Implementation and use of whole exome sequencing for metastatic solid cancer

Abstract

Background

Genomically-guided clinical trials are performed across different tumor types sharing genetic mutations, but trial organization remains complex. Here we address the feasibility and utility of routine somatic and constitutional exome analysis in metastatic cancer patients.

Methods

Exoma trial (NCT02840604) is a multicenter, prospective clinical trial. Eligible patients presented a metastatic cancer progressing after at least one line of systemic therapy. Constitutional genetics testing required geneticist consultation. Somatic and germline exome analysis was restricted to 317 genes. Variants were classified and molecular tumor board made therapeutic recommendations based on ESMO guidelines. Primary endpoint was the feasibility of the approach evaluated by the proportion of patient that received a therapeutic proposal.

Findings

Between May 2016 and October 2018, 506 patients were included. Median time required for tumor sample reception was 8 days. Median time from sample reception to results was 52 days. Somatic analysis was performed for 456 patients (90.1%). Both somatic and constitutional analyses were successfully performed for 386 patients (76.3%). In total, 342 patients (75%) received a therapeutic proposal. Genetic susceptibility to cancer was found in 35 (9%) patients. Only, 79 patients (23.1%) were treated with NGS matched therapy mainly PI3K/AKT/mTOR inhibitors 22 (27.8%), followed by PARP inhibitors 19 (24.1%), antiangiogenics 17 (21.5%), MEK inhibitors 7 (8.9%) and immunotherapy 5 (6.3%). Matched treatment was finally stopped because of disease progression 50 (63%), treatment toxicity 18 (23%), patients’ death 4 (5%). PFS2/PFS1 ratio was > 1,3 for 23,5% of patients treated with the NGS matched therapy and 23,7% of patients treated with standard therapy.

Interpretation

Study shows that exome analysis is feasible in cancer routine care. This strategy improves detection of genetic predispositions and enhances access to target therapies. However, no differences were observed between PFS ratios of patients treated with matched therapy versus standard therapy.

Funding

This work was funding by the centre Georges Francois Leclerc

Research in context.

Evidence before this study

Precision medicine is a new era in the field of cancer therapy. In many cancer types, driver mutations could be targeted by small molecules, leading to a high response rate and better survival. Such mutation could be find in many cancer type thus leading to the hypothesis that large molecular screening may help physician to find better treatment. Many clinical trials test this hypothesis in tertiary cancer center. These trials use large somatic NGS panel or whole exome. Advantage of Exome technics is that this test will not require technical implementation if new genes have and can be used to find genetic predisposition to cancer. However, such strategy is complex because it requires a good organization between oncogeneticists and oncologists to improve patient information and care. The complexity of clinical NGS testing has prevented many hospitals and laboratories from routine usage such large genomic testing.

We here address the feasibility and utility of routine somatic and constitutional exome analysis in a prospective cohort of metastatic cancer patients that received at least one line of chemotherapy.

Added value of this study

This Study shows that exome analysis is feasible in cancer routine care. This strategy improves detection of genetic predispositions and enhances access to target therapies. However, such strategy did not improve patient outcome.

Implications of all the available evidence

The disappointing results of this study underline our superficial understanding of the mechanisms in cancer evolution and cancer heterogeneity. It also highlights the fact that even with the existing targeted agents, if our comprehension is probably superficial. Probably multiomics strategies and the incorporation of novel technologies like RNA-sequencing, whole genome sequencing and circulating cell-free DNA detection should be emphasized for future studies in order to estimate the possibility of novel targets and potential agents for these targets. Alternatively, exploration of particular “extraordinary response” or surprising failure of target therapies may help us to better understand cancer biology.

Alt-text: Unlabelled box

1. Introduction

High-through put next-generation sequencing gives a new insight in the molecular landscape of cancer. Molecular profiling underlines that a same tumor type can contain variable molecular subgroups with different molecular properties. Importantly, particular mutation and related active molecular pathways lead to the identification of druggable targets. Over recent years, based on this concept, oncology has served as a paragon for the application of clinical genomics to treatment of disease [1]. In many cancer types, driver mutations could be targeted by small molecules, leading to a high response rate and better survival. In certain cancer types like colorectal cancer, lung cancer and melanoma, molecular profiling has become standard practice to search for targetable mutations [2], [3], [4]. This work is currently translated in the concept of precision medicine where genomically-guided clinical trials have begun to evaluate the efficacy of molecularly-targeted therapies across different tumor types with shared genetic mutations [5].

