A Comprehensive Tumor Molecular Prole Analysis in Clinical Practice: A Single Center’s Experience

Background: Tumor molecular prole is of great importance for the detection of biomarkers of response to targeted treatment due to the increased availability, with concomitant reduction of cost, of Next Generation Sequencing technology (NGS). In parallel to targeted therapies’, immunotherapies are also evolving, revolutionizing cancer therapy, with Programmed Death-ligand 1 (PD-L1), Microsatellite Instability (MSI), and Tumor Mutational Burden (TMB) analysis being the biomarkers employed most commonly. Methods: In the present study, a 161 gene NGS panel, containing the majority of clinically signicant genes for cancer treatment selection, was used for tumor molecular prole analysis. A variety of tumor types have been analyzed, including aggressive and hard to treat cancers such as pancreatic cancer. Besides, the clinical utility of immunotherapy biomarkers (TMB, MSI, PD-L1), was also studied. Results: Molecular prole analysis was conducted in 610 cancer patients, while in 393 of them a at least one biomarker for immunotherapy response was requested. At least one actionable alteration was detected in 77.70% of the patients. 54.59% of them received information related to on-label or off-label treatment (Tiers 1A.1, 1A.2, 2B, and 2C.1) and 21.31% received a variant that could be used for clinical trial inclusion. The addition to immunotherapy biomarker to targeted biomarkers’ analysis in 191 cases increased the number of patients with an on-label treatment recommendation by 22.40%, while an option for on-label or off-label treatment was provided in 71.35% of the cases. Conclusions: Tumor molecular prole analysis by NGS is a rst-tier methodology for a variety of tumor types, which provides critical information for treatment decision making in cancer patients. Importantly, simultaneous analysis for targeted therapy and immunotherapy biomarkers could lead to a better tumor characterization and provide actionable information in the majority of patients. Moreover, our results indicate that one in two patients is eligible for ICI treatment based on the biomarkers’ analysis. However, when these analyses are performed, the challenge is their


Background
In recent years, technological advances and active research have permitted extensive tumor molecular characterization and have revealed a variety of tumorigenic pathways presenting tumor-speci c alterations. These distinctive molecular characteristics of cancer cells can be targeted as they represent the malignant cell's Achille's heel, without affecting the healthy ones. To this regard, of great importance was the previous knowledge gained by large scale studies that used various, advanced technologies to obtain a comprehensive understanding of the tumor molecular pro le (1).
Tumor molecular pro le is nowadays becoming a reality mainly due to the increased availability, with concomitant reduction of cost of the Next Generation Sequencing technology (NGS) method. The term personalized medicine in anticancer treatment has emerged, indicating the need to treat each patient based on his/her tumor's speci c characteristics (2). The individualization of treatment strategy entails the use of biomarkers that are those quanti able characteristics that can be related to cancer prognosis and prediction of treatment response (2)(3)(4).
The number of laboratories applying high throughput sequencing analysis is continuously increasing, in parallel with the increased request by the clinicians for such analysis. The frequently insu cient in amount of good quality tissue specimen, coupled with the increasing number of approved targeted agents, make the simultaneous analysis of multiple biomarkers using multigene panels imperative. Thus, advanced technology solves one of the most signi cant limitations of tissue testing. Of note, optimal para n embedding procedure remains crucial for obtaining accurate NGS results (2,3).
Currently, in parallel to targeted therapies, an increasing armamentarium of immunotherapy agents also emerging, revolutionizing cancer therapy. The high cost and toxicity that often accompanies immunotherapeutic agents mandate the use of appropriate biomarkers for selecting patients more likely to bene t from them.
The most widely used biomarker is currently PD-L1 expression, assessed by Immunohistochemistry (IHC) (7). However, it is well known that this is not an ideal biomarker since it is not related to treatment response in many tumor types, while it is clearly not the sole predictor of response to check point inhibition. Moreover, even for those tumors with a proven utility for PDL-1 IHC testing, such as lung cancer, several questions regarding methodology and cut offs remain (7,8).
Additionally, microsatellite instability (MSI) has also been associated with response to anti-PD-L1 treatment with pembrolizumab receiving approval for MSI-H tumors (9)(10)(11). Of note, MSI was the rst tumor agnostic biomarker that had ever shown e ciency regardless of tumor type. However, the presence of MSI varies among tumor types with the rate of MSI-H tumors ranging from 10-15% for colon cancer to 0% in others such as lung cancer (12). Thus, still, the majority of responders will not be identi ed by it.
Hence, the enrichment of biomarkers for the identi cation of patients eligible for immunotherapy administration is required.
Several additional biomarkers of immune response have been proposed and are currently under investigation while it seems that their combined use could increase the predictive value of the information obtained (13,14). Among the most studied ones is the analysis of Tumor Mutational Burden (TMB) that measures the number of somatic mutations present in a tumor sample. It has been shown in several studies and clinical trials that the greater the number of somatic alterations identi ed the greater the probability of response to immune treatment (15)(16)(17). It has been reported that TMB cutoff values associated with improved survival from immunotherapy treatment vary signi cantly between cancer types (17). Nevertheless, in the majority of studies and clinical trials, a cut off of 10 muts/MB is used (18)(19)(20).
Furthermore, the clinical utility of TMB as a predictive biomarker for anti-PD1 treatment administration has been shown in the KEYNOTE 158 study leading to the tumor agnostic approval by the USA FDA of pembrolizumab for metastatic untreatable solid tumors with tissue TMB value of ≥ 10 muts/MB (21).
The present study aimed to reveal the applicability and utility of tumor pro le analysis in clinical practice, using a pan-cancer NGS panel for cancer treatment selection. The panel used in this study analyses 161 single genes using the Oncomine Technology (Thermo Fischer Scienti c) and was selected based on the amount of actionable information contained the robustness of the assay and its relatively low cost which enables its use in clinical practice. A variety of tumor types have been analyzed, including aggressive and hard to treat cancers such as pancreatic cancer. In addition, the clinical utility of immunotherapy biomarkers (TMB, MSI, PD-L1) was also explored.

