Clinical and molecular characterization of patients with cancer of unknown primary in the modern era

Abstract

Background

On the basis of historical data, patients with cancer of unknown primary (CUP) are generally assumed to have a dismal prognosis with overall survival of less than 1 year. Treatment is typically cytotoxic chemotherapy guided by histologic features and the pattern of metastatic spread. The purpose of this study was to provide a clinical and pathologic description of patients with CUP in the modern era, to define the frequency of clinically actionable molecular alterations in this population, to determine how molecular testing can alter therapeutic decisions, and to investigate novel uses of next-generation sequencing in the evaluation and treatment of patients with CUP.

Patients and methods

Under Institutional Review Board approval, we identified all CUP patients evaluated at our institution over a recent 2-year period. We documented demographic information, clinical outcomes, pathologic evaluations, next-generation sequencing of available tumor tissue, use of targeted therapies, and clinical trial enrollment.

Results

We identified 333 patients with a diagnosis of CUP evaluated at our institution from 1 January 2014 through 30 June 2016. Of these patients, 150 had targeted next-generation sequencing carried out on available tissue. Median overall survival in this cohort was 13 months. Forty-five of 150 (30%) patients had potentially targetable genomic alterations identified by tumor molecular profiling, and 15 of 150 (10%) received targeted therapies. Dominant mutation signatures were identified in 21 of 150 (14%), largely implicating exogenous mutagen exposures such as ultraviolet radiation and tobacco.

Conclusions

Patients with CUP represent a heterogeneous population, harboring a variety of potentially targetable alterations. Next-generation sequencing may provide an opportunity for CUP patients to benefit from novel personalized therapies.

Introduction

More than 80 000 people in the United States are diagnosed annually with cancer of unknown primary (CUP), defined by the National Comprehensive Cancer Network (NCCN) as ‘histologically proven metastatic malignant tumors whose primary site cannot be identified during pretreatment evaluation’ [1, 2]. Historically, this disease portends a poor prognosis with median overall survival (OS) of less than 1 year [3]. There is growing interest in applying novel molecular methodologies to allow more personalized and targeted treatments in cancer in general and this disease in particular.

Most molecular research in CUP has focused on identifying a primary disease site using gene expression patterns, and several commercial assays are in use [4]. The frequency with which their use results in treatment changes or whether their use improves outcomes remains unclear, and their routine use is not recommended [2].

Tumor genetic testing to identify clinically actionable mutations, increasingly using next-generation sequencing (NGS), is now part of standard management of patients with many cancers. Two descriptive studies have examined NGS in patients with CUP. In one small study, 12/16 patients were deemed to have potentially actionable alterations after 701 genes were examined using NGS [5]. In another study of 200 tumors from patients reportedly with CUP, analysis using the FoundationOne^® (Cambridge, Massachusetts, USA) NGS assay revealed that 169/200 had potentially actionable alterations [6]. Major limitations of these prior studies include the limited clinical follow-up data, the broad definition of ‘potentially actionable’, and more specifically the failure of prior studies to stratify individual genetic alterations on the basis of varying levels of evidence supporting targeted therapies.

Additionally, most studies that defined the prognosis and therapeutic paradigms of CUP were conducted a considerable time ago. The sensitivity and accuracy of diagnostic imaging and the number of therapeutic options available have increased since much of this literature was written. The purpose of our study was to provide a clinical and pathologic description of a modern cohort of patients with CUP, to define the frequency of potentially actionable molecular alterations using NGS, to determine how molecular testing affected therapeutic decisions and outcomes in this population, and to investigate the novel uses of NGS to personalize treatments.

Methods

Patient inclusion

This cohort included patients evaluated at the Memorial Sloan Kettering Cancer Center (MSKCC) between January 2014 and June 2016 based on International Classification of Diseases (ICD)-9 and ICD-10 billing codes for CUP. From these patients, a comprehensive chart review was carried out to exclude patients who had a confirmed primary disease site if immunohistochemical and histologic evaluations of tumors and imaging were consistent with a primary disease site. Patients with tumors with the histologic diagnosis of sarcoma and melanoma were excluded. Those remaining patients without a confirmed primary site were included in the analysis. Demographic and clinical information regarding sex, race, age, survival, treatment, pathology, and molecular testing were obtained. Data cut-off was 1 May 2017. Additionally, it was determined that patients who underwent NGS were identified as having CUP after a comprehensive chart review with the same criteria described above.

