symposium article
Annals of Oncology 23 (Supplement 10): x271–x277, 2012
doi:10.1093/annonc/mds357
Diagnostic work-up of carcinoma of unknown primary:
from immunohistochemistry to molecular profiling
K. A. Oien1 & J. L. Dennis2
University of Glasgow; Institute of Cancer Sciences, Glasgow; 2St George’s, University of London; Medical Biomics Centre, London, UK
Carcinoma of unknown primary (CUP) remains a common and challenging clinical problem. The aim of diagnostic
work-up in CUP is to classify as specifically as possible the cancer affecting the patient, according to the broad tumour
type, subtype and, where possible, site of origin. This classification currently best predicts patient outcome and guides
optimal treatment. A stepwise approach to diagnostic work-up is described. Although pathology is based on
morphology, the assessment of tissue-specific genes through immunohistochemistry (IHC) substantially helps tumour
classification at each diagnostic step. For IHC in CUP, recent improvements include more standardised approaches
and marker panels plus new markers. Tissue-specific genes are also being used in CUP work-up through molecular
profiling. Large-scale profiles of hundreds of tumours of different types have been generated, compared and used to
generate diagnostic algorithms. Commercial tests for CUP classification have been developed at the mRNA and
microRNA (miRNA) levels and validated in metastatic tumours and CUPs. While currently optimal pathology and IHC
remain the ‘gold standard’ for CUP diagnostic work-up, and full clinical correlation is vital, the molecular tests appear to
perform well: in the main diagnostic challenge of undifferentiated or poorly differentiated tumours, molecular profiling
performs as well as or better than IHC.
Key words: cancer, immunohistochemistry, metastatic, molecular profiling, pathology, unknown primary
introduction
This paper focusses on the contribution of
immunohistochemistry (IHC) and molecular profiling to the
diagnostic work-up of carcinoma, or cancer, of unknown
primary (CUP) in pathology. This paper is for the 2012
conference of the European Society of Medical Oncology
(ESMO). It is, therefore, aimed primarily at medical
oncologists, but may also be of value to pathologists, other
clinicians and patients.
CUP is a common and important clinical problem as
discussed in recent clinical reviews [1, 2], guidelines [3, 4]
and conferences [5]. Approximately 15% of all cancers first
present, that is, cause symptoms, with metastases rather than
with the primary tumour. In approximately two-thirds of
these, the primary tumour becomes obvious early on during
investigation [6]. The remaining cases are CUP; in a
minority a primary site will be identified over time. Biopsies
for pathology may be taken at any stage in the diagnostic
process, from initial presentation or later, once other
investigations have been negative: diagnostic difficulty
generally increases in the latter.
*Correspondence to: Dr K. A Oien, Cancer Research UK Beatson Laboratories, Institute
of Cancer Sciences, College of Medical Veterinary and Life Sciences, University of
Glasgow, Garscube Estate, Switchback Road, Glasgow G61 1BD, UK. Tel: +44-141330-3506; Fax: +44-141-330-4127; E-mail: [email protected]
pathological aim in CUP: cancer
classification by tumour site and type
Cancer classification has traditionally been based on
anatomical location(s) and tumour morphology [6]; these have
been the best available guide for cancer patient management.
Almost all biomarkers used in pathology for CUP are aimed at
establishing cancer type, subtype and site, thus are diagnostic:
this includes IHC and molecular profiling. As more specific
therapies emerge, prognostic and predictive biomarkers may
also become important in CUP.
In CUP, the most common sites of metastasis and thus of
biopsy are: solid organs including liver, lung, bone and brain;
lymph nodes especially cervical, inguinal and axillary; and
peritoneal and pleural serous cavities [1, 2]. In pathology, once
a biopsy is obtained, the presence of malignancy must first be
confirmed. Thereafter, a stepwise approach can be taken to
identify the broad tumour type, then tumour subtype, and, if
adenocarcinoma, the likely site of origin, as shown in Table 1.
At each step, IHC can help [6–9].
The broad tumour type for almost all ‘true’ metastatic CUPs
is carcinoma. CUP is a diagnosis of exclusion, since many
studies exclude other tumour types including lymphoma,
melanoma and sarcoma, as well as unusual primary rather
than metastatic tumours [1, 2]. Because these other cancer
types nevertheless often enter the clinical and pathological
differential diagnosis, they must still be considered.
Within carcinoma, the most common subtypes in CUP are
adenocarcinoma (60%), squamous carcinoma (5%),
© The Author 2012. Published by Oxford University Press on behalf of the European Society for Medical Oncology.
All rights reserved. For permissions, please email: [email protected].
symposium
article
1
symposium article
Annals of Oncology
Table 1. Cancer classification by tumour type, subtype and primary site; and useful immunohistochemistry (IHC)
Stepwise approach to CUP work-up
Step 1: identify broad cancer type
Carcinoma
Melanoma
Lymphoma/leukaemia
Sarcoma
(Neuro-glial tumours)
Step 2: if carcinoma or related, then identify its subtype
Adenocarcinoma
Squamous carcinoma
(Transitional cell carcinoma)
Neuroendocrine carcinoma
Solid organ carcinoma: kidney (renal)
Solid organ carcinoma: liver (hepatocellular)
Solid organ carcinoma: thyroid
Solid organ carcinoma: adrenal
(Germ cell tumour)
(Mesothelioma)
Step 3: If adenocarcinoma, then predict possible primary site(s)
e.g. lung, pancreas, colon, stomach, breast, ovary, prostate
Useful immunohistochemical biomarkers
Cytokeratins and other epithelial markers, e.g. AE1/3, CK7, CK20, CK5, EMA
S100, Melan-A, HMB45
CLA, CD20, CD3, CD138, CD30 etc.