Patients
In the present study, 629 cancer patients were referred by their treating oncologist for extensive molecular pro le analysis from November 2017 to April 2020. Informed consent was obtained from all patients participating in the study. Information concerning sex, age, and tumor histology was obtained, while the pathology report was available in all cases. In addition to molecular analysis for targeted treatment selection, analysis for at least one immunotherapy biomarker (PDL-1, MSI, TMB) was also requested in 395 patients. The analysis was performed using the most recent tissue specimen available.

Tissue selection and nucleic acid isolation
Genomic DNA and RNA were isolated from formalin-xed and para n-embedded (FFPE) tumor biopsies using the MagMAX™ Total Nucleic Acid Isolation Kit (Thermo Fischer Scienti c) according to the Manufacturer's instructions. The nucleic acid isolation was conducted in the areas of the FFPE block with the majority of tumor cell content (TCC), as indicated by experienced pathologists in Hematoxylin and eosin-stained sections. Minimum required TCC was over 20%, in a tumor area of > 4mm 2 .

Next Generation Sequencing
Whenever tumor molecular pro le analysis for targeted therapies was requested, Oncomine Comprehensive Assay v3 (OCAv3) (Thermo Fischer Scienti c) was performed, which is an ampliconbased targeted NGS assay. This assay allows the identi cation of various mutation types such as Single nucleotide Variants (SNVs), insertion-deletions (indels), Copy Number Variations (CNVs), and gene fusions, from 161 unique genes. Run metrics were accessed in the Torrent Suite™ software, using the coverage analysis plugin v5.0.4.0. NGS data analysis was completed with the Ion Reporter™ 5.10.1.0 software (Thermo Fisher Scienti c) using the manufacturer's provided work ow (Oncomine Comprehensive v3 -w3.2 -DNA and Fusions -Single Sample). Furthermore, the analysis software Sequence Pilot (version 4.3.0, JSI medical systems, Ettenheim, Germany) was used for variant annotation.
Tumor Mutational Burden analysis was carried out using the Oncomine Tumor Mutation Load Assay (Thermo Fischer Scienti c). The assay used is a targeted NGS assay, with 1.65MB of genomic coverage (1.2MB exonic) that analyzes 409 genes to provide accurate quantitation of somatic mutations used for tumor mutation burden calculation, in FFPE tissues.
TMB was calculated using the Ion reporter pipeline that utilizes a custom variant calling and germline variant ltering to accurately calculate the number of exonic somatic mutations per MB (Oncomine Tumor Mutation Load -w2.0 -DNA -Single Sample).
Microsatellite analysis was conducted using the Ion AmpliSeq™ Microsatellite Instability Panel (Thermo Fischer Scienti c) which is an NGS based assay analyzing 76 markers to assess Microsatellite Instability (MSI) status in tumor-only and tumor-normal samples as indicated by the manufacturer. Analysis of the sequencing output from this panel was carried out using the "MSICall" plugin in the Torrent Suite.

Variant Classi cation
Variants were classi ed according to their predictive value using the four-tiered system jointly recommended by the Association for Molecular Pathology (AMP), the American College of Medical Genetics (ACMG), the American Society of Clinical Oncology (ASCO) and the College of American Pathologists (CAP) for the classi cation of somatic variants (22). The Joint consensus recommendation system proposed by these major scienti c institutions classi es the variants based on their clinical signi cance in 4 tiers 1-4. Tier 1 variants are of the most substantial clinical signi cance and are subdivided to those related to sensitivity or resistance to FDA approved treatments (Tier 1A.1), those proposed by professional guidelines to have predictive value (Tier 1A.2), and those with a strong consensus concerning their predictive signi cance (Tier 1B). The Tier 2 class involves biomarkers with potential clinical relevance. It can be subdivided in variants related to an approved treatment for a different tumor type (Tier 2C.1), variants related to investigational treatments that can be used as an inclusion criterion for patients' enrollment in clinical trials (Tier 2C.1), and variants that have shown predictive value in preclinical studies (Tier 2D). Finally, the 3 and 4 Tiers, include biomarkers of unknown clinical signi cance and the benign or likely benign ones respectively. (22,23).

