A quantitative immunoassay for lung cancer biomarker CIZ1b in patient plasma

Objectives Non-invasive tests for early detection of lung cancer are an important unmet clinical need. CIZ1b plasma biomarker can discriminate stage 1 lung cancer from within high-risk groups with clinically useful accuracy, with ROC AUCs in excess of 0.9 for two independent retrospective cohorts, and could therefore meet this need. Our aim was to characterise the native state of the biomarker and develop a quantitative immunoassay. Design and methods Selective denaturation, preparative electrophoresis and mass spectrometry of human plasma were used to characterise the biomarker and interaction partners. A sandwich ELISA was generated, and specificity for CIZ1b biomarker tested on lung cancer patient plasma. Results CIZ1b biomarker is a denaturation-resistant complex between a C-terminal fragment of CIZ1 bearing the CIZ1b epitope specified by alternative splicing of exon14, and fibrinogen alpha chain. Reconstitution of the biomarker epitope with purified fibrinogen and CIZ1b, but not CIZ1a (non-alternatively spliced exon 14) confirmed the specificity of the results. The endogenous complex is highly stable in lung cancer plasma and can be quantified by pairing of a CIZ1b exon-junction specific antibody with detection of fibrinogen. Application of this sandwich ELISA to a prospectively collected development set of plasmas reveals the same level of accuracy as the western blot used to validate the discriminatory capability of the biomarker. Conclusions Unexpected and unusual molecular structure of CIZ1b in native plasma has complicated immunoassay design, and delayed translation of this promising biomarker. However, CIZ1b can now be measured using a high-throughput, hospital-friendly sandwich ELISA format, overcoming an important barrier to further clinical development and application of this blood test for early stage lung cancer.


Introduction
A unique protein epitope, created by alternative-splicing of the CIZ1 gene, is present in the plasma fraction of blood from patients with lung cancer, making it a strong candidate for high impact, high throughput early detection of lung cancer. We showed previously that detection of this b-variant form of CIZ1 (CIZ1b) is sensitive enough to allow accurate identification of patients with stage 1 disease, evidenced by an archived set of cases and matched high-risk controls, including benign lung nodules, inflammatory diseases of the lung, asthma, COPD and age-matched smokers [1]. Biomarker detection is reproducible and quantifiable in less than a microliter of plasma by western blot, giving very favourable outputs in two independent archived cohorts, with 170 and 160 samples respectively. For set 1, mean CIZ1b level in individuals without diagnosed tumours established a threshold that correctly classified 98% of small cell lung cancers (SCLC) and non-SCLC patients (receiver operator characteristic AUC 0.958). Within set 2, comparison of stage 1 non-SCLC with asymptomatic age-matched smokers, or individuals with benign lung nodules, correctly classified 95% of patients (AUCs 0.913, 0.905), with overall specificity of 76%, and 71% respectively. Moreover, using the mean of controls in set 1, we achieved 95% sensitivity among stage 1 non-SCLC patients in set 2 with 74% specificity, demonstrating the robustness of the classification. Thus, CIZ1b performs with clinically useful accuracy, and could potentially satisfy the unmet need for a circulating biomarker for early detection of lung cancer.
However, clinical development of this promising blood test requires translation of western blot-based analysis of denatured samples, into a high-throughput immunoassay format, compatible with near native epitopes. The unusual and complicated composition of the biomarker has limited timely translation to immunoassay format, however based on detailed analysis of its identity in lung cancer patient plasma we now report development of a quantitative sandwich ELISA. We also report underpinning information on biomarker stability, and unexpected findings about the native format of CIZ1b, and its interaction partners in human blood. The data presented here defines the CIZ1b plasma biomarker as a small CIZ1 fragment spanning the exon 14b/15 junction stably complexed with fibrinogen alpha chain, and has lead to proof of concept validation of a quantitative sandwich ELISA that can detect the CIZ1b/fibrinogen alpha complex in lung cancer patients, via a simple immunoassay of blood.
In intact cells, the CIZ1 protein is part of the nuclear matrix (NM), a biochemically defined fraction that resists extraction from the nucleus and which is thought to play a role in spatially organising nuclear process such as DNA replication [2]. CIZ1 plays a role in DNA replication via a mechanism that involves tethering regulators of DNA replication, including cyclins, to the NM [3,4]. Since it was first identified [5], CIZ1 has been linked with neurological disorders including Alzheimer's disease [6] and cervical dystonia [7], and a rapidly growing range of cancers including the pediatric tumours medulloblastoma [8] and Ewing's tumour [9], and more common adult onset cancers including breast [10], lung [1], colon [11], prostate [12] and liver cancers [13], in some cases showing promise as independent prognostic [11] or diagnostic indicators [1], and in others implicated in the proliferation characteristics of cultured cancer cells [14,15]. Notably, CIZ1 transcripts are alternatively spliced to yield 25 (Aceview 2012) or more [16] variants. Most are not fully characterised but some, including CIZ1b, appear to be inaccurate splicing events expressed only in tumours [1], while other 'normal' variants are associated with disease when expressed in the wrong developmental context [9,17]. Our previous functional analysis of CIZ1 informed our understanding of the altered forms detected in lung cancer libraries, and led us to focus on CIZ1b, and then to identify CIZ1b protein in blood plasma. However, while we now understand much about CIZ1s function inside the nucleus, we do not yet know whether circulating CIZ1b has biological significance in the blood, other than as a very potent marker of malignant lung tumours.