Biostatistical analysis

OS was measured from the date of pathologic diagnosis to date of death. Survival rates and median OS were estimated using the Kaplan–Meier method. Log rank test was used to determine survival difference between groups.

Genomic analysis and interpretation

Based on treating physicians’ recommendations, patients were referred for molecular testing using archival tumor tissue. NGS was carried out within the context of an Institutional Review Board–approved protocol using MSK-IMPACT (NCT01775072), a deep-coverage hybridization capture-based assay encompassing 341 (later expanded to 410) cancer-associated genes. MSK-IMPACT can detect missense mutations, indels, copy number alterations, and select gene fusions [7, 8]. Analysis of tumor and normal (typically from blood) DNA was carried out to allow for accurate somatic mutation calling. Determination of clinical actionability and the level of evidence that a particular molecular alteration was predictive of drug response were characterized using OncoKB, a curated knowledge base of the functions and treatment implications of somatic mutations (http://oncokb.org) [9].

For whole exome sequencing (WES), DNA sequence libraries for six tumor and matched normal samples from patients with no oncogenic driver mutations were recaptured using the Agilent SureSelect Exome Kit (v3) and sequenced on an Illumina HiSeq 2500^® to an average coverage of >200×. Sequencing reads were aligned to the human genome (hg19) and analyzed as described previously [10]. Candidate mutations were filtered for various criteria including recurrence in historical normals, total and allele-level coverage, and known systematic artifacts. All remaining mutation calls were manually reviewed to retain only high confidence variants. The WES data analysis pipeline can be found at https://github.com/soccin/BIC-variants_pipeline. Germline variants were called and assessed for pathogenicity as described previously [11].

For RNA sequencing, formalin-fixed paraffin-embedded tissue blocks from the same tumors analyzed by MSK-IMPACT were obtained for RNA extraction and sequencing. Hematoxylin and eosin–stained slides were reviewed for tumor cellularity. Using the TruSight RNA Pan-Cancer Panel (Illumina^®), 250 ng of extracted total RNA was prepared and sequenced on an Illumina HiSeq 2000^® as paired-end 100-bp reads. The panel consisted of 1385 key cancer genes for gene expression, variant calling, and fusion detection with known and novel gene fusion partners. Sequencing reads were mapped to the human hg19 genome using STAR aligner [12]. Somatic fusion calling was carried out using two fusion detection software tools: FusionCatcher and JAFFA [13, 14].

Results

Clinical characteristics for patients with CUP

Between January 2014 and June 2016, 333 patients with CUP were evaluated at MSKCC. Mean age at diagnosis was 65 years, and the majority had an Eastern Cooperative Oncology Group (ECOG) performance status (PS) of 0–1 as documented at the time of their initial consultation (Table 1). 38% of patients were seen at MSKCC for two or fewer visits (Table 1). Adenocarcinoma (46%) was the most frequent histology (Table 2). Median OS was 13 months and varied across different histologies (Table 2). Patients with ECOG PS 0 had the longest OS (33 months), and those with ECOG PS 3–4 had the shortest OS (3 months) (Table 2).

Table 1.

Clinical characteristics of patients with cancer of unknown primary seen at Memorial Sloan Kettering Cancer Center between 1 January 2014 and 30 June 2016

	N	%
Sex
Female	158	47
Male	175	53
Race
White	266	79
Black	19	6
Asian, Native American, or Alaska Native	16	5
Unknown	32	10
Mean age at diagnosis (range)	65 (26–91)
Year diagnosed
2014	146	44
2015	137	41
2016	50	15
Survival status
Alive	168	50
Dead	165	50
Care at MSKCC
≤ 2 visits at MSKCC	128	38
> 2 visits at MSKCC	205	62
ECOG performance status
0	82	25
1	144	43
2	33	10
3/4	27	8
Unknown	47	14
Total	333	100

MSKCC, Memorial Sloan Kettering Cancer Center; ECOG, Eastern Cooperative Oncology Group.

Table 2.