Gene panel comparison
The clinical utility of the 161 gene panel used in this study was evaluated by its comparison to gene panels comprising a smaller gene number. Thus, we simulated the alterations that would have been obtained in our cohort if the analysis was conducted by two hotspot gene panels previously used in our laboratory, a 24 and a 50 gene panel (24) (Additional le 1). Both panels also included the analysis of 6 fusion driver genes (ALK, ROS1, RET, NTRK1, NTRK2 and NTRK3) analyzed at the RNA level.
Furthermore, in order to investigate if the number of genes analyzed is adequate for implementation in clinical practice, or if by increasing the number of genes tested a more informative result could be obtained, we compared the actionability of the results obtained from this panel to those obtained using a more comprehensive tumor panel that utilizes the same NGS technology. The panel implemented for this evaluation was the Oncomine Comprehensive plus assay (Thermo Fischer Scienti c) that analyses the full coding sequence of 313 genes, hotspot analysis of 169 genes, CNV of 313 genes (most of them also analyzed for SNV and indels). Furthermore, it includes RNA analysis for 51 fusion drivers genes (38 of them also analyzed at the DNA level), adding up to a total of 514 unique genes present in this panel. The intra-panel comparison was performed through a retrospective analysis of genomic data from The Pancancer Analysis of Whole Genomes (PCAWG) study (25).
The web-based Xena Browser was used for visualization and exploration of the data (26,27). More speci cally available data sets from specimens with coding driver alterations information, including single nucleotide variations (SNVs) and small insertions-deletions (indel) and with consensus wholegenome copy number data as well as consensus fusion calls were downloaded and explored. The 990 specimens with information concerning all three types of alterations available were selected. Subsequently, we simulated the results that would have been obtained if this analysis had been performed using the gene sets included in the aforementioned panels and we explored the magnitude of the clinically actionable information obtained in each case. Variant classi cation and biomarker interpretation were performed as described above. For the copy number variation analysis, only the 43 genes of the Oncomine Comprehensive Panel v3 and the 333 of the Oncomine comprehensive plus panel with focal copy number variations were included. In order to resemble the cutoff values used in everyday practice in our laboratory, a threshold of >7 copies was used for considering a sample positive for copy number ampli cation and a threshold of <1 copy for considering a gene loss (28).

PD-L1 expression by immunohistochemistry
For the majority of tumors analyzed (such as lung, colorectal, pancreatic and ovarian cancer) as well as for tumors of unknown primary origin, the level of PD-L1 protein expression was de ned as the percentage of viable tumor cells (TC) showing partial or complete membrane staining at any intensity. Furthermore, in some cases, the percentage of tumor In ltrating Immune Cells (IC) showing staining at any intensity was also calculated (29)(30)(31). In case of bladder, urothelial, and cervical carcinomas, PD-L1 was calculated through the Combined Positive Score (CPS) which is the percentage of positive cells (tumor, lymphocytes, and macrophages) showing partial or complete membrane staining at any intensity (32,33). In case of Head and Neck Squamous Cell Carcinoma, both CPS and TC values were calculated (34). The analysis was conducted using the Immunohistochemistry (IHC) assay VENTANA PD-L1 (SP263) Assay (Roche Diagnostic) that utilizes the Monoclonal Mouse Anti-PD-L1, Clone SP263 accompanied by OptiView DAB IHC Detection Kit on a VENTANA BenchMark Series automated staining instrument.
For breast cancer patients, the VENTANA Monoclonal Mouse Anti-PD-L1, Clone SP142 antibody was used. The level of expression of the PD-L1 protein was de ned as the percentage of tumor-in ltrating Immune Cells showing staining at any intensity (35).

Physicians Survey
In order to investigate the utility of a multi-biomarker analysis in clinical practice and if the results obtained from such approach have an impact in clinical decision making, a questionnaire was given to the referring oncologists, asking whether based on their experience, they consider such analysis useful for patients with the following tumor histological type: Lung, Colorectal, Breast, Ovarian, Prostate, and rare or unknown origin tumors. It was a multichoice survey with the following options of response: a. Useful, b. in the metastatic setting only c. not useful and d. I do not know/not respond.

Statistical analysis
Statistics were performed with SPSS (version 20. IBM SPSS STATISTICS). The p-values were based on Fisher's Exact Test. A p-value <0.05 was considered to be statistically signi cant. Box plots were created using the Plotly.js charting library.