Biomarker characterisation
Published analysis of CIZ1b was carried out with anti-peptide rabbit polyclonal antibody 2B, which was raised against a short unique peptide spanning the alternatively spliced 14b-15 junction (Fig. 1A), with sequence DEEEIEVRSRDIS (junction indicated by underlined valine). The strategy for generation, validation and purification was described previously [1]. The western blot reactivity profile for 2B includes a 65-70 kDa entity in SDS-PAGE of plasma from lung cancer patients, plus a 55 kDa entity in plasma from all people (Fig. 1B, and Fig. S1A). The cancerspecific 65-70 kDa species is the CIZ1b biomarker that is referred to here and previously [1]. Crucially, unlike plasma, we were unable to detect this protein in serum samples from the same lung cancer patients (Fig. 1C), suggesting that CIZ1b biomarker is sequestered in coagulated blood.
Notably, the 55 kDa generic band co-migrates with CIZ1 species detected with anti-CIZ1 exon 17 antibodies, and an anti-CIZ1 exon 8 antibody (Fig. 1B), identifying the 55 kDa species as a variant or proteolytic fragment of full-length hCIZ1 (which has a predicted full length MW of 99 kDa, NCBI Reference Sequence: NM_012127.2). The width of the bands detected via exon 8 and exon 17 is different, possibly because the exon 8 epitope lies within an alternatively spiced region, while detection via exon 17 would reveal species with or without alternative splicing at exon 8. Thus, at least one form of the CIZ1 protein is present in the circulatory system in cancer and non-cancer plasma samples alike, suggesting it to be part of normal physiology. Exon 17 and exon 8 epitopes are not present in the 65-70 kDa species recognised by CIZ1b antibody 2B in cancer patients, which means that sequences both upstream and downstream of the CIZ1b epitope are missing from this protein, raising questions about its identity. The data suggest that the 65-70 kDa species is either a complicated CIZ1 splice variant, or an SDS-stable complex between a CIZ1b fragment of CIZ1 and a carrier protein (Fig. 1D).
In order to understand the relationship between cancer-specific 65-70 and generic 55, and to ask whether 55 i) incorporates the 14b/15 junction or ii) is detected by 2B via epitopes that immediately flank the junction, we generated a further set of rabbit antibodies using the same immunisation protocol as for 2B. This yielded antibody 043, which cleanly and uniquely reacts with the 65-70 kDa band in cancer patients (Fig. 1B, Fig. S1A, B). Thus 043 is a much cleaner detection tool than 2B with potential for application in other assay formats, including ELISA. Lack of recognition of the 55 kDa species suggests less contribution of junction flanking sequences to its epitope compared to 2B, and suggests that the 55 kDa species does not contain the CIZ1b junction.
Direct comparison of the ability of these two antibodies to discriminate lung cancer patients by western blot shows them both to be discriminatory, with a high degree of correlation (Fig. 1E), though notably they do not share exactly the same profile across plasma test set A. Thus, there is some variation in their epitopes but a common ability to discriminate patients with lung cancer from those without. There is also a small advantage in integrating their output to maximise selectivity (raw data in SI Data set 1).