Median overall survival for patients by histology and Eastern Cooperative Oncology Group Performance Status

	N (%)	Deaths	Median OS (months)	95% CI	Number with MSK-IMPACT (%)
Histology
Adenocarcinoma	153 (46)	86	11	8.83 to 14.52	64 (43)
Carcinoma NOS	81 (24)	39	13	9.16 to 19.93	48 (32)
Squamous cell carcinoma	53 (16)	15	33	25.48 to NA	13 (9)
Neuroendocrine cancer	20 (6)	12	9	7 to NA	12 (8)
Neoplasm NOS	12 (4)	6	13	4.17 to NA	6 (4)
Carcinoma with sarcomatoid features	8 (2)	5	5	4.07 to NA	2 (1)
Other	6 (2)	2	17	8.47 to NA	5 (3)
Total	333 (100)	132	13		150 (100)
ECOG performance status
0	82 (25)	24	33	20.89 to NA
1	144 (43)	72	12	9.92 to 18
2	33 (10)	19	8	3.09 to 16.78
3/4	27 (8)	21	3	1.64 to 8.14
Unknown	47 (14)
Total	333 (100)

CI, confidence interval; NA, not available; NOS, not otherwise specified; OS, overall survival; ECOG, Eastern Cooperative Oncology Group.

Molecular characterization of patients with CUP

One hundred and fifty patients (45%) underwent MSK-IMPACT [8]. Average unique sequence coverage was 732× (range 166× to 1398×). The average number of non-silent mutations per sample was 9.3 (range 0–99). The most commonly mutated genes were TP53, KRAS, CDKN2A, KEAP1, and SMARCA4 (Figure 1). Using the OncoKB knowledge base to classify each alteration for clinical actionability, 137/150 (91%) patients harbored at least 1 alteration with a known or likely oncogenic function, with an average of 3.1 oncogenic alterations per tumor (range 0–9). However, the fraction of patients with clinically actionable mutations was smaller.

Figure 1.

Molecular alterations seen in cancers of unknown primary.

Potentially actionable mutations are assigned by the OncoKB classification to one of the following four levels based on the level of evidence that a specific molecular alteration is predictive of drug response: level 1 alterations are FDA-approved biomarkers for particular drugs for a certain indication, level 2 alterations are FDA-approved biomarkers for particular drugs in another indication, level 3 alterations include those for which clinical evidence exists to link the alteration to a drug response for another indication, and level 4 alterations are those for which preclinical evidence exists to link the alteration to a drug response. Since the primary site was unknown, no level 1 alterations were present. Twenty-seven level 2 alterations were identified in 23 tumors, the most common of which were ERBB2 amplification and BRAF V600E mutation; 27 level 3 alterations were identified in 25 tumors, the most common of which were PIK3CA mutations (Table 3). Forty-five tumors harbored at least one clinically actionable level 2 and/or level 3 alteration. Thirty-two level 4 alterations were identified in 38 tumors, the most common of which were KRAS mutations. Of 105 patients without a potentially clinically actionable level 2 or level 3 alteration, 13 had no genetic alterations identified.

Table 3.

Level 2 and 3 alterations in CUP

Alteration level	Alteration seen	N	Number receiving targeted therapy	Time on treatment (months)	Histology
Level 2	BRAF V600E	6	4	<1 to 5	AdenoCa (3)
					Ca NOS (2)
					Squamous cell Ca (1)
	BRCA1 G1567*	1			AdenoCa
	BRCA1-ELAC2 fusion	1			Ca NOS
	BRCA2 E731Gfs*19	1			AdenoCa
	BRCA2 deletion	1			Neuroendocrine ca
	CDK4 amplification	2			Ca NOS (1)
					Squamous cell Ca (1)
	ERBB2 amplification	7	1	5	AdenoCa (4)
					Ca NOS (3)
	GOPC-ROS1 fusion	1			AdenoCa
	KIF5B-ALK fusion	1	1	5	Ca NOS
	MET amplification	2			AdenoCa (1)
					Ca NOS (1)
	MET D1010N	1			Ca NOS
	NCOA4-RET fusion	1	1	7	AdenoCa
	TSC1/2 alteration	2a			Ca NOS (1)
					Other (1)
	Total level 2 alterations	27	7
Level 3	AKT1 E17K	3	1	4	AdenoCa (1)
					Ca NOS (1)
					Neuroendocrine Ca (1)
	ERBB2 Y772_A775DUP	1	1	Unknown	AdenoCa
	ERBB2 S310F/Y	2	1	2	AdenoCa(1)
					Ca NOS (1)
	FGFR1 amplification	2			AdenoCa (1)
					Ca NOS (1)
	FGFR2/3 fusion	4b	1	8 (ongoing)	AdenoCa (3)
					Ca NOS (1)
	IDH2 R172W	1			AdenoCa
	IRF2BP2-NTRK1 fusion	1	1	14 (ongoing)	Ca NOS
	MDM2 amplification	1			Squamous cell Ca
	PIK3CA mutations	10c	1	2	AdenoCa (6)
					Ca NOS (2)
					Neuroendocrine Ca (1)
					Squamous cell Ca (1)
	PTCH1 alteration	2d			Basal cell Ca (1)
					Neuroendocrine Ca (1)
	Total level 3 alterations	27	6