Molecular Analysis for Targeted Therapy
In the present study, 629 tumor tissues were subjected to targeted treatment biomarkers' analysis, using a 161 gene NGS panel. Successful molecular analysis was achieved in 610 of the 629 patients analyzed, while in 19 (3.03%) cases, no results could be obtained due to low DNA/RNA quality or quantity. The tumor types analyzed included common tumor types with targeted treatment available, such as lung, breast and colorectal cancer, but also various hard to treat diseases such as pancreatic, ovarian, prostate, brain cancers, sarcomas, cholangiocarcinomas, and others ( Figure 1).
The mean age of test requisition was 60 years. In total, 936 pathogenic variants in 112 genes were detected in 472 patients (additional le 2). Of those, 85.15% were single nucleotide Variants (SNVs) or a small insertions-deletions (indels) detected at the DNA level, while 3.31% of the variants concerned gene fusions and 11.54% Copy Number Variations (CNVs). 11.54% of the 936 variants identi ed were classi ed as Tier 1, 86.75% of them as Tier 2 and 1.71% as Tier 3 ( Figure 2). At least one variant was detected in 77.38% of the cases. 34.98% of the individuals analyzed carried one genomic alteration, while 23.81% and 19.87% carried two and three or more mutations respectively.
The main reason for multigene test request was the assignment of the appropriate treatment based on patients' molecular pro le. Thus, patients were apportioned based on the clinical signi cance of the alterations detected. In the case of multiple mutations present in the same patient, the variant with the higher level of evidence (LoE) was used for establishing the patient's category. Using this biomarkerde ned categorization, 54.59% of the patients analyzed received information that is related to on-label or off-label treatment (Tiers 1A.1, 1A.2, 1B, and 2C.1). Additionally, the variant detected could be used as a criterion for inclusion in clinical trials (2C.2) or is under investigation in preclinical studies (2D) in 21.48% and 1.80% of the cases respectively. Furthermore, 5.90% of the patients harbored a variant associated with resistance to treatment (1A.1R, 1A.2R) ( Figure 3). As expected, the most frequently mutated gene in this cohort was the gatekeeper TP53 gene, followed by the KRAS and PIK3CA genes (Additional le 3).
These genes were mutated in 36.39%, 24.75% and 10.98% of the patients, respectively ( Figure 4). Furthermore, 7.38% of the patients carried an alteration in a gene involved in the homologous recombination pathway. This type of alterations could be used as predictive biomarkers of response to PARP inhibitors (PARPi) treatment (36,37).

Tissue speci c tumor Molecular pro le
In order to evaluate if molecular pro le analysis is more useful in speci c tumor types compared to others, the mutation frequency and clinical signi cance of the variants detected were calculated for the most common tumor types analyzed in our cohort.

I. Pancreatic Cancer
In the present study, 118 patients undertaking tumor molecular analysis had a diagnosis of pancreatic cancer. KRAS mutation was the prevalent mutated gene in this tumor type, with a mutation frequency of 74.57%. In 64.41% of the patients, an alteration in this gene was the nding with the higher LoE. However, other gene alterations with predictive value (2C.1) coexisted in 10.16% of the KRAS mutant patients. Moreover, in 6 cases (5.08%), the mutation detected was in an HR gene (1 ATM, 2 PALB2, 1 CDK12, 1 FANCA, 1 NBN) with evidence of response to PARPi. Additional variants with associated to off-label treatments were detected in FGFR1 & 4, HER2, MET, PIK3CA and POLE genes ( Figure 5, additional le 4).
Furthermore, 2 patients (1.69%) carried a somatic mutation related to an on-label drug or with strong evidence of actionability. These mutations were detected in genes of the mismatch repair complex (MLH1 and MSH2) and were indicative of microsatellite instability and thus response to immunotherapy.

II. Lung Cancer
In the 67 Lung cancer, patients tested an alteration was detected in 86.57% of the cases ( gure 5). The variant identi ed was related to an FDA approved treatment in 20.89% of the patients. These variants concerned EGFR, BRAF (p.V600) and HER2 mutations in percentages of 8.96%, 4.48% and 1.49%, respectively. Moreover, ALK and RET translocations were detected in 1.49% and 4.48% of the cases, respectively. EGFR TKI resistance-conferring KRAS mutations (Tier 1A.2) were detected in 26.87% of the cases. Apart from these established biomarkers, the expanded gene panel analysis was able to detect additional mutations in multiple other genes with 2C.1 evidence of predictive value in 16.42% of the cases (Additional le 5). Unexpectedly, 6 of the patients (8.95%) carried a mutation in a gene related to PARP inhibitor therapy.

III. Breast Cancer
In the 62 Breast Cancer Patients included in our cohort, a pathogenic variant was found in 80.65% of the cases. A Tier 1 variant was detected in 41.94% of the patients, while in 9.68% a Tier 2C.1 variant, related to off-label treatment, was identi ed. The most prevalent altered gene in these patients was the PIK3CA gene, with 33.87% mutation rate. Additionally, an HR gene alteration was present in 9.68% of the tumors analyzed (Additional le 6).