Currently, clinical trials include large somatic NGS panel or using constitutional and somatic analysis of large panel genes or whole exome [6]. One advantage of Exome technics is that this test will not require technical implementation if new genes have to be analyzed. The use of constitutional analysis is helpful to find genetic predisposition to cancer in addition to finding targeted therapies. However, such strategy is complex because it requires a good organization between oncogeneticists and oncologists to improve patient information and care. In addition, the analysis of somatic and constitutional mutations supports the discovery of unknown genetic variants present in tumor DNA. The targetable relevance of such mutations is not addressed at the present time. The complexity of clinical NGS testing has prevented many hospitals and laboratories from routine usage of large NGS analysis. Currently, molecular NGS profiling trials are only performed in expert centers because of complexity and time required for bioinformatics analysis.

We here address the feasibility and utility of routine somatic and constitutional exome analysis in a prospective cohort of metastatic cancer patients that received at least one line of chemotherapy.

2. Patients and methods

2.1. Study design and procedure

The Exoma trial (NCT02840604) is a multicentric, prospective clinical trial. The Trial was approved by the ethical comitee called (Comité protection des personnes Est). The trial accrued patients between May 2016 and October 2018. Participating principal investigators were located at Dijon Cancer Center (centre Georges Francois Leclerc), Dijon University Hospital, Besancon University Hospital and Chalon sur Saone Private Hospital (Clinique Sainte Marie). Genomic analyses were performed at the Georges-Francois Leclerc Cancer Center, in the Genomic and Immunotherapy Medical Institute, in Dijon. This study aimed to show that exome analysis is feasible in patient routine care and improves access to target therapies and detection of genetic cancer predisposition. Patients were eligible if they presented a locally advanced non-operable or metastatic cancer that had progressed during at least one line of systemic therapy. We only included patients with non-curable diseases. All patients provided a signed informed consent for the trial and genomic analysis. After informed consent, patients had a consultation with a geneticist to explain the consequences of a constitutional genetic testing. Only after this consultation patient could accept or refuse the blood sample for constitutional exome analysis. This trial protocol was approved by an institutional review committee and done in accordance with the Declaration of Helsinki. Study was reported according to CONSORT Checklist.

2.2. Sample selection

After signed informed consent, physicians selected an archival tumor sample of less than one year (primary or metastasis) for genomic analysis. At the discretion of the physician, a new tumor biopsy could be proposed to the patient. Tumor cellularity was assessed by a senior pathologist on a hematoxylin and eosin slide from the same biopsy core used for nucleic acid extraction and molecular analysis.

2.2.1. DNA isolation

DNA was isolated from archival tumor tissue using the Maxwell 16 FFPE Plus LEV DNA Purification kit (Promega, Madison, WI, USA). DNA from whole blood (germline DNA) was isolated using the Maxwell 16 Blood DNA Purification Kit (Promega) following the manufacturer's instructions. Quantity of extracted genomic DNA was assessed by a fluorimetric method with a Qubit device.

2.3. Whole exome capture and sequencing

Two hundred ng of genomic DNA were used for library preparation, using the Agilent SureSelectXT reagent kit (Agilent Technologies, Santa Clara, USA). The totality of enriched library was used in the hybridization and captured with the SureSelect All Exon v5 or v6 (Agilent Technologies) baits. Following hybridization, the captured libraries were purified according to the manufacturer's recommendations and amplified by polymerase chain reaction (12 cycles). Normalized libraries were pooled and DNA was sequenced on an Illumina NextSeq500 device using 2 × 111-bp paired-end reads and multiplexed. Tumor and germline DNA sequencing generated mean target coverages of 78X and 90X respectively, and a mean of more than 90% of the target sequence was covered with a read depth of at least 10X for somatic DNA.

2.4. Exome analysis pipeline

Raw DNA sequencing data were aligned to the hg19 genome build using the Burrows-Wheeler Aligner (BWA) version 0.7.15. Duplicates were marked with Picard version 2.5.0. Base quality scores recalibration and variant calling were performed using GATK tools version 3.6.