Stability
The published [1] and current data support the assertion that CIZ1b is a potent circulating biomarker for lung cancer with significant potential for application in clinical practice. However to realise this potential, it must be robust enough to withstand variations in sample processing and collection regimes that occur in busy hospitals. Therefore, we tested the stability of the 65-70 kDa cancer-selective CIZ1b species in plasma and whole blood, and compared it with the 55 kDa species, using antibody 2B. The data show stability in isolated plasma over at least 6 h when incubated at 37°C, with gradual decay in both species thereafter (Fig. S2A). Similarly, there was relatively little change during an hour at temperatures up to 50°C (Fig. S2B), and stability over at least 3 freezethaw cycles (Fig. S2C). Furthermore, when whole blood was left on the bench for up to 24 h prior to fractionation there was no loss of signal ( Fig. S2D), all of which indicate it to be a robust biomarker suitable for application in a range of clinical settings.

Complex dissociation
Under native conditions both 2B and 043 detect a complex in excess of 720 kDa that co-migrates with an abundant protein in plasma (shown for 043 in Fig. S3A), but there is very little discrimination between cancer and non-cancer plasmas. Upon further separation through a second dimension under denaturing conditions the cancer-specific 65-70 kDa species is revealed and (for 2B) separated from the 55 kDa generic band (Fig. S3B). Thus, cancer-specific detection of CIZ1b in a completely native immunoassay format is unlikely even with 043. We therefore dissected out the three major denaturing influences (heat, SDS, reducing agent) that reveal cancer-specific signal and evaluated their effect on the mobility and discrimination of CIZ1b by 043. This showed that 1% SDS shifts the major reactive species from N720 kDa in native gel, to approximately 340 kDa (Fig. S3C, right), revealing cancer-specific signal, as well as an unspecific band at~200 kDa in both cancer and non-cancer samples. Increasing the incubation temperature has a detrimental effect on the cancer-specific species, with maximum differential achieved at 37°C. Inclusion of reducing agent further shifted the mobility of the cancer-specific signal to the expected position of 65-70 kDa (Fig. S3C, left). Notably, higher concentrations of reducing agent resulted in loss of CIZ1b signal (Fig. S3D,E). Taken    together, these data suggest that the 65-70 kDa entity is itself a complex that is resistant to standard SDS-PAGE denaturing conditions (10 mM DTT) but not more aggressive treatments (500 mM DTT), and that it is normally present within a higher order complex of 340 kDa, which is itself released by SDS from a still bigger complex in excess of 720 kDa (illustrated in Fig. 2A). In fact, fractionation of plasma to enrich for the exosomal compartment differentially partitioned the 55 kDa species (detected via exon 17) and 65-70 kDa species (detected via 043), with the 65-70 kDa species detected exclusively in the pellet, which is a complex fraction that includes exosomes (Fig. S4A,B).