IV. Other Cancers
In the 44 patients with Colorectal cancer, the mutation rate was 84.09% ( Figure 5, additional le 7).
Eighteen patients (40.91%) carried a mutation in one of the RAS genes which are biomarkers of resistance to EGFR antibodies treatment (38,39). Additionally, three patients carried a targetable BRAF somatic mutation. One PMS2 positive tumor mutation was proven to be of germline origin, and thus it was considered eligible for immunotherapy treatment.
Among the 34 patients with prostate cancer, at least one somatic alteration was identi ed in 72.73% of them ( Figure 5). In 5 cases, the mutation detected was in an HR gene (14.71%). Furthermore, 87.88% of the 33 patients with ovarian cancer, carried at least one somatic alteration. Four patients carried a mutation in BRCA1/2 genes, which are biomarkers of response to PARPi therapy, while in four patients, somatic mutations in off-label biomarkers were identi ed. Concerning brain tumors, the mutation rate was 77.78%. An alteration with associated potentially signi cant predictive biomarker was detected in 16 patients (62.96%) ( Figure 5). However, in this tumor histology, the multigene analysis seems to confer not only predictive but also prognostic/diagnostic information (40,41). Genes with diagnostic signi cance are used by the World Health Organization Classi cation of Tumors of the Central Nervous System. For example, IDH1 and IDH2 mutations are used for distinguishing primary from secondary gliomas, while the simultaneous presence of IDH1/2 and TP53 alterations are distinctive of the diffuse astrocytoma histology (40).
Concerning the other histological types, even if the number of patients tested is small, it seems that in tumors of the endometrium (18 cases), esophagus (7 cases) and in cholangiocarcinoma (25 cases) the mutation rate is relatively high (94.44%, 71.43% and 72.00% respectively). On the contrary low mutation rates are observed in gastric tumors (30 cases), hepatocellular carcinomas (11 cases) as well as in the 30 sarcomas analyzed (64.71%, 54.55% and 50.00% respectively).

Panel comparison
The genetic information obtained by the 161 gene panel used in this study compared to that obtained from panels containing fewer genes was evaluated. At this regard, we conducted a simulation of the alterations that would have been detected if two smaller hotspot panels, of 24 and 50 genes respectively, had been used in the 610 patients analyzed (additional le 8).
If the 24 gene panel had been used in our cohort, a clinically signi cant variant (Tier 1 and 2) would have been detected in 58.85% of the cases. In comparison, this percentage would have been 62.62% by using the 50 gene panel. However, these rates are much lower than the 77.70% obtained by the 161 gene panel. Furthermore, considering the on-label and off-label biomarkers, the larger panel managed to detect 14.12% and 10.67% more on/off-label treatment-related biomarkers compared to the 24 and 50 gene panel respectively ( Figure 6).
In order to evaluate if the number of genes analyzed is adequate for implementation in clinical practice, or if by increasing the number of genes tested a more informative result would have been obtained, we compared the actionability of our panel with a more comprehensive panel containing 501 DNA genes and 51 fusion drivers genes (38 of them also analyzed at the DNA level), for a total of 514 unique genes present in this panel (additional le 9).
Among the 990 patients with DNA sequencing results available, an SNV or indel alteration to a driver gene Thus, the increase in the number of genes analyzed seems to increase the yield of patients who could bene t from targeted treatments.

Physicians Survey
Additionally, in order to investigate the implementation of tumor molecular pro le analysis among physicians, a questionnaire was sent to referral oncologists asking whether they consider useful, such analysis for treatment decision making in various tumor types. 61 physicians responded to the survey. By far, the tumor type with the majority of positive responses was lung cancer, with 100% of the physicians responding that multigene panel should be performed for such tumor type ( Table 2).
For colorectal cancer patients, a multigene analysis was considered useful in the primary or metastatic setting by 95.08% of the participants. For breast, ovarian, prostate and pancreatic cancers, the NGS utility was recognized by 80.33%, 80.32%, 90.16% and 95.08% of the participants respectively.

Immunotherapy biomarkers analysis
Tumor testing can give information for the selection of both appropriate targeted treatment and immunotherapy. The most known immunotherapies biomarkers are TMB, PD-L1 and MSI analysis. In the cohort of 610 patients with successful NGS testing for targeted therapy, 395 also requested TMB analysis.
PD-L1 testing was performed in 198 cases, and MSI analysis in 201 patients. In 192 cases, all three immunologic biomarkers were analyzed (additional File 10) with successful analysis for all of them achieved in 191 cases.

Tumor Mutation Burden
Among the 395 patients with TMB analyzed, 14 cases (3.54%) could not receive a result due to the low quality of the genetic material analyzed. In these cases a high proportion (>60) of variants consistent with de-amination artifacts was detected, and thus these sequencing result could not be evaluated for TMB analysis, as indicated by the manufacturer (42). A successful TMB calculation was obtained for the remaining 381 patients. In accordance to previous studies, no association of TMB and PD-L1 values was observed ( Figure 11) (43,44).

Microsatellite instability
Microsatellite instability was detected in 8 out of the 206 tumors tested (3.88%), while for one tumor the analysis failed due to the low quality of the genetic material obtained. Patients with tumors showing MSI high status had a diagnosis of Ovarian cancer, Pancreatic cancer, Colorectal cancer, Prostate cancer, Gastric cancer and Sarcoma. In 2 cases, the tumor instability was linked to hereditary mutations in MMR genes (MSH2 and PMS2). TMB analysis data were also available in 7 of these patients with 6 of them showing high TMB value (>13.46muts/MB). Thus a strong correlation between TMB and MSI was observed with MSI high tumors showing higher median TMB values, in accordance with previous studies (45,46). However, it should be noted that among the 193 MSI stable patients with TMB data available, high TMB values were also observed in 42 cases (Figure 11). This is of great importance, given the higher rate of positivity for this biomarker and the strong evidence of predictive value; thus, its use could identify more patients eligible for immunotherapy uptake.