Vimentin, actin, desmin, S100, c-kit etc.
(Specific markers)
CK7, CK20, PSA plus other adenocarcinoma markers
CK5, p63
(TCC: CK7&20, urothelin)
Chromogranin, CD56, synaptophysin, TTF-1
RCC, CD10, PAX8, Napsin A
Hepar1, CD10, glypican-3
TTF-1, thyroglobulin, PAX8
Melan-A, inhibin
OCT4, PLAP, HCG, AFP
Calretinin, mesothelin, WT1, D2-40
See separate table
AFP, alpha fetoprotein; CLA, common leucocyte antigen; EMA, epithelial membrane antigen; HCG, human chorionic gonadotrophin; PAX8, paired box
gene 8; RCC, renal cell carcinoma marker; WT1, Wilms' tumour protein.
Based on [6–9]. The markers in bold are especially useful. For tumour diagnosis, IHC results must be integrated with the morphology, other IHC markers
and clinical findings: for example, CK20 is positive in colorectal adenocarcinoma but also in transitional cell carcinoma and Merkel cell carcinoma.
neuroendocrine carcinoma (5%) and poorly differentiated
carcinoma (30%); the latter is often grouped with
adenocarcinoma, and together they make up ∼90% of CUPs
[1, 2, 6]. Neuroendocrine carcinoma comprises both poorly
differentiated tumours, including small cell carcinoma, and
well differentiated neuroendocrine tumours, including the old
category of carcinoid tumour. Other carcinoma subtypes
include carcinomas of solid organs, including hepatocellular,
renal, adrenal and thyroid; and transitional cell carcinoma
which is often grouped broadly with squamous carcinoma.
Related tumours, which may appear similar to carcinoma,
include germ cell tumours and mesothelioma [6].
For adenocarcinoma, the most common primary sites in
CUP are lung and pancreas (both ∼25%), then colon, stomach
and oesophagus, breast, ovary and prostate [1, 2], as shown by
clinical follow-up or autopsy. Certain metastatic sites are more
likely to harbour metastases from particular primary sites,
which can aid diagnosis; and this enables CUPs to be divided
into good and poor prognosis categories: for example,
compared with CUPs overall, CUP in loco-regional lymph
nodes has a better prognosis and CUP with multiple liver
metastases generally has a worse prognosis, as detailed in
clinical reviews [1, 2].
Specific treatment is possible for certain CUPs, including
(lymphoma,) neuroendocrine carcinoma, colorectal
adenocarcinoma and other ‘good prognosis’ CUPs [1, 2]. Probably
the pathologist’s main aim in CUP diagnostic work-up is optimal
tumour classification, to enable the oncologist to identify patients
with treatable and/or good prognosis tumours. As new therapies
emerge, so the important tumour classes to identify may change:
our diagnostic work-up therefore needs to be flexible.
x | Oien and Dennis
CUP biomarkers: specific tissues
and issues
Before applying biomarkers in CUP, let us consider what these
biomarkers represent. Cancer classification is based on the
differences in the appearance of different cancers and on their
resemblance to the corresponding normal tissues. Tissues are
aggregates of cells of similar type and function. Differences
between tissues, normal or malignant, are based on the
differences in their gene expression. The human genome
contains ∼30 000 protein-coding genes [10]. Approximately
12 000 genes are active, i.e. expressed at mRNA and protein
levels, in each tissue [10]. Gene expression depends on the
underlying DNA sequence and is regulated at multiple levels
including epigenetic and via microRNAs (miRNAs).
Of the 12 000 genes active in each tissue, 8000 are widely
expressed and involved in basic cellular functions, e.g. protein
and energy production or cell proliferation. A subset of active
genes is specific to one or a few tissue types, related to its
mature function, i.e. its differentiation. Such tissue-specific or
tissue-restricted genes are often regulatory genes or protein
products [10]. Regulatory genes include transcription factors,
especially homeobox genes controlling tissue development and
maintenance, e.g. thyroid transcription factor-1 (TTF-1) in
lung and thyroid. Protein products may be secreted or
expressed in or on the cell and include cytokeratins, e.g. CK7
and CK20 and prostate-specific antigen (PSA) [6].
Just as tumours resemble morphologically the tissue from
which they were derived, so tumours generally still express
some tissue-specific genes, not only in primary cancers but
also in metastases [11]. This is demonstrated using the
Volume 23 | Supplement 10 | September 2012
symposium article
Annals of Oncology
bioinformatics technique of unsupervised clustering applied to
cancer mRNA or miRNA gene expression profiles. Samples
group (cluster) together by similarity: those with the most
similar gene expression cluster most closely. In such
experiments, cancers of one histological type or subtype cluster
together and with the corresponding normal tissue; likewise
paired primary and metastatic tumours usually cluster together
[11]. Clustering is due at least partly to tissue-specific genes,
which explains their utility as diagnostic cancer biomarkers.