For SNV (Single Nucleotide Variation), annotation was performed using the VariantStudio Illumina software. Filters of candidate variants included: coverage depth of 10X or greater and a variant nucleotide allelic fraction in tumor DNA greater than 5%.

2.5. Determination of tumor mutational burden per Mb

Whole exome sequencing data were used to generate tumor mutation burden per Mb for each patient. Tumor mutational burden (TMB) corresponds to the total number of missense and indel somatic-specific mutations divided by the number of megabases (Mb) of the target sequences of the SureSelect All Exon v5 or v6 baits (≈ 50.6Mb). Tumor mutations were identified from paired exomes by subtracting SNV observed in germline exome from SNV observed in the corresponding somatic exome.

2.6. Analysis of somatic mutations

We limited our analysis to 317 genes (Table 1). This list is adapted from Foundation One [7]. We used the knowledge database of somatic mutations Cosmic v.64 released on 2013, March 26th, to classify each selected variant as ‘pathogenic’, ‘probable pathogenicity’ ‘unknown pathogenicity’, or ‘benign’ variants. For each detected and annotated variant we retained for interpretation only variants annotated as pathogenic or likely pathogenic. Unknown variants were retained when present in somatic analysis only and located in a critical domain of the protein. Each therapeutic proposal was then classified using a homemade classification approved by our molecular tumor board: Grade A: positive phase II or III, Grade B: no data in this disease but positive data in other disease or case reports, Grade C: in vitro experiments. After publication of ESCAT recommendation we replaced our grade by ESCAT because of their strong similarity [8].

Table 1.

Patient clinical and tumor pathological characteristics.

Clinicopathologic characteristics
Age at inclusion. years
	Median [range]	65 [24–94]
Sex. n (%)
	Male	202 (39,9)
	Female	304 (60,1)
WHO status at inclusion. n (%)
	0	166 (33,5)
	1	258 (52)
	2	58 (11,7)
	3	14 (2,8)
	Unknown	10
Histology of the primary tumor. n (%)
	Adenocarcinoma	372 (73.4)
	Squamous carcinoma	35 (7)
	Sarcoma	12 (2,4)
	Melanoma	4 (0,8)
	Other	75 (14,8)
	Unknown	8 (1.6)
Metastatic status at inclusion. n (%)
	No metastasis	64 (12,6)
	Metastastis	442 (87,4)
Number of metastastic sites
	Median [range]	2 [1], [2], [3], [4], [5], [6], [7], [8]
Number of chemothery lines. n (%)
	0	7 (1,4)
	1	91 (18)
	2	105 (20,8)
	3 or more	303 (60)

2.7. Analysis of constitutional mutations

On the basis of the same list of 317 genes [7], we performed the analysis in both whole blood cells DNA and tumor DNA to determine whether gene variant was present constitutionally or only in tumor sample. We limited constitutional analysis on 26 genes upon recommendation of our geneticists coming from American College of Medical Genetics gene list (https://www-ncbi-nlm-nih-gov/clinvar/docs/acmg/) (Table 1). Filters of candidate variants included: coverage depth of 10X or greater and a variant nucleotide allelic fraction in tumor DNA greater than 5%.

If the variant was constitutional, we determined its frequency in the general population using EXAC and dbSNP population databases, its presence in diseases databases and reviewed the corresponding available bibliography.

When the analysis indicated possible cancer susceptibility, the results were given on the clinical reports and explained to the patient by a geneticist in order to offer adapted follow-up.

2.8. Statistical hypotheses and analysis

The Exoma trial aimed to show that exome analysis was feasible in patient routine care and improved access to target therapies and detection of genetic cancer predisposition. The primary objective was that more than 30% of included patient could receive a therapeutic proposal. In order to estimate this proportion with a precision of the 95% confidence interval of 4%, 506 patients will have to be included in the study.

In addition, to assess the clinical impact of NGS adjusted therapies, we examined clinical results, as in the Von Hoff model [9], the PFS2/PFS1 ratio for the patients treated after NGS analysis results. This ratio corresponds to the comparison of the progression-free survival on matched therapy (PFS2) with the progression-free survival for the most recent therapy, on which the patient had just experienced progression (PFS1). Progression-free survival on matched treatment (PFS2) was defined as the time from start of treatment to progression, as defined by RECIST 1.1, clinical progression, or death from any cause. Progression-free survival on prior therapy (PFS1) was defined as the time from start of the last prior treatment to progression as defined by RECIST 1.1 or clinical progression [10]. Each patient is his own control.