I. Molecular pro le analysis
In the present study, 629 cancer patients have been referred by their treating physician for biomarkers' analysis using a 161 gene NGS panel. In 610 of them, a successful tumor molecular pro le was obtained with at least one actionable variant (Tier 1-2)) being detected in 77.70% of the cases. All pathogenic variants were categorized based on their clinical signi cance, and only Tier 1 and 2 variants were reported since variants of unknown signi cance, and the benign/likely benign ones were considered confusing rather than useful for the treatment course information. In 54.59% of the patients, the information obtained could be used for on-label or off-label treatment reception (Tiers 1A.1, 1A.2, 1B, and 2C.1) while 21.31% of the cases received a variant that could be used for clinical trials inclusion.
Tests offering comprehensive tumor molecular pro ling are currently being requested by a steadily increasing number of oncologists, especially for patients with limited treatment options available. A good implementation of tissue analysis in treatment decision making was observed in the survey conducted in this study among oncologists. More than 80% of the participating physicians consider clinically useful the tissue NGS analysis for a variety of common tumor types. This percentage was increased to 100% for tumors with many targeted treatment options available such as lung cancer and for tumors with few treatments available such as tumors of unknown origin or rare tumors.
Advances in sequencing technologies and NGS platforms throughput have permitted simultaneous analysis of multiple tumor biomarkers at an adequate time frame to be tailored t in the design of the treatment plan and at an affordable cost for the patients. The information obtained can be used to address targeted treatment, immunotherapies or in case of negative results traditional treatment approaches. Various studies have shown the e cacy of gene-directed treatment compared to the unselected treatment assignment (47)(48)(49). In the IMPACT (Initiative for Molecular Pro ling and Advanced Cancer Therapy) study the overall response rate (ORR), and the time-to-treatment failure (TTF) were higher in patients with a molecular aberration that received a matched treatment compared to those who received unmatched treatment (50). Similarly, in the IMPACT/COMPACT trial, the response rate of patients treated according to their genotype had an overall response rate superior compared to those treated on  (53). The rst tumor agnostic therapy with a biomarker included receiving FDA approval was the PD-1 inhibitor Pembrolizumab, which was approved for patients with MSI unstable tumors (53). Subsequently, TRK inhibitor therapy gained approval in NTRK fusion-positive cancers independently from the tumor's histology (54)(55)(56). Even though the clinical value of these biomarkers cannot be disputed, the percentage of patients positive for these biomarkers is relatively small. For example, in our study, only 8 out of the 198 patients analyzed presented microsatellite instability. This biomarker seems to be more signi cant for colorectal cancer patients, where it is present in 10-15% of the cases, while it is of no use for other tumor types, where it is rarely detected (12). Similarly, the frequency of NTRK fusions in solid tumors of adults is extremely rare in certain tumor types (54,57). Consequently, no positive NTRK tumor was detected in our cohort.
On the other hand, there are agents, with associated biomarkers, that have shown activity in a variety of tumor types. PARP inhibitors are a typical example of such agents having already received approval for Ovarian, Breast, Pancreatic and Prostate cancer patients harboring BRCA1/2 mutations (https://www.fda.gov). Apart from BRCA1/2 mutations, other genes involved in the same pathway of homologous recombination seem to be adequate biomarkers of response to such agents, with several clinical trials investigating the expansion of PARPi targeting biomarkers (36,(58)(59)(60)(61) (www.clinicaltrials.org). These efforts led to the recent approval of the PARP inhibitor Olaparib for metastatic castrate-resistant prostate cancer patients with mutations in other HR genes besides BRCA1/2, increasing the percentage of patients with a potential predictive biomarker result who could bene t from that treatment (59,62). Thus, multigene analysis providing comprehensive information about the mutational status of HR genes should be used for better identi cation of responders to such therapy. In our cohort, 7.38% of the patients carried an alteration in an HR gene, with certain tumors showing increased levels of these alterations, such as breast cancer (9.68%), ovarian cancer (20%) and prostate cancer (14.71%). Moreover, the majority (75.56%) of the HR-positive patients, carried an HR gene mutation in a non BRCA1/2 gene, indicating the necessity of gene panel analysis for the identi cation of patients eligible for PARPi treatment.
An important issue when a multigene analysis is requested is the number of genes that should be included in such analysis and whether analyzing so many genes is offering more confusion than solutions in the physicians' search for an appropriate targeted treatment option for their patients. Despite all the advantages, there is much skepticism concerning the use of a personalized selection of appropriate treatment. A rst di culty in using broad tumor molecular pro le analysis for treatment selection is the unavailability in some cases of appropriate tumor tissue to perform the analysis. This could be due to the low quantity/quality of the tissue available or to its inaccessibility in some inoperable tumor types (63,64). It has been shown that among patients enrolled in tumor-directed treatments, only 70-90% of them had adequate tissue quantity/quality to achieve a successful molecular pro le (65). The technology used in our study permits tumor molecular pro le analysis from a limited quantity of genetic material. Hence, in our cohort more than 97% of the tumor samples were successfully analyzed.
Furthermore, such analysis can provide an immense quantity of genetic data that needs to be appropriately analyzed and interpreted. Thus, the role of bioinformatics analysis is becoming major to provide accurate molecular analysis results (66). Moreover, standardization of variant annotation and reporting could facilitate the understanding of the results obtained and increase their reliability. In our experience, in the majority of cases with ndings associated to off-label treatments recommendations, the long lasting procedures required for the off-label approval of the suggested treatment from the local National Drug Organization for Medicines ("EOF"), or for clinical trial enrollment, often challenged the utilization of the results, especially in cases with advanced disease, requiring immediate management.
While, it is standard practice to perform accurate pre-and post-test counseling prior to a genetic testing for hereditary cancer susceptibility, this is not the case for somatic mutation analysis (67). However, it is crucial for patients referred for tumor genetic analysis to be accurately informed about the necessity and the possible results obtained by tumor testing. Thus, it would be wise to provide an accurate pre-test counseling, explaining the advantages and limitations of the analysis, with particular regard to avoiding raising too high the expectations, since only a subset of patients will receive useful information. In such a way, a patient's disappointment in case of non-informative results will be reduced, and their post-test management will be facilitated. Furthermore, patients should be informed about the possibility of identifying a variant in a gene with known germline mutations. Especially variant's detected in high percentages (>40%), are considered of suspicious germline origin. Since this analysis cannot discriminate between germline and somatic variations, the clari cation of a variant's origin requires the analysis of patients' healthy tissue, usually blood or saliva. In our cohort, 17 patients with a family history of cancer, requested blood analysis for suspicious germline variant identi ed in tissue. In 14 of them (82.35%) the germline origin of the tissue alteration was con rmed (Additional le 11).