This raises two general issues for diagnostic work-up in
CUP. First, tissue-specific gene expression is better retained in
well-differentiated than in poorly differentiated cancers [11]:
the latter are thus harder to diagnose. Second, metastatic
tumours are usually harder to diagnose than the corresponding
primaries [12, 13]; metastases might have lower expression of
tissue-specific genes than the primary tumour.
IHC for CUP
IHC is fundamental to cancer classification in pathology. Most
IHC biomarkers have been identified on a candidate basis, as
single genes involved in a particular process. For most
diagnostic purposes, however, IHC antibodies are used in a
panel, including markers expected to be positive and negative
in different tumours. This should mean that no single aberrant
IHC stain causes incorrect diagnosis. IHC results depend on
both the staining technique and microscopic interpretation:
both may vary and their optimal performance is crucial.
Table 1 presents IHC markers used in tumour typing and
carcinoma subtyping [7, 14, 15]. Most markers are familiar [6];
newer ones include OCT4, a transcription factor expressed in
germ cell tumours, and D2-40, found in mesothelioma and
other tumours [6].
Table 2 presents IHC markers for the prediction of primary
site in adenocarcinoma. These include the classic CK7 and
CK20, and newcomers such as Napsin A, a lung aspartic
protease [9]; paired box gene 8 (PAX8), a Paired-boX
transcription factor regulating gynaecological tissues, kidney
and thyroid [8]; and NKX3.1, a prostatic tumour suppressor
gene [16]. The ESMO guidelines for CUP suggest a similar
IHC work-up [3]. The NICE (National Institute for Clinical
Excellence) guidelines from the UK suggest a minimum of
CK7, CK20, PSA or oestrogen receptor (ER), placental alkaline
phosphatase (PLAP) and TTF-1 [4].
Primary site can also be predicted in well-differentiated
neuroendocrine tumours, in which TTF-1, CDX2 or PDX1
positivity suggests lung, gastrointestinal or pancreatic origin,
respectively [9]. TTF-1 positivity is not site-specific in poorly
differentiated neuroendocrine carcinoma. No IHC biomarkers
are established to predict the primary site in squamous
carcinoma although some have been suggested [17].
diagnostic work-up for CUP including
IHC: performance and practice
Pathology is an interpretative, thus subjective, discipline. In
practice, the stepwise diagnostic approach is often
subconscious. Not all steps may be needed; at each step, the
pathologist chooses whether to use IHC and, if so, which
markers; and different pathologists vary in approach. Faced
with a biopsy from multiple liver tumours, one pathologist
may report metastatic adenocarcinoma based solely on the
H&E appearance of epithelioid cells with focal glandular
structures. Another may confirm carcinoma using IHC with
positive cytokeratin and negative S100 (Table 1); then may, in
the same or a second ‘round’ of IHC, use differential
cytokeratins to confirm adenocarcinoma and exclude other
carcinomas; and finally may apply markers of primary site
(Table 2).
CUP biopsy or cytology specimens are usually small: it may
be difficult to cut 25 sections from a tissue core. Each ‘round’
of IHC takes 1–2 days. Therefore, only limited IHC markers
can be tested: the average in CUP is 7–8 [18, 19] (range 0–27)
[20]. Marker selection is thus crucial; one barrier to correct
tumour classification is simply not considering and applying
the most appropriate markers.
Two common diagnostic difficulties in CUP work-up are
poorly differentiated or undifferentiated cancer and better
differentiated carcinoma, especially adenocarcinoma, without
an obvious primary site. In either, there may be one likely
diagnosis, multiple differential diagnoses [20] or the diagnosis
may be truly unknown. For poorly differentiated carcinoma, a
conscious stepwise approach helps to ensure that all
appropriate tumour classes are considered; and therefore that
Table 2. Prediction of the primary site of adenocarcinoma using selected immunohistochemistry (IHC)
PSA or NKX3.1 TTF-1 or Napsin A GCDFP-15 or mammaglobin WT1 PAX8 ER CA125 Mesothelin CK7 CDX2 and/or CK20
Prostate
Lung
Breast
Ovary serous
Ovary mucinous
Pancreas
Stomach
Colon
+
−
−
−
−
−
−
−
−
+
−
−
−
−
−
−
−
−
±
−
−
−
−
−
−
−
−
+
−
−
−
−
−
−
−
+
∓
−
−
±
±
∓
−
−
−
−
∓
∓
+
∓
±
−
−
−
∓
−
+
∓
±
∓
−
−
+
+
+
∓
+
±
∓
−
−
−
−
∓
∓
∓
+
GCDFP-15, gross cystic disease protein 15; WT1, Wilms' tumour protein.
+ = ≥90%, ± = 50%–90%, ∓ = 10%–50%, − = ≤10%.
Based on [6, 8, 9, 16].
Volume 23 | Supplement 10 | September 2012
doi:10.1093/annonc/mds357 | x
symposium article
the IHC panel(s) selected will enable their diagnosis. For
adenocarcinoma, pancreatic and gastric origins are especially
difficult to establish because their morphology and IHC are
suggestive but not specific, and diagnostic dilemmas about the
primary site are often pairwise, including pancreatico-biliary
versus gastric, gastric versus colorectal and pancreatico-biliary
versus ovarian [6]. Diagnostic difficulties exist when no
specific, or a few non-specific, IHC markers are positive; when
IHC is hard to interpret due to insufficient tissue, necrosis or
poor staining; or when IHC results conflict with the
morphology or clinical scenario. The ‘unmet clinical need’ for
molecular profiling is to achieve better classification in these
difficult-to-diagnose tumours, especially where treatment
options include tumour-specific versus empiric CUP therapy
[21].