The matching score for each patients was calculated as the number of characterized DNA alterations affected by the drug (or drugs) proposed divided by the total number of characterized alterations.

Point estimates and the associated 95% CIs were provided. Standard statistical tests including the chi-squared test and Fisher's exact test for categorical data and the t-test for continuous data were applied. PFS were analyzed by the Kaplan–Meier estimate. The log-rank test and Cox proportional hazards model (Wald test) were applied to test the effect of covariates on PFS. The assumption of the Cox regression was validated.

Statistical analyses were performed in SAS 9.4 and R 3.2.2.

3. Results

3.1. Population characteristics

Between May 2016 and October 2018, 506 patients were included in the EXOMA clinical trial. From this cohort, we could obtain tumor tissue and isolated DNA in 456 cases (90.1%). flow-chart is represented in Fig. 1a.

Fig. 1.

a. Flow chart. b. Main tumor types in the trial.

The analysis could not be performed in 50 cases (9.9%), due to insufficient tumor content or DNA and we have excluded samples that did not meet post-sequencing quality control criteria. Altogether, we successfully sequenced somatic DNA for 456 patients (90.1%). For constitutive analysis, 452 patients (89.3% of all patients) met an oncogeneticist to be informed and give consent for the constitutional analysis; 54 patients (10.5%) did not (patient's refusal). Among the 452 patients who had an oncogeneticist consultation, 16 patients (3.1%) refused the analysis and data was missing for 3 patients. In total, both somatic and constitutional analyses were available for 386 patients (76.3%).

We included a mean of 16.7 patients per month. The median time for reception of tumor sample was 8 days [0–379]. The median turnaround time from sample reception to results was 52 days [3-339]. For patients with available tumor sample, 385 samples (84.4%) came from archival sample. A new biopsy was performed for 71 patients (15.6%).

The main tumor type was breast cancer (21.5%), followed by colorectal (14.8%) and pancreatic cancer (14.2%) which reflected the classical recruitment of metastatic patients in including centers (Fig. 1b). Table 2 summarizes the clinical characteristics of the included patients.

Table 2.

List of genes used for somatic and constitutional analyses. Genes displayed in red are those used for constitutional analysis.

