidelines and thus not assigned an ESMO-ESCAT score. This difference in classification is most apparent for alterations in
KRAS, which was the second most frequently mutated gene in our cohort. While OncoKB suggested MEK inhibitors as a treatment for
KRAS mutant cancers, most early clinical trials did not find any clinical benefit of MEK inhibitor monotherapy in gastrointestinal cancers [
24]. In contrast, only
KRAS G12C was classified as an actionable target by the ESMO guidelines, which is supported by several clinical trials [
25‐
27]. Overall,
KRAS G12C mutations are rare in cancers of the digestive tract and we only detected one case in a patient with pancreatic cancer. For this patient, sotorasib was recommended by the molecular tumor board, but treatment with olaparib was initiated due to concurrent
BRCA1/2 mutations. However, as other genotype-specific or pan-KRAS inhibitors are currently undergoing (pre-)clinical testing [
28], the importance of
KRAS mutations as an actionable target will steadily increase.
In general, it is challenging to compare the frequencies of detected actionable alterations between panel gene sequencing studies, as the actionability can be assessed by many different approaches. Besides the OncoKB database, several other public databases (e.g. Clinical Interpretation of Variants in Cancer [
29], Cancer Genome Interpreter [
30]), and commercial resources (e.g. Jackson Laboratory Clinical Knowledge Base [
31]) exist that assess actionability of genetic alterations. Depending on the evidencelevels that are defined as thresholds by these databases and the depth of literature research, the actionability of genetic alterations might be assessed differently. Furthermore, in many past studies, treatment selection was guided by the recommendations of the local molecular tumor boards or clinical experts [
32,
33], providing another layer of heterogeneity. To reduce this heterogeneity of recommendations, an important step would be to establish publicly accessible, transparent databases that collect clinical and pre-clinical evidence for molecular-directed therapies. These databases would enable a stronger unification of treatment recommendations, especially for druggable mutations that occur at low frequencies and for which clinical evidence is rare. An ongoing data curation would be necessary as well as regular validation of recommendations by expert panels.
Lastly, the frequency of actionable alterations also depends on the composition of tumor entities in the respective cohorts. Several recent retrospective studies have shown that panel sequencing can uncover actionable mutations across different solid tumors at high rates [
12,
13]. However, cancers of the digestive system only account for a small fraction in these studies and the frequency of actionable alterations is less clear in this subgroup. The large proportion of CRC and CCC in our cohort might introduce a bias towards cancer types of the digestive system that have a comparatively high frequency of actionable mutations.
Due to its retrospective design, our study has some limitations. First, we included three panel gene sequencing assay types which differed in their coverage of genetic alterations. We also observed differences in the usage of these assays depending on the tumor entity. The selection of the assay type was not based on predefined algorithms but rather on individual clinical decisions. Thus, comparing frequencies of specific mutations between tumor types must be regarded with care, as the data might be biased by differential selection of panel sequencing assays. To overcome this problem in clinical practice, guidelines that clearly define the optimal time point and assay type for specific tumor entities should be implemented. Secondly, performance of panel gene sequencing assays depends on the quality of the tissue material [
34]. In some cases, FFPE tissue from primary cancers were used that have been stored over longer time periods. This might affect the quality of DNA and specifically of RNA, leading to lower sensitivity for specific alterations such as gene fusions [
35]. Lastly, our data indicates that those patients who received a genomics-directed therapy had an overall high clinical benefit. However, this observation is biased by the large proportion of patients with dMMR CRC who received immune checkpoint inhibitors which are highly effective in this tumor subtype [
36]. In contrast, the clinical benefit of ivosidenib in CCC (median PFS 2.4 months in two patients) or cetuximab/encorafenib in CRC (median PFS 3.4 months in 4 four patients) was less pronounced, but matching results of the respective trials (median PFS of 2.7 months for ivosidenib [
37]and median PFS 4.2 for cetuximab/encorafenib [
38].
Despite these limitations, our study shows that selected patients can benefit greatly from molecular-directed therapies. We report two cases of mutations with low evidence for actionability based on the OncoKB database (
BRCA2 S497L and
ERBB3 G284R), but for which our local molecular tumor board suggested therapeutic options that led to sustained tumor responses. Based on the results of the POLO trial, patients with pancreatic carcinoma with germline loss-of-function
BRCA2 mutations benefit from olaparib maintenance therapy following disease control with first-line oxaliplatin-based chemotherapy regimen [
39,
40]. While the trial only included deleterious and likely deleterious germline mutations, the clinical impact of non-synonymous germline mutations such as
BRCA2 S497L remain unknown [
41]. Here, we report the case of a patient with metastatic pancreatic carcinoma with a combination of germline
BRCA1 E1161Ffs*3 and
BRCA2 S497L mutations that benefit from treatment with olaparib monotherapy after failure of several lines of preceding systemic therapies. Missense mutations of
BRCA2 can moderately increase the risk for breast cancer [
42], but it is unknown if they also modulate the response to PARP inhibitors. Considering the durable response that we observed in our patient case, further preclinical studies should be performed to decipher the actionability of
BRCA2 missense mutations.
ERBB3 mutations can occur in various cancer entities, but are overall rare events [
43]. Mutations of
ERBB3 can induce activation of the MAPK pathway, and a previous case report showed that targeting
ERBB3 G284R mutant breast cancer with trastuzumab plus lapatinib can lead to disease control [
44]. The
ERBB3 G284R mutation is considered an activating alteration [
45], and to our knowledge, our case shows for the first time that treatment with trastuzumab/lapatinib can result in prolonged disease control in metastatic CCC with this specific
ERBB3 mutation.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.