II. Immunotherapy biomarkers
Analysis of the tumor's molecular pro le useful as it is, it seems to be just another piece of the puzzle, since comprehensive tumor pro le should include both biomarkers to guide treatment decision making for both targeted therapy as well as for immunotherapy. Thus, the physician having more biomarkers in his disposition could better comprehend the tumor's biology and decide whether targeted therapy or immunotherapy matches better in each case. In our cohort analysis of biomarkers for both immunotherapy and targeted therapy, was requested in 395 patients, with TMB being the most common immunotherapy biomarker requested. All three biomarkers' analysis was successful in 191 cases. 25.20% of the 381 patients tested had a TMB value >10muts/MB and thus were eligible for ICI treatment.
The median TMB values observed in our population were slightly increased compared to those observed in previous studies (45,68). This could be attributed to methodological differences and to the fact that in the majority of cases the patients analyzed have received more than one treatment lines, commonly chemotherapy, which is known to increase tumor's mutation load (69). Similarly, to our study a TMB positivity rate of 21.1% was observed in a recent study analyzing immunotherapy biomarkers in 48.782 clinical samples (70). TMB has emerged as a promising biomarker of response to such treatments, and several clinical trials have shown that both blood and tissue samples TMB can effectively be used (15,17,19,71). Moreover, the recent approval of anti-PD1 treatment Pembrolizumab for metastatic cancer patients harboring a TMB value >10mut/ΜΒ renders the analysis of such biomarker indispensable for treatment selection strategy.
However, this biomarker has also limitations since TMB calculation methods can differ between different assays, while the gene content of the methodology used seems to affect the TMB values obtained (72)(73)(74). Furthermore, the cut-off values for this marker are not yet fully established. All these issues are addressed from the International harmonization initiatives led by Friends of Cancer Research (FOCR) and the Qualitätssicherungs-Initiative Pathologie (QuIP) (72)(73)(74).
Concerning the other immunotherapy biomarkers, analyzed in this study (PD-L1 and MSI), they could assist in a more accurate patients' selection for treatment with Checkpoint inhibitors. PD-L1 expression, measured by immunohistochemistry methods is the most widely used biomarker and the rst to be approved for treatment with checkpoint inhibitors (7). Nevertheless, it is not applicable in many tumor types, and its sensitivity and speci city in identifying patients eligible for immunotherapy have also been questioned (7,8,19,(75)(76)(77). Moreover, while MSI analysis seems to be an appropriate biomarker, its low incidence in the majority of tumor types limits its clinical utility in the majority of neoplasms. In our cohort microsatellite instability was observed in just 3.88% of the cases; thus, it cannot stand alone as an immunotherapy biomarker, rendering the addition of other biomarkers indispensable to increase the number of patients who could bene t from such treatments.
The incidence of TMB positivity is superior to that of MSI (25.20% compared to 3.88%). Furthermore, in 21.88% (42/193) of the MSI stable cases, a TMB value of >10muts/MB was observed; thus, these patients could receive ICI based on the TMB result only. Moreover, no association between TMB and PD-L1 values was observed. This is in agreement with previous studies, indicating lack of association between median values of these biomarkers. However, in accordance to a recent study, a higher TMB positivity rate was observed in the TMB high group. (70). The TMB positivity rate among the PD-L1 positive patients was 33.77% (26/77) compared to 15.79% (18/114) in the PD-L1 negative group (p=0.005). Importantly, it has been reported that patients with positive values for both TMB and PD-L1 could have greater bene t from such treatment compared to those showing positivity for only one of these biomarkers (43,44).
Collectively, among the 191 patients with all three immunotherapy biomarkers tested, ICIs option based on TMB result could be considered in 44 patients (23.04%), 26 of them with simultaneous PD-L1 positivity.
As it can be seen in the Venn diagram ( Figure 13) showing the correlation among these biomarkers in 191 patients tested with all three biomarkers, 50.26% of the cases had at least one positive biomarker. A positive result for both PD-L1 and TMB was seen in 13.61% of the cases (with simultaneous MSI high result in 3 cases). In 2 patients concomitant TMB and MSI high values were observed (1.05%). An additional 35.60% % of the patients could receive immunotherapy-based one either TMB or PDL-1 or MSI positivity (8.38%, 26.70%, 0.52% respectively).
The analysis of immunotherapy biomarkers, though, does not seem to be the only determinant of response to ICI, since the tumor mutational status also seems to have a signi cant in uence on the probability of response. For example, several studies have shown the reduced e cacy of ICIs in Non-Small Cell Lung cancer patients harboring EGFR mutations and ALK rearrangements (43,78,79). The absence of such targetable alteration could direct the treatment strategy to immunotherapy in these malignancies. In addition, it has been shown that alterations in certain genes, such as KRAS, TP53, MET, ARID1A and others are enriched in immunotherapy responsive patients. Thus, their identi cation could lead to such treatment option (80)(81)(82).
Moreover, alterations in DNA repair genes such as the MMR genes, POLE and HR genes have been shown to have a positive predictive effect and are correlated to increased TMB values (83)(84)(85)(86). In contrast, other gene alterations such as JAK1/2 and STK11/LKB1, KEAP1 and PTEN mutations are related to resistance to PD-1 Blockade (80,(87)(88)(89). Interestingly, in our study, 2 of the patients with TMB high values and one patient with PD-L1 positive result also harbored an STK11 mutation. In none of these cases, immunotherapy response was achieved.
Thus, the addition of immunotherapy biomarkers to tumor molecular pro ling seems to be a one-way road in order to achieve a comprehensive tumor characterization and provide the right treatment option for each patient. Moreover, the simultaneous analysis of such biomarkers, leads to the increase of patients with an on-label treatment recommendation by 22.40%. By combining immunotherapy and targeted therapy biomarkers, 71.35% of the patients analyzed received information related to on-label or off-label treatments. This is obviously improved compared to the 50.52% of on/off-label biomarkers achieved by analyzing only the molecular pro le of the tumor in the same patient cohort.
Nowadays it seems that the tissue is not the issue anymore, since NGS technological advantages permit the simultaneous analysis of many targets from limited tissue material, achieving to analyze up to 97% of the tissue samples as in the present study. The challenge, though, when these analyses are performed is their implementation in clinical practice. Thus, the results obtained must be appropriately comprehended and adopted for the designation of the treatment selection strategy, which can be achieved through interdiscipline collaboration. To this regard, of great use would be the presence of a multidisciplinary Molecular Tumor Board that could assist in the accurate interpretation of the ndings obtained from such complex NGS analysis and provide therapeutic recommendations based on all available clinical data for each individual patient (90)(91)(92).