Identification
These observations enable a two-step gel-purification strategy for the 65-70 kDa CIZ1b band. First, the 340 kDa species was isolated from two independent lung cancer patient plasmas (Fig. 3A), followed by further separation in the presence of reducing agent (Fig. 3B). The 65-70 kDa band corresponding to the cancer-specific western blot signal was excised and digested with either trypsin or Asp-N endoprotease, for identification by mass spectrometry (MS). For both enzymatic digestions and two patients, Mascot database searching returned a single significant identification (P b 0.05), corresponding to Uniprot P02671; human fibrinogen alpha chain. The endogenous fibrinogen molecule is a hexameric glycoprotein with a pre-cleavage formula mass of 326 kDa and relative mobility of~340 kDa, comprised of two sets of three different chains (α, β, and γ) of 67, 51 and 45 kDa respectively [18].
Notably, CIZ1b was not identified in the 65-70 kDa band after digestion with either enzyme. To determine whether a diagnostic CIZ1b fragment could be observed by MS if present, a short synthetic CIZ1b peptide (b13) and long CIZ1b peptide (b66, Table S1) were digested with Asp-N, which generates the diagnostic fragment DEEEIEVRSR by Fig. 3. Reconstitution of CIZ1b biomarker from purified components. A) Two independent lung cancer patient plasmas were separated after incubation with 1% SDS (no reducing agent), in order to isolate the 340 kDa band from stained gels (left). Band identity was verified by Ponceau S stain followed by western blot with anti-CIZ1b antibody 043 of a parallel gel (right). B) Gel slices were soaked in 2% SDS-PAGE loading buffer (with 200 mM βME) and separated through a denaturing gel to recover the 043-reactive species at 70 kDa. Note that sample loaded as a gel slice is slightly retardation compared to markers which are loaded in solution. Coomassie blue staining revealed 5 dominant bands including one at 70 kDa, which was isolated and digested with either trypsin (cancer plasma 1, C1) or AspN (cancer plasma 2, C2). C) Western blot of normal human plasma (non-cancer, N) showing reconstitution of CIZ1b epitope via complex formation between a carrier protein in plasma and synthetic CIZ1b peptides of the indicated lengths, but not equivalent CIZ1a peptide (Table S1). Note that free peptides migrate in reducing SDS-PAGE with a relative mobility 3 times expected (~21 kDa instead of 7.6 kDa for b66,~5 kDa instead of 1.6 for b13, see  (Table S1), probed with CIZ1b antibodies as indicated. Under non-reducing conditions fibrinogen migrates with apparent molecular weight of~340 kDa and this co-migrates with CIZ1b signal. E) Direct ELISA showing mean A450 nm (triplicate analysis with SEM) generated by CIZ1b antibodies 2B or 043 as indicated, using molar equivalents of preformed peptide/fibrinogen complex as analyte, as described in Methods. **t-test P b 0.05. cutting either side of the CIZ1b exon junction (Fig. S5A). The peptide was positively identified post digestion of b13 and b66 by MALDI-MS/ MS, though at low intensity relative to other peptides in the digest (shown for b66 in Fig. S5B). Further analysis was performed by LC-MS/MS using both data dependent acquisition (DDA) and targeted selection reaction monitoring (SRM), which focused the analysis to cycle through fragmentation of the expected 2 + and 3 + ions of DEEEIEVRSR, excluding all other precursors, to maximise specificity and sensitivity for this analyte. DEEEIEVRSR was identified in both the DDA and SRM analyses, indicating that the diagnostic peptide is observable by MS when excised from chemically synthesized peptide (SRM chromatogram, Fig. S5C). However, when the same peptide was spiked into normal human plasma to reconstitute CIZ1b (see later and Fig. S5D), the diagnostic DEEEIEVRSR peptide was not detected in the excised 65-70 kDa band (SRM chromatogram, Fig. S5C). Lack of identification could reflect: i) secondary modifications that alter the expected mass of the CIZ1b diagnostic peptide or impair enzymatic digestion, ii) the more abundant proteins in the sample, primarily fibrinogen alpha chain, causing ion suppression or iii) absolute abundance falling below the limit of detection for the instrumentation after the losses incurred from the PAGE and digestion steps. It should be noted that the peptide is highly acidic, which is likely to reduce ionization efficiency in positive-mode MS, so that even with SRM a small drop in abundance could fall below the limit of detection. We concluded that if we cannot reliably detect the spiked peptide at physiologically relevant levels we are unlikely to detect endogenous CIZ1b, and so followed alternative approaches.

Epitope reconstitution
In order to confirm that the 65-70 kDa cancer-specific epitope does indeed contain CIZ1b sequences, we adopted an epitope reconstitution strategy. To achieve this a pair of synthetic peptides spanning the exon14/15 junction (CIZ1b13, and CIZ1a22 which includes the alternatively spliced intervening sequence, Table S1) were incubated with normal human plasma (Fig. 3C), and the output compared to lung cancer patient plasma. The CIZ1b peptide, but not CIZ1a, generated 043 and 2B-reactive epitope of comparable mobility to the endogenous epitope (65-70 kDa), which is far in excess of the relative molecular mass of free peptide and not present in either of the individual components of the reaction (peptide lanes 6, 7, or plasma lane 3). This suggests that CIZ1b peptide forms an SDS/DTT-resistant complex with a carrier protein in human plasma, recreating the CIZ1b biomarker recognised by 2B and 043. Since we did not detect a mobility difference between endogenous (cancer plasma lane 1) and reconstituted epitope, the data also imply that the CIZ1b fragment contributing to the 65-70 kDa cancer-specific species is very short, in the order of 10-20 amino acids in length. Furthermore, individual molecules may not necessarily have uniform molecular boundaries, possibly accounting for the oftendiffuse nature of the band (Fig. S1A).
A molar equivalent of the much longer CIZ1b peptide (b66 , Table S1) reconstituted reactive epitope less efficiently, and was detectable only by 2B (Fig. 3C lane 2, and at higher concentration in Fig. S5D). With this peptide the reconstituted signal migrated with greater relative molecular mass, at~80 kDa, consistent with the increased size of the contributing peptide. Importantly, reconstituted CIZ1b epitope, like the endogenous epitope, resists standard denaturing SDS-PAGE conditions (10 mM DTT, 2% SDS, 90°C), but is dissociated by a more aggressive reducing environment (Fig. S3E). Taken together, the data suggest that, in vivo, the cancer-specific epitope is composed of a relatively short fragment of CIZ1 encompassing the exon14b/15 junction, mounted on fibrinogen alpha chain, which itself makes up most of the mass of the 65-70 kDa species. To further confirm this, epitope reconstitution experiments were performed with purified human fibrinogen (Fig. 3D). For both antibodies, reactive epitope was generated at the mobility of fibrinogen complex (340 kDa in non-reducing gels) and the specificity of the signal was confirmed by lack of reactivity with equivalent CIZ1a peptide. Direct ELISA using human fibrinogen complexed with molar equivalents of CIZ1a and CIZ1b peptides (Fig. 3E) confirmed three things; i) the specificity of epitope reconstitution with peptide b13 over a22, ii) negligible signal with fibrinogen alone, and iii) greater reactivity when peptide is mounted on fibrinogen. Moreover, inclusion of the very short CIZ1b peptide b7 (Table S1) revealed differences between 2B and 043; 2B recognises b7 while 043 does not in a manner that is unrelated to the presence of fibrinogen. The existence of CIZ1b epitope in a complex with fibrinogen is consistent with our earlier observations that the biomarker is depleted from serum samples isolated from clotted blood (Fig. 1C). Thus, future clinical evaluation and application of this biomarker must avoid serum samples, and ensure stable use of anticoagulants.