ABL1	ABL2	AKT1	AKT2	AKT3	ALK	AMER1	ANAPC2	APC	AR
ARAF	ARFRP1	ARID1A	ARID2	ASXL1	ATM	ATR	ATRX	AURKA	AURKB
AXIN1	AXIN2	AXL	BAP1	BARD1	BCL2	BCL6	BCOR	BCORL1	BLM
BRAF	BRCA1	BRCA2	BRIP1	BTK	BUB1B	C11orf30	CARD11	CBFB	CBL
CCND1	CCND2	CCND3	CCNE1	CD79A	CD79B	CDC25A	CDC34	CDC73	CDH1
CDK12	CDK4	CDK5RAP1	CDK6	CDK8	CDKN1B	CDKN2A	CDKN2C	CDKN3	CEBPA
CHEK1	CHEK2	CIC	CKIT	CREBBP	CRKL	CRLF2	CSF1R	CTCF	CTNNA1
CTNNB1	CUL1	CUL2	CUL3	DAXX	DDR2	DICER1	DNMT3A	DOT1L	DPYD
E2F1	EGFR	EP300	EPCAM	EPHA2	EPHA3	EPHA5	EPHB1	ERBB2	ERBB3
ERBB4	ERCC1	ERCC2	ERCC3	ERCC4	ERCC5	ERCC6	ERCC8	ERG	ESR1
ESR2	EZH2	FAM46C	FANCA	FANCC	FANCD2	FANCE	FANCF	FANCG	FANCL
FBXW7	FGF10	FGF14	FGF19	FGF23	FGF3	FGF4	FGF6	FGFR1	FGFR2
FGFR3	FGFR4	FLCN	FLT1	FLT3	FLT4	FOXL2	GATA1	GATA2	GATA3
GID4	GLI1	GLI2	GLI3	GNA11	GNA13	GNAQ	GNAS	GREM1	GRIN2A
GRP	GSK3A	GSK3B	HGF	HRAS	HSP90AA1	IDH1	IDH2	IGF1R	IKBKE
IKZF1	IL7R	INHBA	INPP4A	INPP4B	IRF4	IRS2	JAK1	JAK2	JAK3
JUN	KAT6A	KDM5A	KDM5C	KDM6A	KDR	KEAP1	KIT	KLHL6	KRAS
LCK	LRP1B	MAP2K1	MAP2K2	MAP2K3	MAP2K4	MAP3K1	MAPK1	MAPK3	MCL1
MDM2	MDM4	MED12	MEF2B	MEN1	MET	MITF	MLH1	MLH3	MPL
MSH2	MSH6	MTOR	MUTYH	MYC	MYCL1	MYCN	MYD88	NBN	NF1
NF2	NFE2L2	NFKBIA	NKX2-1	NLRP3	NOTCH1	NOTCH2	NPM1	NRAS	NTHL1
NTRK1	NTRK2	NTRK3	NUP93	PAK3	PALB2	PARP1	PARP2	PAX5	PBRM1
PDGFRA	PDGFRB	PDK1	PGR	PIK3CA	PIK3CG	PIK3R1	PIK3R2	PMS1	PMS2
POLD	POLE	PPP2R1A	PRDM1	PRKACA	PRKACB	PRKAR1A	PRKDC	PRSS1	PRUNE2
PTCH1	PTCH2	PTEN	PTPN11	RAD50	RAD51B	RAD51C	RAD51D	RAD54L	RAF1
RARA	RB1	RET	RICTOR	RNF43	ROS1	RPTOR	RUNX1	SDHAF2	SDHB
SDHC	SDHD	SETD2	SF3B1	SHH	SKP2	SLC28A1	SLC29A1	SMAD1	SMAD2
SMAD3	SMAD4	SMAD5	SMARCA4	SMARCB1	SMO	SOCS1	SOX10	SPARC	SPEN
SPOP	SRC	STAG2	STAT3	STAT4	STK11	SUFU	SUZ12	TERT	TET2
TGFBR2	THRA	THRB	TNFAIP3	TNFRSF14	TOP1	TP53	TP53BP1	TSC1	TSC2
TSHR	TYMS	UIMC1	VHL	WISP3	WNT	WT1	XPO1	XRCC1	XRCC2
XRCC3	XRCC4	XRCC5	XRCC6	YES1	ZNF217	ZNF703

3.2. Landscape of constitutional and somatic mutations

For somatic analysis we limited our analysis to 317 genes (Table 1). The three most frequently tumor altered genes in the EXOMA cohort were TP53 (38.6%), KRAS (18%) and PIK3CA (13.8%) (Fig. 2a).

Fig. 2.

Genomic characteristics. a. List of top mutated genes. b. top mutated genes in main cancers. c. Tumor mutational burden value across tumor type. d. relation between Tumor mutational burden and alterations in DNA repair pathways. e. List of constitutional alterations in actionable genes. (d,e: Lines represent median and interquartile ranges); for panel d * mean p value< 0.05 (Mann-Whitney test).

Among the 5 main cancers, TP53 mutations were the most prevalent: 52.9% of patients in colorectal cancer, 49.2% of patients in pancreatic cancer, 48.6% of patients with ovarian cancer, 35.9% of patients with breast cancer, and 33.3% of patients with NSCLC. TP53 mutations coding consequences were mainly missense variants (68%) and frameshift variants (16%).

KRAS mutations were the second most prevalent: 50.8% of patients with pancreatic cancer, 44.3% of patients with colorectal cancer and 23.3% of patients of NSCLC. KRAS mutations coding consequences were mainly missense variants (99%).

The third most prevalent mutations were PIK3CA mutations, present for 24.3% of patients with breast cancer and 13.5% for patients with ovarian cancer (Fig. 2b). PIK3CA mutations coding consequences were mainly missense variants (94%).