Conclusions
The NGS analysis conducted in this study offered actionable information (Tier1 and 2) in 77.70% of the 610 patients with tumor molecular pro le analysis available. Moreover, simultaneous analysis for targeted therapy and immunotherapy biomarkers resulted in a better tumor characterization and provided actionable information in 83.25% of the 191 patients tested, with one to two patients being eligible for ICI treatment based on the biomarkers' analysis. Thus, the comprehensive analysis of these biomarkers increased the number of patients with a treatment-related nding and contributed to a more individualized approach for cancer treatment. In conclusion, the present study has shown that the implementation of molecular pro ling using appropriate pan-cancer panels in clinical practice is feasible. Of signi cance, is the appropriate comprehension of the molecular results obtained from such analysis and their proper utilization for designing the treatment selection strategy, which can be achieved through inter-discipline collaboration.

Declarations
Ethics approval and consent to participate The study was approved by the ethical committee of "Bioclinic Thessaloniki" Hospital. All patients gave informed consent for molecular analysis in blood and tissue, in accordance with the Declaration of Helsinki. All patients included in the study have consented for publication of the data generated by the analysis performed.

Consent for publication
All patients included in the study have consented for publication of the data generated by the analysis performed.
Availability of data and materials All data generated or analyzed during this study are included in this published article and its supplementary information les.

Competing interests
The authors EI, NT, AT, VM, GK, EK, EB, GT, DF and GN declare that are employees in GENEKOR MSA. The other authors declare no con icts of interest.
Funding GENEKOR MSA provided support in the form of salaries for authors EI, NT, AT, VM, GK, EK, EB, GT, DF and GN but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The speci c roles of these authors are articulated in the 'author contributions' section.     The Top 20 most frequently altered genes in the cohort analyzed. Patients with biomarkers related to clinical trials, 2D: Patients with biomarkers with preclinical evidence of predictive value, 3: Patients harboring alterations with con icting evidence of cancer association.     TMB-MSI and TMB-PD-L1 correlation.

Figure 12
Patients' categorization based on the level of evidence of the most clinically signi cant variant associated with response, with and without the use of immunotherapy biomarkers. Biomarkers associated with resistance were excluded from this analysis.