CIZ1b immunoassay
Based on knowledge of the composition of CIZ1b biomarker, we designed a sandwich immunoassay format capable of quantitative detection in partially denatured plasma. Antigen capture with CIZ1b antibody 043 from 5 μl of plasma (Fig. S6A) and detection via fibrinogen alpha chain, generated a cancer-specific signal, illustrated with assay development set B, described in Methods. Specifically, results by western blot with 043 generated ROC AUC 0.823 (P = 0.0002) and by ELISA with 043/anti-fibrinogen of 0.830 (P = 0.0007, Fig. 4A,B). Thus, the two methods generate very similar outcomes, and the data is significantly correlated (Fig. S6B).
Set B, which represents patients with a wide range of conditions presenting at York Hospital respiratory medicine clinics, was also used to evaluate the contribution of the CIZ1b component of the assay, since fibrinogen alone has been reported to have some potential as a biomarker for lung cancer [19]. Quantitative detection of fibrinogen in the linear range (Fig. S6C,D) returned some discriminatory capability across this set, though relatively poor ROC AUC of 0.628 (P = 0.225). This is consistent with western blot evaluation of fibrinogen (Fig. S1B), which indicated little difference between lung cancer patients and those without malignant disease. Thus capture of CIZ1b is the element that contributes the majority of the cancer-selectivity to this assay. We further tested the specificity of the configuration by analysing Set B after capture with 043, but substituting anti-fibrinogen detector antibody with anti-human IgG (Fig. 4D). Results indicate no discrimination between patients and control plasmas, confirming that anti-fibrinogen is an informative detector reagent that specifically detects a cancer-selective complex retrieved via CIZ1b. Finally, measurement of total IgG levels across set B by direct ELISA (Fig. 4E) was used to normalize the output generated by 043/fibrinogen, leading to a marginal improvement in discrimination and ROC AUC 0.845 (P = 0.0002, Fig. 4F). Thus, we report here a straightforward sandwich ELISA format suitable for high-throughput application on hospital platforms, that can reliably quantify circulating CIZ1b biomarker, and which can be used to identify patients with early stage lung cancer.