^aOne with TSC1 Y185* and one with TSC2 X212_splice.

^bOne of each of the following: KCNH7-FGFR2 fusion, KIF14-FGFR2 fusion, INA-FGFR2 fusion, and WHSC1-FGFR3 fusion.

^cFour with PIK3CA H1047R, two with PIK3CA E542K, one with each of the following: PIK3CA N1044K, PIK3CA Q546R, PIK3CA R88Q, and PIK3CA P539R.

^dOne with each of the following: PTCH1 G1093*, PTCH1 G492*.

Ca, carcinoma; NOS, not otherwise specified.

For patients with tumors harboring ERBB2 amplifications, immunohistochemical staining and imaging studies were not suggestive of a breast or gynecologic primary diagnosis. One patient did have a history of breast cancer and vulvovaginal Paget’s disease; however, a primary site could not be determined even after comparison of her current disease to her prior breast cancer and vulvovaginal Paget’s disease specimens. For those four patients with FGFR2/3 fusions, there was little evidence to suggest a diagnosis of cholangiocarcinoma. Specifically, one had no evidence of liver disease and the other three had widespread metastatic disease involving multiple organs and the absence of a dominant liver mass.

Whole exome and RNA sequencing of patients with CUP with no known driver identified

For the 13/150 tumors with no clear oncogenic driver alteration identified by MSK-IMPACT, WES was carried out on tumors with adequate tissue and tumor purity. WES of 5 tumors revealed a median of 23 non-synonymous protein-coding mutations (range 12–41). Computational analyses identified one tumor with a VSTM1 A33fs mutation, a gene involved in leukocyte differentiation and implicated in leukemia pathogenesis, and a PTPRA X302 splice site alteration, a gene implicated in EGFR signaling and a molecular alteration seen in gastric and lung cancers [15–18]. Both mutations were truncating and accompanied by deletion of the wild-type allele, suggesting bi-allelic somatic inactivation of both genes. A second tumor had a RIMS2 R1188C mutation, a codon that is mutated in esophageal, thyroid, uterine, and lung cancers, although of unclear clinical significance [17, 18]. Assessment of 7/13 patients who consented for germline analysis revealed inherited pathogenic variants in two cases: MUTYH R274Q and CHEK2 T367Mfs*15. RNA-Seq analysis of an additional tumor retrospectively revealed a BRD4-NUT fusion, identified in the majority of patients with NUT midline carcinoma. Preclinical data suggest this fusion may confer sensitivity to bromodomain inhibitors (NCT01587703, NCT02630251, NCT02698176, NCT02419417, and NCT02259114). Taken together, these results suggest the potential benefit of comprehensive molecular profiling and integrated somatic and germline analysis in this population, although this hypothesis remains to be tested.