How does IHC currently perform in cancer classification? A
recent meta-analysis identified only five to six large, ‘blinded’
studies of multiple (>3) IHC markers across many (>50)
different tumours [12]. For an individual diagnostic marker,
specificity should be over 96% and ideally 99% [6, 22];
sensitivity should be as high as possible, but at least over 50%.
Across the meta-analysis, IHC sensitivity was remarkably
consistent, ∼82% in mixed primary and metastatic tumours
and 66% in metastases alone [12]. This confirms that
metastatic tumours are more difficult to classify than primaries
by IHC and provides a baseline above which molecular
profiling should perform to be potentially clinically useful.
molecular profiling for CUP:
development and characteristics of tests
Large-scale molecular profiling applicable to CUP has been
achieved at mRNA, miRNA, DNA and epigenetic levels. Three
tests, based on mRNA or miRNA, are commercially available:
Pathwork Tissue of Origin (TOO) test [23], bioTheranostics’
Cancer Type ID (CTID) [20] and miRview mets2 [24]. Their
development, characteristics, validation and clinical impact will
be described and compared [25–29].
The principle underlying these molecular CUP tests is that:
‘different tissue types have distinct RNA profiles’ [23]. To
develop each test, gene expression profiles were generated for
hundreds of different tumours. Using bioinformatics, a subset
of discriminatory genes was identified and diagnostic
algorithms built for cancer classification. Knowledge of the
biopsy site enables distinction of genes expressed by the
tumour and by the organ biopsied [30]. All three tests have
evolved significantly since their initial development and are
now termed ‘second generation’.
The TOO test analyses 2000 mRNAs by microarray [23]. Its
database contains 2140 tumours of 58 types and subtypes,
grouped into 15 classes: breast, bladder, colorectal, gastric,
testicular germ cell, hepatocellular, kidney, non-small-cell lung,
non-Hodgkin’s lymphoma, melanoma, ovarian, pancreatic,
prostate, sarcoma and thyroid [23]. The TOO test reports
similarity scores (SS) compared with each of the 15 tumour
classes: for one sample, the 15 SS total 100. The higher the SS,
the more likely that diagnosis is: thus a prediction with SS > 60
agrees with 90% of reference diagnoses; as SS falls, agreement
x | Oien and Dennis
Annals of Oncology
declines until SS < 5 rules out that tumour class with >99%
confidence [23, 31]. Most reports provide one highly likely
diagnosis and rule out at least 12 tumour classes. Additional
tests, based on 10 more tumour types and subtypes, enable the
separation of ovary from endometrium, and lung from head
and neck cancer: primary site of squamous carcinoma is thus
also being predicted by molecular profiling.
CTID analyses 92 mRNAs by RT-PCR [20]. Its database
contains 2206 tumours of 30 main types and 54 subtypes [20].
(This database was also used by Agendia in the Netherlands to
develop a microarray-based test, CUPPrint, which is published
but no longer marketed [28].) CTID reports one main cancer
class, e.g. ovary, and subclass, e.g. ovary serous, with its
probability. CTID also reports any other main cancer classes
with >5% probability and those with <5% probability, which
are ruled out [20].
miRview mets2 analyses 64 miRNAs by microarray [24]. Its
dataset contains 1282 tumours of 42 types and subtypes [24].
Each test miRNA profile is subjected to two classification
algorithms, with separate outputs: if these diagnoses agree,
confidence is higher. miRview mets2 reports one or two of
either the 42 tumour types or seven combined classes:
sarcoma, kidney, thyroid, neuroendocrine lung, germ cell,
astrocytic or oligodendroglial, and pancreatico-biliary
adenocarcinoma [24].
All three tests can use formalin-fixed, paraffin-embedded
tissue or cytology specimens [20, 24, 31]. Minimal tumour is
required: only a few histological sections, less than for most
CUP IHC. Most (>60%) of the sample should comprise viable
tumour; if necessary, the commercial laboratory can perform
microdissection [20]. On average, 10% of specimens do not
process successfully, i.e. do not yield an RNA profile, due to
insufficient tissue, poor preservation and/or extensive necrosis.
All three tests appear reproducible [23, 24]. They cost currently
∼$3000–4000: equivalent IHC costs ∼$100–200 [6].
molecular profiling for CUP: validation
in known tumours
Molecular tests thus developed require validation in
independent tumour sets, containing primary tumours and/or
metastases from known primaries; poorly differentiated
tumours are more realistic. Only a few validation studies have
been published for the second generation tests, some as
abstracts only. Overall, all three tests have a high specificity of
≥99%. Their sensitivity in tumours of known origin ranges
from 72% to 95%; sensitivity is often but not always lower in
metastases than in primary tumours [25].
The TOO test showed 87% sensitivity in 283 poorly
differentiated or undifferentiated primary tumours and 91%
sensitivity in 179 known metastases [23]. In further studies, the
TOO test showed 94% sensitivity in 17 malignant effusions by
cytology [31] and 95% sensitivity in 37 difficult primary and
metastatic cases [32].