Discussion
Primary lung cancer remains the main cause of cancer death (1.3 million pa worldwide). Early detection of lung cancer is important because the only curative therapy today is surgical resection of earlystage tumours [20]. Crucially, for patients whose lung cancers are detected at stage 1 the outcome is substantially better than average. For stage 1A non-small cell lung cancer (NSCLC), the 5-year survival is 58-73% and for stage 1B 43-58%. However, NSCLC is usually diagnosed later when outcomes are significantly worse (stage 2A 36-46%, stage 2B 25-36%, stage 3A 19-24%, stage 3B 7-9%, stage 4 2-13% [21]). Screening programs using chest X-ray (CXR) or low-dose computed tomography (LDCT) can identify the disease at an early stage, and therefore have the potential to improve survival. The largest study concluded that a significant reduction (20%) in mortality can be achieved among highrisk populations by regular screening of pre-symptomatic individuals, however around a quarter of those screened were referred for invasive follow-up procedures to investigate suspicious imaging results, and around 96% of these turned out to be false positive [22]. Similar findings were made by an independent trial using a population-based questionnaire to identify high-risk individuals [23]. Thus both the personal and economic costs of screening make stand-alone imaging strategies difficult to justify. Further refined patient selection criteria coupled with a non-invasive technology that can eliminate a substantial proportion of false positive results, could add impact to early detection initiatives, by reducing over-diagnosis and at the same time reducing demand for LDCT.
Two potential technologies that could meet this need are breath analysis [24] and circulating biomarkers. A number of studies suggest candidate biomarkers, usually panels of tumour proteins [25,26] or autoantibodies [27], and an associated risk index. There is also much exploration of circulating DNA, RNA and methylation [28]. However, we know of no other publicly available data set that evaluates a single analyte blood-based biomarker on cohorts with stage 1 disease, with the degree of accuracy and stability seen for the CIZ1b biomarker [1]. Furthermore, to-date none of the candidates have been validated on prospectively collected cohorts, or directly compared to LDCT. The present technology will facilitate such comparison for CIZ1b, by enabling high-throughput measurement on a hospital-friendly platform.
Clinical biomarkers have been instrumental in reducing mortality in cancer patients [29], however lack of understanding of the capability of biomarker tests can result in inappropriate use and lead to over-investigation of disease that is incorrectly assumed to be malignant [30]. Furthermore the value of some has been questioned especially in relation to their use in the initial diagnosis of cancer [29,31]. Lack of clarity usually arises when a marker is found in other pathologies or when, like PSA, it is not specific for neoplastic cells. While further investigation will be required for CIZ1b we already know that it can distinguish between early stage malignant disease and non-malignant conditions that may appear similar at presentation [1].
The CIZ1b epitope that we measure in the blood may be the residue of degraded circulating tumour cells, or it may be a biologically active form that is secreted by cells. In support of the second possibility, nonvariant epitopes of CIZ1 are consistently detected in plasma from people with or without cancer indicating that presence in the blood is part of normal CIZ1 biology. Crucially, the CIZ1b form can be robustly measured in microliter quantities of plasma suggesting that it is not a rare protein in lung cancer patients. Moreover, it is an exceptionally stable analyte, which is a fundamental requirement for any biomarker in routine clinical use. By definition, nuclear matrix proteins (those that remain in the residual nucleus after extraction of chromatin, RNA, lipids and saltsensitive proteins) are highly stable entities. CIZ1 is part of this fraction [3] though it is also detected as punctate staining in the cytosol, and in some cell types decorating the plasma membrane (unpublished).
The finding that CIZ1b epitope is part of a dramatically truncated fragment of CIZ1 that is stably associated with fibrinogen is unexpected. While it is documented that certain bacterial proteins, such as extracellular fibrinogen-binding protein (Efb) of S. aureus [32], can impact on fibrinogen-dependent platelet aggregation, it is not known at this point whether CIZ1b fragment has any impact on functionality, though Fig. 4. Immunoassay validation using lung cancer plasma and controls in development set B (n = 39 non-cancer, 13 lung cancer). A) CIZ1b western blot with antibody 043. B) Sandwich ELISA format for CIZ1b antigen in plasma, using 043 capture antibody and anti-fibrinogen detector antibody. Comparison of the two data sets is in Fig. S6B. C) Quantitative detection of fibrinogen in the same samples using paired fibrinogen antibodies (calibration is in Fig. S6C). D) Substitution of anti-human IgG for anti-fibrinogen in a sandwich format similar to that shown in B. E) Detection of IgG in plasma by direct ELISA. F) Normalization of the data in C against the data in G. In all cases, receiver operating characteristic curves (ROC) indicate the relationship between true positive and false positive fraction (black points) and 95% confidence interval of the fitted ROC curve (grey points), and area under the curve (AUC) values in bold. Box and whisker plots display minimum and maximum, lower, median, and upper quartile, and outliers. the long CIZ1 peptides used here (b66, a74) do share some similarity with the consensus binding sequence of Efb. While a state of hypercoagulability is well-known to be associated with malignancy including primary lung cancers [33], the effect of CIZ1b remains to be investigated.
A number of retrospective studies have suggested that fibrinogen levels may offer information that can be used to identify or monitor the course of disease, for example as part of a panel of analytes involved in haemostasis and angiogenesis in breast cancer patients [34], or via measurement of fibrinogen in urine samples from bladder cancer patients [35]. Moreover, a recent study of 84,000 individuals that looked for elevation of three inflammatory biomarkers (CRP, fibrinogen and leukocyte count) observed an increased risk of colorectal, lung and breast cancer, with greatest risk associated with elevation of all three [19]. While we were not able to observe elevated fibrinogen by western blot in our cohorts, ROC AUCs for ELISA output for Set B did suggest some information content with respect to discrimination of the patients with confirmed cancer, though this was not statistically significant and is unlikely to provide sufficient discriminatory power on its own.
In summary, the data presented here and previously [1] show that CIZ1b is a strong candidate cancer-specific, single marker capable of identifying early stage lung cancer within at-risk groups, without resort to invasive procedures. The technological developments now reported bring a low-cost, minimally invasive and accurate blood test a step closer, with potential to significantly improve lung cancer survival by improving early detection rates.