We could determine the tumor mutational burden (TMB) in 313 patients for which both somatic and constitutional exome analysis were available. TMB is the number of coding and non-coding mutations divided by the length of the sequencing design. The median TMB was 5.1 mutations per Mb (range 0.6–54). The tumor type with higher TMB was NSCLC, a median of 6 mutations per Mb. The tumor type with the lower TMB was the ovarian cancer with a median of 4.4 mutation per Mb p = 0.15 (Mann-Whitney test) (Fig. 2c). We observed a strong correlation between mutations in DNA repair genes (either somatic or constitutional) and TMB (Fig. 2d).

For 386 patients (76.3%) we could perform constitutional exome analysis. We limited our analysis on 26 genes known to be related to increase risk of cancer (Table 1). We observed 361 variants, 197 neutral or benign, 129 variants of unknown significance and 35 deleterious variants Nine patients required a new consultation by a geneticist for cancer predisposition that where not discovered before inclusion in this clinical trial. Fig. 2e shows the impact of constitutional alterations in actionable genes.

3.3. Clinical actionability and utility

All WES analyses were discussed at the molecular tumor board. A therapeutic proposal was done if there was an open clinical trial testing a drug which targets the mutation, or if there was an approved drug available for the relevant disease or for another disease known to target the mutated gene or the related activated pathway. The decision was based on discussion made at the tumor board with a basic proposal established according to Target database (https://software.broadinstitute.org/cancer/cga/target), then decision was classified using international recommendations provided by ESMO Scale of Clinical Actionability for molecular Targets (ESCAT) [8]. In addition, each mutation could have a particular biological impact which was categorized in four categories (i.e. pathogenic, likely pathogenic, unknown significance or benign) [11]. In most trials, only class I and II (pathogenic, likely pathogenic) were selected for therapeutic proposal, while class III (mutation of unknown significance) were excluded. In this study, class III variants were also retained for therapeutic proposal. As shown in Table 3, we selected for recommendation grade I to III ESCAT level of evidence and for some cases discussed grade IV level of evidence. This recommendation was made in around 40% of cases with class III variant (variant of unknown significance) (Table 3).

Table 3.

Classification of variants and therapeutic proposals.

Classification
Therapeutic proposal. n (%)
	I	18 (4.2)
	II	50 (11.6)
	III	238 (55.2)
	IV	125 (29)
Variant. n (%)
	I - II	262 (60.5)
	III	171 (39.5)

Among the patients with a WES analysis available, a total of 342 patients (75%) received at least one therapeutic proposal based on this analysis. Therapeutic proposal goes from 1 to 5 per patients. In this cohort, 79 patients (23.1%) were treated at progression with a NGS matched therapy: mainly by PI3K/AKT/mTOR inhibitors (27.8%), followed by PARP inhibitors (24.1%), antiangiogenics (21.5%), MEK inhibitors (8.9%) and immunotherapy (6.3%). Matched treatment was finally stopped because of disease progression (63%), treatment toxicity (23%), patients’ death (5%) or other reasons (9%). Table 4 summarizes the treatment received by the patients after NGS analysis. Supplementary Table 1 and Supplementary Figure 1a summarized data for patients that received matched therapy and Supplementary Fig. 1b. summarize their PFS using Kaplan-Meier curves. In contrast, 263 patients (76.9%) were not treated at progression by a treatment based on NGS analysis: 149 patients (56.7%) were treated according to the oncologist's choice, most of the time with standard of care therapy; 114 patients (43.3%) were further untreated, due to palliative care for 66 patients (57.9%), death for 40 patients (35.1%) and other reasons for 8 patients (7%). The Cox univariate analysis of patients that received matched therapy underline that male, poor performance status, pancreatic and colorectal cancer were associated with poor outcome (Supplementary Table 2).

Table 4.

Therapeutic proposal.