CTID showed 83% sensitivity in 187 known primary
tumours [20]. In an independent study, CTID showed
sensitivities of 87% for tumour typing and 82% for subtyping
in 790 known primary and metastatic cancers [33]. Sensitivity
Volume 23 | Supplement 10 | September 2012
Annals of Oncology
was similar in primary and metastatic cancers, in tumours
across the range of differentiation and with limited tissue [33].
In further studies, CTID showed sensitivities of 78% for
tumour typing and 72% for subtyping in 132 high-grade
known metastatic cancers [18] and predicted the primary site
in 95% of 75 primary and metastatic neuroendocrine tumours
[34].
mirView mets2 showed 85% sensitivity in 509 known
primary and metastatic tumours [24]; in the 82% of samples in
which both algorithms predicted the same diagnosis, sensitivity
was 90%.
molecular profiling for CUP: application
in CUP
We will now consider the performance of these second
generation tests in unknown cases. By definition, CUP
generally lacks a definitive diagnosis, but the molecular
prediction may be compared with the initial clinical diagnosis,
with the pathology and IHC prediction and with the eventual
diagnosis after follow-up and/or autopsy.
The TOO test yielded a prediction in 96% of 45 CUPs [21],
including 11 lung, 6 pancreas, 6 sarcomas, 5 ovary and 4 colon.
Most appeared clinically appropriate; those which did not
include five sarcoma predictions, in which the tumours showed
cytokeratin positivity by IHC, suggesting carcinoma.
CTID yielded a prediction in 91% of 815 submitted cancers
of indeterminate or unknown primary [19]. CTID agreed with
submitted diagnoses in 74%–78% of 300 mainly poorly
differentiated and metastatic submitted cancers [20].
Predictions included more rare tumours than expected,
including cholangiocarcinoma, small intestinal
adenocarcinoma, ovarian mucinous tumour, neuroendocrine
carcinoma and Merkel cell carcinoma [20]. In a third study of
CUP patients in whom the primary site became known, CTID
provided 20 predictions: 75% were correct; 10% were
indeterminate, truly being lung and 15% appeared incorrect
with predictions being sarcoma, germ cell and intestinal
adenocarcinoma and true diagnoses being lung, pancreatic and
gastro-oesophageal cancers [35].
miRview mets2 agreed with clinico-pathological data in 88%
of 55 brain CUPs [24]. In a second study of 92 CUP patients,
miRview mets2 agreed in 92% of 84 assessable cases: the seven
incorrect predictions had a final diagnosis of lung, head and
neck or pancreas [36].
molecular profiling for CUP: limitations
and strengths compared with IHC
Each molecular test may have diagnostic difficulties made
evident through careful validation studies, of which few have
yet been published for miRview mets2. The TOO test has two
specific potential issues. First, its results are reported as one of
15 classes; the tumour subtype at a given site does not appear
to be described, although the original database included most
subtypes. Second, the latter lacked certain tumour types
(termed ‘off-panel’), including neuroendocrine tumours,
cholangiocarcinoma and mesothelioma [23], which is
Volume 23 | Supplement 10 | September 2012
symposium article
important because neuroendocrine tumours are ‘good
prognosis’ CUPs and cholangiocarcinoma enters the
differential diagnosis for liver metastases. As the test
information states [37], ‘It is not possible to distinguish a
tumour type if not included in the original diagnostic panel.’
One cannot know in advance whether a tumour is off-panel;
and off-panel tumours yield indeterminate or incorrect results
[37].
Certain diagnoses may be difficult. For the TOO test,
sarcoma predictions appeared incorrect in five of six CUPs
[21] and 16 of 40 known tumours [23]. For all three tests, but
perhaps most for CTID and least for miRview mets2, common
incorrect diagnoses include pancreatic, colonic and gastrooesophageal cancer and their separation, and lung cancer [33,
35]. The problem appears less in second than in first
generation tests [13, 23, 30, 35, 37, 38]. Diagnostic difficulty is
greater when the tumour is poorly differentiated or has atypical
morphology for its site; and the test may be confounded by
insufficient tissue, ‘off-panel’ cases or normal tissue from the
biopsy site [30, 37].
The strengths of molecular profiling are shown by
comparison with IHC in two of the above studies. CTID was
more sensitive than optimal IHC in 132 high-grade metastases,
for tumour typing (78% versus 68%) and subtyping (72%
versus 61%) [18]. CTID was more sensitive than IHC for lung,
bladder and breast, and of similar sensitivity for
gastrointestinal and kidney tumours [18]. miRview mets2
agreed with the final diagnosis in 92% of 84 CUPs, compared
with 70% for IHC [36]. These IHC sensitivities are similar to
those in the IHC meta-analysis [12]. In these studies,
molecular profiling was more accurate than IHC in metastases,
confirming its potential clinical utility.
molecular profiling for CUP: clinical
impact
What difference might these molecular tests make to diagnosis
and management [19, 20, 39, 40]? One good example is a
retrospective analysis of 107 patients with poorly differentiated
or metastatic cancer who had undergone extensive clinicopathological evaluation [39]. TOO testing affected the working
diagnosis as follows: 14 patients still had site unspecified; 30
changed from unspecified to a specified site; 36 still had the
same specified site; 24 had a change of the specified site; and 3
had a specified site changed to unspecified. Overall, the
diagnosis changed in 57 (53%) and clinical management
changed in 72 patients (65%), including: ruling in or out
further investigations; changing specific treatment; and
facilitating hospice referral. Confirming a diagnosis can be as
helpful as changing it [39]. The figures appear similar for
CTID which yielded a confirmatory prediction in 47%–55%
and a new diagnosis in 36%–37% of up to 815 submitted cases
[19, 20].