Methods
Please see supplemental methods for details of plasma sets A and B, gels and sample preparation conditions, epitope reconstitution and mass spectrometry.

Antibodies for western blot
CIZ1b was detected using positively and negatively affinity purified anti-peptide rabbit pAb 2B [1], or newly developed reagent 043 protein A fraction. CIZ1 exon 17 was detected with a goat pAb against peptides in human exon 17 (Table S1) or NB74624, and exon 8 with NB74623 (both Novus Biologicals). Plasminogen was detected with Ab98262 (Abcam) and fibrinogen with goat pAb F8512 (Sigma) or mouse mAb 120-10068 (Novus Biologicals).

Enzyme linked immunosorbent assays
CIZ1b-fibrinogen alpha chain complex was detected by sandwich ELISA after partial denaturation of 5 μl of plasma in 100 μl of 0.5% tween20, 1% SDS in PBS, supplemented with 0.5 mM PMSF and incubation at 20°C for 30 min with repeated vortexing. Plates (Nunc Maxisorp 442404) were coated with 1 μg of 043 CIZ1b pAb protein A fraction (Pierce 20356), overnight at 4°C in 100 μl of 50 mM carbonate pH 9.6, then blocked for 2 h at room temperature with 300 μl of filtered 1% BSA (Jackson ImmunoResearch 1000162) in PBS. After three washes in 300 μl of 0.05% tween20 in PBS, analyte was added for 2 h at room temperature with gentle shaking. After five washes, captured complex was detected with sheep anti-fibrinogen pAb (AF4786 R&D systems) followed by anti-sheep HRP (Jackson ImmunoResearch 213-032-177), and results developed using 100 μl Sure Blue (KPL 520001) with detection at A450 nm using a FLUOstar OPTIMA plate reader (BMG Labtech). For measurement of fibrinogen capture antibody was chicken IgY (1 μg/well, Genway 15-288-22856), and analyte concentration was reduced five orders of magnitude by serial dilution. For measurement of human IgG in direct ELISA, 0.5 μl of plasma was coated directly onto plates in 100 μl of 50 mM carbonate buffer pH 9.6, blocked, washed and detected with anti-IgG (Jackson ImmunoResearch 709-035-149). For direct ELISA of synthetic analyte, 0.5 μg purified fibrinogen (Sigma F3879) complexed with 100 pmol peptide was coated onto plates in 100 μl of 50 mM carbonate pH 9.6 as above, probe with 043 or 2B and detected with anti-rabbit HRP (Jackson ImmunoResearch).

Conflict of interest
This work was carried out at the University of York by University staff supported by a research contract with Cizzle Biotech. Cizzle is a spin out from the University of York funded by Yorkshire Cancer Research. DC is a minority shareholder in Cizzle Biotech.

Author contributions
GH, DW, OTJ, AH, EA, RC, AD and DC performed the research. DC designed the research and wrote the paper, with input from AD, and OTJ. DW, OTJ and JW collected human plasma samples.

Funders
This work was funded by Cizzle Biotech (CizzleRC13-16) with additional support from the Wellcome Trust through C2D2 at the University of York (ISSF award 097829/Z/11/A) and the Jean Shanks Foundation (HYMS2011).