NGS-based proposals
Patients with therapeutic proposal after NGS analysis. n (%)		N = =456
	Yes	342 (75)
	No	114 (25)
Treatment based on NGS analysis at disease progression. n (%)		N = =342
	Yes	79 (23.1)
	No	263 (76.9)
Type of treatment delivered according to NGS analysis		N = =79
	PIK3/AKT/mTOR inhibitors	22 (27.8)
	PARP inhibitors	19 (24.1)
	Antiangiogenics	17 (21.5)
	MEK inhibitors	7 (8.9)
	Immunotherapy	5 (6.3)
	Other	9 (11.4)
End of treatment according to NGS analysis
	Progression disease	34 (43)
	Toxicity	5 (6.3)
	Death	4 (5.1)
	Other	9 (11.4)
	Unknown	27 (34.2)
Reason for no treatment delivery		N = =263
	Other therapy	149 (56.7)
	No further therapy	114 (43.3)
	Palliative care	66 (57.9)
	Death	40 (35.1)
	Other	8 (7)

The median PFS2 of patients treated with targeted therapy and non-targeted therapy were respectively 2.5 months (95% CI [2.2; 3.7]) and 2.4 months (95% CI [2.1; 3.3]). To assess the clinical relevance of such strategy we compared PFS1/PFS2 ratio in patient that received the targeted therapy and the non-targeted therapy.

We define PFS1 as the duration in months of the treatment prior to NGS analysis results, and PFS2 as the duration in months of the therapy after NGS analysis results (matched or non-matched on NGS). A ratio PFS2/PFS1 greater than 1.3 was considered as clinically significant [9]. Table 5 summarizes the data for patients treated with or without NGS-based therapy (those patients with both PFS1 and PFS2 data available). No difference was observed between the proportion of non-NGS-based treated patients and NGS-based treated patients with a ratio greater than 1.3 p = 0.8 (Chi square test).

Table 5.

Summary of survival results. PFS: progression-free survival.

	Patients treated with non-NGS-based therapy	Patients with NGS based therapy	Patients with PFS2 and PFS1 data
	N = 89	N = 48	N = 137
PFS of the 1st treatment - months	3.6 [0.5–33.8]	3.3 [0.5–10.1]	3.5 [0.5–33.8]
PFS of the 2nd treatment - months	2.1 [0.03–18]	2.3 [0.2–10.1]	2.2 [0.03–18]
Ratio 2nd PFS/ 1st PFS
Median [range]	0,6 [0.003–3,6]	0,6 [0.1–6,1]	0.6 [0.003–6.1]
⩽ 1.3	66 (74%)	36 (75%)	102 (74%)
> 1.3	23 (26%)	12 (25%)	35 (26%)

The primary tumor type, the type of NGS-based treatment delivered, the therapeutic class of NGS-based treatment, were not associated with a different PFS2/PFS1 ratio. Moreover, a higher matching score (> 0,5) was not associated with better PFS2/PFS1p = 0.43 (Wilcoxon rank test).

4. Discussion

Several molecular profiling trials took place in recent years. Among those, Von Hoff et al. compared in 2010 the PFS2/PFS1 ratio, taking patients as their own controls, and considered there was a clinical benefit for a ratio ≥ 1.3. This study used a panel of genes for molecular testing. In this trial, the median PFS was 2.9 months and 27% of patients have PFS2/PFS1 ratio ≥ 1.3. Our study observed very similar results, but in contrast to this study we did not observe increased benefit of target therapies in breast cancer [9]. MOSCATO prospective trial also using a gene panel, showed a clinical benefit of the precision medicine strategy [12]. More than 1000 patients were included, with a new biopsy mandated by the protocol. Most tumor samples presented an actionable molecular alteration, and 199 patients were treated with a targeted therapy matched to a genomic alteration, mostly by inclusion in phase I/II trials. The PFS2/PFS1 ratio was >1.3 in 33% of the patients. This study underlined objective responses and improved overall survival with the matched treatment. Similarly, a US large-scale, prospective clinical sequencing imitative using MSK-IMPACT gene panel, performed matched tumor and normal sequences from a cohort of more than 10,000 patients with advanced cancer. 37% of patients harbored a clinically relevant alteration, and 11% were subsequently enrolled on a genomically matched clinical trial [1]. Interestingly, the top 3 mutated genes were also TP53 (41% of patients) KRAS (15%) and PIK3CA, consistent with our findings and confirming the role of these pathways in cancer biology. However, these promising results were not confirmed in a randomized trial. In particular, the SHIVA trial attributed randomly to patients with a molecular alterations detected using a gene panel, focused on three mains pathways (hormone receptor, PI3K/AKT/mTOR, RAF/MEK) a matched molecularly targeted agent or a treatment at physician's choice [13]. No median PFS difference was highlighted similarly to what was found in our study.