The impact of molecular tests on the patient outcome is less
clear: studies are few and not randomised. For CTID, CUP
patients whose molecular profiles suggested colorectal tumour,
and who then received colorectal-specific therapy, had survival
times longer than historical CUP controls but similar to
doi:10.1093/annonc/mds357 | x
symposium article
patients with known metastatic colorectal cancer [41]. In a
further study, CTID was used to direct site-specific therapy in
CUP patients, yielding a median overall survival (12.2 mo)
better than survival for historical controls receiving empirical
CUP therapy [42]. These studies suggest that molecular
profiling may lead to improved survival in those treated
accordingly; but further studies are needed to confirm and
clarify the benefit.
Cost and logistics also need consideration: commercial
molecular profiling is an external test, which is more expensive
than IHC but less expensive than many therapies.
molecular profiling for CUP: further tests
Most other molecular profiling approaches have been described
in a single paper, with initial validation, often including CUP
specimens. Some are broad-based assays, classifying all likely
tumours, like the commercial tests [43–45]; others are more
specific, classifying tumour subsets [14, 46, 47]. Most assays
involve pre-specified gene sets [14, 43, 44, 46, 47], like the
commercial tests, and have been developed similarly; a few are
more flexible [48, 49]. Most have sensitivities of 78%–90% [43,
44, 48], similar to the commercial tests.
mRNA-based assays include a broad 79-gene assay with 13
classes and 89% accuracy [43]; and a specific 10-gene RT-PCR
assay, for six primary sites of adenocarcinoma [46, 47]. Other
assays are based on miRNA [44] and methylation [45]. Two
more papers describe de novo prediction of tumour class by
comparison with a tumour database but without using
predefined gene sets, based on mRNA [48] or high-throughput
sequencing [49].
Lastly, Centeno et al. described an integrated stepwise
approach with diagnostic steps involving: identification of
carcinoma; subtyping into adenocarcinoma, neuroendocrine
carcinoma, squamous carcinoma and urothelial carcinoma;
then prediction of primary site in each subtype. Each step is
based on the morphology and IHC plus, where necessary,
mRNA profiling using separate gene sets: at each step, accuracy
was 87%–90%. This paper demonstrated the integration of
(subset) molecular profiling into a standard stepwise pathology
approach [14].
further integration of tests, conclusions
and future
It can be seen that pathology with IHC and molecular profiling
are not completely independent, but represent ‘different sides
of the same coin’. First, IHC and molecular profiles use many
of the same tissue-specific genes, including PAX8, CDX2, PSA
and S100 [6, 13, 38]. Second, they perform in parallel: tumour
types easy to distinguish on morphology and IHC are also
distinct on molecular testing; tumours hard to diagnose on
morphology and IHC may also be challenging for molecular
tests, e.g. pancreatic adenocarcinomas. Given these similarities,
where can molecular profiling make an impact?
Pathology, with IHC where necessary, is likely to retain its
place as gold standard in tumour classification, especially
where the tumour is primary and/or is well or moderately
x | Oien and Dennis
Annals of Oncology
differentiated and/or has classical IHC results and appropriate
clinical findings; no publications have suggested otherwise.
However, in already well worked-up poorly differentiated and/
or metastatic tumours, including CUP, molecular profiling
performs well, with sensitivities of 72%–95% and may
outperform optimal IHC by 10%–20%, according to initial
studies. For CUP patients, molecular profiling may change the
diagnosis in around half, and affects management in most.
Molecular profiling could thus contribute to diagnosis of
poorly differentiated and/or metastatic tumours. In these, the
diagnosis may be truly unknown or there may be an
intractable differential diagnosis of two or more possibilities
[20]. It is not clear whether profiling could be a valuable firstline option where tumour tissue or turnaround time is limited.
Profiling appears unhelpful in necrotic tumour. Of the CUP
molecular tests, a few address specific differential diagnoses
and are, therefore, directly integrated with pathology. Most,
including the commercial tests, are ‘stand-alone’, for diagnosis
across almost all likely tumour types; their integration involves
comparison of the molecular prediction with the morphology
and IHC and a careful overall interpretation to ensure that the
diagnosis fits the clinical context [37].
For the future, further studies are needed on the
performance and comparison of molecular profiling and IHC
in large cohorts of poorly differentiated and/or metastatic
tumours, including CUPs; and the impact of molecular
profiling and subsequent specific therapy on survival and other
outcome measures in patients with CUP. IHC and molecular
classifiers should be flexible to enable tumour classes to change
as new therapies emerge; ideally, further development should
include better classification of the remaining difficult-todiagnose tumours, e.g. pancreatico-biliary and gastric.