Treatment outcomes for patients with potentially targetable alterations in CUP

Of 150 patients with tumors who underwent MSK-IMPACT, 45 had tumors with a total of 54 potentially actionable mutations (27 level 2 alterations in 23 total patients and 27 level 3 alterations in 25 patients). Of these 45 patients, 13 received targeted therapies. For those 23 patients with tumors harboring at least one level 2 alteration, 7 received targeted therapies matched to the genomic alteration (Table 3). These patients received targeted therapies previously shown to be active in patients with BRAF V600E mutations, ERBB2 amplification, KIF5B-ALK fusion, and NCOA4-RET fusion (Table 3). For those 25 patients with at least one level 3 alteration, 6 received targeted therapies, including 1 with each of the following alterations: AKT1 E17K mutation, ERBB2 Y772_A775 duplication, ERBB2 S310Y mutation, INA-FGFR2 fusion, IRF2BP2-NTRK1 fusion, and PIK3CA E542K mutation (Table 3). Many patients with potentially targetable alterations did not receive targeted therapies for a variety of reasons (Table 4). Of note, an additional two patients received targeted therapies for molecular alterations of unclear significance, one with an ERBB2 V697L mutation who received a kinase inhibitor targeting ERBB2 and another with a BRCA2 C2332F mutation treated with a PARP inhibitor.

Table 4.

Reasons why patients with potentially targetable alterations did not receive targeted therapy

Reason for not receiving targeted therapy	N
No evidence of disease, on observation, or receiving and responding to first-line therapy	10
Poor performance status/Ineligible for trial	11
No FDA-approved drug or basket trial available at the time of MSK-IMPACT testing	2a
Not offered	3b
Not receiving care at MSKCC	6
Total	32

^aTwo patients had CDK4 amplifications identified on MSK-IMPACT testing; however, both patients died before the CDK4 inhibitor palbociclib was approved and no appropriate basket study was available at the time of result return.

^bOne patient with MET amplification, one patient with MET D1010N mutation, and one with a BRCA2 deletion.

For 15 patients who received a targeted therapy, results were variable, with time to treatment failure (TTF) ranging from less than 1 month to 14 months, with several patients remaining on targeted therapy at the time of data cut-off (Table 3). Of the four patients with tumors harboring BRAF V600E who received therapy, TTF ranged from less than 1 month to 5 months; for one patient with a tumor harboring ERBB2 amplification, ERBB2-directed treatment was continued for 5 months. Patients benefiting the longest from targeted therapy in this cohort had tumors with targetable genomic rearrangements, including a KIF5B-ALK fusion (5 months), NCOA4-RET fusion (7 months), INA-FGFR2 fusion (8 months, ongoing), and IRF2BP2-NTRK1 fusion (14 months, ongoing) (Table 3).

Mutation signatures identified using NGS

Finally, we explored the potential diagnostic and therapeutic implications for of additional genomic features. We identified a dominant mutation signature in 21 tumors, using the class and flanking nucleotide context of all observed silent and non-silent mutations (Figure 1) [19]. This included seven tumors with evidence of tobacco-related damage, all in patients who were current or former smokers, and six tumors with a signature of ultraviolet radiation, most commonly associated with melanoma. Given the proven efficacy of immunotherapy in lung cancer and melanoma, these mutation signatures could potentially lend support to immunotherapy for these patients [20, 21].

Discussion

This is the largest clinical and molecular description of patients with CUP in the modern era with accompanying detailed chart review and molecular classifications. Historical data suggest that OS ranges from 3 to 8 months; however, our data note OS of 13 months. This favorable outcome must be interpreted in the context that this series was a single-institution cohort, including patients fit enough for consultation at a tertiary care center and with the majority of patients presenting with good PS. Patients with squamous cell CUP had the longest median OS; however, 33/53 of these tumors represented patients with disease limited to the head and neck and who were treated definitively with chemotherapy and radiation therapy, likely reflecting a less aggressive and potentially curable population. The median survival of the other 20 squamous cell patients was 14 months. Thirty percent of this cohort were patients with carcinoma NOS, neoplasm NOS, and other histologies, likely reflecting a referral bias to a tertiary care center of patients with nonspecific and challenging pathologic diagnoses.

Prior studies had suggested that the majority of CUPs have potentially actionable alterations. Eighty-five percent of tumors from 200 patients reportedly with CUP whose tumors underwent the FoundationOne^® assay were reported to have an actionable alteration[6]. A major difference in this prior analysis was the use of an especially broad definition of actionability that failed to consider the level of evidence supporting the use of a molecular event as a predictive biomarker, thus overstating the practical clinical utility of genomic profiling in patients with CUP. For example, no distinction was made between alterations such as BRAF V600E, which is a standard of care biomarker for an FDA-approved drug, and mutations in the RAS family of genes, an oncogenic target for which no effective therapy is currently available. Using a more restrictive definition of actionability as defined here, we found that potentially actionable alterations are present in tumors from patients with CUP but less frequently than previously reported. Applying the same criteria for actionability in this paper to the alterations identified in the 200 patients with CUP tested by the FoundationOne^® assay, we identified level 2 or 3 actionable alterations in 34% of tumors, consistent with the findings reported there.