In contrast to gene panels, WES is less used in precision medicine. In 2015, Beltran et al. enrolled prospectively 97 patients to analyze both tumor and normal tissue using WES [14]. WES provided informative results in 91 cases (94%), but matched therapy was used in only 5 patients (5%).

Most, trials suggested that archived formol fixed paraffin embed sections could induce DNA fragmentation and mutations which may flow the results [15]. Tumor heterogeneity and tumor evolution during therapies may also flow results. Accordingly, fresh tumor biopsy is often recommended in most trials. In addition to a better estimate of tumor heterogeneity, multiple biopsies or biopsy of progressive metastases are recommended. Another strategy is to use deep sequencing using panels to better estimate heterogeneity. In this study, to add exome analysis to routine care we have used the available archived paraffin embedded tumors because some patients may refuse biopsies or such additional biopsy would have been difficult or dangerous. Tumor DNA extraction and NGS analysis using archived biopsies was only not possible in 50 cases (9.9%), mostly due to insufficient DNA quantity or quality, thus suggesting approach feasibility. Mutation frequency and clinical results are very similar to previous studies, validating the efficacy of exome approach on archived paraffin embedded sections. In addition, this strategy permits to find new genetic cancer predisposition which improve patients and relatives care. However, no differences in term of PFS or PFS2/PFS1 ratios were observed between NGS-based therapy and non-NGS-based therapy group in this study. So, like previous studies, we have failed to demonstrate the superiority of precision medicine in comparison to classical treatment. A major problem in this study is its real life design. For instance, a significant part of patients was heavily pretreated (60% of the patients had at least 3 lines of therapy), and some were not further treated despite available therapeutic propositions following NGS analysis due to physical impairment. Notably 33% of patients with disease progression and therapeutic proposal did not receive further therapy. These data support the conclusion that such strategy is not adapted to heavily pretreated patients with risk of rapid performance status deterioration. Likewise, in the recent I-PREDICT trial, patients were treated with combo-therapies of target agents. In this trial substantial numbers of patients dropped off from target therapies, mostly due to disease deterioration with hospice placement or demise [16]. Therefore, we believe that precision medicine approaches should be initiated earlier in the course of the disease. During study follow-up in our structure, we felt the need to adjust the molecular tumor board, proposing early patients’ inclusion and to stimulate earlier treatment with target therapies.

The recent I-PREDICT [16] and WINTHER [17] trials invest the matching score strategies to improve patients selection. These studies raise the intuitive hypothesis that the more the therapy target mutations the more effective is the therapy. However, in EXOMA study of the matching score failed to demonstrate that patients treated with drugs that target more that 50% of targetable hits gain a benefit from this strategy. A major difference between I-PREDICT and EXOMA is that few patients in this trial received combo-therapies of target agents that may improve benefit from treatment.

The pitfall of exome sequencing is the absence of translocation and fusion detection. Despite this drawback, these results are similar to panel sequencing in terms of clinical efficacy. Probably RNA sequencing may be required in the future to assess fusion. In addition, such strategy could be helpful to take into account stromal microenvironment information and expression data [17]. In addition, determination of mutational signature should be implemented to help decision [18]. In conclusion, this trial contributes to assess the feasibility of tumor genomic profiling in routine care. However, patient treated with target therapies did not get clinical benefit from this strategy. The disappointing results of this study underline our superficial understanding of the mechanisms in cancer evolution and cancer heterogeneity. It also highlights the fact that even with the existing targeted agents, if our comprehension of the interactions in cancer cells, their evolution the interpatient and intra patient heterogeneity is not deeper, we may not improve current cancer treatment with precision medicine. In a technical point of view, the archived paraffin embedded tumors set a limit in the prospective nature of this trial because could they induce DNA fragmentation and mutations. Probably multiomics strategies and the incorporation of novel technologies like RNA-sequencing, whole genome sequencing and circulating cell-free DNA detection should be emphasized for future studies in order to estimate the possibility of novel targets and potential agents for these targets. Alternatively, exploration of particular “extraordinary response” or surprising failure of target therapies may help us to better understand cancer biology. All these efforts are important to improve Cancer Precision Medicine.

Declaration of competing interest

Other authors declare no relevant conflict of interests related to this publication.