Recent developments in molecular profiling and IHC mean
that the unknown primary is gradually becoming more known;
we hope that these improvements will continue over the next
decade in parallel with advances in treatment and outcomes
for patients with CUP.
disclosure
KAO has no financial interests in any company mentioned in
this paper; she is conducting a research study, which includes
collaborations with the company BioTheranostics and with the
research group of Professor David Bowtell, which are
mentioned in this paper; their contribution is intellectual and
to provide molecular analyses but not otherwise financial. JLD
declares no conflicts of interest.
references
1. Pavlidis N, Pentheroudakis G. Cancer of unknown primary site. Lancet 2012;
379: 1428–1435.
2. Massard C, Loriot Y, Fizazi K. Carcinomas of an unknown primary origin––
diagnosis and treatment. Nat Rev Clin Oncol 2011; 8: 701–710.
3. Fizazi K, Greco FA, Pavlidis N et al. Cancers of unknown primary site: ESMO
clinical practice guidelines for diagnosis, treatment and follow-up. Ann Oncol
2011; 22(Suppl 6): vi64–vi68.
4. National Institute for Health and Clinical Excellence. Metastatic Malignant Disease
of Unknown Primary Origin: Diagnosis and Management of Metastatic Malignant
Volume 23 | Supplement 10 | September 2012
Annals of Oncology
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
Disease of Unknown Primary Origin. London: National Institute for Health and
Clinical Excellence 2010; CG104.
Greco FA, Oien K, Erlander M et al. Cancer of unknown primary: progress in the
search for improved and rapid diagnosis leading toward superior patient
outcomes. Ann Oncol 2012; 23: 298–304.
Oien KA. Pathologic evaluation of unknown primary cancer. Semin Oncol 2009;
36: 8–37.
Bahrami A, Truong LD, Ro JY. Undifferentiated tumor––true identity by
immunohistochemistry. Arch Pathol Lab Med 2008; 132: 326–348.
Laury AR, Perets R, Piao H et al. A comprehensive analysis of PAX8 expression
in human epithelial tumors. Am J Surg Pathol 2011; 35: 816–826.
Turner BM, Cagle PT, Sainz IM et al. Napsin A, a new marker for lung
adenocarcinoma, is complementary and more sensitive and specific than thyroid
transcription factor 1 in the differential diagnosis of primary pulmonary carcinoma
evaluation of 1674 cases by tissue microarray. Arch Pathol Lab Med 2012; 136:
163–171.
Ramskold D, Wang ET, Burge CB et al. An abundance of ubiquitously expressed
genes revealed by tissue transcriptome sequence data. PLoS Comput Biol 2009;
5: e1000598.
Gevaert O, Daemen A, De Moor B et al. A taxonomy of epithelial human cancer
and their metastases. BMC Med Genomics 2009; 2: 69.
Anderson GG, Weiss LM. Determining tissue of origin for metastatic cancers:
meta-analysis and literature review of immunohistochemistry performance. Appl
Immunohistochem Mol Morphol 2010; 18: 3–8.
Monzon FA, Lyons-Weiler M, Buturovic LJ et al. Multicenter Validation of a
1,550-Gene Expression Profile for Identification of Tumor Tissue of Origin. J Clin
Oncol 2009; 27: 2503–2508.
Centeno BA, Bloom G, Chen DT et al. Hybrid model integrating
immunohistochemistry and expression profiling for the classification of
carcinomas of unknown primary site. J Mol Diagn 2010; 12: 476–486.
Hashimoto K, Sasajima Y, Ando M et al. Immunohistochemical profile for
unknown primary adenocarcinoma. PLoS One 2012; 7: e31181.
Gurel B, Ali TZ, Montgomery EA et al. NKX3.1 as a marker of prostatic origin in
metastatic tumors. Am J Surg Pathol 2010; 34: 1097–1105.
Pereira TC, Share SM, Magalhaes AV et al. Can we tell the site of origin of
metastatic squamous cell carcinoma? An immunohistochemical tissue microarray
study of 194 cases. Appl Immunohistochem Mol Morphol 2011; 19: 10–14.
Weiss LM, Chu PG, Schroeder BE et al. Blinded comparator study of
immunohistochemistry (IC) versus a 92-gene cancer classifier in the diagnosis of
primary site in metastatic tumors. J Clin Oncol 2012; 30: e21019.
Schroeder BE, Laouri M, Chen E et al. Pathological diagnoses in cases of
indeterminate or unknown primary submitted for molecular tumor profiling. Lab
Invest 2012; 92: 105A–106A.
Erlander MG, Ma X-J, Kesty NC et al. Performance and clinical evaluation of the
92-gene real-time PCR assay for tumor classification. J Mol Diagn 2011; 13:
493–503.
Hainsworth JD, Pillai R, Henner WD et al. Molecular tumor profiling in the
diagnosis of patients with carcinoma of unknown primary site: retrospective
evaluation of gene microarray assay. J Mol Biomarkers Diagn 2011; 2: 106.
Dennis JL, Hvidsten TR, Wit EC et al. Markers of adenocarcinoma characteristic
of the site of origin: development of a diagnostic algorithm. Clin Cancer Res
2005; 11: 3766–3772.
Pillai R, Deeter R, Rigl CT et al. Validation and reproducibility of a microarraybased gene expression test for tumor identification in formalin-fixed, paraffinembedded specimens. J Mol Diagn 2011; 13: 48–56.
Meiri E, Mueller WC, Rosenwald S et al. A second-generation microRNA-based
assay for diagnosing tumor tissue origin. Oncologist 2012; 17: 801–812.