A minority of patients with potentially actionable alterations received a targeted agent, owing to several factors. Many were unable to receive investigational therapies due to poor PS or rapid clinical decline. Others did not pursue treatment at our institution, and limited records were available to confirm whether a patient received targeted therapy, a limitation of this analysis from a tertiary care center. Given that CUP is not an indication for any targeted agent and that targeted therapies can cost $10 000 or more per month, the expense of these medications remains a barrier for many patients.

Our results suggest that patients with CUP may potentially benefit from access to targeted therapies and basket trials in which eligibility is based on molecular features rather than primary disease site. For the minority of patients who received targeted therapies, the patients receiving the most benefit were those with tumors harboring genomic rearrangements, consistent with promising outcomes from recent trials targeting genomic rearrangements [22, 23]. These results support the potential utility of broader DNA- and RNA-based profiling technologies capable of detecting clinically significant gene fusions.

Although relatively few patients had their treatment altered as a result of molecular testing and achieved more than 6 months of tumor control, the total benefit of genomic profiling remains hard to quantify in this population. NGS can lead to identification of a likely primary site for some patients who initially appear to have CUP. For instance, identification of genetic alterations seen recurrently in cholangiocarcinoma such as FGFR1/2 fusions with novel partners and mutations in BAP1, IDH1, and IDH2 in the presence of a dominant intrahepatic mass can clarify a diagnosis of cholangiocarcinoma and facilitate entry into molecularly selected clinical trials for patients with cholangiocarcinoma. In this cohort, we excluded those patients with a dominant hepatic mass and pathologic review consistent with cholangiocarcinoma. Among the six patients with BRAF V600E mutations, one had a dominant liver lesion that could be consistent with a diagnosis of cholangiocarcinoma; however, pathology revealed a high-grade carcinoma with squamous features and immunohistochemical staining negative for CK7, an unusual pattern for cholangiocarcinoma. Of the remaining five patients, two had no liver involvement, two had widely metastatic disease including multiple liver lesions measuring between 1 and 2.5 cm, and one did have extensive hepatic involvement of her disease with multiple metastatic lesions throughout her liver which could be consistent with cholangiocarcinoma but could also be consistent with another unknown primary with extensive hepatic metastases.

Beyond the identification of a targeted genetic event, the identification of mutation signatures associated with ultraviolet light and tobacco would suggest that immunotherapy should be considered for such patients, given the dramatic and often sustained responses to immunotherapy in lung cancer and melanoma [24–27]. Further research is needed to determine whether molecular signatures can be used to select patients for treatments with immunotherapy independent of site of origin.

The complexity and depth of information garnered with NGS allows for further analysis of CUP tumors beyond currently characterized mutation signatures. A more complex, bioinformatics-based classifier using additional features such as broad and focal copy number alterations and recurrent driver mutations could help clarify primary disease site, results that could be further explored with additional directed immunohistochemical or imaging analysis. A focused immunohistochemical evaluation could also allow for judicious use of the small biopsy specimens available in this population. How the benefit of extensive immunohistochemical characterization of CUP compares to genomic characterization for directing oncologic therapy is the subject of ongoing research at our institution.

The challenges in pursuing a personalized medicine paradigm for patients with CUP are many. While patients and clinicians continue to obtain increasing amounts of information regarding the presence of key molecular alterations in these tumors, in only a minority of CUP patients are actionable alterations identified that can meaningfully change treatment opportunities given our current therapeutic armamentarium. Even fewer of these patients ultimately receive targeted therapies. However, for some patients, these targeted therapies have provided sustained disease stability and shrinkage, warranting broader testing using NGS, particularly among those patients either fit enough for a clinical trial with an investigational agent or able to survive the arduous and time-consuming process of receiving an investigational medication through compassionate use. Our understanding of the use of molecular data in this field is growing, and our ability to act on these results is rapidly evolving. The presence of defined mutation signatures and novel alterations continue to provide opportunities to explore new treatments and potentially improved outcomes for patients with CUP.