Takei H, Monzon FA. Gene-expression assays and personalized cancer care:
tissue-of-origin test for cancer of unknown primary origin. Pers Med 2011; 8:
429–436.
Monzon FA, Koen TJ. Diagnosis of metastatic neoplasms: molecular approaches
for identification of tissue of origin. Arch Pathol Lab Med 2010; 134: 216–224.
Handorf CR. Gene expression analysis and immunohistochemistry in evaluation of
cancer of unknown primary: time for a patient-centered approach. J Natl Compr
Cancer Netw 2011; 9: 1415–1420.
Volume 23 | Supplement 10 | September 2012
symposium article
28. Bender RA, Erlander MG. Molecular classification of unknown primary cancer.
Semin Oncol 2009; 36: 38–43.
29. Monzon FA, Dumur CI. Diagnosis of uncertain primary tumors with the pathwork
tissue-of-origin test. Expert Rev Mol Diagn 2010; 10: 17–25.
30. Dumur CI, Fuller CE, Blevins TL et al. Clinical verification of the performance of
the pathwork tissue of origin test: utility and limitations. Am J Clin Pathol 2011;
136: 924–933.
31. Stancel GA, Coffey D, Alvarez K et al. Identification of tissue of origin in body
fluid specimens using a gene expression microarray assay. Cancer Cytopathol
2012; 120: 62–70.
32. Grenert JP, Smith A, Ruan W et al. Gene expression profiling from formalin-fixed,
paraffin-embedded tissue for tumor diagnosis. Clin Chim Acta 2011; 412:
1462–1464.
33. Kerr SE, Schnabel CA, Sullivan PS et al Multisite Validation Study To Determine
Performance Characteristics of a 92-gene Molecular Cancer Classifier. Clin
Cancer Res 2012; 18: 3592–3960.
34. Kerr SE, Schnabel CA, Sullivan PS et al. Use of a 92-gene molecular classifier to
predict the site of origin for primary and metastatic tumors with neuroendocrine
differentiation. Lab Invest 2012; 92: 147A.
35. Greco FA, Spigel DR, Yardley DA et al. Molecular profiling in unknown primary
cancer: accuracy of tissue of origin prediction. Oncologist 2010; 15: 500–506.
36. Sanden M, Pentheroudakis G, St Cyr B et al. A novel microRNA-based test
demonstrate above 90% accuracy in classification of metastatic tumors from
patients diagnosed with carcinoma of unknown primary. Lab Invest 2012; 92:
456A.
37. Beck AH, Rodriguez-Paris J, Zehnder J et al. Evaluation of a gene expression
microarray-based assay to determine tissue type of origin on a diverse set of 49
malignancies. Am J Surg Pathol 2011; 35: 1030–1037.
38. Ma XJ, Patel R, Wang XQ et al. Molecular classification of human cancers using
a 92-gene real-time quantitative polymerase chain reaction assay. Arch Pathol
Lab Med 2006; 130: 465–473.
39. Nystrom SJ, Hornberger JC, Varadhachary GR et al Clinical utility of geneexpression profiling for tumor-site origin in patients with metastatic or poorly
differentiated cancer: impact on diagnosis, treatment, and survival. Oncotarget
2012; 3: 620–628.
40. Laouri M, Halks-Miller M, Henner WD et al. Potential clinical utility of geneexpression profiling in identifying tumors of uncertain origin. Pers Med 2011; 8:
615–622.
41. Hainsworth JD, Schnabel CA, Erlander MG et al. A retrospective study of
treatment outcomes in patients with carcinoma of unknown primary site and
a colorectal cancer molecular profile. Clin Colorectal Cancer 2012; 11:
112–118.
42. Greco FA, Rubin MS, Boccia RV et al. Molecular gene expression profiling to
predict the tissue of origin and direct site-specific therapy in patients ( pts) with
carcinoma of unknown primary site (CUP): results of a prospective Sarah Cannon
Research Institute (SCRI) trial. J Clin Oncol 2012; 42: e10530.
43. Tothill RW, Kowalczyk A, Rischin D et al. An expression-based site of origin
diagnostic method designed for clinical application to cancer of unknown origin.
Cancer Res 2005; 65: 4031–4040.
44. Ferracin M, Pedriali M, Veronese A et al. MicroRNA profiling for the identification
of cancers with unknown primary tissue-of-origin. J Pathol 2011; 225: 43–53.
45. Fernandez AF, Assenov Y, Martin-Subero JI et al. A DNA methylation fingerprint
of 1628 human samples. Genome Res 2012; 22: 407–419.
46. Talantov D, Baden J, Jatkoe T et al. A quantitative reverse transcriptasepolymerase chain reaction assay to identify metastatic carcinoma tissue of origin.
J Mol Diagn 2006; 8: 320–329.
47. Varadhachary GR, Talantov D, Raber MN et al. Molecular profiling of carcinoma
of unknown primary and correlation with clinical evaluation. J Clin Oncol 2008;
26: 4442–4448.
48. Ojala KA, Kilpinen SK, Kallioniemi OP. Classification of unknown primary tumors
with a data-driven method based on a large microarray reference database.
Genome Med 2011; 3: 63.
49. Quon G, Morris Q. ISOLATE: a computational strategy for identifying the primary
origin of cancers using high-throughput sequencing. Bioinformatics 2009; 25:
2882–2889.
doi:10.1093/annonc/mds357 | x